net/ipv4/igmp.c

   1 /*
   2  *      Linux NET3:     Internet Group Management Protocol  [IGMP]
   3  *
   4  *      This code implements the IGMP protocol as defined in RFC1112. There has
   5  *      been a further revision of this protocol since which is now supported.
   6  *
   7  *      If you have trouble with this module be careful what gcc you have used,
   8  *      the older version didn't come out right using gcc 2.5.8, the newer one
   9  *      seems to fall out with gcc 2.6.2.
  10  *
  11  *      Authors:
  12  *              Alan Cox <alan@lxorguk.ukuu.org.uk>
  13  *
  14  *      This program is free software; you can redistribute it and/or
  15  *      modify it under the terms of the GNU General Public License
  16  *      as published by the Free Software Foundation; either version
  17  *      2 of the License, or (at your option) any later version.
  18  *
  19  *      Fixes:
  20  *
  21  *              Alan Cox        :       Added lots of __inline__ to optimise
  22  *                                      the memory usage of all the tiny little
  23  *                                      functions.
  24  *              Alan Cox        :       Dumped the header building experiment.
  25  *              Alan Cox        :       Minor tweaks ready for multicast routing
  26  *                                      and extended IGMP protocol.
  27  *              Alan Cox        :       Removed a load of inline directives. Gcc 2.5.8
  28  *                                      writes utterly bogus code otherwise (sigh)
  29  *                                      fixed IGMP loopback to behave in the manner
  30  *                                      desired by mrouted, fixed the fact it has been
  31  *                                      broken since 1.3.6 and cleaned up a few minor
  32  *                                      points.
  33  *
  34  *              Chih-Jen Chang  :       Tried to revise IGMP to Version 2
  35  *              Tsu-Sheng Tsao          E-mail: chihjenc@scf.usc.edu and tsusheng@scf.usc.edu
  36  *                                      The enhancements are mainly based on Steve Deering's
  37  *                                      ipmulti-3.5 source code.
  38  *              Chih-Jen Chang  :       Added the igmp_get_mrouter_info and
  39  *              Tsu-Sheng Tsao          igmp_set_mrouter_info to keep track of
  40  *                                      the mrouted version on that device.
  41  *              Chih-Jen Chang  :       Added the max_resp_time parameter to
  42  *              Tsu-Sheng Tsao          igmp_heard_query(). Using this parameter
  43  *                                      to identify the multicast router version
  44  *                                      and do what the IGMP version 2 specified.
  45  *              Chih-Jen Chang  :       Added a timer to revert to IGMP V2 router
  46  *              Tsu-Sheng Tsao          if the specified time expired.
  47  *              Alan Cox        :       Stop IGMP from 0.0.0.0 being accepted.
  48  *              Alan Cox        :       Use GFP_ATOMIC in the right places.
  49  *              Christian Daudt :       igmp timer wasn't set for local group
  50  *                                      memberships but was being deleted,
  51  *                                      which caused a "del_timer() called
  52  *                                      from %p with timer not initialized\n"
  53  *                                      message (960131).
  54  *              Christian Daudt :       removed del_timer from
  55  *                                      igmp_timer_expire function (960205).
  56  *             Christian Daudt :       igmp_heard_report now only calls
  57  *                                     igmp_timer_expire if tm->running is
  58  *                                     true (960216).
  59  *              Malcolm Beattie :       ttl comparison wrong in igmp_rcv made
  60  *                                      igmp_heard_query never trigger. Expiry
  61  *                                      miscalculation fixed in igmp_heard_query
  62  *                                      and random() made to return unsigned to
  63  *                                      prevent negative expiry times.
  64  *              Alexey Kuznetsov:       Wrong group leaving behaviour, backport
  65  *                                      fix from pending 2.1.x patches.
  66  *              Alan Cox:               Forget to enable FDDI support earlier.
  67  *              Alexey Kuznetsov:       Fixed leaving groups on device down.
  68  *              Alexey Kuznetsov:       Accordance to igmp-v2-06 draft.
  69  *              David L Stevens:        IGMPv3 support, with help from
  70  *                                      Vinay Kulkarni
  71  */
  72
  73 #include <linux/module.h>
  74 #include <linux/slab.h>
  75 #include <asm/uaccess.h>
  76 #include <linux/types.h>
  77 #include <linux/kernel.h>
  78 #include <linux/jiffies.h>
  79 #include <linux/string.h>
  80 #include <linux/socket.h>
  81 #include <linux/sockios.h>
  82 #include <linux/in.h>
  83 #include <linux/inet.h>
  84 #include <linux/netdevice.h>
  85 #include <linux/skbuff.h>
  86 #include <linux/inetdevice.h>
  87 #include <linux/igmp.h>
  88 #include <linux/if_arp.h>
  89 #include <linux/rtnetlink.h>
  90 #include <linux/times.h>
  91 #include <linux/pkt_sched.h>
  92 #include <linux/byteorder/generic.h>
  93
  94 #include <net/net_namespace.h>
  95 #include <net/arp.h>
  96 #include <net/ip.h>
  97 #include <net/protocol.h>
  98 #include <net/route.h>
  99 #include <net/sock.h>
 100 #include <net/checksum.h>
 101 #include <net/inet_common.h>
 102 #include <linux/netfilter_ipv4.h>
 103 #ifdef CONFIG_IP_MROUTE
 104 #include <linux/mroute.h>
 105 #endif
 106 #ifdef CONFIG_PROC_FS
 107 #include <linux/proc_fs.h>
 108 #include <linux/seq_file.h>
 109 #endif
 110
 111 #ifdef CONFIG_IP_MULTICAST
 112 /* Parameter names and values are taken from igmp-v2-06 draft */
 113
 114 #define IGMP_V1_ROUTER_PRESENT_TIMEOUT          (400*HZ)
 115 #define IGMP_V2_ROUTER_PRESENT_TIMEOUT          (400*HZ)
 116 #define IGMP_V2_UNSOLICITED_REPORT_INTERVAL     (10*HZ)
 117 #define IGMP_V3_UNSOLICITED_REPORT_INTERVAL     (1*HZ)
 118 #define IGMP_QUERY_RESPONSE_INTERVAL            (10*HZ)
 119 #define IGMP_QUERY_ROBUSTNESS_VARIABLE          2
 120
 121
 122 #define IGMP_INITIAL_REPORT_DELAY               (1)
 123
 124 /* IGMP_INITIAL_REPORT_DELAY is not from IGMP specs!
 125  * IGMP specs require to report membership immediately after
 126  * joining a group, but we delay the first report by a
 127  * small interval. It seems more natural and still does not
 128  * contradict to specs provided this delay is small enough.
 129  */
 130
 131 #define IGMP_V1_SEEN(in_dev) \
 132         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 1 || \
 133          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 1 || \
 134          ((in_dev)->mr_v1_seen && \
 135           time_before(jiffies, (in_dev)->mr_v1_seen)))
 136 #define IGMP_V2_SEEN(in_dev) \
 137         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 2 || \
 138          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 2 || \
 139          ((in_dev)->mr_v2_seen && \
 140           time_before(jiffies, (in_dev)->mr_v2_seen)))
 141
 142 static int unsolicited_report_interval(struct in_device *in_dev)
 143 {
 144         int interval_ms, interval_jiffies;
 145
 146         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 147                 interval_ms = IN_DEV_CONF_GET(
 148                         in_dev,
 149                         IGMPV2_UNSOLICITED_REPORT_INTERVAL);
 150         else /* v3 */
 151                 interval_ms = IN_DEV_CONF_GET(
 152                         in_dev,
 153                         IGMPV3_UNSOLICITED_REPORT_INTERVAL);
 154
 155         interval_jiffies = msecs_to_jiffies(interval_ms);
 156
 157         /* _timer functions can't handle a delay of 0 jiffies so ensure
 158          *  we always return a positive value.
 159          */
 160         if (interval_jiffies <= 0)
 161                 interval_jiffies = 1;
 162         return interval_jiffies;
 163 }
 164
 165 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im);
 166 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im);
 167 static void igmpv3_clear_delrec(struct in_device *in_dev);
 168 static int sf_setstate(struct ip_mc_list *pmc);
 169 static void sf_markstate(struct ip_mc_list *pmc);
 170 #endif
 171 static void ip_mc_clear_src(struct ip_mc_list *pmc);
 172 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
 173                          int sfcount, __be32 *psfsrc, int delta);
 174
 175 static void ip_ma_put(struct ip_mc_list *im)
 176 {
 177         if (atomic_dec_and_test(&im->refcnt)) {
 178                 in_dev_put(im->interface);
 179                 kfree_rcu(im, rcu);
 180         }
 181 }
 182
 183 #define for_each_pmc_rcu(in_dev, pmc)                           \
 184         for (pmc = rcu_dereference(in_dev->mc_list);            \
 185              pmc != NULL;                                       \
 186              pmc = rcu_dereference(pmc->next_rcu))
 187
 188 #define for_each_pmc_rtnl(in_dev, pmc)                          \
 189         for (pmc = rtnl_dereference(in_dev->mc_list);           \
 190              pmc != NULL;                                       \
 191              pmc = rtnl_dereference(pmc->next_rcu))
 192
 193 static void ip_sf_list_clear_all(struct ip_sf_list *psf)
 194 {
 195         struct ip_sf_list *next;
 196
 197         while (psf) {
 198                 next = psf->sf_next;
 199                 kfree(psf);
 200                 psf = next;
 201         }
 202 }
 203
 204 #ifdef CONFIG_IP_MULTICAST
 205
 206 /*
 207  *      Timer management
 208  */
 209
 210 static void igmp_stop_timer(struct ip_mc_list *im)
 211 {
 212         spin_lock_bh(&im->lock);
 213         if (del_timer(&im->timer))
 214                 atomic_dec(&im->refcnt);
 215         im->tm_running = 0;
 216         im->reporter = 0;
 217         im->unsolicit_count = 0;
 218         spin_unlock_bh(&im->lock);
 219 }
 220
 221 /* It must be called with locked im->lock */
 222 static void igmp_start_timer(struct ip_mc_list *im, int max_delay)
 223 {
 224         int tv = prandom_u32() % max_delay;
 225
 226         im->tm_running = 1;
 227         if (!mod_timer(&im->timer, jiffies+tv+2))
 228                 atomic_inc(&im->refcnt);
 229 }
 230
 231 static void igmp_gq_start_timer(struct in_device *in_dev)
 232 {
 233         int tv = prandom_u32() % in_dev->mr_maxdelay;
 234         unsigned long exp = jiffies + tv + 2;
 235
 236         if (in_dev->mr_gq_running &&
 237             time_after_eq(exp, (in_dev->mr_gq_timer).expires))
 238                 return;
 239
 240         in_dev->mr_gq_running = 1;
 241         if (!mod_timer(&in_dev->mr_gq_timer, exp))
 242                 in_dev_hold(in_dev);
 243 }
 244
 245 static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
 246 {
 247         int tv = prandom_u32() % delay;
 248
 249         if (!mod_timer(&in_dev->mr_ifc_timer, jiffies+tv+2))
 250                 in_dev_hold(in_dev);
 251 }
 252
 253 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
 254 {
 255         spin_lock_bh(&im->lock);
 256         im->unsolicit_count = 0;
 257         if (del_timer(&im->timer)) {
 258                 if ((long)(im->timer.expires-jiffies) < max_delay) {
 259                         add_timer(&im->timer);
 260                         im->tm_running = 1;
 261                         spin_unlock_bh(&im->lock);
 262                         return;
 263                 }
 264                 atomic_dec(&im->refcnt);
 265         }
 266         igmp_start_timer(im, max_delay);
 267         spin_unlock_bh(&im->lock);
 268 }
 269
 270
 271 /*
 272  *      Send an IGMP report.
 273  */
 274
 275 #define IGMP_SIZE (sizeof(struct igmphdr)+sizeof(struct iphdr)+4)
 276
 277
 278 static int is_in(struct ip_mc_list *pmc, struct ip_sf_list *psf, int type,
 279         int gdeleted, int sdeleted)
 280 {
 281         switch (type) {
 282         case IGMPV3_MODE_IS_INCLUDE:
 283         case IGMPV3_MODE_IS_EXCLUDE:
 284                 if (gdeleted || sdeleted)
 285                         return 0;
 286                 if (!(pmc->gsquery && !psf->sf_gsresp)) {
 287                         if (pmc->sfmode == MCAST_INCLUDE)
 288                                 return 1;
 289                         /* don't include if this source is excluded
 290                          * in all filters
 291                          */
 292                         if (psf->sf_count[MCAST_INCLUDE])
 293                                 return type == IGMPV3_MODE_IS_INCLUDE;
 294                         return pmc->sfcount[MCAST_EXCLUDE] ==
 295                                 psf->sf_count[MCAST_EXCLUDE];
 296                 }
 297                 return 0;
 298         case IGMPV3_CHANGE_TO_INCLUDE:
 299                 if (gdeleted || sdeleted)
 300                         return 0;
 301                 return psf->sf_count[MCAST_INCLUDE] != 0;
 302         case IGMPV3_CHANGE_TO_EXCLUDE:
 303                 if (gdeleted || sdeleted)
 304                         return 0;
 305                 if (pmc->sfcount[MCAST_EXCLUDE] == 0 ||
 306                     psf->sf_count[MCAST_INCLUDE])
 307                         return 0;
 308                 return pmc->sfcount[MCAST_EXCLUDE] ==
 309                         psf->sf_count[MCAST_EXCLUDE];
 310         case IGMPV3_ALLOW_NEW_SOURCES:
 311                 if (gdeleted || !psf->sf_crcount)
 312                         return 0;
 313                 return (pmc->sfmode == MCAST_INCLUDE) ^ sdeleted;
 314         case IGMPV3_BLOCK_OLD_SOURCES:
 315                 if (pmc->sfmode == MCAST_INCLUDE)
 316                         return gdeleted || (psf->sf_crcount && sdeleted);
 317                 return psf->sf_crcount && !gdeleted && !sdeleted;
 318         }
 319         return 0;
 320 }
 321
 322 static int
 323 igmp_scount(struct ip_mc_list *pmc, int type, int gdeleted, int sdeleted)
 324 {
 325         struct ip_sf_list *psf;
 326         int scount = 0;
 327
 328         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 329                 if (!is_in(pmc, psf, type, gdeleted, sdeleted))
 330                         continue;
 331                 scount++;
 332         }
 333         return scount;
 334 }
 335
 336 /* source address selection per RFC 3376 section 4.2.13 */
 337 static __be32 igmpv3_get_srcaddr(struct net_device *dev,
 338                                  const struct flowi4 *fl4)
 339 {
 340         struct in_device *in_dev = __in_dev_get_rcu(dev);
 341
 342         if (!in_dev)
 343                 return htonl(INADDR_ANY);
 344
 345         for_ifa(in_dev) {
 346                 if (fl4->saddr == ifa->ifa_local)
 347                         return fl4->saddr;
 348         } endfor_ifa(in_dev);
 349
 350         return htonl(INADDR_ANY);
 351 }
 352
 353 static struct sk_buff *igmpv3_newpack(struct net_device *dev, unsigned int mtu)
 354 {
 355         struct sk_buff *skb;
 356         struct rtable *rt;
 357         struct iphdr *pip;
 358         struct igmpv3_report *pig;
 359         struct net *net = dev_net(dev);
 360         struct flowi4 fl4;
 361         int hlen = LL_RESERVED_SPACE(dev);
 362         int tlen = dev->needed_tailroom;
 363         unsigned int size = mtu;
 364
 365         while (1) {
 366                 skb = alloc_skb(size + hlen + tlen,
 367                                 GFP_ATOMIC | __GFP_NOWARN);
 368                 if (skb)
 369                         break;
 370                 size >>= 1;
 371                 if (size < 256)
 372                         return NULL;
 373         }
 374         skb->priority = TC_PRIO_CONTROL;
 375
 376         rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
 377                                    0, 0,
 378                                    IPPROTO_IGMP, 0, dev->ifindex);
 379         if (IS_ERR(rt)) {
 380                 kfree_skb(skb);
 381                 return NULL;
 382         }
 383
 384         skb_dst_set(skb, &rt->dst);
 385         skb->dev = dev;
 386
 387         skb_reserve(skb, hlen);
 388         skb_tailroom_reserve(skb, mtu, tlen);
 389
 390         skb_reset_network_header(skb);
 391         pip = ip_hdr(skb);
 392         skb_put(skb, sizeof(struct iphdr) + 4);
 393
 394         pip->version  = 4;
 395         pip->ihl      = (sizeof(struct iphdr)+4)>>2;
 396         pip->tos      = 0xc0;
 397         pip->frag_off = htons(IP_DF);
 398         pip->ttl      = 1;
 399         pip->daddr    = fl4.daddr;
 400
 401         rcu_read_lock();
 402         pip->saddr    = igmpv3_get_srcaddr(dev, &fl4);
 403         rcu_read_unlock();
 404
 405         pip->protocol = IPPROTO_IGMP;
 406         pip->tot_len  = 0;      /* filled in later */
 407         ip_select_ident(net, skb, NULL);
 408         ((u8 *)&pip[1])[0] = IPOPT_RA;
 409         ((u8 *)&pip[1])[1] = 4;
 410         ((u8 *)&pip[1])[2] = 0;
 411         ((u8 *)&pip[1])[3] = 0;
 412
 413         skb->transport_header = skb->network_header + sizeof(struct iphdr) + 4;
 414         skb_put(skb, sizeof(*pig));
 415         pig = igmpv3_report_hdr(skb);
 416         pig->type = IGMPV3_HOST_MEMBERSHIP_REPORT;
 417         pig->resv1 = 0;
 418         pig->csum = 0;
 419         pig->resv2 = 0;
 420         pig->ngrec = 0;
 421         return skb;
 422 }
 423
 424 static int igmpv3_sendpack(struct sk_buff *skb)
 425 {
 426         struct igmphdr *pig = igmp_hdr(skb);
 427         const int igmplen = skb_tail_pointer(skb) - skb_transport_header(skb);
 428
 429         pig->csum = ip_compute_csum(igmp_hdr(skb), igmplen);
 430
 431         return ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 432 }
 433
 434 static int grec_size(struct ip_mc_list *pmc, int type, int gdel, int sdel)
 435 {
 436         return sizeof(struct igmpv3_grec) + 4*igmp_scount(pmc, type, gdel, sdel);
 437 }
 438
 439 static struct sk_buff *add_grhead(struct sk_buff *skb, struct ip_mc_list *pmc,
 440         int type, struct igmpv3_grec **ppgr, unsigned int mtu)
 441 {
 442         struct net_device *dev = pmc->interface->dev;
 443         struct igmpv3_report *pih;
 444         struct igmpv3_grec *pgr;
 445
 446         if (!skb) {
 447                 skb = igmpv3_newpack(dev, mtu);
 448                 if (!skb)
 449                         return NULL;
 450         }
 451         pgr = (struct igmpv3_grec *)skb_put(skb, sizeof(struct igmpv3_grec));
 452         pgr->grec_type = type;
 453         pgr->grec_auxwords = 0;
 454         pgr->grec_nsrcs = 0;
 455         pgr->grec_mca = pmc->multiaddr;
 456         pih = igmpv3_report_hdr(skb);
 457         pih->ngrec = htons(ntohs(pih->ngrec)+1);
 458         *ppgr = pgr;
 459         return skb;
 460 }
 461
 462 #define AVAILABLE(skb)  ((skb) ? skb_availroom(skb) : 0)
 463
 464 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
 465         int type, int gdeleted, int sdeleted)
 466 {
 467         struct net_device *dev = pmc->interface->dev;
 468         struct net *net = dev_net(dev);
 469         struct igmpv3_report *pih;
 470         struct igmpv3_grec *pgr = NULL;
 471         struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
 472         int scount, stotal, first, isquery, truncate;
 473         unsigned int mtu;
 474
 475         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 476                 return skb;
 477         if (ipv4_is_local_multicast(pmc->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
 478                 return skb;
 479
 480         mtu = READ_ONCE(dev->mtu);
 481         if (mtu < IPV4_MIN_MTU)
 482                 return skb;
 483
 484         isquery = type == IGMPV3_MODE_IS_INCLUDE ||
 485                   type == IGMPV3_MODE_IS_EXCLUDE;
 486         truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
 487                     type == IGMPV3_CHANGE_TO_EXCLUDE;
 488
 489         stotal = scount = 0;
 490
 491         psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
 492
 493         if (!*psf_list)
 494                 goto empty_source;
 495
 496         pih = skb ? igmpv3_report_hdr(skb) : NULL;
 497
 498         /* EX and TO_EX get a fresh packet, if needed */
 499         if (truncate) {
 500                 if (pih && pih->ngrec &&
 501                     AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
 502                         if (skb)
 503                                 igmpv3_sendpack(skb);
 504                         skb = igmpv3_newpack(dev, mtu);
 505                 }
 506         }
 507         first = 1;
 508         psf_prev = NULL;
 509         for (psf = *psf_list; psf; psf = psf_next) {
 510                 __be32 *psrc;
 511
 512                 psf_next = psf->sf_next;
 513
 514                 if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
 515                         psf_prev = psf;
 516                         continue;
 517                 }
 518
 519                 /* Based on RFC3376 5.1. Should not send source-list change
 520                  * records when there is a filter mode change.
 521                  */
 522                 if (((gdeleted && pmc->sfmode == MCAST_EXCLUDE) ||
 523                      (!gdeleted && pmc->crcount)) &&
 524                     (type == IGMPV3_ALLOW_NEW_SOURCES ||
 525                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount)
 526                         goto decrease_sf_crcount;
 527
 528                 /* clear marks on query responses */
 529                 if (isquery)
 530                         psf->sf_gsresp = 0;
 531
 532                 if (AVAILABLE(skb) < sizeof(__be32) +
 533                     first*sizeof(struct igmpv3_grec)) {
 534                         if (truncate && !first)
 535                                 break;   /* truncate these */
 536                         if (pgr)
 537                                 pgr->grec_nsrcs = htons(scount);
 538                         if (skb)
 539                                 igmpv3_sendpack(skb);
 540                         skb = igmpv3_newpack(dev, mtu);
 541                         first = 1;
 542                         scount = 0;
 543                 }
 544                 if (first) {
 545                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 546                         first = 0;
 547                 }
 548                 if (!skb)
 549                         return NULL;
 550                 psrc = (__be32 *)skb_put(skb, sizeof(__be32));
 551                 *psrc = psf->sf_inaddr;
 552                 scount++; stotal++;
 553                 if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
 554                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
 555 decrease_sf_crcount:
 556                         psf->sf_crcount--;
 557                         if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
 558                                 if (psf_prev)
 559                                         psf_prev->sf_next = psf->sf_next;
 560                                 else
 561                                         *psf_list = psf->sf_next;
 562                                 kfree(psf);
 563                                 continue;
 564                         }
 565                 }
 566                 psf_prev = psf;
 567         }
 568
 569 empty_source:
 570         if (!stotal) {
 571                 if (type == IGMPV3_ALLOW_NEW_SOURCES ||
 572                     type == IGMPV3_BLOCK_OLD_SOURCES)
 573                         return skb;
 574                 if (pmc->crcount || isquery) {
 575                         /* make sure we have room for group header */
 576                         if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
 577                                 igmpv3_sendpack(skb);
 578                                 skb = NULL; /* add_grhead will get a new one */
 579                         }
 580                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 581                 }
 582         }
 583         if (pgr)
 584                 pgr->grec_nsrcs = htons(scount);
 585
 586         if (isquery)
 587                 pmc->gsquery = 0;       /* clear query state on report */
 588         return skb;
 589 }
 590
 591 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
 592 {
 593         struct sk_buff *skb = NULL;
 594         struct net *net = dev_net(in_dev->dev);
 595         int type;
 596
 597         if (!pmc) {
 598                 rcu_read_lock();
 599                 for_each_pmc_rcu(in_dev, pmc) {
 600                         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 601                                 continue;
 602                         if (ipv4_is_local_multicast(pmc->multiaddr) &&
 603                              !net->ipv4.sysctl_igmp_llm_reports)
 604                                 continue;
 605                         spin_lock_bh(&pmc->lock);
 606                         if (pmc->sfcount[MCAST_EXCLUDE])
 607                                 type = IGMPV3_MODE_IS_EXCLUDE;
 608                         else
 609                                 type = IGMPV3_MODE_IS_INCLUDE;
 610                         skb = add_grec(skb, pmc, type, 0, 0);
 611                         spin_unlock_bh(&pmc->lock);
 612                 }
 613                 rcu_read_unlock();
 614         } else {
 615                 spin_lock_bh(&pmc->lock);
 616                 if (pmc->sfcount[MCAST_EXCLUDE])
 617                         type = IGMPV3_MODE_IS_EXCLUDE;
 618                 else
 619                         type = IGMPV3_MODE_IS_INCLUDE;
 620                 skb = add_grec(skb, pmc, type, 0, 0);
 621                 spin_unlock_bh(&pmc->lock);
 622         }
 623         if (!skb)
 624                 return 0;
 625         return igmpv3_sendpack(skb);
 626 }
 627
 628 /*
 629  * remove zero-count source records from a source filter list
 630  */
 631 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
 632 {
 633         struct ip_sf_list *psf_prev, *psf_next, *psf;
 634
 635         psf_prev = NULL;
 636         for (psf = *ppsf; psf; psf = psf_next) {
 637                 psf_next = psf->sf_next;
 638                 if (psf->sf_crcount == 0) {
 639                         if (psf_prev)
 640                                 psf_prev->sf_next = psf->sf_next;
 641                         else
 642                                 *ppsf = psf->sf_next;
 643                         kfree(psf);
 644                 } else
 645                         psf_prev = psf;
 646         }
 647 }
 648
 649 static void kfree_pmc(struct ip_mc_list *pmc)
 650 {
 651         ip_sf_list_clear_all(pmc->sources);
 652         ip_sf_list_clear_all(pmc->tomb);
 653         kfree(pmc);
 654 }
 655
 656 static void igmpv3_send_cr(struct in_device *in_dev)
 657 {
 658         struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
 659         struct sk_buff *skb = NULL;
 660         int type, dtype;
 661
 662         rcu_read_lock();
 663         spin_lock_bh(&in_dev->mc_tomb_lock);
 664
 665         /* deleted MCA's */
 666         pmc_prev = NULL;
 667         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
 668                 pmc_next = pmc->next;
 669                 if (pmc->sfmode == MCAST_INCLUDE) {
 670                         type = IGMPV3_BLOCK_OLD_SOURCES;
 671                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 672                         skb = add_grec(skb, pmc, type, 1, 0);
 673                         skb = add_grec(skb, pmc, dtype, 1, 1);
 674                 }
 675                 if (pmc->crcount) {
 676                         if (pmc->sfmode == MCAST_EXCLUDE) {
 677                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 678                                 skb = add_grec(skb, pmc, type, 1, 0);
 679                         }
 680                         pmc->crcount--;
 681                         if (pmc->crcount == 0) {
 682                                 igmpv3_clear_zeros(&pmc->tomb);
 683                                 igmpv3_clear_zeros(&pmc->sources);
 684                         }
 685                 }
 686                 if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
 687                         if (pmc_prev)
 688                                 pmc_prev->next = pmc_next;
 689                         else
 690                                 in_dev->mc_tomb = pmc_next;
 691                         in_dev_put(pmc->interface);
 692                         kfree_pmc(pmc);
 693                 } else
 694                         pmc_prev = pmc;
 695         }
 696         spin_unlock_bh(&in_dev->mc_tomb_lock);
 697
 698         /* change recs */
 699         for_each_pmc_rcu(in_dev, pmc) {
 700                 spin_lock_bh(&pmc->lock);
 701                 if (pmc->sfcount[MCAST_EXCLUDE]) {
 702                         type = IGMPV3_BLOCK_OLD_SOURCES;
 703                         dtype = IGMPV3_ALLOW_NEW_SOURCES;
 704                 } else {
 705                         type = IGMPV3_ALLOW_NEW_SOURCES;
 706                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 707                 }
 708                 skb = add_grec(skb, pmc, type, 0, 0);
 709                 skb = add_grec(skb, pmc, dtype, 0, 1);  /* deleted sources */
 710
 711                 /* filter mode changes */
 712                 if (pmc->crcount) {
 713                         if (pmc->sfmode == MCAST_EXCLUDE)
 714                                 type = IGMPV3_CHANGE_TO_EXCLUDE;
 715                         else
 716                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 717                         skb = add_grec(skb, pmc, type, 0, 0);
 718                         pmc->crcount--;
 719                 }
 720                 spin_unlock_bh(&pmc->lock);
 721         }
 722         rcu_read_unlock();
 723
 724         if (!skb)
 725                 return;
 726         (void) igmpv3_sendpack(skb);
 727 }
 728
 729 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
 730         int type)
 731 {
 732         struct sk_buff *skb;
 733         struct iphdr *iph;
 734         struct igmphdr *ih;
 735         struct rtable *rt;
 736         struct net_device *dev = in_dev->dev;
 737         struct net *net = dev_net(dev);
 738         __be32  group = pmc ? pmc->multiaddr : 0;
 739         struct flowi4 fl4;
 740         __be32  dst;
 741         int hlen, tlen;
 742
 743         if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
 744                 return igmpv3_send_report(in_dev, pmc);
 745
 746         if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
 747                 return 0;
 748
 749         if (type == IGMP_HOST_LEAVE_MESSAGE)
 750                 dst = IGMP_ALL_ROUTER;
 751         else
 752                 dst = group;
 753
 754         rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
 755                                    0, 0,
 756                                    IPPROTO_IGMP, 0, dev->ifindex);
 757         if (IS_ERR(rt))
 758                 return -1;
 759
 760         hlen = LL_RESERVED_SPACE(dev);
 761         tlen = dev->needed_tailroom;
 762         skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
 763         if (!skb) {
 764                 ip_rt_put(rt);
 765                 return -1;
 766         }
 767         skb->priority = TC_PRIO_CONTROL;
 768
 769         skb_dst_set(skb, &rt->dst);
 770
 771         skb_reserve(skb, hlen);
 772
 773         skb_reset_network_header(skb);
 774         iph = ip_hdr(skb);
 775         skb_put(skb, sizeof(struct iphdr) + 4);
 776
 777         iph->version  = 4;
 778         iph->ihl      = (sizeof(struct iphdr)+4)>>2;
 779         iph->tos      = 0xc0;
 780         iph->frag_off = htons(IP_DF);
 781         iph->ttl      = 1;
 782         iph->daddr    = dst;
 783         iph->saddr    = fl4.saddr;
 784         iph->protocol = IPPROTO_IGMP;
 785         ip_select_ident(net, skb, NULL);
 786         ((u8 *)&iph[1])[0] = IPOPT_RA;
 787         ((u8 *)&iph[1])[1] = 4;
 788         ((u8 *)&iph[1])[2] = 0;
 789         ((u8 *)&iph[1])[3] = 0;
 790
 791         ih = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
 792         ih->type = type;
 793         ih->code = 0;
 794         ih->csum = 0;
 795         ih->group = group;
 796         ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
 797
 798         return ip_local_out(net, skb->sk, skb);
 799 }
 800
 801 static void igmp_gq_timer_expire(unsigned long data)
 802 {
 803         struct in_device *in_dev = (struct in_device *)data;
 804
 805         in_dev->mr_gq_running = 0;
 806         igmpv3_send_report(in_dev, NULL);
 807         in_dev_put(in_dev);
 808 }
 809
 810 static void igmp_ifc_timer_expire(unsigned long data)
 811 {
 812         struct in_device *in_dev = (struct in_device *)data;
 813
 814         igmpv3_send_cr(in_dev);
 815         if (in_dev->mr_ifc_count) {
 816                 in_dev->mr_ifc_count--;
 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         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(unsigned long data)
 834 {
 835         struct ip_mc_list *im = (struct ip_mc_list *)data;
 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) {
 842                 im->unsolicit_count--;
 843                 igmp_start_timer(im, unsolicited_report_interval(in_dev));
 844         }
 845         im->reporter = 1;
 846         spin_unlock(&im->lock);
 847
 848         if (IGMP_V1_SEEN(in_dev))
 849                 igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
 850         else if (IGMP_V2_SEEN(in_dev))
 851                 igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
 852         else
 853                 igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
 854
 855         ip_ma_put(im);
 856 }
 857
 858 /* mark EXCLUDE-mode sources */
 859 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 860 {
 861         struct ip_sf_list *psf;
 862         int i, scount;
 863
 864         scount = 0;
 865         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 866                 if (scount == nsrcs)
 867                         break;
 868                 for (i = 0; i < nsrcs; i++) {
 869                         /* skip inactive filters */
 870                         if (psf->sf_count[MCAST_INCLUDE] ||
 871                             pmc->sfcount[MCAST_EXCLUDE] !=
 872                             psf->sf_count[MCAST_EXCLUDE])
 873                                 break;
 874                         if (srcs[i] == psf->sf_inaddr) {
 875                                 scount++;
 876                                 break;
 877                         }
 878                 }
 879         }
 880         pmc->gsquery = 0;
 881         if (scount == nsrcs)    /* all sources excluded */
 882                 return 0;
 883         return 1;
 884 }
 885
 886 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 887 {
 888         struct ip_sf_list *psf;
 889         int i, scount;
 890
 891         if (pmc->sfmode == MCAST_EXCLUDE)
 892                 return igmp_xmarksources(pmc, nsrcs, srcs);
 893
 894         /* mark INCLUDE-mode sources */
 895         scount = 0;
 896         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 897                 if (scount == nsrcs)
 898                         break;
 899                 for (i = 0; i < nsrcs; i++)
 900                         if (srcs[i] == psf->sf_inaddr) {
 901                                 psf->sf_gsresp = 1;
 902                                 scount++;
 903                                 break;
 904                         }
 905         }
 906         if (!scount) {
 907                 pmc->gsquery = 0;
 908                 return 0;
 909         }
 910         pmc->gsquery = 1;
 911         return 1;
 912 }
 913
 914 /* return true if packet was dropped */
 915 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
 916 {
 917         struct ip_mc_list *im;
 918         struct net *net = dev_net(in_dev->dev);
 919
 920         /* Timers are only set for non-local groups */
 921
 922         if (group == IGMP_ALL_HOSTS)
 923                 return false;
 924         if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
 925                 return false;
 926
 927         rcu_read_lock();
 928         for_each_pmc_rcu(in_dev, im) {
 929                 if (im->multiaddr == group) {
 930                         igmp_stop_timer(im);
 931                         break;
 932                 }
 933         }
 934         rcu_read_unlock();
 935         return false;
 936 }
 937
 938 /* return true if packet was dropped */
 939 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
 940         int len)
 941 {
 942         struct igmphdr          *ih = igmp_hdr(skb);
 943         struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
 944         struct ip_mc_list       *im;
 945         __be32                  group = ih->group;
 946         int                     max_delay;
 947         int                     mark = 0;
 948         struct net              *net = dev_net(in_dev->dev);
 949
 950
 951         if (len == 8) {
 952                 if (ih->code == 0) {
 953                         /* Alas, old v1 router presents here. */
 954
 955                         max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 956                         in_dev->mr_v1_seen = jiffies +
 957                                 IGMP_V1_ROUTER_PRESENT_TIMEOUT;
 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                                 IGMP_V2_ROUTER_PRESENT_TIMEOUT;
 964                 }
 965                 /* cancel the interface change timer */
 966                 in_dev->mr_ifc_count = 0;
 967                 if (del_timer(&in_dev->mr_ifc_timer))
 968                         __in_dev_put(in_dev);
 969                 /* clear deleted report items */
 970                 igmpv3_clear_delrec(in_dev);
 971         } else if (len < 12) {
 972                 return true;    /* ignore bogus packet; freed by caller */
 973         } else if (IGMP_V1_SEEN(in_dev)) {
 974                 /* This is a v3 query with v1 queriers present */
 975                 max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 976                 group = 0;
 977         } else if (IGMP_V2_SEEN(in_dev)) {
 978                 /* this is a v3 query with v2 queriers present;
 979                  * Interpretation of the max_delay code is problematic here.
 980                  * A real v2 host would use ih_code directly, while v3 has a
 981                  * different encoding. We use the v3 encoding as more likely
 982                  * to be intended in a v3 query.
 983                  */
 984                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 985                 if (!max_delay)
 986                         max_delay = 1;  /* can't mod w/ 0 */
 987         } else { /* v3 */
 988                 if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
 989                         return true;
 990
 991                 ih3 = igmpv3_query_hdr(skb);
 992                 if (ih3->nsrcs) {
 993                         if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
 994                                            + ntohs(ih3->nsrcs)*sizeof(__be32)))
 995                                 return true;
 996                         ih3 = igmpv3_query_hdr(skb);
 997                 }
 998
 999                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
1000                 if (!max_delay)
1001                         max_delay = 1;  /* can't mod w/ 0 */
1002                 in_dev->mr_maxdelay = max_delay;
1003                 if (ih3->qrv)
1004                         in_dev->mr_qrv = ih3->qrv;
1005                 if (!group) { /* general query */
1006                         if (ih3->nsrcs)
1007                                 return true;    /* no sources allowed */
1008                         igmp_gq_start_timer(in_dev);
1009                         return false;
1010                 }
1011                 /* mark sources to include, if group & source-specific */
1012                 mark = ih3->nsrcs != 0;
1013         }
1014
1015         /*
1016          * - Start the timers in all of our membership records
1017          *   that the query applies to for the interface on
1018          *   which the query arrived excl. those that belong
1019          *   to a "local" group (224.0.0.X)
1020          * - For timers already running check if they need to
1021          *   be reset.
1022          * - Use the igmp->igmp_code field as the maximum
1023          *   delay possible
1024          */
1025         rcu_read_lock();
1026         for_each_pmc_rcu(in_dev, im) {
1027                 int changed;
1028
1029                 if (group && group != im->multiaddr)
1030                         continue;
1031                 if (im->multiaddr == IGMP_ALL_HOSTS)
1032                         continue;
1033                 if (ipv4_is_local_multicast(im->multiaddr) &&
1034                     !net->ipv4.sysctl_igmp_llm_reports)
1035                         continue;
1036                 spin_lock_bh(&im->lock);
1037                 if (im->tm_running)
1038                         im->gsquery = im->gsquery && mark;
1039                 else
1040                         im->gsquery = mark;
1041                 changed = !im->gsquery ||
1042                         igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
1043                 spin_unlock_bh(&im->lock);
1044                 if (changed)
1045                         igmp_mod_timer(im, max_delay);
1046         }
1047         rcu_read_unlock();
1048         return false;
1049 }
1050
1051 /* called in rcu_read_lock() section */
1052 int igmp_rcv(struct sk_buff *skb)
1053 {
1054         /* This basically follows the spec line by line -- see RFC1112 */
1055         struct igmphdr *ih;
1056         struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
1057         int len = skb->len;
1058         bool dropped = true;
1059
1060         if (!in_dev)
1061                 goto drop;
1062
1063         if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
1064                 goto drop;
1065
1066         if (skb_checksum_simple_validate(skb))
1067                 goto drop;
1068
1069         ih = igmp_hdr(skb);
1070         switch (ih->type) {
1071         case IGMP_HOST_MEMBERSHIP_QUERY:
1072                 dropped = igmp_heard_query(in_dev, skb, len);
1073                 break;
1074         case IGMP_HOST_MEMBERSHIP_REPORT:
1075         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1076                 /* Is it our report looped back? */
1077                 if (rt_is_output_route(skb_rtable(skb)))
1078                         break;
1079                 /* don't rely on MC router hearing unicast reports */
1080                 if (skb->pkt_type == PACKET_MULTICAST ||
1081                     skb->pkt_type == PACKET_BROADCAST)
1082                         dropped = igmp_heard_report(in_dev, ih->group);
1083                 break;
1084         case IGMP_PIM:
1085 #ifdef CONFIG_IP_PIMSM_V1
1086                 return pim_rcv_v1(skb);
1087 #endif
1088         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1089         case IGMP_DVMRP:
1090         case IGMP_TRACE:
1091         case IGMP_HOST_LEAVE_MESSAGE:
1092         case IGMP_MTRACE:
1093         case IGMP_MTRACE_RESP:
1094                 break;
1095         default:
1096                 break;
1097         }
1098
1099 drop:
1100         if (dropped)
1101                 kfree_skb(skb);
1102         else
1103                 consume_skb(skb);
1104         return 0;
1105 }
1106
1107 #endif
1108
1109
1110 /*
1111  *      Add a filter to a device
1112  */
1113
1114 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1115 {
1116         char buf[MAX_ADDR_LEN];
1117         struct net_device *dev = in_dev->dev;
1118
1119         /* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1120            We will get multicast token leakage, when IFF_MULTICAST
1121            is changed. This check should be done in ndo_set_rx_mode
1122            routine. Something sort of:
1123            if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1124            --ANK
1125            */
1126         if (arp_mc_map(addr, buf, dev, 0) == 0)
1127                 dev_mc_add(dev, buf);
1128 }
1129
1130 /*
1131  *      Remove a filter from a device
1132  */
1133
1134 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1135 {
1136         char buf[MAX_ADDR_LEN];
1137         struct net_device *dev = in_dev->dev;
1138
1139         if (arp_mc_map(addr, buf, dev, 0) == 0)
1140                 dev_mc_del(dev, buf);
1141 }
1142
1143 #ifdef CONFIG_IP_MULTICAST
1144 /*
1145  * deleted ip_mc_list manipulation
1146  */
1147 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1148 {
1149         struct ip_mc_list *pmc;
1150         struct net *net = dev_net(in_dev->dev);
1151
1152         /* this is an "ip_mc_list" for convenience; only the fields below
1153          * are actually used. In particular, the refcnt and users are not
1154          * used for management of the delete list. Using the same structure
1155          * for deleted items allows change reports to use common code with
1156          * non-deleted or query-response MCA's.
1157          */
1158         pmc = kzalloc(sizeof(*pmc), GFP_KERNEL);
1159         if (!pmc)
1160                 return;
1161         spin_lock_init(&pmc->lock);
1162         spin_lock_bh(&im->lock);
1163         pmc->interface = im->interface;
1164         in_dev_hold(in_dev);
1165         pmc->multiaddr = im->multiaddr;
1166         pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1167         pmc->sfmode = im->sfmode;
1168         if (pmc->sfmode == MCAST_INCLUDE) {
1169                 struct ip_sf_list *psf;
1170
1171                 pmc->tomb = im->tomb;
1172                 pmc->sources = im->sources;
1173                 im->tomb = im->sources = NULL;
1174                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1175                         psf->sf_crcount = pmc->crcount;
1176         }
1177         spin_unlock_bh(&im->lock);
1178
1179         spin_lock_bh(&in_dev->mc_tomb_lock);
1180         pmc->next = in_dev->mc_tomb;
1181         in_dev->mc_tomb = pmc;
1182         spin_unlock_bh(&in_dev->mc_tomb_lock);
1183 }
1184
1185 /*
1186  * restore ip_mc_list deleted records
1187  */
1188 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1189 {
1190         struct ip_mc_list *pmc, *pmc_prev;
1191         struct ip_sf_list *psf;
1192         struct net *net = dev_net(in_dev->dev);
1193         __be32 multiaddr = im->multiaddr;
1194
1195         spin_lock_bh(&in_dev->mc_tomb_lock);
1196         pmc_prev = NULL;
1197         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1198                 if (pmc->multiaddr == multiaddr)
1199                         break;
1200                 pmc_prev = pmc;
1201         }
1202         if (pmc) {
1203                 if (pmc_prev)
1204                         pmc_prev->next = pmc->next;
1205                 else
1206                         in_dev->mc_tomb = pmc->next;
1207         }
1208         spin_unlock_bh(&in_dev->mc_tomb_lock);
1209
1210         spin_lock_bh(&im->lock);
1211         if (pmc) {
1212                 im->interface = pmc->interface;
1213                 im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1214                 if (im->sfmode == MCAST_INCLUDE) {
1215                         swap(im->tomb, pmc->tomb);
1216                         swap(im->sources, pmc->sources);
1217                         for (psf = im->sources; psf; psf = psf->sf_next)
1218                                 psf->sf_crcount = im->crcount;
1219                 }
1220                 in_dev_put(pmc->interface);
1221                 kfree_pmc(pmc);
1222         }
1223         spin_unlock_bh(&im->lock);
1224 }
1225
1226 /*
1227  * flush ip_mc_list deleted records
1228  */
1229 static void igmpv3_clear_delrec(struct in_device *in_dev)
1230 {
1231         struct ip_mc_list *pmc, *nextpmc;
1232
1233         spin_lock_bh(&in_dev->mc_tomb_lock);
1234         pmc = in_dev->mc_tomb;
1235         in_dev->mc_tomb = NULL;
1236         spin_unlock_bh(&in_dev->mc_tomb_lock);
1237
1238         for (; pmc; pmc = nextpmc) {
1239                 nextpmc = pmc->next;
1240                 ip_mc_clear_src(pmc);
1241                 in_dev_put(pmc->interface);
1242                 kfree_pmc(pmc);
1243         }
1244         /* clear dead sources, too */
1245         rcu_read_lock();
1246         for_each_pmc_rcu(in_dev, pmc) {
1247                 struct ip_sf_list *psf;
1248
1249                 spin_lock_bh(&pmc->lock);
1250                 psf = pmc->tomb;
1251                 pmc->tomb = NULL;
1252                 spin_unlock_bh(&pmc->lock);
1253                 ip_sf_list_clear_all(psf);
1254         }
1255         rcu_read_unlock();
1256 }
1257 #endif
1258
1259 static void igmp_group_dropped(struct ip_mc_list *im)
1260 {
1261         struct in_device *in_dev = im->interface;
1262 #ifdef CONFIG_IP_MULTICAST
1263         struct net *net = dev_net(in_dev->dev);
1264         int reporter;
1265 #endif
1266
1267         if (im->loaded) {
1268                 im->loaded = 0;
1269                 ip_mc_filter_del(in_dev, im->multiaddr);
1270         }
1271
1272 #ifdef CONFIG_IP_MULTICAST
1273         if (im->multiaddr == IGMP_ALL_HOSTS)
1274                 return;
1275         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1276                 return;
1277
1278         reporter = im->reporter;
1279         igmp_stop_timer(im);
1280
1281         if (!in_dev->dead) {
1282                 if (IGMP_V1_SEEN(in_dev))
1283                         return;
1284                 if (IGMP_V2_SEEN(in_dev)) {
1285                         if (reporter)
1286                                 igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1287                         return;
1288                 }
1289                 /* IGMPv3 */
1290                 igmpv3_add_delrec(in_dev, im);
1291
1292                 igmp_ifc_event(in_dev);
1293         }
1294 #endif
1295 }
1296
1297 static void igmp_group_added(struct ip_mc_list *im)
1298 {
1299         struct in_device *in_dev = im->interface;
1300 #ifdef CONFIG_IP_MULTICAST
1301         struct net *net = dev_net(in_dev->dev);
1302 #endif
1303
1304         if (im->loaded == 0) {
1305                 im->loaded = 1;
1306                 ip_mc_filter_add(in_dev, im->multiaddr);
1307         }
1308
1309 #ifdef CONFIG_IP_MULTICAST
1310         if (im->multiaddr == IGMP_ALL_HOSTS)
1311                 return;
1312         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1313                 return;
1314
1315         if (in_dev->dead)
1316                 return;
1317         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1318                 spin_lock_bh(&im->lock);
1319                 igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1320                 spin_unlock_bh(&im->lock);
1321                 return;
1322         }
1323         /* else, v3 */
1324
1325         im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1326         igmp_ifc_event(in_dev);
1327 #endif
1328 }
1329
1330
1331 /*
1332  *      Multicast list managers
1333  */
1334
1335 static u32 ip_mc_hash(const struct ip_mc_list *im)
1336 {
1337         return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1338 }
1339
1340 static void ip_mc_hash_add(struct in_device *in_dev,
1341                            struct ip_mc_list *im)
1342 {
1343         struct ip_mc_list __rcu **mc_hash;
1344         u32 hash;
1345
1346         mc_hash = rtnl_dereference(in_dev->mc_hash);
1347         if (mc_hash) {
1348                 hash = ip_mc_hash(im);
1349                 im->next_hash = mc_hash[hash];
1350                 rcu_assign_pointer(mc_hash[hash], im);
1351                 return;
1352         }
1353
1354         /* do not use a hash table for small number of items */
1355         if (in_dev->mc_count < 4)
1356                 return;
1357
1358         mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1359                           GFP_KERNEL);
1360         if (!mc_hash)
1361                 return;
1362
1363         for_each_pmc_rtnl(in_dev, im) {
1364                 hash = ip_mc_hash(im);
1365                 im->next_hash = mc_hash[hash];
1366                 RCU_INIT_POINTER(mc_hash[hash], im);
1367         }
1368
1369         rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1370 }
1371
1372 static void ip_mc_hash_remove(struct in_device *in_dev,
1373                               struct ip_mc_list *im)
1374 {
1375         struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1376         struct ip_mc_list *aux;
1377
1378         if (!mc_hash)
1379                 return;
1380         mc_hash += ip_mc_hash(im);
1381         while ((aux = rtnl_dereference(*mc_hash)) != im)
1382                 mc_hash = &aux->next_hash;
1383         *mc_hash = im->next_hash;
1384 }
1385
1386
1387 /*
1388  *      A socket has joined a multicast group on device dev.
1389  */
1390
1391 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1392 {
1393         struct ip_mc_list *im;
1394 #ifdef CONFIG_IP_MULTICAST
1395         struct net *net = dev_net(in_dev->dev);
1396 #endif
1397
1398         ASSERT_RTNL();
1399
1400         for_each_pmc_rtnl(in_dev, im) {
1401                 if (im->multiaddr == addr) {
1402                         im->users++;
1403                         ip_mc_add_src(in_dev, &addr, MCAST_EXCLUDE, 0, NULL, 0);
1404                         goto out;
1405                 }
1406         }
1407
1408         im = kzalloc(sizeof(*im), GFP_KERNEL);
1409         if (!im)
1410                 goto out;
1411
1412         im->users = 1;
1413         im->interface = in_dev;
1414         in_dev_hold(in_dev);
1415         im->multiaddr = addr;
1416         /* initial mode is (EX, empty) */
1417         im->sfmode = MCAST_EXCLUDE;
1418         im->sfcount[MCAST_EXCLUDE] = 1;
1419         atomic_set(&im->refcnt, 1);
1420         spin_lock_init(&im->lock);
1421 #ifdef CONFIG_IP_MULTICAST
1422         setup_timer(&im->timer, igmp_timer_expire, (unsigned long)im);
1423         im->unsolicit_count = net->ipv4.sysctl_igmp_qrv;
1424 #endif
1425
1426         im->next_rcu = in_dev->mc_list;
1427         in_dev->mc_count++;
1428         rcu_assign_pointer(in_dev->mc_list, im);
1429
1430         ip_mc_hash_add(in_dev, im);
1431
1432 #ifdef CONFIG_IP_MULTICAST
1433         igmpv3_del_delrec(in_dev, im);
1434 #endif
1435         igmp_group_added(im);
1436         if (!in_dev->dead)
1437                 ip_rt_multicast_event(in_dev);
1438 out:
1439         return;
1440 }
1441 EXPORT_SYMBOL(ip_mc_inc_group);
1442
1443 static int ip_mc_check_iphdr(struct sk_buff *skb)
1444 {
1445         const struct iphdr *iph;
1446         unsigned int len;
1447         unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1448
1449         if (!pskb_may_pull(skb, offset))
1450                 return -EINVAL;
1451
1452         iph = ip_hdr(skb);
1453
1454         if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1455                 return -EINVAL;
1456
1457         offset += ip_hdrlen(skb) - sizeof(*iph);
1458
1459         if (!pskb_may_pull(skb, offset))
1460                 return -EINVAL;
1461
1462         iph = ip_hdr(skb);
1463
1464         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1465                 return -EINVAL;
1466
1467         len = skb_network_offset(skb) + ntohs(iph->tot_len);
1468         if (skb->len < len || len < offset)
1469                 return -EINVAL;
1470
1471         skb_set_transport_header(skb, offset);
1472
1473         return 0;
1474 }
1475
1476 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1477 {
1478         unsigned int len = skb_transport_offset(skb);
1479
1480         len += sizeof(struct igmpv3_report);
1481
1482         return pskb_may_pull(skb, len) ? 0 : -EINVAL;
1483 }
1484
1485 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1486 {
1487         unsigned int len = skb_transport_offset(skb);
1488
1489         len += sizeof(struct igmphdr);
1490         if (skb->len < len)
1491                 return -EINVAL;
1492
1493         /* IGMPv{1,2}? */
1494         if (skb->len != len) {
1495                 /* or IGMPv3? */
1496                 len += sizeof(struct igmpv3_query) - sizeof(struct igmphdr);
1497                 if (skb->len < len || !pskb_may_pull(skb, len))
1498                         return -EINVAL;
1499         }
1500
1501         /* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1502          * all-systems destination addresses (224.0.0.1) for general queries
1503          */
1504         if (!igmp_hdr(skb)->group &&
1505             ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1506                 return -EINVAL;
1507
1508         return 0;
1509 }
1510
1511 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1512 {
1513         switch (igmp_hdr(skb)->type) {
1514         case IGMP_HOST_LEAVE_MESSAGE:
1515         case IGMP_HOST_MEMBERSHIP_REPORT:
1516         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1517                 /* fall through */
1518                 return 0;
1519         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1520                 return ip_mc_check_igmp_reportv3(skb);
1521         case IGMP_HOST_MEMBERSHIP_QUERY:
1522                 return ip_mc_check_igmp_query(skb);
1523         default:
1524                 return -ENOMSG;
1525         }
1526 }
1527
1528 static inline __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1529 {
1530         return skb_checksum_simple_validate(skb);
1531 }
1532
1533 static int __ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1534
1535 {
1536         struct sk_buff *skb_chk;
1537         unsigned int transport_len;
1538         unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1539         int ret = -EINVAL;
1540
1541         transport_len = ntohs(ip_hdr(skb)->tot_len) - ip_hdrlen(skb);
1542
1543         skb_chk = skb_checksum_trimmed(skb, transport_len,
1544                                        ip_mc_validate_checksum);
1545         if (!skb_chk)
1546                 goto err;
1547
1548         if (!pskb_may_pull(skb_chk, len))
1549                 goto err;
1550
1551         ret = ip_mc_check_igmp_msg(skb_chk);
1552         if (ret)
1553                 goto err;
1554
1555         if (skb_trimmed)
1556                 *skb_trimmed = skb_chk;
1557         /* free now unneeded clone */
1558         else if (skb_chk != skb)
1559                 kfree_skb(skb_chk);
1560
1561         ret = 0;
1562
1563 err:
1564         if (ret && skb_chk && skb_chk != skb)
1565                 kfree_skb(skb_chk);
1566
1567         return ret;
1568 }
1569
1570 /**
1571  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1572  * @skb: the skb to validate
1573  * @skb_trimmed: to store an skb pointer trimmed to IPv4 packet tail (optional)
1574  *
1575  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1576  * skb transport header accordingly and returns zero.
1577  *
1578  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1579  *  standard
1580  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1581  * -ENOMEM: A memory allocation failure happened.
1582  *
1583  * Optionally, an skb pointer might be provided via skb_trimmed (or set it
1584  * to NULL): After parsing an IGMP packet successfully it will point to
1585  * an skb which has its tail aligned to the IP packet end. This might
1586  * either be the originally provided skb or a trimmed, cloned version if
1587  * the skb frame had data beyond the IP packet. A cloned skb allows us
1588  * to leave the original skb and its full frame unchanged (which might be
1589  * desirable for layer 2 frame jugglers).
1590  *
1591  * Caller needs to set the skb network header and free any returned skb if it
1592  * differs from the provided skb.
1593  */
1594 int ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1595 {
1596         int ret = ip_mc_check_iphdr(skb);
1597
1598         if (ret < 0)
1599                 return ret;
1600
1601         if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1602                 return -ENOMSG;
1603
1604         return __ip_mc_check_igmp(skb, skb_trimmed);
1605 }
1606 EXPORT_SYMBOL(ip_mc_check_igmp);
1607
1608 /*
1609  *      Resend IGMP JOIN report; used by netdev notifier.
1610  */
1611 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1612 {
1613 #ifdef CONFIG_IP_MULTICAST
1614         struct ip_mc_list *im;
1615         int type;
1616         struct net *net = dev_net(in_dev->dev);
1617
1618         ASSERT_RTNL();
1619
1620         for_each_pmc_rtnl(in_dev, im) {
1621                 if (im->multiaddr == IGMP_ALL_HOSTS)
1622                         continue;
1623                 if (ipv4_is_local_multicast(im->multiaddr) &&
1624                     !net->ipv4.sysctl_igmp_llm_reports)
1625                         continue;
1626
1627                 /* a failover is happening and switches
1628                  * must be notified immediately
1629                  */
1630                 if (IGMP_V1_SEEN(in_dev))
1631                         type = IGMP_HOST_MEMBERSHIP_REPORT;
1632                 else if (IGMP_V2_SEEN(in_dev))
1633                         type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1634                 else
1635                         type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1636                 igmp_send_report(in_dev, im, type);
1637         }
1638 #endif
1639 }
1640
1641 /*
1642  *      A socket has left a multicast group on device dev
1643  */
1644
1645 void ip_mc_dec_group(struct in_device *in_dev, __be32 addr)
1646 {
1647         struct ip_mc_list *i;
1648         struct ip_mc_list __rcu **ip;
1649
1650         ASSERT_RTNL();
1651
1652         for (ip = &in_dev->mc_list;
1653              (i = rtnl_dereference(*ip)) != NULL;
1654              ip = &i->next_rcu) {
1655                 if (i->multiaddr == addr) {
1656                         if (--i->users == 0) {
1657                                 ip_mc_hash_remove(in_dev, i);
1658                                 *ip = i->next_rcu;
1659                                 in_dev->mc_count--;
1660                                 igmp_group_dropped(i);
1661                                 ip_mc_clear_src(i);
1662
1663                                 if (!in_dev->dead)
1664                                         ip_rt_multicast_event(in_dev);
1665
1666                                 ip_ma_put(i);
1667                                 return;
1668                         }
1669                         break;
1670                 }
1671         }
1672 }
1673 EXPORT_SYMBOL(ip_mc_dec_group);
1674
1675 /* Device changing type */
1676
1677 void ip_mc_unmap(struct in_device *in_dev)
1678 {
1679         struct ip_mc_list *pmc;
1680
1681         ASSERT_RTNL();
1682
1683         for_each_pmc_rtnl(in_dev, pmc)
1684                 igmp_group_dropped(pmc);
1685 }
1686
1687 void ip_mc_remap(struct in_device *in_dev)
1688 {
1689         struct ip_mc_list *pmc;
1690
1691         ASSERT_RTNL();
1692
1693         for_each_pmc_rtnl(in_dev, pmc) {
1694 #ifdef CONFIG_IP_MULTICAST
1695                 igmpv3_del_delrec(in_dev, pmc);
1696 #endif
1697                 igmp_group_added(pmc);
1698         }
1699 }
1700
1701 /* Device going down */
1702
1703 void ip_mc_down(struct in_device *in_dev)
1704 {
1705         struct ip_mc_list *pmc;
1706
1707         ASSERT_RTNL();
1708
1709         for_each_pmc_rtnl(in_dev, pmc)
1710                 igmp_group_dropped(pmc);
1711
1712 #ifdef CONFIG_IP_MULTICAST
1713         in_dev->mr_ifc_count = 0;
1714         if (del_timer(&in_dev->mr_ifc_timer))
1715                 __in_dev_put(in_dev);
1716         in_dev->mr_gq_running = 0;
1717         if (del_timer(&in_dev->mr_gq_timer))
1718                 __in_dev_put(in_dev);
1719 #endif
1720
1721         ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1722 }
1723
1724 void ip_mc_init_dev(struct in_device *in_dev)
1725 {
1726 #ifdef CONFIG_IP_MULTICAST
1727         struct net *net = dev_net(in_dev->dev);
1728 #endif
1729         ASSERT_RTNL();
1730
1731 #ifdef CONFIG_IP_MULTICAST
1732         setup_timer(&in_dev->mr_gq_timer, igmp_gq_timer_expire,
1733                         (unsigned long)in_dev);
1734         setup_timer(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire,
1735                         (unsigned long)in_dev);
1736         in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1737 #endif
1738
1739         spin_lock_init(&in_dev->mc_tomb_lock);
1740 }
1741
1742 /* Device going up */
1743
1744 void ip_mc_up(struct in_device *in_dev)
1745 {
1746         struct ip_mc_list *pmc;
1747 #ifdef CONFIG_IP_MULTICAST
1748         struct net *net = dev_net(in_dev->dev);
1749 #endif
1750
1751         ASSERT_RTNL();
1752
1753 #ifdef CONFIG_IP_MULTICAST
1754         in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1755 #endif
1756         ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1757
1758         for_each_pmc_rtnl(in_dev, pmc) {
1759 #ifdef CONFIG_IP_MULTICAST
1760                 igmpv3_del_delrec(in_dev, pmc);
1761 #endif
1762                 igmp_group_added(pmc);
1763         }
1764 }
1765
1766 /*
1767  *      Device is about to be destroyed: clean up.
1768  */
1769
1770 void ip_mc_destroy_dev(struct in_device *in_dev)
1771 {
1772         struct ip_mc_list *i;
1773
1774         ASSERT_RTNL();
1775
1776         /* Deactivate timers */
1777         ip_mc_down(in_dev);
1778 #ifdef CONFIG_IP_MULTICAST
1779         igmpv3_clear_delrec(in_dev);
1780 #endif
1781
1782         while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1783                 in_dev->mc_list = i->next_rcu;
1784                 in_dev->mc_count--;
1785                 ip_mc_clear_src(i);
1786                 ip_ma_put(i);
1787         }
1788 }
1789
1790 /* RTNL is locked */
1791 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1792 {
1793         struct net_device *dev = NULL;
1794         struct in_device *idev = NULL;
1795
1796         if (imr->imr_ifindex) {
1797                 idev = inetdev_by_index(net, imr->imr_ifindex);
1798                 return idev;
1799         }
1800         if (imr->imr_address.s_addr) {
1801                 dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1802                 if (!dev)
1803                         return NULL;
1804         }
1805
1806         if (!dev) {
1807                 struct rtable *rt = ip_route_output(net,
1808                                                     imr->imr_multiaddr.s_addr,
1809                                                     0, 0, 0);
1810                 if (!IS_ERR(rt)) {
1811                         dev = rt->dst.dev;
1812                         ip_rt_put(rt);
1813                 }
1814         }
1815         if (dev) {
1816                 imr->imr_ifindex = dev->ifindex;
1817                 idev = __in_dev_get_rtnl(dev);
1818         }
1819         return idev;
1820 }
1821
1822 /*
1823  *      Join a socket to a group
1824  */
1825
1826 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1827         __be32 *psfsrc)
1828 {
1829         struct ip_sf_list *psf, *psf_prev;
1830         int rv = 0;
1831
1832         psf_prev = NULL;
1833         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1834                 if (psf->sf_inaddr == *psfsrc)
1835                         break;
1836                 psf_prev = psf;
1837         }
1838         if (!psf || psf->sf_count[sfmode] == 0) {
1839                 /* source filter not found, or count wrong =>  bug */
1840                 return -ESRCH;
1841         }
1842         psf->sf_count[sfmode]--;
1843         if (psf->sf_count[sfmode] == 0) {
1844                 ip_rt_multicast_event(pmc->interface);
1845         }
1846         if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1847 #ifdef CONFIG_IP_MULTICAST
1848                 struct in_device *in_dev = pmc->interface;
1849                 struct net *net = dev_net(in_dev->dev);
1850 #endif
1851
1852                 /* no more filters for this source */
1853                 if (psf_prev)
1854                         psf_prev->sf_next = psf->sf_next;
1855                 else
1856                         pmc->sources = psf->sf_next;
1857 #ifdef CONFIG_IP_MULTICAST
1858                 if (psf->sf_oldin &&
1859                     !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1860                         psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1861                         psf->sf_next = pmc->tomb;
1862                         pmc->tomb = psf;
1863                         rv = 1;
1864                 } else
1865 #endif
1866                         kfree(psf);
1867         }
1868         return rv;
1869 }
1870
1871 #ifndef CONFIG_IP_MULTICAST
1872 #define igmp_ifc_event(x)       do { } while (0)
1873 #endif
1874
1875 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1876                          int sfcount, __be32 *psfsrc, int delta)
1877 {
1878         struct ip_mc_list *pmc;
1879         int     changerec = 0;
1880         int     i, err;
1881
1882         if (!in_dev)
1883                 return -ENODEV;
1884         rcu_read_lock();
1885         for_each_pmc_rcu(in_dev, pmc) {
1886                 if (*pmca == pmc->multiaddr)
1887                         break;
1888         }
1889         if (!pmc) {
1890                 /* MCA not found?? bug */
1891                 rcu_read_unlock();
1892                 return -ESRCH;
1893         }
1894         spin_lock_bh(&pmc->lock);
1895         rcu_read_unlock();
1896 #ifdef CONFIG_IP_MULTICAST
1897         sf_markstate(pmc);
1898 #endif
1899         if (!delta) {
1900                 err = -EINVAL;
1901                 if (!pmc->sfcount[sfmode])
1902                         goto out_unlock;
1903                 pmc->sfcount[sfmode]--;
1904         }
1905         err = 0;
1906         for (i = 0; i < sfcount; i++) {
1907                 int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1908
1909                 changerec |= rv > 0;
1910                 if (!err && rv < 0)
1911                         err = rv;
1912         }
1913         if (pmc->sfmode == MCAST_EXCLUDE &&
1914             pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1915             pmc->sfcount[MCAST_INCLUDE]) {
1916 #ifdef CONFIG_IP_MULTICAST
1917                 struct ip_sf_list *psf;
1918                 struct net *net = dev_net(in_dev->dev);
1919 #endif
1920
1921                 /* filter mode change */
1922                 pmc->sfmode = MCAST_INCLUDE;
1923 #ifdef CONFIG_IP_MULTICAST
1924                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1925                 in_dev->mr_ifc_count = pmc->crcount;
1926                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1927                         psf->sf_crcount = 0;
1928                 igmp_ifc_event(pmc->interface);
1929         } else if (sf_setstate(pmc) || changerec) {
1930                 igmp_ifc_event(pmc->interface);
1931 #endif
1932         }
1933 out_unlock:
1934         spin_unlock_bh(&pmc->lock);
1935         return err;
1936 }
1937
1938 /*
1939  * Add multicast single-source filter to the interface list
1940  */
1941 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1942         __be32 *psfsrc)
1943 {
1944         struct ip_sf_list *psf, *psf_prev;
1945
1946         psf_prev = NULL;
1947         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1948                 if (psf->sf_inaddr == *psfsrc)
1949                         break;
1950                 psf_prev = psf;
1951         }
1952         if (!psf) {
1953                 psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1954                 if (!psf)
1955                         return -ENOBUFS;
1956                 psf->sf_inaddr = *psfsrc;
1957                 if (psf_prev) {
1958                         psf_prev->sf_next = psf;
1959                 } else
1960                         pmc->sources = psf;
1961         }
1962         psf->sf_count[sfmode]++;
1963         if (psf->sf_count[sfmode] == 1) {
1964                 ip_rt_multicast_event(pmc->interface);
1965         }
1966         return 0;
1967 }
1968
1969 #ifdef CONFIG_IP_MULTICAST
1970 static void sf_markstate(struct ip_mc_list *pmc)
1971 {
1972         struct ip_sf_list *psf;
1973         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1974
1975         for (psf = pmc->sources; psf; psf = psf->sf_next)
1976                 if (pmc->sfcount[MCAST_EXCLUDE]) {
1977                         psf->sf_oldin = mca_xcount ==
1978                                 psf->sf_count[MCAST_EXCLUDE] &&
1979                                 !psf->sf_count[MCAST_INCLUDE];
1980                 } else
1981                         psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
1982 }
1983
1984 static int sf_setstate(struct ip_mc_list *pmc)
1985 {
1986         struct ip_sf_list *psf, *dpsf;
1987         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1988         int qrv = pmc->interface->mr_qrv;
1989         int new_in, rv;
1990
1991         rv = 0;
1992         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1993                 if (pmc->sfcount[MCAST_EXCLUDE]) {
1994                         new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
1995                                 !psf->sf_count[MCAST_INCLUDE];
1996                 } else
1997                         new_in = psf->sf_count[MCAST_INCLUDE] != 0;
1998                 if (new_in) {
1999                         if (!psf->sf_oldin) {
2000                                 struct ip_sf_list *prev = NULL;
2001
2002                                 for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
2003                                         if (dpsf->sf_inaddr == psf->sf_inaddr)
2004                                                 break;
2005                                         prev = dpsf;
2006                                 }
2007                                 if (dpsf) {
2008                                         if (prev)
2009                                                 prev->sf_next = dpsf->sf_next;
2010                                         else
2011                                                 pmc->tomb = dpsf->sf_next;
2012                                         kfree(dpsf);
2013                                 }
2014                                 psf->sf_crcount = qrv;
2015                                 rv++;
2016                         }
2017                 } else if (psf->sf_oldin) {
2018
2019                         psf->sf_crcount = 0;
2020                         /*
2021                          * add or update "delete" records if an active filter
2022                          * is now inactive
2023                          */
2024                         for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
2025                                 if (dpsf->sf_inaddr == psf->sf_inaddr)
2026                                         break;
2027                         if (!dpsf) {
2028                                 dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
2029                                 if (!dpsf)
2030                                         continue;
2031                                 *dpsf = *psf;
2032                                 /* pmc->lock held by callers */
2033                                 dpsf->sf_next = pmc->tomb;
2034                                 pmc->tomb = dpsf;
2035                         }
2036                         dpsf->sf_crcount = qrv;
2037                         rv++;
2038                 }
2039         }
2040         return rv;
2041 }
2042 #endif
2043
2044 /*
2045  * Add multicast source filter list to the interface list
2046  */
2047 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2048                          int sfcount, __be32 *psfsrc, int delta)
2049 {
2050         struct ip_mc_list *pmc;
2051         int     isexclude;
2052         int     i, err;
2053
2054         if (!in_dev)
2055                 return -ENODEV;
2056         rcu_read_lock();
2057         for_each_pmc_rcu(in_dev, pmc) {
2058                 if (*pmca == pmc->multiaddr)
2059                         break;
2060         }
2061         if (!pmc) {
2062                 /* MCA not found?? bug */
2063                 rcu_read_unlock();
2064                 return -ESRCH;
2065         }
2066         spin_lock_bh(&pmc->lock);
2067         rcu_read_unlock();
2068
2069 #ifdef CONFIG_IP_MULTICAST
2070         sf_markstate(pmc);
2071 #endif
2072         isexclude = pmc->sfmode == MCAST_EXCLUDE;
2073         if (!delta)
2074                 pmc->sfcount[sfmode]++;
2075         err = 0;
2076         for (i = 0; i < sfcount; i++) {
2077                 err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2078                 if (err)
2079                         break;
2080         }
2081         if (err) {
2082                 int j;
2083
2084                 if (!delta)
2085                         pmc->sfcount[sfmode]--;
2086                 for (j = 0; j < i; j++)
2087                         (void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2088         } else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2089 #ifdef CONFIG_IP_MULTICAST
2090                 struct ip_sf_list *psf;
2091                 struct net *net = dev_net(pmc->interface->dev);
2092                 in_dev = pmc->interface;
2093 #endif
2094
2095                 /* filter mode change */
2096                 if (pmc->sfcount[MCAST_EXCLUDE])
2097                         pmc->sfmode = MCAST_EXCLUDE;
2098                 else if (pmc->sfcount[MCAST_INCLUDE])
2099                         pmc->sfmode = MCAST_INCLUDE;
2100 #ifdef CONFIG_IP_MULTICAST
2101                 /* else no filters; keep old mode for reports */
2102
2103                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
2104                 in_dev->mr_ifc_count = pmc->crcount;
2105                 for (psf = pmc->sources; psf; psf = psf->sf_next)
2106                         psf->sf_crcount = 0;
2107                 igmp_ifc_event(in_dev);
2108         } else if (sf_setstate(pmc)) {
2109                 igmp_ifc_event(in_dev);
2110 #endif
2111         }
2112         spin_unlock_bh(&pmc->lock);
2113         return err;
2114 }
2115
2116 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2117 {
2118         struct ip_sf_list *tomb, *sources;
2119
2120         spin_lock_bh(&pmc->lock);
2121         tomb = pmc->tomb;
2122         pmc->tomb = NULL;
2123         sources = pmc->sources;
2124         pmc->sources = NULL;
2125         pmc->sfmode = MCAST_EXCLUDE;
2126         pmc->sfcount[MCAST_INCLUDE] = 0;
2127         pmc->sfcount[MCAST_EXCLUDE] = 1;
2128         spin_unlock_bh(&pmc->lock);
2129
2130         ip_sf_list_clear_all(tomb);
2131         ip_sf_list_clear_all(sources);
2132 }
2133
2134 /* Join a multicast group
2135  */
2136
2137 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2138 {
2139         __be32 addr = imr->imr_multiaddr.s_addr;
2140         struct ip_mc_socklist *iml, *i;
2141         struct in_device *in_dev;
2142         struct inet_sock *inet = inet_sk(sk);
2143         struct net *net = sock_net(sk);
2144         int ifindex;
2145         int count = 0;
2146         int err;
2147
2148         ASSERT_RTNL();
2149
2150         if (!ipv4_is_multicast(addr))
2151                 return -EINVAL;
2152
2153         in_dev = ip_mc_find_dev(net, imr);
2154
2155         if (!in_dev) {
2156                 err = -ENODEV;
2157                 goto done;
2158         }
2159
2160         err = -EADDRINUSE;
2161         ifindex = imr->imr_ifindex;
2162         for_each_pmc_rtnl(inet, i) {
2163                 if (i->multi.imr_multiaddr.s_addr == addr &&
2164                     i->multi.imr_ifindex == ifindex)
2165                         goto done;
2166                 count++;
2167         }
2168         err = -ENOBUFS;
2169         if (count >= net->ipv4.sysctl_igmp_max_memberships)
2170                 goto done;
2171         iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2172         if (!iml)
2173                 goto done;
2174
2175         memcpy(&iml->multi, imr, sizeof(*imr));
2176         iml->next_rcu = inet->mc_list;
2177         iml->sflist = NULL;
2178         iml->sfmode = MCAST_EXCLUDE;
2179         rcu_assign_pointer(inet->mc_list, iml);
2180         ip_mc_inc_group(in_dev, addr);
2181         err = 0;
2182 done:
2183         return err;
2184 }
2185 EXPORT_SYMBOL(ip_mc_join_group);
2186
2187 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2188                            struct in_device *in_dev)
2189 {
2190         struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2191         int err;
2192
2193         if (!psf) {
2194                 /* any-source empty exclude case */
2195                 return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2196                         iml->sfmode, 0, NULL, 0);
2197         }
2198         err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2199                         iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2200         RCU_INIT_POINTER(iml->sflist, NULL);
2201         /* decrease mem now to avoid the memleak warning */
2202         atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
2203         kfree_rcu(psf, rcu);
2204         return err;
2205 }
2206
2207 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2208 {
2209         struct inet_sock *inet = inet_sk(sk);
2210         struct ip_mc_socklist *iml;
2211         struct ip_mc_socklist __rcu **imlp;
2212         struct in_device *in_dev;
2213         struct net *net = sock_net(sk);
2214         __be32 group = imr->imr_multiaddr.s_addr;
2215         u32 ifindex;
2216         int ret = -EADDRNOTAVAIL;
2217
2218         ASSERT_RTNL();
2219
2220         in_dev = ip_mc_find_dev(net, imr);
2221         if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2222                 ret = -ENODEV;
2223                 goto out;
2224         }
2225         ifindex = imr->imr_ifindex;
2226         for (imlp = &inet->mc_list;
2227              (iml = rtnl_dereference(*imlp)) != NULL;
2228              imlp = &iml->next_rcu) {
2229                 if (iml->multi.imr_multiaddr.s_addr != group)
2230                         continue;
2231                 if (ifindex) {
2232                         if (iml->multi.imr_ifindex != ifindex)
2233                                 continue;
2234                 } else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2235                                 iml->multi.imr_address.s_addr)
2236                         continue;
2237
2238                 (void) ip_mc_leave_src(sk, iml, in_dev);
2239
2240                 *imlp = iml->next_rcu;
2241
2242                 if (in_dev)
2243                         ip_mc_dec_group(in_dev, group);
2244
2245                 /* decrease mem now to avoid the memleak warning */
2246                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2247                 kfree_rcu(iml, rcu);
2248                 return 0;
2249         }
2250 out:
2251         return ret;
2252 }
2253 EXPORT_SYMBOL(ip_mc_leave_group);
2254
2255 int ip_mc_source(int add, int omode, struct sock *sk, struct
2256         ip_mreq_source *mreqs, int ifindex)
2257 {
2258         int err;
2259         struct ip_mreqn imr;
2260         __be32 addr = mreqs->imr_multiaddr;
2261         struct ip_mc_socklist *pmc;
2262         struct in_device *in_dev = NULL;
2263         struct inet_sock *inet = inet_sk(sk);
2264         struct ip_sf_socklist *psl;
2265         struct net *net = sock_net(sk);
2266         int leavegroup = 0;
2267         int i, j, rv;
2268
2269         if (!ipv4_is_multicast(addr))
2270                 return -EINVAL;
2271
2272         ASSERT_RTNL();
2273
2274         imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2275         imr.imr_address.s_addr = mreqs->imr_interface;
2276         imr.imr_ifindex = ifindex;
2277         in_dev = ip_mc_find_dev(net, &imr);
2278
2279         if (!in_dev) {
2280                 err = -ENODEV;
2281                 goto done;
2282         }
2283         err = -EADDRNOTAVAIL;
2284
2285         for_each_pmc_rtnl(inet, pmc) {
2286                 if ((pmc->multi.imr_multiaddr.s_addr ==
2287                      imr.imr_multiaddr.s_addr) &&
2288                     (pmc->multi.imr_ifindex == imr.imr_ifindex))
2289                         break;
2290         }
2291         if (!pmc) {             /* must have a prior join */
2292                 err = -EINVAL;
2293                 goto done;
2294         }
2295         /* if a source filter was set, must be the same mode as before */
2296         if (pmc->sflist) {
2297                 if (pmc->sfmode != omode) {
2298                         err = -EINVAL;
2299                         goto done;
2300                 }
2301         } else if (pmc->sfmode != omode) {
2302                 /* allow mode switches for empty-set filters */
2303                 ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2304                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2305                         NULL, 0);
2306                 pmc->sfmode = omode;
2307         }
2308
2309         psl = rtnl_dereference(pmc->sflist);
2310         if (!add) {
2311                 if (!psl)
2312                         goto done;      /* err = -EADDRNOTAVAIL */
2313                 rv = !0;
2314                 for (i = 0; i < psl->sl_count; i++) {
2315                         rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2316                                 sizeof(__be32));
2317                         if (rv == 0)
2318                                 break;
2319                 }
2320                 if (rv)         /* source not found */
2321                         goto done;      /* err = -EADDRNOTAVAIL */
2322
2323                 /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2324                 if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2325                         leavegroup = 1;
2326                         goto done;
2327                 }
2328
2329                 /* update the interface filter */
2330                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2331                         &mreqs->imr_sourceaddr, 1);
2332
2333                 for (j = i+1; j < psl->sl_count; j++)
2334                         psl->sl_addr[j-1] = psl->sl_addr[j];
2335                 psl->sl_count--;
2336                 err = 0;
2337                 goto done;
2338         }
2339         /* else, add a new source to the filter */
2340
2341         if (psl && psl->sl_count >= net->ipv4.sysctl_igmp_max_msf) {
2342                 err = -ENOBUFS;
2343                 goto done;
2344         }
2345         if (!psl || psl->sl_count == psl->sl_max) {
2346                 struct ip_sf_socklist *newpsl;
2347                 int count = IP_SFBLOCK;
2348
2349                 if (psl)
2350                         count += psl->sl_max;
2351                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2352                 if (!newpsl) {
2353                         err = -ENOBUFS;
2354                         goto done;
2355                 }
2356                 newpsl->sl_max = count;
2357                 newpsl->sl_count = count - IP_SFBLOCK;
2358                 if (psl) {
2359                         for (i = 0; i < psl->sl_count; i++)
2360                                 newpsl->sl_addr[i] = psl->sl_addr[i];
2361                         /* decrease mem now to avoid the memleak warning */
2362                         atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2363                 }
2364                 rcu_assign_pointer(pmc->sflist, newpsl);
2365                 if (psl)
2366                         kfree_rcu(psl, rcu);
2367                 psl = newpsl;
2368         }
2369         rv = 1; /* > 0 for insert logic below if sl_count is 0 */
2370         for (i = 0; i < psl->sl_count; i++) {
2371                 rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2372                         sizeof(__be32));
2373                 if (rv == 0)
2374                         break;
2375         }
2376         if (rv == 0)            /* address already there is an error */
2377                 goto done;
2378         for (j = psl->sl_count-1; j >= i; j--)
2379                 psl->sl_addr[j+1] = psl->sl_addr[j];
2380         psl->sl_addr[i] = mreqs->imr_sourceaddr;
2381         psl->sl_count++;
2382         err = 0;
2383         /* update the interface list */
2384         ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2385                 &mreqs->imr_sourceaddr, 1);
2386 done:
2387         if (leavegroup)
2388                 err = ip_mc_leave_group(sk, &imr);
2389         return err;
2390 }
2391
2392 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2393 {
2394         int err = 0;
2395         struct ip_mreqn imr;
2396         __be32 addr = msf->imsf_multiaddr;
2397         struct ip_mc_socklist *pmc;
2398         struct in_device *in_dev;
2399         struct inet_sock *inet = inet_sk(sk);
2400         struct ip_sf_socklist *newpsl, *psl;
2401         struct net *net = sock_net(sk);
2402         int leavegroup = 0;
2403
2404         if (!ipv4_is_multicast(addr))
2405                 return -EINVAL;
2406         if (msf->imsf_fmode != MCAST_INCLUDE &&
2407             msf->imsf_fmode != MCAST_EXCLUDE)
2408                 return -EINVAL;
2409
2410         ASSERT_RTNL();
2411
2412         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2413         imr.imr_address.s_addr = msf->imsf_interface;
2414         imr.imr_ifindex = ifindex;
2415         in_dev = ip_mc_find_dev(net, &imr);
2416
2417         if (!in_dev) {
2418                 err = -ENODEV;
2419                 goto done;
2420         }
2421
2422         /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2423         if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2424                 leavegroup = 1;
2425                 goto done;
2426         }
2427
2428         for_each_pmc_rtnl(inet, pmc) {
2429                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2430                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2431                         break;
2432         }
2433         if (!pmc) {             /* must have a prior join */
2434                 err = -EINVAL;
2435                 goto done;
2436         }
2437         if (msf->imsf_numsrc) {
2438                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2439                                                            GFP_KERNEL);
2440                 if (!newpsl) {
2441                         err = -ENOBUFS;
2442                         goto done;
2443                 }
2444                 newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2445                 memcpy(newpsl->sl_addr, msf->imsf_slist,
2446                         msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2447                 err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2448                         msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2449                 if (err) {
2450                         sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2451                         goto done;
2452                 }
2453         } else {
2454                 newpsl = NULL;
2455                 (void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2456                                      msf->imsf_fmode, 0, NULL, 0);
2457         }
2458         psl = rtnl_dereference(pmc->sflist);
2459         if (psl) {
2460                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2461                         psl->sl_count, psl->sl_addr, 0);
2462                 /* decrease mem now to avoid the memleak warning */
2463                 atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2464         } else {
2465                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2466                         0, NULL, 0);
2467         }
2468         rcu_assign_pointer(pmc->sflist, newpsl);
2469         if (psl)
2470                 kfree_rcu(psl, rcu);
2471         pmc->sfmode = msf->imsf_fmode;
2472         err = 0;
2473 done:
2474         if (leavegroup)
2475                 err = ip_mc_leave_group(sk, &imr);
2476         return err;
2477 }
2478
2479 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2480         struct ip_msfilter __user *optval, int __user *optlen)
2481 {
2482         int err, len, count, copycount;
2483         struct ip_mreqn imr;
2484         __be32 addr = msf->imsf_multiaddr;
2485         struct ip_mc_socklist *pmc;
2486         struct in_device *in_dev;
2487         struct inet_sock *inet = inet_sk(sk);
2488         struct ip_sf_socklist *psl;
2489         struct net *net = sock_net(sk);
2490
2491         ASSERT_RTNL();
2492
2493         if (!ipv4_is_multicast(addr))
2494                 return -EINVAL;
2495
2496         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2497         imr.imr_address.s_addr = msf->imsf_interface;
2498         imr.imr_ifindex = 0;
2499         in_dev = ip_mc_find_dev(net, &imr);
2500
2501         if (!in_dev) {
2502                 err = -ENODEV;
2503                 goto done;
2504         }
2505         err = -EADDRNOTAVAIL;
2506
2507         for_each_pmc_rtnl(inet, pmc) {
2508                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2509                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2510                         break;
2511         }
2512         if (!pmc)               /* must have a prior join */
2513                 goto done;
2514         msf->imsf_fmode = pmc->sfmode;
2515         psl = rtnl_dereference(pmc->sflist);
2516         if (!psl) {
2517                 len = 0;
2518                 count = 0;
2519         } else {
2520                 count = psl->sl_count;
2521         }
2522         copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2523         len = copycount * sizeof(psl->sl_addr[0]);
2524         msf->imsf_numsrc = count;
2525         if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2526             copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2527                 return -EFAULT;
2528         }
2529         if (len &&
2530             copy_to_user(&optval->imsf_slist[0], psl->sl_addr, len))
2531                 return -EFAULT;
2532         return 0;
2533 done:
2534         return err;
2535 }
2536
2537 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2538         struct group_filter __user *optval, int __user *optlen)
2539 {
2540         int err, i, count, copycount;
2541         struct sockaddr_in *psin;
2542         __be32 addr;
2543         struct ip_mc_socklist *pmc;
2544         struct inet_sock *inet = inet_sk(sk);
2545         struct ip_sf_socklist *psl;
2546
2547         ASSERT_RTNL();
2548
2549         psin = (struct sockaddr_in *)&gsf->gf_group;
2550         if (psin->sin_family != AF_INET)
2551                 return -EINVAL;
2552         addr = psin->sin_addr.s_addr;
2553         if (!ipv4_is_multicast(addr))
2554                 return -EINVAL;
2555
2556         err = -EADDRNOTAVAIL;
2557
2558         for_each_pmc_rtnl(inet, pmc) {
2559                 if (pmc->multi.imr_multiaddr.s_addr == addr &&
2560                     pmc->multi.imr_ifindex == gsf->gf_interface)
2561                         break;
2562         }
2563         if (!pmc)               /* must have a prior join */
2564                 goto done;
2565         gsf->gf_fmode = pmc->sfmode;
2566         psl = rtnl_dereference(pmc->sflist);
2567         count = psl ? psl->sl_count : 0;
2568         copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2569         gsf->gf_numsrc = count;
2570         if (put_user(GROUP_FILTER_SIZE(copycount), optlen) ||
2571             copy_to_user(optval, gsf, GROUP_FILTER_SIZE(0))) {
2572                 return -EFAULT;
2573         }
2574         for (i = 0; i < copycount; i++) {
2575                 struct sockaddr_storage ss;
2576
2577                 psin = (struct sockaddr_in *)&ss;
2578                 memset(&ss, 0, sizeof(ss));
2579                 psin->sin_family = AF_INET;
2580                 psin->sin_addr.s_addr = psl->sl_addr[i];
2581                 if (copy_to_user(&optval->gf_slist[i], &ss, sizeof(ss)))
2582                         return -EFAULT;
2583         }
2584         return 0;
2585 done:
2586         return err;
2587 }
2588
2589 /*
2590  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2591  */
2592 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr, int dif)
2593 {
2594         struct inet_sock *inet = inet_sk(sk);
2595         struct ip_mc_socklist *pmc;
2596         struct ip_sf_socklist *psl;
2597         int i;
2598         int ret;
2599
2600         ret = 1;
2601         if (!ipv4_is_multicast(loc_addr))
2602                 goto out;
2603
2604         rcu_read_lock();
2605         for_each_pmc_rcu(inet, pmc) {
2606                 if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2607                     pmc->multi.imr_ifindex == dif)
2608                         break;
2609         }
2610         ret = inet->mc_all;
2611         if (!pmc)
2612                 goto unlock;
2613         psl = rcu_dereference(pmc->sflist);
2614         ret = (pmc->sfmode == MCAST_EXCLUDE);
2615         if (!psl)
2616                 goto unlock;
2617
2618         for (i = 0; i < psl->sl_count; i++) {
2619                 if (psl->sl_addr[i] == rmt_addr)
2620                         break;
2621         }
2622         ret = 0;
2623         if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2624                 goto unlock;
2625         if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2626                 goto unlock;
2627         ret = 1;
2628 unlock:
2629         rcu_read_unlock();
2630 out:
2631         return ret;
2632 }
2633
2634 /*
2635  *      A socket is closing.
2636  */
2637
2638 void ip_mc_drop_socket(struct sock *sk)
2639 {
2640         struct inet_sock *inet = inet_sk(sk);
2641         struct ip_mc_socklist *iml;
2642         struct net *net = sock_net(sk);
2643
2644         if (!inet->mc_list)
2645                 return;
2646
2647         rtnl_lock();
2648         while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2649                 struct in_device *in_dev;
2650
2651                 inet->mc_list = iml->next_rcu;
2652                 in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2653                 (void) ip_mc_leave_src(sk, iml, in_dev);
2654                 if (in_dev)
2655                         ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2656                 /* decrease mem now to avoid the memleak warning */
2657                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2658                 kfree_rcu(iml, rcu);
2659         }
2660         rtnl_unlock();
2661 }
2662
2663 /* called with rcu_read_lock() */
2664 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2665 {
2666         struct ip_mc_list *im;
2667         struct ip_mc_list __rcu **mc_hash;
2668         struct ip_sf_list *psf;
2669         int rv = 0;
2670
2671         mc_hash = rcu_dereference(in_dev->mc_hash);
2672         if (mc_hash) {
2673                 u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2674
2675                 for (im = rcu_dereference(mc_hash[hash]);
2676                      im != NULL;
2677                      im = rcu_dereference(im->next_hash)) {
2678                         if (im->multiaddr == mc_addr)
2679                                 break;
2680                 }
2681         } else {
2682                 for_each_pmc_rcu(in_dev, im) {
2683                         if (im->multiaddr == mc_addr)
2684                                 break;
2685                 }
2686         }
2687         if (im && proto == IPPROTO_IGMP) {
2688                 rv = 1;
2689         } else if (im) {
2690                 if (src_addr) {
2691                         spin_lock_bh(&im->lock);
2692                         for (psf = im->sources; psf; psf = psf->sf_next) {
2693                                 if (psf->sf_inaddr == src_addr)
2694                                         break;
2695                         }
2696                         if (psf)
2697                                 rv = psf->sf_count[MCAST_INCLUDE] ||
2698                                         psf->sf_count[MCAST_EXCLUDE] !=
2699                                         im->sfcount[MCAST_EXCLUDE];
2700                         else
2701                                 rv = im->sfcount[MCAST_EXCLUDE] != 0;
2702                         spin_unlock_bh(&im->lock);
2703                 } else
2704                         rv = 1; /* unspecified source; tentatively allow */
2705         }
2706         return rv;
2707 }
2708
2709 #if defined(CONFIG_PROC_FS)
2710 struct igmp_mc_iter_state {
2711         struct seq_net_private p;
2712         struct net_device *dev;
2713         struct in_device *in_dev;
2714 };
2715
2716 #define igmp_mc_seq_private(seq)        ((struct igmp_mc_iter_state *)(seq)->private)
2717
2718 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2719 {
2720         struct net *net = seq_file_net(seq);
2721         struct ip_mc_list *im = NULL;
2722         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2723
2724         state->in_dev = NULL;
2725         for_each_netdev_rcu(net, state->dev) {
2726                 struct in_device *in_dev;
2727
2728                 in_dev = __in_dev_get_rcu(state->dev);
2729                 if (!in_dev)
2730                         continue;
2731                 im = rcu_dereference(in_dev->mc_list);
2732                 if (im) {
2733                         state->in_dev = in_dev;
2734                         break;
2735                 }
2736         }
2737         return im;
2738 }
2739
2740 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2741 {
2742         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2743
2744         im = rcu_dereference(im->next_rcu);
2745         while (!im) {
2746                 state->dev = next_net_device_rcu(state->dev);
2747                 if (!state->dev) {
2748                         state->in_dev = NULL;
2749                         break;
2750                 }
2751                 state->in_dev = __in_dev_get_rcu(state->dev);
2752                 if (!state->in_dev)
2753                         continue;
2754                 im = rcu_dereference(state->in_dev->mc_list);
2755         }
2756         return im;
2757 }
2758
2759 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2760 {
2761         struct ip_mc_list *im = igmp_mc_get_first(seq);
2762         if (im)
2763                 while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2764                         --pos;
2765         return pos ? NULL : im;
2766 }
2767
2768 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2769         __acquires(rcu)
2770 {
2771         rcu_read_lock();
2772         return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2773 }
2774
2775 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2776 {
2777         struct ip_mc_list *im;
2778         if (v == SEQ_START_TOKEN)
2779                 im = igmp_mc_get_first(seq);
2780         else
2781                 im = igmp_mc_get_next(seq, v);
2782         ++*pos;
2783         return im;
2784 }
2785
2786 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2787         __releases(rcu)
2788 {
2789         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2790
2791         state->in_dev = NULL;
2792         state->dev = NULL;
2793         rcu_read_unlock();
2794 }
2795
2796 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2797 {
2798         if (v == SEQ_START_TOKEN)
2799                 seq_puts(seq,
2800                          "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2801         else {
2802                 struct ip_mc_list *im = (struct ip_mc_list *)v;
2803                 struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2804                 char   *querier;
2805                 long delta;
2806
2807 #ifdef CONFIG_IP_MULTICAST
2808                 querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2809                           IGMP_V2_SEEN(state->in_dev) ? "V2" :
2810                           "V3";
2811 #else
2812                 querier = "NONE";
2813 #endif
2814
2815                 if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2816                         seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2817                                    state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2818                 }
2819
2820                 delta = im->timer.expires - jiffies;
2821                 seq_printf(seq,
2822                            "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2823                            im->multiaddr, im->users,
2824                            im->tm_running,
2825                            im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2826                            im->reporter);
2827         }
2828         return 0;
2829 }
2830
2831 static const struct seq_operations igmp_mc_seq_ops = {
2832         .start  =       igmp_mc_seq_start,
2833         .next   =       igmp_mc_seq_next,
2834         .stop   =       igmp_mc_seq_stop,
2835         .show   =       igmp_mc_seq_show,
2836 };
2837
2838 static int igmp_mc_seq_open(struct inode *inode, struct file *file)
2839 {
2840         return seq_open_net(inode, file, &igmp_mc_seq_ops,
2841                         sizeof(struct igmp_mc_iter_state));
2842 }
2843
2844 static const struct file_operations igmp_mc_seq_fops = {
2845         .owner          =       THIS_MODULE,
2846         .open           =       igmp_mc_seq_open,
2847         .read           =       seq_read,
2848         .llseek         =       seq_lseek,
2849         .release        =       seq_release_net,
2850 };
2851
2852 struct igmp_mcf_iter_state {
2853         struct seq_net_private p;
2854         struct net_device *dev;
2855         struct in_device *idev;
2856         struct ip_mc_list *im;
2857 };
2858
2859 #define igmp_mcf_seq_private(seq)       ((struct igmp_mcf_iter_state *)(seq)->private)
2860
2861 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2862 {
2863         struct net *net = seq_file_net(seq);
2864         struct ip_sf_list *psf = NULL;
2865         struct ip_mc_list *im = NULL;
2866         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2867
2868         state->idev = NULL;
2869         state->im = NULL;
2870         for_each_netdev_rcu(net, state->dev) {
2871                 struct in_device *idev;
2872                 idev = __in_dev_get_rcu(state->dev);
2873                 if (unlikely(!idev))
2874                         continue;
2875                 im = rcu_dereference(idev->mc_list);
2876                 if (likely(im)) {
2877                         spin_lock_bh(&im->lock);
2878                         psf = im->sources;
2879                         if (likely(psf)) {
2880                                 state->im = im;
2881                                 state->idev = idev;
2882                                 break;
2883                         }
2884                         spin_unlock_bh(&im->lock);
2885                 }
2886         }
2887         return psf;
2888 }
2889
2890 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2891 {
2892         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2893
2894         psf = psf->sf_next;
2895         while (!psf) {
2896                 spin_unlock_bh(&state->im->lock);
2897                 state->im = state->im->next;
2898                 while (!state->im) {
2899                         state->dev = next_net_device_rcu(state->dev);
2900                         if (!state->dev) {
2901                                 state->idev = NULL;
2902                                 goto out;
2903                         }
2904                         state->idev = __in_dev_get_rcu(state->dev);
2905                         if (!state->idev)
2906                                 continue;
2907                         state->im = rcu_dereference(state->idev->mc_list);
2908                 }
2909                 if (!state->im)
2910                         break;
2911                 spin_lock_bh(&state->im->lock);
2912                 psf = state->im->sources;
2913         }
2914 out:
2915         return psf;
2916 }
2917
2918 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2919 {
2920         struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2921         if (psf)
2922                 while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2923                         --pos;
2924         return pos ? NULL : psf;
2925 }
2926
2927 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2928         __acquires(rcu)
2929 {
2930         rcu_read_lock();
2931         return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2932 }
2933
2934 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2935 {
2936         struct ip_sf_list *psf;
2937         if (v == SEQ_START_TOKEN)
2938                 psf = igmp_mcf_get_first(seq);
2939         else
2940                 psf = igmp_mcf_get_next(seq, v);
2941         ++*pos;
2942         return psf;
2943 }
2944
2945 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2946         __releases(rcu)
2947 {
2948         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2949         if (likely(state->im)) {
2950                 spin_unlock_bh(&state->im->lock);
2951                 state->im = NULL;
2952         }
2953         state->idev = NULL;
2954         state->dev = NULL;
2955         rcu_read_unlock();
2956 }
2957
2958 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2959 {
2960         struct ip_sf_list *psf = (struct ip_sf_list *)v;
2961         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2962
2963         if (v == SEQ_START_TOKEN) {
2964                 seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
2965         } else {
2966                 seq_printf(seq,
2967                            "%3d %6.6s 0x%08x "
2968                            "0x%08x %6lu %6lu\n",
2969                            state->dev->ifindex, state->dev->name,
2970                            ntohl(state->im->multiaddr),
2971                            ntohl(psf->sf_inaddr),
2972                            psf->sf_count[MCAST_INCLUDE],
2973                            psf->sf_count[MCAST_EXCLUDE]);
2974         }
2975         return 0;
2976 }
2977
2978 static const struct seq_operations igmp_mcf_seq_ops = {
2979         .start  =       igmp_mcf_seq_start,
2980         .next   =       igmp_mcf_seq_next,
2981         .stop   =       igmp_mcf_seq_stop,
2982         .show   =       igmp_mcf_seq_show,
2983 };
2984
2985 static int igmp_mcf_seq_open(struct inode *inode, struct file *file)
2986 {
2987         return seq_open_net(inode, file, &igmp_mcf_seq_ops,
2988                         sizeof(struct igmp_mcf_iter_state));
2989 }
2990
2991 static const struct file_operations igmp_mcf_seq_fops = {
2992         .owner          =       THIS_MODULE,
2993         .open           =       igmp_mcf_seq_open,
2994         .read           =       seq_read,
2995         .llseek         =       seq_lseek,
2996         .release        =       seq_release_net,
2997 };
2998
2999 static int __net_init igmp_net_init(struct net *net)
3000 {
3001         struct proc_dir_entry *pde;
3002         int err;
3003
3004         pde = proc_create("igmp", S_IRUGO, net->proc_net, &igmp_mc_seq_fops);
3005         if (!pde)
3006                 goto out_igmp;
3007         pde = proc_create("mcfilter", S_IRUGO, net->proc_net,
3008                           &igmp_mcf_seq_fops);
3009         if (!pde)
3010                 goto out_mcfilter;
3011         err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
3012                                    SOCK_DGRAM, 0, net);
3013         if (err < 0) {
3014                 pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
3015                        err);
3016                 goto out_sock;
3017         }
3018
3019         return 0;
3020
3021 out_sock:
3022         remove_proc_entry("mcfilter", net->proc_net);
3023 out_mcfilter:
3024         remove_proc_entry("igmp", net->proc_net);
3025 out_igmp:
3026         return -ENOMEM;
3027 }
3028
3029 static void __net_exit igmp_net_exit(struct net *net)
3030 {
3031         remove_proc_entry("mcfilter", net->proc_net);
3032         remove_proc_entry("igmp", net->proc_net);
3033         inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
3034 }
3035
3036 static struct pernet_operations igmp_net_ops = {
3037         .init = igmp_net_init,
3038         .exit = igmp_net_exit,
3039 };
3040 #endif
3041
3042 static int igmp_netdev_event(struct notifier_block *this,
3043                              unsigned long event, void *ptr)
3044 {
3045         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3046         struct in_device *in_dev;
3047
3048         switch (event) {
3049         case NETDEV_RESEND_IGMP:
3050                 in_dev = __in_dev_get_rtnl(dev);
3051                 if (in_dev)
3052                         ip_mc_rejoin_groups(in_dev);
3053                 break;
3054         default:
3055                 break;
3056         }
3057         return NOTIFY_DONE;
3058 }
3059
3060 static struct notifier_block igmp_notifier = {
3061         .notifier_call = igmp_netdev_event,
3062 };
3063
3064 int __init igmp_mc_init(void)
3065 {
3066 #if defined(CONFIG_PROC_FS)
3067         int err;
3068
3069         err = register_pernet_subsys(&igmp_net_ops);
3070         if (err)
3071                 return err;
3072         err = register_netdevice_notifier(&igmp_notifier);
3073         if (err)
3074                 goto reg_notif_fail;
3075         return 0;
3076
3077 reg_notif_fail:
3078         unregister_pernet_subsys(&igmp_net_ops);
3079         return err;
3080 #else
3081         return register_netdevice_notifier(&igmp_notifier);
3082 #endif
3083 }