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                         im->tomb = pmc->tomb;
1216                         pmc->tomb = NULL;
1217
1218                         im->sources = pmc->sources;
1219                         pmc->sources = NULL;
1220
1221                         for (psf = im->sources; psf; psf = psf->sf_next)
1222                                 psf->sf_crcount = im->crcount;
1223                 }
1224                 in_dev_put(pmc->interface);
1225                 kfree_pmc(pmc);
1226         }
1227         spin_unlock_bh(&im->lock);
1228 }
1229
1230 /*
1231  * flush ip_mc_list deleted records
1232  */
1233 static void igmpv3_clear_delrec(struct in_device *in_dev)
1234 {
1235         struct ip_mc_list *pmc, *nextpmc;
1236
1237         spin_lock_bh(&in_dev->mc_tomb_lock);
1238         pmc = in_dev->mc_tomb;
1239         in_dev->mc_tomb = NULL;
1240         spin_unlock_bh(&in_dev->mc_tomb_lock);
1241
1242         for (; pmc; pmc = nextpmc) {
1243                 nextpmc = pmc->next;
1244                 ip_mc_clear_src(pmc);
1245                 in_dev_put(pmc->interface);
1246                 kfree_pmc(pmc);
1247         }
1248         /* clear dead sources, too */
1249         rcu_read_lock();
1250         for_each_pmc_rcu(in_dev, pmc) {
1251                 struct ip_sf_list *psf;
1252
1253                 spin_lock_bh(&pmc->lock);
1254                 psf = pmc->tomb;
1255                 pmc->tomb = NULL;
1256                 spin_unlock_bh(&pmc->lock);
1257                 ip_sf_list_clear_all(psf);
1258         }
1259         rcu_read_unlock();
1260 }
1261 #endif
1262
1263 static void igmp_group_dropped(struct ip_mc_list *im)
1264 {
1265         struct in_device *in_dev = im->interface;
1266 #ifdef CONFIG_IP_MULTICAST
1267         struct net *net = dev_net(in_dev->dev);
1268         int reporter;
1269 #endif
1270
1271         if (im->loaded) {
1272                 im->loaded = 0;
1273                 ip_mc_filter_del(in_dev, im->multiaddr);
1274         }
1275
1276 #ifdef CONFIG_IP_MULTICAST
1277         if (im->multiaddr == IGMP_ALL_HOSTS)
1278                 return;
1279         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1280                 return;
1281
1282         reporter = im->reporter;
1283         igmp_stop_timer(im);
1284
1285         if (!in_dev->dead) {
1286                 if (IGMP_V1_SEEN(in_dev))
1287                         return;
1288                 if (IGMP_V2_SEEN(in_dev)) {
1289                         if (reporter)
1290                                 igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1291                         return;
1292                 }
1293                 /* IGMPv3 */
1294                 igmpv3_add_delrec(in_dev, im);
1295
1296                 igmp_ifc_event(in_dev);
1297         }
1298 #endif
1299 }
1300
1301 static void igmp_group_added(struct ip_mc_list *im)
1302 {
1303         struct in_device *in_dev = im->interface;
1304 #ifdef CONFIG_IP_MULTICAST
1305         struct net *net = dev_net(in_dev->dev);
1306 #endif
1307
1308         if (im->loaded == 0) {
1309                 im->loaded = 1;
1310                 ip_mc_filter_add(in_dev, im->multiaddr);
1311         }
1312
1313 #ifdef CONFIG_IP_MULTICAST
1314         if (im->multiaddr == IGMP_ALL_HOSTS)
1315                 return;
1316         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1317                 return;
1318
1319         if (in_dev->dead)
1320                 return;
1321         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1322                 spin_lock_bh(&im->lock);
1323                 igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1324                 spin_unlock_bh(&im->lock);
1325                 return;
1326         }
1327         /* else, v3 */
1328
1329         im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1330         igmp_ifc_event(in_dev);
1331 #endif
1332 }
1333
1334
1335 /*
1336  *      Multicast list managers
1337  */
1338
1339 static u32 ip_mc_hash(const struct ip_mc_list *im)
1340 {
1341         return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1342 }
1343
1344 static void ip_mc_hash_add(struct in_device *in_dev,
1345                            struct ip_mc_list *im)
1346 {
1347         struct ip_mc_list __rcu **mc_hash;
1348         u32 hash;
1349
1350         mc_hash = rtnl_dereference(in_dev->mc_hash);
1351         if (mc_hash) {
1352                 hash = ip_mc_hash(im);
1353                 im->next_hash = mc_hash[hash];
1354                 rcu_assign_pointer(mc_hash[hash], im);
1355                 return;
1356         }
1357
1358         /* do not use a hash table for small number of items */
1359         if (in_dev->mc_count < 4)
1360                 return;
1361
1362         mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1363                           GFP_KERNEL);
1364         if (!mc_hash)
1365                 return;
1366
1367         for_each_pmc_rtnl(in_dev, im) {
1368                 hash = ip_mc_hash(im);
1369                 im->next_hash = mc_hash[hash];
1370                 RCU_INIT_POINTER(mc_hash[hash], im);
1371         }
1372
1373         rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1374 }
1375
1376 static void ip_mc_hash_remove(struct in_device *in_dev,
1377                               struct ip_mc_list *im)
1378 {
1379         struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1380         struct ip_mc_list *aux;
1381
1382         if (!mc_hash)
1383                 return;
1384         mc_hash += ip_mc_hash(im);
1385         while ((aux = rtnl_dereference(*mc_hash)) != im)
1386                 mc_hash = &aux->next_hash;
1387         *mc_hash = im->next_hash;
1388 }
1389
1390
1391 /*
1392  *      A socket has joined a multicast group on device dev.
1393  */
1394
1395 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1396 {
1397         struct ip_mc_list *im;
1398 #ifdef CONFIG_IP_MULTICAST
1399         struct net *net = dev_net(in_dev->dev);
1400 #endif
1401
1402         ASSERT_RTNL();
1403
1404         for_each_pmc_rtnl(in_dev, im) {
1405                 if (im->multiaddr == addr) {
1406                         im->users++;
1407                         ip_mc_add_src(in_dev, &addr, MCAST_EXCLUDE, 0, NULL, 0);
1408                         goto out;
1409                 }
1410         }
1411
1412         im = kzalloc(sizeof(*im), GFP_KERNEL);
1413         if (!im)
1414                 goto out;
1415
1416         im->users = 1;
1417         im->interface = in_dev;
1418         in_dev_hold(in_dev);
1419         im->multiaddr = addr;
1420         /* initial mode is (EX, empty) */
1421         im->sfmode = MCAST_EXCLUDE;
1422         im->sfcount[MCAST_EXCLUDE] = 1;
1423         atomic_set(&im->refcnt, 1);
1424         spin_lock_init(&im->lock);
1425 #ifdef CONFIG_IP_MULTICAST
1426         setup_timer(&im->timer, igmp_timer_expire, (unsigned long)im);
1427         im->unsolicit_count = net->ipv4.sysctl_igmp_qrv;
1428 #endif
1429
1430         im->next_rcu = in_dev->mc_list;
1431         in_dev->mc_count++;
1432         rcu_assign_pointer(in_dev->mc_list, im);
1433
1434         ip_mc_hash_add(in_dev, im);
1435
1436 #ifdef CONFIG_IP_MULTICAST
1437         igmpv3_del_delrec(in_dev, im);
1438 #endif
1439         igmp_group_added(im);
1440         if (!in_dev->dead)
1441                 ip_rt_multicast_event(in_dev);
1442 out:
1443         return;
1444 }
1445 EXPORT_SYMBOL(ip_mc_inc_group);
1446
1447 static int ip_mc_check_iphdr(struct sk_buff *skb)
1448 {
1449         const struct iphdr *iph;
1450         unsigned int len;
1451         unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1452
1453         if (!pskb_may_pull(skb, offset))
1454                 return -EINVAL;
1455
1456         iph = ip_hdr(skb);
1457
1458         if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1459                 return -EINVAL;
1460
1461         offset += ip_hdrlen(skb) - sizeof(*iph);
1462
1463         if (!pskb_may_pull(skb, offset))
1464                 return -EINVAL;
1465
1466         iph = ip_hdr(skb);
1467
1468         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1469                 return -EINVAL;
1470
1471         len = skb_network_offset(skb) + ntohs(iph->tot_len);
1472         if (skb->len < len || len < offset)
1473                 return -EINVAL;
1474
1475         skb_set_transport_header(skb, offset);
1476
1477         return 0;
1478 }
1479
1480 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1481 {
1482         unsigned int len = skb_transport_offset(skb);
1483
1484         len += sizeof(struct igmpv3_report);
1485
1486         return pskb_may_pull(skb, len) ? 0 : -EINVAL;
1487 }
1488
1489 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1490 {
1491         unsigned int len = skb_transport_offset(skb);
1492
1493         len += sizeof(struct igmphdr);
1494         if (skb->len < len)
1495                 return -EINVAL;
1496
1497         /* IGMPv{1,2}? */
1498         if (skb->len != len) {
1499                 /* or IGMPv3? */
1500                 len += sizeof(struct igmpv3_query) - sizeof(struct igmphdr);
1501                 if (skb->len < len || !pskb_may_pull(skb, len))
1502                         return -EINVAL;
1503         }
1504
1505         /* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1506          * all-systems destination addresses (224.0.0.1) for general queries
1507          */
1508         if (!igmp_hdr(skb)->group &&
1509             ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1510                 return -EINVAL;
1511
1512         return 0;
1513 }
1514
1515 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1516 {
1517         switch (igmp_hdr(skb)->type) {
1518         case IGMP_HOST_LEAVE_MESSAGE:
1519         case IGMP_HOST_MEMBERSHIP_REPORT:
1520         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1521                 /* fall through */
1522                 return 0;
1523         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1524                 return ip_mc_check_igmp_reportv3(skb);
1525         case IGMP_HOST_MEMBERSHIP_QUERY:
1526                 return ip_mc_check_igmp_query(skb);
1527         default:
1528                 return -ENOMSG;
1529         }
1530 }
1531
1532 static inline __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1533 {
1534         return skb_checksum_simple_validate(skb);
1535 }
1536
1537 static int __ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1538
1539 {
1540         struct sk_buff *skb_chk;
1541         unsigned int transport_len;
1542         unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1543         int ret = -EINVAL;
1544
1545         transport_len = ntohs(ip_hdr(skb)->tot_len) - ip_hdrlen(skb);
1546
1547         skb_chk = skb_checksum_trimmed(skb, transport_len,
1548                                        ip_mc_validate_checksum);
1549         if (!skb_chk)
1550                 goto err;
1551
1552         if (!pskb_may_pull(skb_chk, len))
1553                 goto err;
1554
1555         ret = ip_mc_check_igmp_msg(skb_chk);
1556         if (ret)
1557                 goto err;
1558
1559         if (skb_trimmed)
1560                 *skb_trimmed = skb_chk;
1561         /* free now unneeded clone */
1562         else if (skb_chk != skb)
1563                 kfree_skb(skb_chk);
1564
1565         ret = 0;
1566
1567 err:
1568         if (ret && skb_chk && skb_chk != skb)
1569                 kfree_skb(skb_chk);
1570
1571         return ret;
1572 }
1573
1574 /**
1575  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1576  * @skb: the skb to validate
1577  * @skb_trimmed: to store an skb pointer trimmed to IPv4 packet tail (optional)
1578  *
1579  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1580  * skb transport header accordingly and returns zero.
1581  *
1582  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1583  *  standard
1584  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1585  * -ENOMEM: A memory allocation failure happened.
1586  *
1587  * Optionally, an skb pointer might be provided via skb_trimmed (or set it
1588  * to NULL): After parsing an IGMP packet successfully it will point to
1589  * an skb which has its tail aligned to the IP packet end. This might
1590  * either be the originally provided skb or a trimmed, cloned version if
1591  * the skb frame had data beyond the IP packet. A cloned skb allows us
1592  * to leave the original skb and its full frame unchanged (which might be
1593  * desirable for layer 2 frame jugglers).
1594  *
1595  * Caller needs to set the skb network header and free any returned skb if it
1596  * differs from the provided skb.
1597  */
1598 int ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1599 {
1600         int ret = ip_mc_check_iphdr(skb);
1601
1602         if (ret < 0)
1603                 return ret;
1604
1605         if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1606                 return -ENOMSG;
1607
1608         return __ip_mc_check_igmp(skb, skb_trimmed);
1609 }
1610 EXPORT_SYMBOL(ip_mc_check_igmp);
1611
1612 /*
1613  *      Resend IGMP JOIN report; used by netdev notifier.
1614  */
1615 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1616 {
1617 #ifdef CONFIG_IP_MULTICAST
1618         struct ip_mc_list *im;
1619         int type;
1620         struct net *net = dev_net(in_dev->dev);
1621
1622         ASSERT_RTNL();
1623
1624         for_each_pmc_rtnl(in_dev, im) {
1625                 if (im->multiaddr == IGMP_ALL_HOSTS)
1626                         continue;
1627                 if (ipv4_is_local_multicast(im->multiaddr) &&
1628                     !net->ipv4.sysctl_igmp_llm_reports)
1629                         continue;
1630
1631                 /* a failover is happening and switches
1632                  * must be notified immediately
1633                  */
1634                 if (IGMP_V1_SEEN(in_dev))
1635                         type = IGMP_HOST_MEMBERSHIP_REPORT;
1636                 else if (IGMP_V2_SEEN(in_dev))
1637                         type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1638                 else
1639                         type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1640                 igmp_send_report(in_dev, im, type);
1641         }
1642 #endif
1643 }
1644
1645 /*
1646  *      A socket has left a multicast group on device dev
1647  */
1648
1649 void ip_mc_dec_group(struct in_device *in_dev, __be32 addr)
1650 {
1651         struct ip_mc_list *i;
1652         struct ip_mc_list __rcu **ip;
1653
1654         ASSERT_RTNL();
1655
1656         for (ip = &in_dev->mc_list;
1657              (i = rtnl_dereference(*ip)) != NULL;
1658              ip = &i->next_rcu) {
1659                 if (i->multiaddr == addr) {
1660                         if (--i->users == 0) {
1661                                 ip_mc_hash_remove(in_dev, i);
1662                                 *ip = i->next_rcu;
1663                                 in_dev->mc_count--;
1664                                 igmp_group_dropped(i);
1665                                 ip_mc_clear_src(i);
1666
1667                                 if (!in_dev->dead)
1668                                         ip_rt_multicast_event(in_dev);
1669
1670                                 ip_ma_put(i);
1671                                 return;
1672                         }
1673                         break;
1674                 }
1675         }
1676 }
1677 EXPORT_SYMBOL(ip_mc_dec_group);
1678
1679 /* Device changing type */
1680
1681 void ip_mc_unmap(struct in_device *in_dev)
1682 {
1683         struct ip_mc_list *pmc;
1684
1685         ASSERT_RTNL();
1686
1687         for_each_pmc_rtnl(in_dev, pmc)
1688                 igmp_group_dropped(pmc);
1689 }
1690
1691 void ip_mc_remap(struct in_device *in_dev)
1692 {
1693         struct ip_mc_list *pmc;
1694
1695         ASSERT_RTNL();
1696
1697         for_each_pmc_rtnl(in_dev, pmc) {
1698 #ifdef CONFIG_IP_MULTICAST
1699                 igmpv3_del_delrec(in_dev, pmc);
1700 #endif
1701                 igmp_group_added(pmc);
1702         }
1703 }
1704
1705 /* Device going down */
1706
1707 void ip_mc_down(struct in_device *in_dev)
1708 {
1709         struct ip_mc_list *pmc;
1710
1711         ASSERT_RTNL();
1712
1713         for_each_pmc_rtnl(in_dev, pmc)
1714                 igmp_group_dropped(pmc);
1715
1716 #ifdef CONFIG_IP_MULTICAST
1717         in_dev->mr_ifc_count = 0;
1718         if (del_timer(&in_dev->mr_ifc_timer))
1719                 __in_dev_put(in_dev);
1720         in_dev->mr_gq_running = 0;
1721         if (del_timer(&in_dev->mr_gq_timer))
1722                 __in_dev_put(in_dev);
1723 #endif
1724
1725         ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1726 }
1727
1728 void ip_mc_init_dev(struct in_device *in_dev)
1729 {
1730 #ifdef CONFIG_IP_MULTICAST
1731         struct net *net = dev_net(in_dev->dev);
1732 #endif
1733         ASSERT_RTNL();
1734
1735 #ifdef CONFIG_IP_MULTICAST
1736         setup_timer(&in_dev->mr_gq_timer, igmp_gq_timer_expire,
1737                         (unsigned long)in_dev);
1738         setup_timer(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire,
1739                         (unsigned long)in_dev);
1740         in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1741 #endif
1742
1743         spin_lock_init(&in_dev->mc_tomb_lock);
1744 }
1745
1746 /* Device going up */
1747
1748 void ip_mc_up(struct in_device *in_dev)
1749 {
1750         struct ip_mc_list *pmc;
1751 #ifdef CONFIG_IP_MULTICAST
1752         struct net *net = dev_net(in_dev->dev);
1753 #endif
1754
1755         ASSERT_RTNL();
1756
1757 #ifdef CONFIG_IP_MULTICAST
1758         in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1759 #endif
1760         ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1761
1762         for_each_pmc_rtnl(in_dev, pmc) {
1763 #ifdef CONFIG_IP_MULTICAST
1764                 igmpv3_del_delrec(in_dev, pmc);
1765 #endif
1766                 igmp_group_added(pmc);
1767         }
1768 }
1769
1770 /*
1771  *      Device is about to be destroyed: clean up.
1772  */
1773
1774 void ip_mc_destroy_dev(struct in_device *in_dev)
1775 {
1776         struct ip_mc_list *i;
1777
1778         ASSERT_RTNL();
1779
1780         /* Deactivate timers */
1781         ip_mc_down(in_dev);
1782 #ifdef CONFIG_IP_MULTICAST
1783         igmpv3_clear_delrec(in_dev);
1784 #endif
1785
1786         while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1787                 in_dev->mc_list = i->next_rcu;
1788                 in_dev->mc_count--;
1789                 ip_ma_put(i);
1790         }
1791 }
1792
1793 /* RTNL is locked */
1794 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1795 {
1796         struct net_device *dev = NULL;
1797         struct in_device *idev = NULL;
1798
1799         if (imr->imr_ifindex) {
1800                 idev = inetdev_by_index(net, imr->imr_ifindex);
1801                 return idev;
1802         }
1803         if (imr->imr_address.s_addr) {
1804                 dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1805                 if (!dev)
1806                         return NULL;
1807         }
1808
1809         if (!dev) {
1810                 struct rtable *rt = ip_route_output(net,
1811                                                     imr->imr_multiaddr.s_addr,
1812                                                     0, 0, 0);
1813                 if (!IS_ERR(rt)) {
1814                         dev = rt->dst.dev;
1815                         ip_rt_put(rt);
1816                 }
1817         }
1818         if (dev) {
1819                 imr->imr_ifindex = dev->ifindex;
1820                 idev = __in_dev_get_rtnl(dev);
1821         }
1822         return idev;
1823 }
1824
1825 /*
1826  *      Join a socket to a group
1827  */
1828
1829 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1830         __be32 *psfsrc)
1831 {
1832         struct ip_sf_list *psf, *psf_prev;
1833         int rv = 0;
1834
1835         psf_prev = NULL;
1836         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1837                 if (psf->sf_inaddr == *psfsrc)
1838                         break;
1839                 psf_prev = psf;
1840         }
1841         if (!psf || psf->sf_count[sfmode] == 0) {
1842                 /* source filter not found, or count wrong =>  bug */
1843                 return -ESRCH;
1844         }
1845         psf->sf_count[sfmode]--;
1846         if (psf->sf_count[sfmode] == 0) {
1847                 ip_rt_multicast_event(pmc->interface);
1848         }
1849         if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1850 #ifdef CONFIG_IP_MULTICAST
1851                 struct in_device *in_dev = pmc->interface;
1852                 struct net *net = dev_net(in_dev->dev);
1853 #endif
1854
1855                 /* no more filters for this source */
1856                 if (psf_prev)
1857                         psf_prev->sf_next = psf->sf_next;
1858                 else
1859                         pmc->sources = psf->sf_next;
1860 #ifdef CONFIG_IP_MULTICAST
1861                 if (psf->sf_oldin &&
1862                     !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1863                         psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1864                         psf->sf_next = pmc->tomb;
1865                         pmc->tomb = psf;
1866                         rv = 1;
1867                 } else
1868 #endif
1869                         kfree(psf);
1870         }
1871         return rv;
1872 }
1873
1874 #ifndef CONFIG_IP_MULTICAST
1875 #define igmp_ifc_event(x)       do { } while (0)
1876 #endif
1877
1878 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1879                          int sfcount, __be32 *psfsrc, int delta)
1880 {
1881         struct ip_mc_list *pmc;
1882         int     changerec = 0;
1883         int     i, err;
1884
1885         if (!in_dev)
1886                 return -ENODEV;
1887         rcu_read_lock();
1888         for_each_pmc_rcu(in_dev, pmc) {
1889                 if (*pmca == pmc->multiaddr)
1890                         break;
1891         }
1892         if (!pmc) {
1893                 /* MCA not found?? bug */
1894                 rcu_read_unlock();
1895                 return -ESRCH;
1896         }
1897         spin_lock_bh(&pmc->lock);
1898         rcu_read_unlock();
1899 #ifdef CONFIG_IP_MULTICAST
1900         sf_markstate(pmc);
1901 #endif
1902         if (!delta) {
1903                 err = -EINVAL;
1904                 if (!pmc->sfcount[sfmode])
1905                         goto out_unlock;
1906                 pmc->sfcount[sfmode]--;
1907         }
1908         err = 0;
1909         for (i = 0; i < sfcount; i++) {
1910                 int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1911
1912                 changerec |= rv > 0;
1913                 if (!err && rv < 0)
1914                         err = rv;
1915         }
1916         if (pmc->sfmode == MCAST_EXCLUDE &&
1917             pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1918             pmc->sfcount[MCAST_INCLUDE]) {
1919 #ifdef CONFIG_IP_MULTICAST
1920                 struct ip_sf_list *psf;
1921                 struct net *net = dev_net(in_dev->dev);
1922 #endif
1923
1924                 /* filter mode change */
1925                 pmc->sfmode = MCAST_INCLUDE;
1926 #ifdef CONFIG_IP_MULTICAST
1927                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1928                 in_dev->mr_ifc_count = pmc->crcount;
1929                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1930                         psf->sf_crcount = 0;
1931                 igmp_ifc_event(pmc->interface);
1932         } else if (sf_setstate(pmc) || changerec) {
1933                 igmp_ifc_event(pmc->interface);
1934 #endif
1935         }
1936 out_unlock:
1937         spin_unlock_bh(&pmc->lock);
1938         return err;
1939 }
1940
1941 /*
1942  * Add multicast single-source filter to the interface list
1943  */
1944 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1945         __be32 *psfsrc)
1946 {
1947         struct ip_sf_list *psf, *psf_prev;
1948
1949         psf_prev = NULL;
1950         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1951                 if (psf->sf_inaddr == *psfsrc)
1952                         break;
1953                 psf_prev = psf;
1954         }
1955         if (!psf) {
1956                 psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1957                 if (!psf)
1958                         return -ENOBUFS;
1959                 psf->sf_inaddr = *psfsrc;
1960                 if (psf_prev) {
1961                         psf_prev->sf_next = psf;
1962                 } else
1963                         pmc->sources = psf;
1964         }
1965         psf->sf_count[sfmode]++;
1966         if (psf->sf_count[sfmode] == 1) {
1967                 ip_rt_multicast_event(pmc->interface);
1968         }
1969         return 0;
1970 }
1971
1972 #ifdef CONFIG_IP_MULTICAST
1973 static void sf_markstate(struct ip_mc_list *pmc)
1974 {
1975         struct ip_sf_list *psf;
1976         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1977
1978         for (psf = pmc->sources; psf; psf = psf->sf_next)
1979                 if (pmc->sfcount[MCAST_EXCLUDE]) {
1980                         psf->sf_oldin = mca_xcount ==
1981                                 psf->sf_count[MCAST_EXCLUDE] &&
1982                                 !psf->sf_count[MCAST_INCLUDE];
1983                 } else
1984                         psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
1985 }
1986
1987 static int sf_setstate(struct ip_mc_list *pmc)
1988 {
1989         struct ip_sf_list *psf, *dpsf;
1990         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1991         int qrv = pmc->interface->mr_qrv;
1992         int new_in, rv;
1993
1994         rv = 0;
1995         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1996                 if (pmc->sfcount[MCAST_EXCLUDE]) {
1997                         new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
1998                                 !psf->sf_count[MCAST_INCLUDE];
1999                 } else
2000                         new_in = psf->sf_count[MCAST_INCLUDE] != 0;
2001                 if (new_in) {
2002                         if (!psf->sf_oldin) {
2003                                 struct ip_sf_list *prev = NULL;
2004
2005                                 for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
2006                                         if (dpsf->sf_inaddr == psf->sf_inaddr)
2007                                                 break;
2008                                         prev = dpsf;
2009                                 }
2010                                 if (dpsf) {
2011                                         if (prev)
2012                                                 prev->sf_next = dpsf->sf_next;
2013                                         else
2014                                                 pmc->tomb = dpsf->sf_next;
2015                                         kfree(dpsf);
2016                                 }
2017                                 psf->sf_crcount = qrv;
2018                                 rv++;
2019                         }
2020                 } else if (psf->sf_oldin) {
2021
2022                         psf->sf_crcount = 0;
2023                         /*
2024                          * add or update "delete" records if an active filter
2025                          * is now inactive
2026                          */
2027                         for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
2028                                 if (dpsf->sf_inaddr == psf->sf_inaddr)
2029                                         break;
2030                         if (!dpsf) {
2031                                 dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
2032                                 if (!dpsf)
2033                                         continue;
2034                                 *dpsf = *psf;
2035                                 /* pmc->lock held by callers */
2036                                 dpsf->sf_next = pmc->tomb;
2037                                 pmc->tomb = dpsf;
2038                         }
2039                         dpsf->sf_crcount = qrv;
2040                         rv++;
2041                 }
2042         }
2043         return rv;
2044 }
2045 #endif
2046
2047 /*
2048  * Add multicast source filter list to the interface list
2049  */
2050 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2051                          int sfcount, __be32 *psfsrc, int delta)
2052 {
2053         struct ip_mc_list *pmc;
2054         int     isexclude;
2055         int     i, err;
2056
2057         if (!in_dev)
2058                 return -ENODEV;
2059         rcu_read_lock();
2060         for_each_pmc_rcu(in_dev, pmc) {
2061                 if (*pmca == pmc->multiaddr)
2062                         break;
2063         }
2064         if (!pmc) {
2065                 /* MCA not found?? bug */
2066                 rcu_read_unlock();
2067                 return -ESRCH;
2068         }
2069         spin_lock_bh(&pmc->lock);
2070         rcu_read_unlock();
2071
2072 #ifdef CONFIG_IP_MULTICAST
2073         sf_markstate(pmc);
2074 #endif
2075         isexclude = pmc->sfmode == MCAST_EXCLUDE;
2076         if (!delta)
2077                 pmc->sfcount[sfmode]++;
2078         err = 0;
2079         for (i = 0; i < sfcount; i++) {
2080                 err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2081                 if (err)
2082                         break;
2083         }
2084         if (err) {
2085                 int j;
2086
2087                 if (!delta)
2088                         pmc->sfcount[sfmode]--;
2089                 for (j = 0; j < i; j++)
2090                         (void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2091         } else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2092 #ifdef CONFIG_IP_MULTICAST
2093                 struct ip_sf_list *psf;
2094                 struct net *net = dev_net(pmc->interface->dev);
2095                 in_dev = pmc->interface;
2096 #endif
2097
2098                 /* filter mode change */
2099                 if (pmc->sfcount[MCAST_EXCLUDE])
2100                         pmc->sfmode = MCAST_EXCLUDE;
2101                 else if (pmc->sfcount[MCAST_INCLUDE])
2102                         pmc->sfmode = MCAST_INCLUDE;
2103 #ifdef CONFIG_IP_MULTICAST
2104                 /* else no filters; keep old mode for reports */
2105
2106                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
2107                 in_dev->mr_ifc_count = pmc->crcount;
2108                 for (psf = pmc->sources; psf; psf = psf->sf_next)
2109                         psf->sf_crcount = 0;
2110                 igmp_ifc_event(in_dev);
2111         } else if (sf_setstate(pmc)) {
2112                 igmp_ifc_event(in_dev);
2113 #endif
2114         }
2115         spin_unlock_bh(&pmc->lock);
2116         return err;
2117 }
2118
2119 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2120 {
2121         struct ip_sf_list *tomb, *sources;
2122
2123         spin_lock_bh(&pmc->lock);
2124         tomb = pmc->tomb;
2125         pmc->tomb = NULL;
2126         sources = pmc->sources;
2127         pmc->sources = NULL;
2128         pmc->sfmode = MCAST_EXCLUDE;
2129         pmc->sfcount[MCAST_INCLUDE] = 0;
2130         pmc->sfcount[MCAST_EXCLUDE] = 1;
2131         spin_unlock_bh(&pmc->lock);
2132
2133         ip_sf_list_clear_all(tomb);
2134         ip_sf_list_clear_all(sources);
2135 }
2136
2137 /* Join a multicast group
2138  */
2139
2140 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2141 {
2142         __be32 addr = imr->imr_multiaddr.s_addr;
2143         struct ip_mc_socklist *iml, *i;
2144         struct in_device *in_dev;
2145         struct inet_sock *inet = inet_sk(sk);
2146         struct net *net = sock_net(sk);
2147         int ifindex;
2148         int count = 0;
2149         int err;
2150
2151         ASSERT_RTNL();
2152
2153         if (!ipv4_is_multicast(addr))
2154                 return -EINVAL;
2155
2156         in_dev = ip_mc_find_dev(net, imr);
2157
2158         if (!in_dev) {
2159                 err = -ENODEV;
2160                 goto done;
2161         }
2162
2163         err = -EADDRINUSE;
2164         ifindex = imr->imr_ifindex;
2165         for_each_pmc_rtnl(inet, i) {
2166                 if (i->multi.imr_multiaddr.s_addr == addr &&
2167                     i->multi.imr_ifindex == ifindex)
2168                         goto done;
2169                 count++;
2170         }
2171         err = -ENOBUFS;
2172         if (count >= net->ipv4.sysctl_igmp_max_memberships)
2173                 goto done;
2174         iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2175         if (!iml)
2176                 goto done;
2177
2178         memcpy(&iml->multi, imr, sizeof(*imr));
2179         iml->next_rcu = inet->mc_list;
2180         iml->sflist = NULL;
2181         iml->sfmode = MCAST_EXCLUDE;
2182         rcu_assign_pointer(inet->mc_list, iml);
2183         ip_mc_inc_group(in_dev, addr);
2184         err = 0;
2185 done:
2186         return err;
2187 }
2188 EXPORT_SYMBOL(ip_mc_join_group);
2189
2190 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2191                            struct in_device *in_dev)
2192 {
2193         struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2194         int err;
2195
2196         if (!psf) {
2197                 /* any-source empty exclude case */
2198                 return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2199                         iml->sfmode, 0, NULL, 0);
2200         }
2201         err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2202                         iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2203         RCU_INIT_POINTER(iml->sflist, NULL);
2204         /* decrease mem now to avoid the memleak warning */
2205         atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
2206         kfree_rcu(psf, rcu);
2207         return err;
2208 }
2209
2210 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2211 {
2212         struct inet_sock *inet = inet_sk(sk);
2213         struct ip_mc_socklist *iml;
2214         struct ip_mc_socklist __rcu **imlp;
2215         struct in_device *in_dev;
2216         struct net *net = sock_net(sk);
2217         __be32 group = imr->imr_multiaddr.s_addr;
2218         u32 ifindex;
2219         int ret = -EADDRNOTAVAIL;
2220
2221         ASSERT_RTNL();
2222
2223         in_dev = ip_mc_find_dev(net, imr);
2224         if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2225                 ret = -ENODEV;
2226                 goto out;
2227         }
2228         ifindex = imr->imr_ifindex;
2229         for (imlp = &inet->mc_list;
2230              (iml = rtnl_dereference(*imlp)) != NULL;
2231              imlp = &iml->next_rcu) {
2232                 if (iml->multi.imr_multiaddr.s_addr != group)
2233                         continue;
2234                 if (ifindex) {
2235                         if (iml->multi.imr_ifindex != ifindex)
2236                                 continue;
2237                 } else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2238                                 iml->multi.imr_address.s_addr)
2239                         continue;
2240
2241                 (void) ip_mc_leave_src(sk, iml, in_dev);
2242
2243                 *imlp = iml->next_rcu;
2244
2245                 if (in_dev)
2246                         ip_mc_dec_group(in_dev, group);
2247
2248                 /* decrease mem now to avoid the memleak warning */
2249                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2250                 kfree_rcu(iml, rcu);
2251                 return 0;
2252         }
2253 out:
2254         return ret;
2255 }
2256 EXPORT_SYMBOL(ip_mc_leave_group);
2257
2258 int ip_mc_source(int add, int omode, struct sock *sk, struct
2259         ip_mreq_source *mreqs, int ifindex)
2260 {
2261         int err;
2262         struct ip_mreqn imr;
2263         __be32 addr = mreqs->imr_multiaddr;
2264         struct ip_mc_socklist *pmc;
2265         struct in_device *in_dev = NULL;
2266         struct inet_sock *inet = inet_sk(sk);
2267         struct ip_sf_socklist *psl;
2268         struct net *net = sock_net(sk);
2269         int leavegroup = 0;
2270         int i, j, rv;
2271
2272         if (!ipv4_is_multicast(addr))
2273                 return -EINVAL;
2274
2275         ASSERT_RTNL();
2276
2277         imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2278         imr.imr_address.s_addr = mreqs->imr_interface;
2279         imr.imr_ifindex = ifindex;
2280         in_dev = ip_mc_find_dev(net, &imr);
2281
2282         if (!in_dev) {
2283                 err = -ENODEV;
2284                 goto done;
2285         }
2286         err = -EADDRNOTAVAIL;
2287
2288         for_each_pmc_rtnl(inet, pmc) {
2289                 if ((pmc->multi.imr_multiaddr.s_addr ==
2290                      imr.imr_multiaddr.s_addr) &&
2291                     (pmc->multi.imr_ifindex == imr.imr_ifindex))
2292                         break;
2293         }
2294         if (!pmc) {             /* must have a prior join */
2295                 err = -EINVAL;
2296                 goto done;
2297         }
2298         /* if a source filter was set, must be the same mode as before */
2299         if (pmc->sflist) {
2300                 if (pmc->sfmode != omode) {
2301                         err = -EINVAL;
2302                         goto done;
2303                 }
2304         } else if (pmc->sfmode != omode) {
2305                 /* allow mode switches for empty-set filters */
2306                 ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2307                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2308                         NULL, 0);
2309                 pmc->sfmode = omode;
2310         }
2311
2312         psl = rtnl_dereference(pmc->sflist);
2313         if (!add) {
2314                 if (!psl)
2315                         goto done;      /* err = -EADDRNOTAVAIL */
2316                 rv = !0;
2317                 for (i = 0; i < psl->sl_count; i++) {
2318                         rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2319                                 sizeof(__be32));
2320                         if (rv == 0)
2321                                 break;
2322                 }
2323                 if (rv)         /* source not found */
2324                         goto done;      /* err = -EADDRNOTAVAIL */
2325
2326                 /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2327                 if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2328                         leavegroup = 1;
2329                         goto done;
2330                 }
2331
2332                 /* update the interface filter */
2333                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2334                         &mreqs->imr_sourceaddr, 1);
2335
2336                 for (j = i+1; j < psl->sl_count; j++)
2337                         psl->sl_addr[j-1] = psl->sl_addr[j];
2338                 psl->sl_count--;
2339                 err = 0;
2340                 goto done;
2341         }
2342         /* else, add a new source to the filter */
2343
2344         if (psl && psl->sl_count >= net->ipv4.sysctl_igmp_max_msf) {
2345                 err = -ENOBUFS;
2346                 goto done;
2347         }
2348         if (!psl || psl->sl_count == psl->sl_max) {
2349                 struct ip_sf_socklist *newpsl;
2350                 int count = IP_SFBLOCK;
2351
2352                 if (psl)
2353                         count += psl->sl_max;
2354                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2355                 if (!newpsl) {
2356                         err = -ENOBUFS;
2357                         goto done;
2358                 }
2359                 newpsl->sl_max = count;
2360                 newpsl->sl_count = count - IP_SFBLOCK;
2361                 if (psl) {
2362                         for (i = 0; i < psl->sl_count; i++)
2363                                 newpsl->sl_addr[i] = psl->sl_addr[i];
2364                         /* decrease mem now to avoid the memleak warning */
2365                         atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2366                         kfree_rcu(psl, rcu);
2367                 }
2368                 rcu_assign_pointer(pmc->sflist, newpsl);
2369                 psl = newpsl;
2370         }
2371         rv = 1; /* > 0 for insert logic below if sl_count is 0 */
2372         for (i = 0; i < psl->sl_count; i++) {
2373                 rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2374                         sizeof(__be32));
2375                 if (rv == 0)
2376                         break;
2377         }
2378         if (rv == 0)            /* address already there is an error */
2379                 goto done;
2380         for (j = psl->sl_count-1; j >= i; j--)
2381                 psl->sl_addr[j+1] = psl->sl_addr[j];
2382         psl->sl_addr[i] = mreqs->imr_sourceaddr;
2383         psl->sl_count++;
2384         err = 0;
2385         /* update the interface list */
2386         ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2387                 &mreqs->imr_sourceaddr, 1);
2388 done:
2389         if (leavegroup)
2390                 err = ip_mc_leave_group(sk, &imr);
2391         return err;
2392 }
2393
2394 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2395 {
2396         int err = 0;
2397         struct ip_mreqn imr;
2398         __be32 addr = msf->imsf_multiaddr;
2399         struct ip_mc_socklist *pmc;
2400         struct in_device *in_dev;
2401         struct inet_sock *inet = inet_sk(sk);
2402         struct ip_sf_socklist *newpsl, *psl;
2403         struct net *net = sock_net(sk);
2404         int leavegroup = 0;
2405
2406         if (!ipv4_is_multicast(addr))
2407                 return -EINVAL;
2408         if (msf->imsf_fmode != MCAST_INCLUDE &&
2409             msf->imsf_fmode != MCAST_EXCLUDE)
2410                 return -EINVAL;
2411
2412         ASSERT_RTNL();
2413
2414         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2415         imr.imr_address.s_addr = msf->imsf_interface;
2416         imr.imr_ifindex = ifindex;
2417         in_dev = ip_mc_find_dev(net, &imr);
2418
2419         if (!in_dev) {
2420                 err = -ENODEV;
2421                 goto done;
2422         }
2423
2424         /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2425         if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2426                 leavegroup = 1;
2427                 goto done;
2428         }
2429
2430         for_each_pmc_rtnl(inet, pmc) {
2431                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2432                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2433                         break;
2434         }
2435         if (!pmc) {             /* must have a prior join */
2436                 err = -EINVAL;
2437                 goto done;
2438         }
2439         if (msf->imsf_numsrc) {
2440                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2441                                                            GFP_KERNEL);
2442                 if (!newpsl) {
2443                         err = -ENOBUFS;
2444                         goto done;
2445                 }
2446                 newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2447                 memcpy(newpsl->sl_addr, msf->imsf_slist,
2448                         msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2449                 err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2450                         msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2451                 if (err) {
2452                         sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2453                         goto done;
2454                 }
2455         } else {
2456                 newpsl = NULL;
2457                 (void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2458                                      msf->imsf_fmode, 0, NULL, 0);
2459         }
2460         psl = rtnl_dereference(pmc->sflist);
2461         if (psl) {
2462                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2463                         psl->sl_count, psl->sl_addr, 0);
2464                 /* decrease mem now to avoid the memleak warning */
2465                 atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2466                 kfree_rcu(psl, rcu);
2467         } else
2468                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2469                         0, NULL, 0);
2470         rcu_assign_pointer(pmc->sflist, newpsl);
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                         for (psf = im->sources; psf; psf = psf->sf_next) {
2692                                 if (psf->sf_inaddr == src_addr)
2693                                         break;
2694                         }
2695                         if (psf)
2696                                 rv = psf->sf_count[MCAST_INCLUDE] ||
2697                                         psf->sf_count[MCAST_EXCLUDE] !=
2698                                         im->sfcount[MCAST_EXCLUDE];
2699                         else
2700                                 rv = im->sfcount[MCAST_EXCLUDE] != 0;
2701                 } else
2702                         rv = 1; /* unspecified source; tentatively allow */
2703         }
2704         return rv;
2705 }
2706
2707 #if defined(CONFIG_PROC_FS)
2708 struct igmp_mc_iter_state {
2709         struct seq_net_private p;
2710         struct net_device *dev;
2711         struct in_device *in_dev;
2712 };
2713
2714 #define igmp_mc_seq_private(seq)        ((struct igmp_mc_iter_state *)(seq)->private)
2715
2716 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2717 {
2718         struct net *net = seq_file_net(seq);
2719         struct ip_mc_list *im = NULL;
2720         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2721
2722         state->in_dev = NULL;
2723         for_each_netdev_rcu(net, state->dev) {
2724                 struct in_device *in_dev;
2725
2726                 in_dev = __in_dev_get_rcu(state->dev);
2727                 if (!in_dev)
2728                         continue;
2729                 im = rcu_dereference(in_dev->mc_list);
2730                 if (im) {
2731                         state->in_dev = in_dev;
2732                         break;
2733                 }
2734         }
2735         return im;
2736 }
2737
2738 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2739 {
2740         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2741
2742         im = rcu_dereference(im->next_rcu);
2743         while (!im) {
2744                 state->dev = next_net_device_rcu(state->dev);
2745                 if (!state->dev) {
2746                         state->in_dev = NULL;
2747                         break;
2748                 }
2749                 state->in_dev = __in_dev_get_rcu(state->dev);
2750                 if (!state->in_dev)
2751                         continue;
2752                 im = rcu_dereference(state->in_dev->mc_list);
2753         }
2754         return im;
2755 }
2756
2757 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2758 {
2759         struct ip_mc_list *im = igmp_mc_get_first(seq);
2760         if (im)
2761                 while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2762                         --pos;
2763         return pos ? NULL : im;
2764 }
2765
2766 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2767         __acquires(rcu)
2768 {
2769         rcu_read_lock();
2770         return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2771 }
2772
2773 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2774 {
2775         struct ip_mc_list *im;
2776         if (v == SEQ_START_TOKEN)
2777                 im = igmp_mc_get_first(seq);
2778         else
2779                 im = igmp_mc_get_next(seq, v);
2780         ++*pos;
2781         return im;
2782 }
2783
2784 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2785         __releases(rcu)
2786 {
2787         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2788
2789         state->in_dev = NULL;
2790         state->dev = NULL;
2791         rcu_read_unlock();
2792 }
2793
2794 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2795 {
2796         if (v == SEQ_START_TOKEN)
2797                 seq_puts(seq,
2798                          "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2799         else {
2800                 struct ip_mc_list *im = (struct ip_mc_list *)v;
2801                 struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2802                 char   *querier;
2803                 long delta;
2804
2805 #ifdef CONFIG_IP_MULTICAST
2806                 querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2807                           IGMP_V2_SEEN(state->in_dev) ? "V2" :
2808                           "V3";
2809 #else
2810                 querier = "NONE";
2811 #endif
2812
2813                 if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2814                         seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2815                                    state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2816                 }
2817
2818                 delta = im->timer.expires - jiffies;
2819                 seq_printf(seq,
2820                            "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2821                            im->multiaddr, im->users,
2822                            im->tm_running,
2823                            im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2824                            im->reporter);
2825         }
2826         return 0;
2827 }
2828
2829 static const struct seq_operations igmp_mc_seq_ops = {
2830         .start  =       igmp_mc_seq_start,
2831         .next   =       igmp_mc_seq_next,
2832         .stop   =       igmp_mc_seq_stop,
2833         .show   =       igmp_mc_seq_show,
2834 };
2835
2836 static int igmp_mc_seq_open(struct inode *inode, struct file *file)
2837 {
2838         return seq_open_net(inode, file, &igmp_mc_seq_ops,
2839                         sizeof(struct igmp_mc_iter_state));
2840 }
2841
2842 static const struct file_operations igmp_mc_seq_fops = {
2843         .owner          =       THIS_MODULE,
2844         .open           =       igmp_mc_seq_open,
2845         .read           =       seq_read,
2846         .llseek         =       seq_lseek,
2847         .release        =       seq_release_net,
2848 };
2849
2850 struct igmp_mcf_iter_state {
2851         struct seq_net_private p;
2852         struct net_device *dev;
2853         struct in_device *idev;
2854         struct ip_mc_list *im;
2855 };
2856
2857 #define igmp_mcf_seq_private(seq)       ((struct igmp_mcf_iter_state *)(seq)->private)
2858
2859 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2860 {
2861         struct net *net = seq_file_net(seq);
2862         struct ip_sf_list *psf = NULL;
2863         struct ip_mc_list *im = NULL;
2864         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2865
2866         state->idev = NULL;
2867         state->im = NULL;
2868         for_each_netdev_rcu(net, state->dev) {
2869                 struct in_device *idev;
2870                 idev = __in_dev_get_rcu(state->dev);
2871                 if (unlikely(!idev))
2872                         continue;
2873                 im = rcu_dereference(idev->mc_list);
2874                 if (likely(im)) {
2875                         spin_lock_bh(&im->lock);
2876                         psf = im->sources;
2877                         if (likely(psf)) {
2878                                 state->im = im;
2879                                 state->idev = idev;
2880                                 break;
2881                         }
2882                         spin_unlock_bh(&im->lock);
2883                 }
2884         }
2885         return psf;
2886 }
2887
2888 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2889 {
2890         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2891
2892         psf = psf->sf_next;
2893         while (!psf) {
2894                 spin_unlock_bh(&state->im->lock);
2895                 state->im = state->im->next;
2896                 while (!state->im) {
2897                         state->dev = next_net_device_rcu(state->dev);
2898                         if (!state->dev) {
2899                                 state->idev = NULL;
2900                                 goto out;
2901                         }
2902                         state->idev = __in_dev_get_rcu(state->dev);
2903                         if (!state->idev)
2904                                 continue;
2905                         state->im = rcu_dereference(state->idev->mc_list);
2906                 }
2907                 if (!state->im)
2908                         break;
2909                 spin_lock_bh(&state->im->lock);
2910                 psf = state->im->sources;
2911         }
2912 out:
2913         return psf;
2914 }
2915
2916 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2917 {
2918         struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2919         if (psf)
2920                 while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2921                         --pos;
2922         return pos ? NULL : psf;
2923 }
2924
2925 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2926         __acquires(rcu)
2927 {
2928         rcu_read_lock();
2929         return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2930 }
2931
2932 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2933 {
2934         struct ip_sf_list *psf;
2935         if (v == SEQ_START_TOKEN)
2936                 psf = igmp_mcf_get_first(seq);
2937         else
2938                 psf = igmp_mcf_get_next(seq, v);
2939         ++*pos;
2940         return psf;
2941 }
2942
2943 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2944         __releases(rcu)
2945 {
2946         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2947         if (likely(state->im)) {
2948                 spin_unlock_bh(&state->im->lock);
2949                 state->im = NULL;
2950         }
2951         state->idev = NULL;
2952         state->dev = NULL;
2953         rcu_read_unlock();
2954 }
2955
2956 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2957 {
2958         struct ip_sf_list *psf = (struct ip_sf_list *)v;
2959         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2960
2961         if (v == SEQ_START_TOKEN) {
2962                 seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
2963         } else {
2964                 seq_printf(seq,
2965                            "%3d %6.6s 0x%08x "
2966                            "0x%08x %6lu %6lu\n",
2967                            state->dev->ifindex, state->dev->name,
2968                            ntohl(state->im->multiaddr),
2969                            ntohl(psf->sf_inaddr),
2970                            psf->sf_count[MCAST_INCLUDE],
2971                            psf->sf_count[MCAST_EXCLUDE]);
2972         }
2973         return 0;
2974 }
2975
2976 static const struct seq_operations igmp_mcf_seq_ops = {
2977         .start  =       igmp_mcf_seq_start,
2978         .next   =       igmp_mcf_seq_next,
2979         .stop   =       igmp_mcf_seq_stop,
2980         .show   =       igmp_mcf_seq_show,
2981 };
2982
2983 static int igmp_mcf_seq_open(struct inode *inode, struct file *file)
2984 {
2985         return seq_open_net(inode, file, &igmp_mcf_seq_ops,
2986                         sizeof(struct igmp_mcf_iter_state));
2987 }
2988
2989 static const struct file_operations igmp_mcf_seq_fops = {
2990         .owner          =       THIS_MODULE,
2991         .open           =       igmp_mcf_seq_open,
2992         .read           =       seq_read,
2993         .llseek         =       seq_lseek,
2994         .release        =       seq_release_net,
2995 };
2996
2997 static int __net_init igmp_net_init(struct net *net)
2998 {
2999         struct proc_dir_entry *pde;
3000         int err;
3001
3002         pde = proc_create("igmp", S_IRUGO, net->proc_net, &igmp_mc_seq_fops);
3003         if (!pde)
3004                 goto out_igmp;
3005         pde = proc_create("mcfilter", S_IRUGO, net->proc_net,
3006                           &igmp_mcf_seq_fops);
3007         if (!pde)
3008                 goto out_mcfilter;
3009         err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
3010                                    SOCK_DGRAM, 0, net);
3011         if (err < 0) {
3012                 pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
3013                        err);
3014                 goto out_sock;
3015         }
3016
3017         return 0;
3018
3019 out_sock:
3020         remove_proc_entry("mcfilter", net->proc_net);
3021 out_mcfilter:
3022         remove_proc_entry("igmp", net->proc_net);
3023 out_igmp:
3024         return -ENOMEM;
3025 }
3026
3027 static void __net_exit igmp_net_exit(struct net *net)
3028 {
3029         remove_proc_entry("mcfilter", net->proc_net);
3030         remove_proc_entry("igmp", net->proc_net);
3031         inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
3032 }
3033
3034 static struct pernet_operations igmp_net_ops = {
3035         .init = igmp_net_init,
3036         .exit = igmp_net_exit,
3037 };
3038 #endif
3039
3040 static int igmp_netdev_event(struct notifier_block *this,
3041                              unsigned long event, void *ptr)
3042 {
3043         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3044         struct in_device *in_dev;
3045
3046         switch (event) {
3047         case NETDEV_RESEND_IGMP:
3048                 in_dev = __in_dev_get_rtnl(dev);
3049                 if (in_dev)
3050                         ip_mc_rejoin_groups(in_dev);
3051                 break;
3052         default:
3053                 break;
3054         }
3055         return NOTIFY_DONE;
3056 }
3057
3058 static struct notifier_block igmp_notifier = {
3059         .notifier_call = igmp_netdev_event,
3060 };
3061
3062 int __init igmp_mc_init(void)
3063 {
3064 #if defined(CONFIG_PROC_FS)
3065         int err;
3066
3067         err = register_pernet_subsys(&igmp_net_ops);
3068         if (err)
3069                 return err;
3070         err = register_netdevice_notifier(&igmp_notifier);
3071         if (err)
3072                 goto reg_notif_fail;
3073         return 0;
3074
3075 reg_notif_fail:
3076         unregister_pernet_subsys(&igmp_net_ops);
3077         return err;
3078 #else
3079         return register_netdevice_notifier(&igmp_notifier);
3080 #endif
3081 }