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 <linux/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
  93 #include <net/net_namespace.h>
  94 #include <net/arp.h>
  95 #include <net/ip.h>
  96 #include <net/protocol.h>
  97 #include <net/route.h>
  98 #include <net/sock.h>
  99 #include <net/checksum.h>
 100 #include <net/inet_common.h>
 101 #include <linux/netfilter_ipv4.h>
 102 #ifdef CONFIG_IP_MROUTE
 103 #include <linux/mroute.h>
 104 #endif
 105 #ifdef CONFIG_PROC_FS
 106 #include <linux/proc_fs.h>
 107 #include <linux/seq_file.h>
 108 #endif
 109
 110 #ifdef CONFIG_IP_MULTICAST
 111 /* Parameter names and values are taken from igmp-v2-06 draft */
 112
 113 #define IGMP_V1_ROUTER_PRESENT_TIMEOUT          (400*HZ)
 114 #define IGMP_V2_ROUTER_PRESENT_TIMEOUT          (400*HZ)
 115 #define IGMP_V2_UNSOLICITED_REPORT_INTERVAL     (10*HZ)
 116 #define IGMP_V3_UNSOLICITED_REPORT_INTERVAL     (1*HZ)
 117 #define IGMP_QUERY_RESPONSE_INTERVAL            (10*HZ)
 118 #define IGMP_QUERY_ROBUSTNESS_VARIABLE          2
 119
 120
 121 #define IGMP_INITIAL_REPORT_DELAY               (1)
 122
 123 /* IGMP_INITIAL_REPORT_DELAY is not from IGMP specs!
 124  * IGMP specs require to report membership immediately after
 125  * joining a group, but we delay the first report by a
 126  * small interval. It seems more natural and still does not
 127  * contradict to specs provided this delay is small enough.
 128  */
 129
 130 #define IGMP_V1_SEEN(in_dev) \
 131         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 1 || \
 132          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 1 || \
 133          ((in_dev)->mr_v1_seen && \
 134           time_before(jiffies, (in_dev)->mr_v1_seen)))
 135 #define IGMP_V2_SEEN(in_dev) \
 136         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 2 || \
 137          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 2 || \
 138          ((in_dev)->mr_v2_seen && \
 139           time_before(jiffies, (in_dev)->mr_v2_seen)))
 140
 141 static int unsolicited_report_interval(struct in_device *in_dev)
 142 {
 143         int interval_ms, interval_jiffies;
 144
 145         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 146                 interval_ms = IN_DEV_CONF_GET(
 147                         in_dev,
 148                         IGMPV2_UNSOLICITED_REPORT_INTERVAL);
 149         else /* v3 */
 150                 interval_ms = IN_DEV_CONF_GET(
 151                         in_dev,
 152                         IGMPV3_UNSOLICITED_REPORT_INTERVAL);
 153
 154         interval_jiffies = msecs_to_jiffies(interval_ms);
 155
 156         /* _timer functions can't handle a delay of 0 jiffies so ensure
 157          *  we always return a positive value.
 158          */
 159         if (interval_jiffies <= 0)
 160                 interval_jiffies = 1;
 161         return interval_jiffies;
 162 }
 163
 164 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im);
 165 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im);
 166 static void igmpv3_clear_delrec(struct in_device *in_dev);
 167 static int sf_setstate(struct ip_mc_list *pmc);
 168 static void sf_markstate(struct ip_mc_list *pmc);
 169 #endif
 170 static void ip_mc_clear_src(struct ip_mc_list *pmc);
 171 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
 172                          int sfcount, __be32 *psfsrc, int delta);
 173
 174 static void ip_ma_put(struct ip_mc_list *im)
 175 {
 176         if (atomic_dec_and_test(&im->refcnt)) {
 177                 in_dev_put(im->interface);
 178                 kfree_rcu(im, rcu);
 179         }
 180 }
 181
 182 #define for_each_pmc_rcu(in_dev, pmc)                           \
 183         for (pmc = rcu_dereference(in_dev->mc_list);            \
 184              pmc != NULL;                                       \
 185              pmc = rcu_dereference(pmc->next_rcu))
 186
 187 #define for_each_pmc_rtnl(in_dev, pmc)                          \
 188         for (pmc = rtnl_dereference(in_dev->mc_list);           \
 189              pmc != NULL;                                       \
 190              pmc = rtnl_dereference(pmc->next_rcu))
 191
 192 #ifdef CONFIG_IP_MULTICAST
 193
 194 /*
 195  *      Timer management
 196  */
 197
 198 static void igmp_stop_timer(struct ip_mc_list *im)
 199 {
 200         spin_lock_bh(&im->lock);
 201         if (del_timer(&im->timer))
 202                 atomic_dec(&im->refcnt);
 203         im->tm_running = 0;
 204         im->reporter = 0;
 205         im->unsolicit_count = 0;
 206         spin_unlock_bh(&im->lock);
 207 }
 208
 209 /* It must be called with locked im->lock */
 210 static void igmp_start_timer(struct ip_mc_list *im, int max_delay)
 211 {
 212         int tv = prandom_u32() % max_delay;
 213
 214         im->tm_running = 1;
 215         if (!mod_timer(&im->timer, jiffies+tv+2))
 216                 atomic_inc(&im->refcnt);
 217 }
 218
 219 static void igmp_gq_start_timer(struct in_device *in_dev)
 220 {
 221         int tv = prandom_u32() % in_dev->mr_maxdelay;
 222         unsigned long exp = jiffies + tv + 2;
 223
 224         if (in_dev->mr_gq_running &&
 225             time_after_eq(exp, (in_dev->mr_gq_timer).expires))
 226                 return;
 227
 228         in_dev->mr_gq_running = 1;
 229         if (!mod_timer(&in_dev->mr_gq_timer, exp))
 230                 in_dev_hold(in_dev);
 231 }
 232
 233 static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
 234 {
 235         int tv = prandom_u32() % delay;
 236
 237         if (!mod_timer(&in_dev->mr_ifc_timer, jiffies+tv+2))
 238                 in_dev_hold(in_dev);
 239 }
 240
 241 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
 242 {
 243         spin_lock_bh(&im->lock);
 244         im->unsolicit_count = 0;
 245         if (del_timer(&im->timer)) {
 246                 if ((long)(im->timer.expires-jiffies) < max_delay) {
 247                         add_timer(&im->timer);
 248                         im->tm_running = 1;
 249                         spin_unlock_bh(&im->lock);
 250                         return;
 251                 }
 252                 atomic_dec(&im->refcnt);
 253         }
 254         igmp_start_timer(im, max_delay);
 255         spin_unlock_bh(&im->lock);
 256 }
 257
 258
 259 /*
 260  *      Send an IGMP report.
 261  */
 262
 263 #define IGMP_SIZE (sizeof(struct igmphdr)+sizeof(struct iphdr)+4)
 264
 265
 266 static int is_in(struct ip_mc_list *pmc, struct ip_sf_list *psf, int type,
 267         int gdeleted, int sdeleted)
 268 {
 269         switch (type) {
 270         case IGMPV3_MODE_IS_INCLUDE:
 271         case IGMPV3_MODE_IS_EXCLUDE:
 272                 if (gdeleted || sdeleted)
 273                         return 0;
 274                 if (!(pmc->gsquery && !psf->sf_gsresp)) {
 275                         if (pmc->sfmode == MCAST_INCLUDE)
 276                                 return 1;
 277                         /* don't include if this source is excluded
 278                          * in all filters
 279                          */
 280                         if (psf->sf_count[MCAST_INCLUDE])
 281                                 return type == IGMPV3_MODE_IS_INCLUDE;
 282                         return pmc->sfcount[MCAST_EXCLUDE] ==
 283                                 psf->sf_count[MCAST_EXCLUDE];
 284                 }
 285                 return 0;
 286         case IGMPV3_CHANGE_TO_INCLUDE:
 287                 if (gdeleted || sdeleted)
 288                         return 0;
 289                 return psf->sf_count[MCAST_INCLUDE] != 0;
 290         case IGMPV3_CHANGE_TO_EXCLUDE:
 291                 if (gdeleted || sdeleted)
 292                         return 0;
 293                 if (pmc->sfcount[MCAST_EXCLUDE] == 0 ||
 294                     psf->sf_count[MCAST_INCLUDE])
 295                         return 0;
 296                 return pmc->sfcount[MCAST_EXCLUDE] ==
 297                         psf->sf_count[MCAST_EXCLUDE];
 298         case IGMPV3_ALLOW_NEW_SOURCES:
 299                 if (gdeleted || !psf->sf_crcount)
 300                         return 0;
 301                 return (pmc->sfmode == MCAST_INCLUDE) ^ sdeleted;
 302         case IGMPV3_BLOCK_OLD_SOURCES:
 303                 if (pmc->sfmode == MCAST_INCLUDE)
 304                         return gdeleted || (psf->sf_crcount && sdeleted);
 305                 return psf->sf_crcount && !gdeleted && !sdeleted;
 306         }
 307         return 0;
 308 }
 309
 310 static int
 311 igmp_scount(struct ip_mc_list *pmc, int type, int gdeleted, int sdeleted)
 312 {
 313         struct ip_sf_list *psf;
 314         int scount = 0;
 315
 316         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 317                 if (!is_in(pmc, psf, type, gdeleted, sdeleted))
 318                         continue;
 319                 scount++;
 320         }
 321         return scount;
 322 }
 323
 324 static struct sk_buff *igmpv3_newpack(struct net_device *dev, unsigned int mtu)
 325 {
 326         struct sk_buff *skb;
 327         struct rtable *rt;
 328         struct iphdr *pip;
 329         struct igmpv3_report *pig;
 330         struct net *net = dev_net(dev);
 331         struct flowi4 fl4;
 332         int hlen = LL_RESERVED_SPACE(dev);
 333         int tlen = dev->needed_tailroom;
 334         unsigned int size = mtu;
 335
 336         while (1) {
 337                 skb = alloc_skb(size + hlen + tlen,
 338                                 GFP_ATOMIC | __GFP_NOWARN);
 339                 if (skb)
 340                         break;
 341                 size >>= 1;
 342                 if (size < 256)
 343                         return NULL;
 344         }
 345         skb->priority = TC_PRIO_CONTROL;
 346
 347         rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
 348                                    0, 0,
 349                                    IPPROTO_IGMP, 0, dev->ifindex);
 350         if (IS_ERR(rt)) {
 351                 kfree_skb(skb);
 352                 return NULL;
 353         }
 354
 355         skb_dst_set(skb, &rt->dst);
 356         skb->dev = dev;
 357
 358         skb_reserve(skb, hlen);
 359         skb_tailroom_reserve(skb, mtu, tlen);
 360
 361         skb_reset_network_header(skb);
 362         pip = ip_hdr(skb);
 363         skb_put(skb, sizeof(struct iphdr) + 4);
 364
 365         pip->version  = 4;
 366         pip->ihl      = (sizeof(struct iphdr)+4)>>2;
 367         pip->tos      = 0xc0;
 368         pip->frag_off = htons(IP_DF);
 369         pip->ttl      = 1;
 370         pip->daddr    = fl4.daddr;
 371         pip->saddr    = fl4.saddr;
 372         pip->protocol = IPPROTO_IGMP;
 373         pip->tot_len  = 0;      /* filled in later */
 374         ip_select_ident(net, skb, NULL);
 375         ((u8 *)&pip[1])[0] = IPOPT_RA;
 376         ((u8 *)&pip[1])[1] = 4;
 377         ((u8 *)&pip[1])[2] = 0;
 378         ((u8 *)&pip[1])[3] = 0;
 379
 380         skb->transport_header = skb->network_header + sizeof(struct iphdr) + 4;
 381         skb_put(skb, sizeof(*pig));
 382         pig = igmpv3_report_hdr(skb);
 383         pig->type = IGMPV3_HOST_MEMBERSHIP_REPORT;
 384         pig->resv1 = 0;
 385         pig->csum = 0;
 386         pig->resv2 = 0;
 387         pig->ngrec = 0;
 388         return skb;
 389 }
 390
 391 static int igmpv3_sendpack(struct sk_buff *skb)
 392 {
 393         struct igmphdr *pig = igmp_hdr(skb);
 394         const int igmplen = skb_tail_pointer(skb) - skb_transport_header(skb);
 395
 396         pig->csum = ip_compute_csum(igmp_hdr(skb), igmplen);
 397
 398         return ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 399 }
 400
 401 static int grec_size(struct ip_mc_list *pmc, int type, int gdel, int sdel)
 402 {
 403         return sizeof(struct igmpv3_grec) + 4*igmp_scount(pmc, type, gdel, sdel);
 404 }
 405
 406 static struct sk_buff *add_grhead(struct sk_buff *skb, struct ip_mc_list *pmc,
 407         int type, struct igmpv3_grec **ppgr)
 408 {
 409         struct net_device *dev = pmc->interface->dev;
 410         struct igmpv3_report *pih;
 411         struct igmpv3_grec *pgr;
 412
 413         if (!skb)
 414                 skb = igmpv3_newpack(dev, dev->mtu);
 415         if (!skb)
 416                 return NULL;
 417         pgr = (struct igmpv3_grec *)skb_put(skb, sizeof(struct igmpv3_grec));
 418         pgr->grec_type = type;
 419         pgr->grec_auxwords = 0;
 420         pgr->grec_nsrcs = 0;
 421         pgr->grec_mca = pmc->multiaddr;
 422         pih = igmpv3_report_hdr(skb);
 423         pih->ngrec = htons(ntohs(pih->ngrec)+1);
 424         *ppgr = pgr;
 425         return skb;
 426 }
 427
 428 #define AVAILABLE(skb)  ((skb) ? skb_availroom(skb) : 0)
 429
 430 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
 431         int type, int gdeleted, int sdeleted)
 432 {
 433         struct net_device *dev = pmc->interface->dev;
 434         struct net *net = dev_net(dev);
 435         struct igmpv3_report *pih;
 436         struct igmpv3_grec *pgr = NULL;
 437         struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
 438         int scount, stotal, first, isquery, truncate;
 439
 440         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 441                 return skb;
 442         if (ipv4_is_local_multicast(pmc->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
 443                 return skb;
 444
 445         isquery = type == IGMPV3_MODE_IS_INCLUDE ||
 446                   type == IGMPV3_MODE_IS_EXCLUDE;
 447         truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
 448                     type == IGMPV3_CHANGE_TO_EXCLUDE;
 449
 450         stotal = scount = 0;
 451
 452         psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
 453
 454         if (!*psf_list)
 455                 goto empty_source;
 456
 457         pih = skb ? igmpv3_report_hdr(skb) : NULL;
 458
 459         /* EX and TO_EX get a fresh packet, if needed */
 460         if (truncate) {
 461                 if (pih && pih->ngrec &&
 462                     AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
 463                         if (skb)
 464                                 igmpv3_sendpack(skb);
 465                         skb = igmpv3_newpack(dev, dev->mtu);
 466                 }
 467         }
 468         first = 1;
 469         psf_prev = NULL;
 470         for (psf = *psf_list; psf; psf = psf_next) {
 471                 __be32 *psrc;
 472
 473                 psf_next = psf->sf_next;
 474
 475                 if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
 476                         psf_prev = psf;
 477                         continue;
 478                 }
 479
 480                 /* Based on RFC3376 5.1. Should not send source-list change
 481                  * records when there is a filter mode change.
 482                  */
 483                 if (((gdeleted && pmc->sfmode == MCAST_EXCLUDE) ||
 484                      (!gdeleted && pmc->crcount)) &&
 485                     (type == IGMPV3_ALLOW_NEW_SOURCES ||
 486                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount)
 487                         goto decrease_sf_crcount;
 488
 489                 /* clear marks on query responses */
 490                 if (isquery)
 491                         psf->sf_gsresp = 0;
 492
 493                 if (AVAILABLE(skb) < sizeof(__be32) +
 494                     first*sizeof(struct igmpv3_grec)) {
 495                         if (truncate && !first)
 496                                 break;   /* truncate these */
 497                         if (pgr)
 498                                 pgr->grec_nsrcs = htons(scount);
 499                         if (skb)
 500                                 igmpv3_sendpack(skb);
 501                         skb = igmpv3_newpack(dev, dev->mtu);
 502                         first = 1;
 503                         scount = 0;
 504                 }
 505                 if (first) {
 506                         skb = add_grhead(skb, pmc, type, &pgr);
 507                         first = 0;
 508                 }
 509                 if (!skb)
 510                         return NULL;
 511                 psrc = (__be32 *)skb_put(skb, sizeof(__be32));
 512                 *psrc = psf->sf_inaddr;
 513                 scount++; stotal++;
 514                 if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
 515                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
 516 decrease_sf_crcount:
 517                         psf->sf_crcount--;
 518                         if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
 519                                 if (psf_prev)
 520                                         psf_prev->sf_next = psf->sf_next;
 521                                 else
 522                                         *psf_list = psf->sf_next;
 523                                 kfree(psf);
 524                                 continue;
 525                         }
 526                 }
 527                 psf_prev = psf;
 528         }
 529
 530 empty_source:
 531         if (!stotal) {
 532                 if (type == IGMPV3_ALLOW_NEW_SOURCES ||
 533                     type == IGMPV3_BLOCK_OLD_SOURCES)
 534                         return skb;
 535                 if (pmc->crcount || isquery) {
 536                         /* make sure we have room for group header */
 537                         if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
 538                                 igmpv3_sendpack(skb);
 539                                 skb = NULL; /* add_grhead will get a new one */
 540                         }
 541                         skb = add_grhead(skb, pmc, type, &pgr);
 542                 }
 543         }
 544         if (pgr)
 545                 pgr->grec_nsrcs = htons(scount);
 546
 547         if (isquery)
 548                 pmc->gsquery = 0;       /* clear query state on report */
 549         return skb;
 550 }
 551
 552 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
 553 {
 554         struct sk_buff *skb = NULL;
 555         struct net *net = dev_net(in_dev->dev);
 556         int type;
 557
 558         if (!pmc) {
 559                 rcu_read_lock();
 560                 for_each_pmc_rcu(in_dev, pmc) {
 561                         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 562                                 continue;
 563                         if (ipv4_is_local_multicast(pmc->multiaddr) &&
 564                              !net->ipv4.sysctl_igmp_llm_reports)
 565                                 continue;
 566                         spin_lock_bh(&pmc->lock);
 567                         if (pmc->sfcount[MCAST_EXCLUDE])
 568                                 type = IGMPV3_MODE_IS_EXCLUDE;
 569                         else
 570                                 type = IGMPV3_MODE_IS_INCLUDE;
 571                         skb = add_grec(skb, pmc, type, 0, 0);
 572                         spin_unlock_bh(&pmc->lock);
 573                 }
 574                 rcu_read_unlock();
 575         } else {
 576                 spin_lock_bh(&pmc->lock);
 577                 if (pmc->sfcount[MCAST_EXCLUDE])
 578                         type = IGMPV3_MODE_IS_EXCLUDE;
 579                 else
 580                         type = IGMPV3_MODE_IS_INCLUDE;
 581                 skb = add_grec(skb, pmc, type, 0, 0);
 582                 spin_unlock_bh(&pmc->lock);
 583         }
 584         if (!skb)
 585                 return 0;
 586         return igmpv3_sendpack(skb);
 587 }
 588
 589 /*
 590  * remove zero-count source records from a source filter list
 591  */
 592 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
 593 {
 594         struct ip_sf_list *psf_prev, *psf_next, *psf;
 595
 596         psf_prev = NULL;
 597         for (psf = *ppsf; psf; psf = psf_next) {
 598                 psf_next = psf->sf_next;
 599                 if (psf->sf_crcount == 0) {
 600                         if (psf_prev)
 601                                 psf_prev->sf_next = psf->sf_next;
 602                         else
 603                                 *ppsf = psf->sf_next;
 604                         kfree(psf);
 605                 } else
 606                         psf_prev = psf;
 607         }
 608 }
 609
 610 static void igmpv3_send_cr(struct in_device *in_dev)
 611 {
 612         struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
 613         struct sk_buff *skb = NULL;
 614         int type, dtype;
 615
 616         rcu_read_lock();
 617         spin_lock_bh(&in_dev->mc_tomb_lock);
 618
 619         /* deleted MCA's */
 620         pmc_prev = NULL;
 621         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
 622                 pmc_next = pmc->next;
 623                 if (pmc->sfmode == MCAST_INCLUDE) {
 624                         type = IGMPV3_BLOCK_OLD_SOURCES;
 625                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 626                         skb = add_grec(skb, pmc, type, 1, 0);
 627                         skb = add_grec(skb, pmc, dtype, 1, 1);
 628                 }
 629                 if (pmc->crcount) {
 630                         if (pmc->sfmode == MCAST_EXCLUDE) {
 631                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 632                                 skb = add_grec(skb, pmc, type, 1, 0);
 633                         }
 634                         pmc->crcount--;
 635                         if (pmc->crcount == 0) {
 636                                 igmpv3_clear_zeros(&pmc->tomb);
 637                                 igmpv3_clear_zeros(&pmc->sources);
 638                         }
 639                 }
 640                 if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
 641                         if (pmc_prev)
 642                                 pmc_prev->next = pmc_next;
 643                         else
 644                                 in_dev->mc_tomb = pmc_next;
 645                         in_dev_put(pmc->interface);
 646                         kfree(pmc);
 647                 } else
 648                         pmc_prev = pmc;
 649         }
 650         spin_unlock_bh(&in_dev->mc_tomb_lock);
 651
 652         /* change recs */
 653         for_each_pmc_rcu(in_dev, pmc) {
 654                 spin_lock_bh(&pmc->lock);
 655                 if (pmc->sfcount[MCAST_EXCLUDE]) {
 656                         type = IGMPV3_BLOCK_OLD_SOURCES;
 657                         dtype = IGMPV3_ALLOW_NEW_SOURCES;
 658                 } else {
 659                         type = IGMPV3_ALLOW_NEW_SOURCES;
 660                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 661                 }
 662                 skb = add_grec(skb, pmc, type, 0, 0);
 663                 skb = add_grec(skb, pmc, dtype, 0, 1);  /* deleted sources */
 664
 665                 /* filter mode changes */
 666                 if (pmc->crcount) {
 667                         if (pmc->sfmode == MCAST_EXCLUDE)
 668                                 type = IGMPV3_CHANGE_TO_EXCLUDE;
 669                         else
 670                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 671                         skb = add_grec(skb, pmc, type, 0, 0);
 672                         pmc->crcount--;
 673                 }
 674                 spin_unlock_bh(&pmc->lock);
 675         }
 676         rcu_read_unlock();
 677
 678         if (!skb)
 679                 return;
 680         (void) igmpv3_sendpack(skb);
 681 }
 682
 683 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
 684         int type)
 685 {
 686         struct sk_buff *skb;
 687         struct iphdr *iph;
 688         struct igmphdr *ih;
 689         struct rtable *rt;
 690         struct net_device *dev = in_dev->dev;
 691         struct net *net = dev_net(dev);
 692         __be32  group = pmc ? pmc->multiaddr : 0;
 693         struct flowi4 fl4;
 694         __be32  dst;
 695         int hlen, tlen;
 696
 697         if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
 698                 return igmpv3_send_report(in_dev, pmc);
 699
 700         if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
 701                 return 0;
 702
 703         if (type == IGMP_HOST_LEAVE_MESSAGE)
 704                 dst = IGMP_ALL_ROUTER;
 705         else
 706                 dst = group;
 707
 708         rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
 709                                    0, 0,
 710                                    IPPROTO_IGMP, 0, dev->ifindex);
 711         if (IS_ERR(rt))
 712                 return -1;
 713
 714         hlen = LL_RESERVED_SPACE(dev);
 715         tlen = dev->needed_tailroom;
 716         skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
 717         if (!skb) {
 718                 ip_rt_put(rt);
 719                 return -1;
 720         }
 721         skb->priority = TC_PRIO_CONTROL;
 722
 723         skb_dst_set(skb, &rt->dst);
 724
 725         skb_reserve(skb, hlen);
 726
 727         skb_reset_network_header(skb);
 728         iph = ip_hdr(skb);
 729         skb_put(skb, sizeof(struct iphdr) + 4);
 730
 731         iph->version  = 4;
 732         iph->ihl      = (sizeof(struct iphdr)+4)>>2;
 733         iph->tos      = 0xc0;
 734         iph->frag_off = htons(IP_DF);
 735         iph->ttl      = 1;
 736         iph->daddr    = dst;
 737         iph->saddr    = fl4.saddr;
 738         iph->protocol = IPPROTO_IGMP;
 739         ip_select_ident(net, skb, NULL);
 740         ((u8 *)&iph[1])[0] = IPOPT_RA;
 741         ((u8 *)&iph[1])[1] = 4;
 742         ((u8 *)&iph[1])[2] = 0;
 743         ((u8 *)&iph[1])[3] = 0;
 744
 745         ih = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
 746         ih->type = type;
 747         ih->code = 0;
 748         ih->csum = 0;
 749         ih->group = group;
 750         ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
 751
 752         return ip_local_out(net, skb->sk, skb);
 753 }
 754
 755 static void igmp_gq_timer_expire(unsigned long data)
 756 {
 757         struct in_device *in_dev = (struct in_device *)data;
 758
 759         in_dev->mr_gq_running = 0;
 760         igmpv3_send_report(in_dev, NULL);
 761         in_dev_put(in_dev);
 762 }
 763
 764 static void igmp_ifc_timer_expire(unsigned long data)
 765 {
 766         struct in_device *in_dev = (struct in_device *)data;
 767
 768         igmpv3_send_cr(in_dev);
 769         if (in_dev->mr_ifc_count) {
 770                 in_dev->mr_ifc_count--;
 771                 igmp_ifc_start_timer(in_dev,
 772                                      unsolicited_report_interval(in_dev));
 773         }
 774         in_dev_put(in_dev);
 775 }
 776
 777 static void igmp_ifc_event(struct in_device *in_dev)
 778 {
 779         struct net *net = dev_net(in_dev->dev);
 780         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 781                 return;
 782         in_dev->mr_ifc_count = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
 783         igmp_ifc_start_timer(in_dev, 1);
 784 }
 785
 786
 787 static void igmp_timer_expire(unsigned long data)
 788 {
 789         struct ip_mc_list *im = (struct ip_mc_list *)data;
 790         struct in_device *in_dev = im->interface;
 791
 792         spin_lock(&im->lock);
 793         im->tm_running = 0;
 794
 795         if (im->unsolicit_count) {
 796                 im->unsolicit_count--;
 797                 igmp_start_timer(im, unsolicited_report_interval(in_dev));
 798         }
 799         im->reporter = 1;
 800         spin_unlock(&im->lock);
 801
 802         if (IGMP_V1_SEEN(in_dev))
 803                 igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
 804         else if (IGMP_V2_SEEN(in_dev))
 805                 igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
 806         else
 807                 igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
 808
 809         ip_ma_put(im);
 810 }
 811
 812 /* mark EXCLUDE-mode sources */
 813 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 814 {
 815         struct ip_sf_list *psf;
 816         int i, scount;
 817
 818         scount = 0;
 819         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 820                 if (scount == nsrcs)
 821                         break;
 822                 for (i = 0; i < nsrcs; i++) {
 823                         /* skip inactive filters */
 824                         if (psf->sf_count[MCAST_INCLUDE] ||
 825                             pmc->sfcount[MCAST_EXCLUDE] !=
 826                             psf->sf_count[MCAST_EXCLUDE])
 827                                 break;
 828                         if (srcs[i] == psf->sf_inaddr) {
 829                                 scount++;
 830                                 break;
 831                         }
 832                 }
 833         }
 834         pmc->gsquery = 0;
 835         if (scount == nsrcs)    /* all sources excluded */
 836                 return 0;
 837         return 1;
 838 }
 839
 840 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 841 {
 842         struct ip_sf_list *psf;
 843         int i, scount;
 844
 845         if (pmc->sfmode == MCAST_EXCLUDE)
 846                 return igmp_xmarksources(pmc, nsrcs, srcs);
 847
 848         /* mark INCLUDE-mode sources */
 849         scount = 0;
 850         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 851                 if (scount == nsrcs)
 852                         break;
 853                 for (i = 0; i < nsrcs; i++)
 854                         if (srcs[i] == psf->sf_inaddr) {
 855                                 psf->sf_gsresp = 1;
 856                                 scount++;
 857                                 break;
 858                         }
 859         }
 860         if (!scount) {
 861                 pmc->gsquery = 0;
 862                 return 0;
 863         }
 864         pmc->gsquery = 1;
 865         return 1;
 866 }
 867
 868 /* return true if packet was dropped */
 869 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
 870 {
 871         struct ip_mc_list *im;
 872         struct net *net = dev_net(in_dev->dev);
 873
 874         /* Timers are only set for non-local groups */
 875
 876         if (group == IGMP_ALL_HOSTS)
 877                 return false;
 878         if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
 879                 return false;
 880
 881         rcu_read_lock();
 882         for_each_pmc_rcu(in_dev, im) {
 883                 if (im->multiaddr == group) {
 884                         igmp_stop_timer(im);
 885                         break;
 886                 }
 887         }
 888         rcu_read_unlock();
 889         return false;
 890 }
 891
 892 /* return true if packet was dropped */
 893 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
 894         int len)
 895 {
 896         struct igmphdr          *ih = igmp_hdr(skb);
 897         struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
 898         struct ip_mc_list       *im;
 899         __be32                  group = ih->group;
 900         int                     max_delay;
 901         int                     mark = 0;
 902         struct net              *net = dev_net(in_dev->dev);
 903
 904
 905         if (len == 8) {
 906                 if (ih->code == 0) {
 907                         /* Alas, old v1 router presents here. */
 908
 909                         max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 910                         in_dev->mr_v1_seen = jiffies +
 911                                 IGMP_V1_ROUTER_PRESENT_TIMEOUT;
 912                         group = 0;
 913                 } else {
 914                         /* v2 router present */
 915                         max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
 916                         in_dev->mr_v2_seen = jiffies +
 917                                 IGMP_V2_ROUTER_PRESENT_TIMEOUT;
 918                 }
 919                 /* cancel the interface change timer */
 920                 in_dev->mr_ifc_count = 0;
 921                 if (del_timer(&in_dev->mr_ifc_timer))
 922                         __in_dev_put(in_dev);
 923                 /* clear deleted report items */
 924                 igmpv3_clear_delrec(in_dev);
 925         } else if (len < 12) {
 926                 return true;    /* ignore bogus packet; freed by caller */
 927         } else if (IGMP_V1_SEEN(in_dev)) {
 928                 /* This is a v3 query with v1 queriers present */
 929                 max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 930                 group = 0;
 931         } else if (IGMP_V2_SEEN(in_dev)) {
 932                 /* this is a v3 query with v2 queriers present;
 933                  * Interpretation of the max_delay code is problematic here.
 934                  * A real v2 host would use ih_code directly, while v3 has a
 935                  * different encoding. We use the v3 encoding as more likely
 936                  * to be intended in a v3 query.
 937                  */
 938                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 939                 if (!max_delay)
 940                         max_delay = 1;  /* can't mod w/ 0 */
 941         } else { /* v3 */
 942                 if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
 943                         return true;
 944
 945                 ih3 = igmpv3_query_hdr(skb);
 946                 if (ih3->nsrcs) {
 947                         if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
 948                                            + ntohs(ih3->nsrcs)*sizeof(__be32)))
 949                                 return true;
 950                         ih3 = igmpv3_query_hdr(skb);
 951                 }
 952
 953                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 954                 if (!max_delay)
 955                         max_delay = 1;  /* can't mod w/ 0 */
 956                 in_dev->mr_maxdelay = max_delay;
 957                 if (ih3->qrv)
 958                         in_dev->mr_qrv = ih3->qrv;
 959                 if (!group) { /* general query */
 960                         if (ih3->nsrcs)
 961                                 return true;    /* no sources allowed */
 962                         igmp_gq_start_timer(in_dev);
 963                         return false;
 964                 }
 965                 /* mark sources to include, if group & source-specific */
 966                 mark = ih3->nsrcs != 0;
 967         }
 968
 969         /*
 970          * - Start the timers in all of our membership records
 971          *   that the query applies to for the interface on
 972          *   which the query arrived excl. those that belong
 973          *   to a "local" group (224.0.0.X)
 974          * - For timers already running check if they need to
 975          *   be reset.
 976          * - Use the igmp->igmp_code field as the maximum
 977          *   delay possible
 978          */
 979         rcu_read_lock();
 980         for_each_pmc_rcu(in_dev, im) {
 981                 int changed;
 982
 983                 if (group && group != im->multiaddr)
 984                         continue;
 985                 if (im->multiaddr == IGMP_ALL_HOSTS)
 986                         continue;
 987                 if (ipv4_is_local_multicast(im->multiaddr) &&
 988                     !net->ipv4.sysctl_igmp_llm_reports)
 989                         continue;
 990                 spin_lock_bh(&im->lock);
 991                 if (im->tm_running)
 992                         im->gsquery = im->gsquery && mark;
 993                 else
 994                         im->gsquery = mark;
 995                 changed = !im->gsquery ||
 996                         igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
 997                 spin_unlock_bh(&im->lock);
 998                 if (changed)
 999                         igmp_mod_timer(im, max_delay);
1000         }
1001         rcu_read_unlock();
1002         return false;
1003 }
1004
1005 /* called in rcu_read_lock() section */
1006 int igmp_rcv(struct sk_buff *skb)
1007 {
1008         /* This basically follows the spec line by line -- see RFC1112 */
1009         struct igmphdr *ih;
1010         struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
1011         int len = skb->len;
1012         bool dropped = true;
1013
1014         if (!in_dev)
1015                 goto drop;
1016
1017         if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
1018                 goto drop;
1019
1020         if (skb_checksum_simple_validate(skb))
1021                 goto drop;
1022
1023         ih = igmp_hdr(skb);
1024         switch (ih->type) {
1025         case IGMP_HOST_MEMBERSHIP_QUERY:
1026                 dropped = igmp_heard_query(in_dev, skb, len);
1027                 break;
1028         case IGMP_HOST_MEMBERSHIP_REPORT:
1029         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1030                 /* Is it our report looped back? */
1031                 if (rt_is_output_route(skb_rtable(skb)))
1032                         break;
1033                 /* don't rely on MC router hearing unicast reports */
1034                 if (skb->pkt_type == PACKET_MULTICAST ||
1035                     skb->pkt_type == PACKET_BROADCAST)
1036                         dropped = igmp_heard_report(in_dev, ih->group);
1037                 break;
1038         case IGMP_PIM:
1039 #ifdef CONFIG_IP_PIMSM_V1
1040                 return pim_rcv_v1(skb);
1041 #endif
1042         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1043         case IGMP_DVMRP:
1044         case IGMP_TRACE:
1045         case IGMP_HOST_LEAVE_MESSAGE:
1046         case IGMP_MTRACE:
1047         case IGMP_MTRACE_RESP:
1048                 break;
1049         default:
1050                 break;
1051         }
1052
1053 drop:
1054         if (dropped)
1055                 kfree_skb(skb);
1056         else
1057                 consume_skb(skb);
1058         return 0;
1059 }
1060
1061 #endif
1062
1063
1064 /*
1065  *      Add a filter to a device
1066  */
1067
1068 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1069 {
1070         char buf[MAX_ADDR_LEN];
1071         struct net_device *dev = in_dev->dev;
1072
1073         /* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1074            We will get multicast token leakage, when IFF_MULTICAST
1075            is changed. This check should be done in ndo_set_rx_mode
1076            routine. Something sort of:
1077            if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1078            --ANK
1079            */
1080         if (arp_mc_map(addr, buf, dev, 0) == 0)
1081                 dev_mc_add(dev, buf);
1082 }
1083
1084 /*
1085  *      Remove a filter from a device
1086  */
1087
1088 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1089 {
1090         char buf[MAX_ADDR_LEN];
1091         struct net_device *dev = in_dev->dev;
1092
1093         if (arp_mc_map(addr, buf, dev, 0) == 0)
1094                 dev_mc_del(dev, buf);
1095 }
1096
1097 #ifdef CONFIG_IP_MULTICAST
1098 /*
1099  * deleted ip_mc_list manipulation
1100  */
1101 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1102 {
1103         struct ip_mc_list *pmc;
1104         struct net *net = dev_net(in_dev->dev);
1105
1106         /* this is an "ip_mc_list" for convenience; only the fields below
1107          * are actually used. In particular, the refcnt and users are not
1108          * used for management of the delete list. Using the same structure
1109          * for deleted items allows change reports to use common code with
1110          * non-deleted or query-response MCA's.
1111          */
1112         pmc = kzalloc(sizeof(*pmc), GFP_KERNEL);
1113         if (!pmc)
1114                 return;
1115         spin_lock_bh(&im->lock);
1116         pmc->interface = im->interface;
1117         in_dev_hold(in_dev);
1118         pmc->multiaddr = im->multiaddr;
1119         pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1120         pmc->sfmode = im->sfmode;
1121         if (pmc->sfmode == MCAST_INCLUDE) {
1122                 struct ip_sf_list *psf;
1123
1124                 pmc->tomb = im->tomb;
1125                 pmc->sources = im->sources;
1126                 im->tomb = im->sources = NULL;
1127                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1128                         psf->sf_crcount = pmc->crcount;
1129         }
1130         spin_unlock_bh(&im->lock);
1131
1132         spin_lock_bh(&in_dev->mc_tomb_lock);
1133         pmc->next = in_dev->mc_tomb;
1134         in_dev->mc_tomb = pmc;
1135         spin_unlock_bh(&in_dev->mc_tomb_lock);
1136 }
1137
1138 /*
1139  * restore ip_mc_list deleted records
1140  */
1141 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1142 {
1143         struct ip_mc_list *pmc, *pmc_prev;
1144         struct ip_sf_list *psf;
1145         struct net *net = dev_net(in_dev->dev);
1146         __be32 multiaddr = im->multiaddr;
1147
1148         spin_lock_bh(&in_dev->mc_tomb_lock);
1149         pmc_prev = NULL;
1150         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1151                 if (pmc->multiaddr == multiaddr)
1152                         break;
1153                 pmc_prev = pmc;
1154         }
1155         if (pmc) {
1156                 if (pmc_prev)
1157                         pmc_prev->next = pmc->next;
1158                 else
1159                         in_dev->mc_tomb = pmc->next;
1160         }
1161         spin_unlock_bh(&in_dev->mc_tomb_lock);
1162
1163         spin_lock_bh(&im->lock);
1164         if (pmc) {
1165                 im->interface = pmc->interface;
1166                 im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1167                 im->sfmode = pmc->sfmode;
1168                 if (pmc->sfmode == MCAST_INCLUDE) {
1169                         im->tomb = pmc->tomb;
1170                         im->sources = pmc->sources;
1171                         for (psf = im->sources; psf; psf = psf->sf_next)
1172                                 psf->sf_crcount = im->crcount;
1173                 }
1174                 in_dev_put(pmc->interface);
1175                 kfree(pmc);
1176         }
1177         spin_unlock_bh(&im->lock);
1178 }
1179
1180 /*
1181  * flush ip_mc_list deleted records
1182  */
1183 static void igmpv3_clear_delrec(struct in_device *in_dev)
1184 {
1185         struct ip_mc_list *pmc, *nextpmc;
1186
1187         spin_lock_bh(&in_dev->mc_tomb_lock);
1188         pmc = in_dev->mc_tomb;
1189         in_dev->mc_tomb = NULL;
1190         spin_unlock_bh(&in_dev->mc_tomb_lock);
1191
1192         for (; pmc; pmc = nextpmc) {
1193                 nextpmc = pmc->next;
1194                 ip_mc_clear_src(pmc);
1195                 in_dev_put(pmc->interface);
1196                 kfree(pmc);
1197         }
1198         /* clear dead sources, too */
1199         rcu_read_lock();
1200         for_each_pmc_rcu(in_dev, pmc) {
1201                 struct ip_sf_list *psf, *psf_next;
1202
1203                 spin_lock_bh(&pmc->lock);
1204                 psf = pmc->tomb;
1205                 pmc->tomb = NULL;
1206                 spin_unlock_bh(&pmc->lock);
1207                 for (; psf; psf = psf_next) {
1208                         psf_next = psf->sf_next;
1209                         kfree(psf);
1210                 }
1211         }
1212         rcu_read_unlock();
1213 }
1214 #endif
1215
1216 static void igmp_group_dropped(struct ip_mc_list *im)
1217 {
1218         struct in_device *in_dev = im->interface;
1219 #ifdef CONFIG_IP_MULTICAST
1220         struct net *net = dev_net(in_dev->dev);
1221         int reporter;
1222 #endif
1223
1224         if (im->loaded) {
1225                 im->loaded = 0;
1226                 ip_mc_filter_del(in_dev, im->multiaddr);
1227         }
1228
1229 #ifdef CONFIG_IP_MULTICAST
1230         if (im->multiaddr == IGMP_ALL_HOSTS)
1231                 return;
1232         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1233                 return;
1234
1235         reporter = im->reporter;
1236         igmp_stop_timer(im);
1237
1238         if (!in_dev->dead) {
1239                 if (IGMP_V1_SEEN(in_dev))
1240                         return;
1241                 if (IGMP_V2_SEEN(in_dev)) {
1242                         if (reporter)
1243                                 igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1244                         return;
1245                 }
1246                 /* IGMPv3 */
1247                 igmpv3_add_delrec(in_dev, im);
1248
1249                 igmp_ifc_event(in_dev);
1250         }
1251 #endif
1252 }
1253
1254 static void igmp_group_added(struct ip_mc_list *im)
1255 {
1256         struct in_device *in_dev = im->interface;
1257 #ifdef CONFIG_IP_MULTICAST
1258         struct net *net = dev_net(in_dev->dev);
1259 #endif
1260
1261         if (im->loaded == 0) {
1262                 im->loaded = 1;
1263                 ip_mc_filter_add(in_dev, im->multiaddr);
1264         }
1265
1266 #ifdef CONFIG_IP_MULTICAST
1267         if (im->multiaddr == IGMP_ALL_HOSTS)
1268                 return;
1269         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1270                 return;
1271
1272         if (in_dev->dead)
1273                 return;
1274         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1275                 spin_lock_bh(&im->lock);
1276                 igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1277                 spin_unlock_bh(&im->lock);
1278                 return;
1279         }
1280         /* else, v3 */
1281
1282         im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1283         igmp_ifc_event(in_dev);
1284 #endif
1285 }
1286
1287
1288 /*
1289  *      Multicast list managers
1290  */
1291
1292 static u32 ip_mc_hash(const struct ip_mc_list *im)
1293 {
1294         return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1295 }
1296
1297 static void ip_mc_hash_add(struct in_device *in_dev,
1298                            struct ip_mc_list *im)
1299 {
1300         struct ip_mc_list __rcu **mc_hash;
1301         u32 hash;
1302
1303         mc_hash = rtnl_dereference(in_dev->mc_hash);
1304         if (mc_hash) {
1305                 hash = ip_mc_hash(im);
1306                 im->next_hash = mc_hash[hash];
1307                 rcu_assign_pointer(mc_hash[hash], im);
1308                 return;
1309         }
1310
1311         /* do not use a hash table for small number of items */
1312         if (in_dev->mc_count < 4)
1313                 return;
1314
1315         mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1316                           GFP_KERNEL);
1317         if (!mc_hash)
1318                 return;
1319
1320         for_each_pmc_rtnl(in_dev, im) {
1321                 hash = ip_mc_hash(im);
1322                 im->next_hash = mc_hash[hash];
1323                 RCU_INIT_POINTER(mc_hash[hash], im);
1324         }
1325
1326         rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1327 }
1328
1329 static void ip_mc_hash_remove(struct in_device *in_dev,
1330                               struct ip_mc_list *im)
1331 {
1332         struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1333         struct ip_mc_list *aux;
1334
1335         if (!mc_hash)
1336                 return;
1337         mc_hash += ip_mc_hash(im);
1338         while ((aux = rtnl_dereference(*mc_hash)) != im)
1339                 mc_hash = &aux->next_hash;
1340         *mc_hash = im->next_hash;
1341 }
1342
1343
1344 /*
1345  *      A socket has joined a multicast group on device dev.
1346  */
1347
1348 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1349 {
1350         struct ip_mc_list *im;
1351 #ifdef CONFIG_IP_MULTICAST
1352         struct net *net = dev_net(in_dev->dev);
1353 #endif
1354
1355         ASSERT_RTNL();
1356
1357         for_each_pmc_rtnl(in_dev, im) {
1358                 if (im->multiaddr == addr) {
1359                         im->users++;
1360                         ip_mc_add_src(in_dev, &addr, MCAST_EXCLUDE, 0, NULL, 0);
1361                         goto out;
1362                 }
1363         }
1364
1365         im = kzalloc(sizeof(*im), GFP_KERNEL);
1366         if (!im)
1367                 goto out;
1368
1369         im->users = 1;
1370         im->interface = in_dev;
1371         in_dev_hold(in_dev);
1372         im->multiaddr = addr;
1373         /* initial mode is (EX, empty) */
1374         im->sfmode = MCAST_EXCLUDE;
1375         im->sfcount[MCAST_EXCLUDE] = 1;
1376         atomic_set(&im->refcnt, 1);
1377         spin_lock_init(&im->lock);
1378 #ifdef CONFIG_IP_MULTICAST
1379         setup_timer(&im->timer, igmp_timer_expire, (unsigned long)im);
1380         im->unsolicit_count = net->ipv4.sysctl_igmp_qrv;
1381 #endif
1382
1383         im->next_rcu = in_dev->mc_list;
1384         in_dev->mc_count++;
1385         rcu_assign_pointer(in_dev->mc_list, im);
1386
1387         ip_mc_hash_add(in_dev, im);
1388
1389 #ifdef CONFIG_IP_MULTICAST
1390         igmpv3_del_delrec(in_dev, im);
1391 #endif
1392         igmp_group_added(im);
1393         if (!in_dev->dead)
1394                 ip_rt_multicast_event(in_dev);
1395 out:
1396         return;
1397 }
1398 EXPORT_SYMBOL(ip_mc_inc_group);
1399
1400 static int ip_mc_check_iphdr(struct sk_buff *skb)
1401 {
1402         const struct iphdr *iph;
1403         unsigned int len;
1404         unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1405
1406         if (!pskb_may_pull(skb, offset))
1407                 return -EINVAL;
1408
1409         iph = ip_hdr(skb);
1410
1411         if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1412                 return -EINVAL;
1413
1414         offset += ip_hdrlen(skb) - sizeof(*iph);
1415
1416         if (!pskb_may_pull(skb, offset))
1417                 return -EINVAL;
1418
1419         iph = ip_hdr(skb);
1420
1421         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1422                 return -EINVAL;
1423
1424         len = skb_network_offset(skb) + ntohs(iph->tot_len);
1425         if (skb->len < len || len < offset)
1426                 return -EINVAL;
1427
1428         skb_set_transport_header(skb, offset);
1429
1430         return 0;
1431 }
1432
1433 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1434 {
1435         unsigned int len = skb_transport_offset(skb);
1436
1437         len += sizeof(struct igmpv3_report);
1438
1439         return pskb_may_pull(skb, len) ? 0 : -EINVAL;
1440 }
1441
1442 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1443 {
1444         unsigned int len = skb_transport_offset(skb);
1445
1446         len += sizeof(struct igmphdr);
1447         if (skb->len < len)
1448                 return -EINVAL;
1449
1450         /* IGMPv{1,2}? */
1451         if (skb->len != len) {
1452                 /* or IGMPv3? */
1453                 len += sizeof(struct igmpv3_query) - sizeof(struct igmphdr);
1454                 if (skb->len < len || !pskb_may_pull(skb, len))
1455                         return -EINVAL;
1456         }
1457
1458         /* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1459          * all-systems destination addresses (224.0.0.1) for general queries
1460          */
1461         if (!igmp_hdr(skb)->group &&
1462             ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1463                 return -EINVAL;
1464
1465         return 0;
1466 }
1467
1468 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1469 {
1470         switch (igmp_hdr(skb)->type) {
1471         case IGMP_HOST_LEAVE_MESSAGE:
1472         case IGMP_HOST_MEMBERSHIP_REPORT:
1473         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1474                 /* fall through */
1475                 return 0;
1476         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1477                 return ip_mc_check_igmp_reportv3(skb);
1478         case IGMP_HOST_MEMBERSHIP_QUERY:
1479                 return ip_mc_check_igmp_query(skb);
1480         default:
1481                 return -ENOMSG;
1482         }
1483 }
1484
1485 static inline __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1486 {
1487         return skb_checksum_simple_validate(skb);
1488 }
1489
1490 static int __ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1491
1492 {
1493         struct sk_buff *skb_chk;
1494         unsigned int transport_len;
1495         unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1496         int ret = -EINVAL;
1497
1498         transport_len = ntohs(ip_hdr(skb)->tot_len) - ip_hdrlen(skb);
1499
1500         skb_chk = skb_checksum_trimmed(skb, transport_len,
1501                                        ip_mc_validate_checksum);
1502         if (!skb_chk)
1503                 goto err;
1504
1505         if (!pskb_may_pull(skb_chk, len))
1506                 goto err;
1507
1508         ret = ip_mc_check_igmp_msg(skb_chk);
1509         if (ret)
1510                 goto err;
1511
1512         if (skb_trimmed)
1513                 *skb_trimmed = skb_chk;
1514         /* free now unneeded clone */
1515         else if (skb_chk != skb)
1516                 kfree_skb(skb_chk);
1517
1518         ret = 0;
1519
1520 err:
1521         if (ret && skb_chk && skb_chk != skb)
1522                 kfree_skb(skb_chk);
1523
1524         return ret;
1525 }
1526
1527 /**
1528  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1529  * @skb: the skb to validate
1530  * @skb_trimmed: to store an skb pointer trimmed to IPv4 packet tail (optional)
1531  *
1532  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1533  * skb transport header accordingly and returns zero.
1534  *
1535  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1536  *  standard
1537  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1538  * -ENOMEM: A memory allocation failure happened.
1539  *
1540  * Optionally, an skb pointer might be provided via skb_trimmed (or set it
1541  * to NULL): After parsing an IGMP packet successfully it will point to
1542  * an skb which has its tail aligned to the IP packet end. This might
1543  * either be the originally provided skb or a trimmed, cloned version if
1544  * the skb frame had data beyond the IP packet. A cloned skb allows us
1545  * to leave the original skb and its full frame unchanged (which might be
1546  * desirable for layer 2 frame jugglers).
1547  *
1548  * Caller needs to set the skb network header and free any returned skb if it
1549  * differs from the provided skb.
1550  */
1551 int ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1552 {
1553         int ret = ip_mc_check_iphdr(skb);
1554
1555         if (ret < 0)
1556                 return ret;
1557
1558         if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1559                 return -ENOMSG;
1560
1561         return __ip_mc_check_igmp(skb, skb_trimmed);
1562 }
1563 EXPORT_SYMBOL(ip_mc_check_igmp);
1564
1565 /*
1566  *      Resend IGMP JOIN report; used by netdev notifier.
1567  */
1568 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1569 {
1570 #ifdef CONFIG_IP_MULTICAST
1571         struct ip_mc_list *im;
1572         int type;
1573         struct net *net = dev_net(in_dev->dev);
1574
1575         ASSERT_RTNL();
1576
1577         for_each_pmc_rtnl(in_dev, im) {
1578                 if (im->multiaddr == IGMP_ALL_HOSTS)
1579                         continue;
1580                 if (ipv4_is_local_multicast(im->multiaddr) &&
1581                     !net->ipv4.sysctl_igmp_llm_reports)
1582                         continue;
1583
1584                 /* a failover is happening and switches
1585                  * must be notified immediately
1586                  */
1587                 if (IGMP_V1_SEEN(in_dev))
1588                         type = IGMP_HOST_MEMBERSHIP_REPORT;
1589                 else if (IGMP_V2_SEEN(in_dev))
1590                         type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1591                 else
1592                         type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1593                 igmp_send_report(in_dev, im, type);
1594         }
1595 #endif
1596 }
1597
1598 /*
1599  *      A socket has left a multicast group on device dev
1600  */
1601
1602 void ip_mc_dec_group(struct in_device *in_dev, __be32 addr)
1603 {
1604         struct ip_mc_list *i;
1605         struct ip_mc_list __rcu **ip;
1606
1607         ASSERT_RTNL();
1608
1609         for (ip = &in_dev->mc_list;
1610              (i = rtnl_dereference(*ip)) != NULL;
1611              ip = &i->next_rcu) {
1612                 if (i->multiaddr == addr) {
1613                         if (--i->users == 0) {
1614                                 ip_mc_hash_remove(in_dev, i);
1615                                 *ip = i->next_rcu;
1616                                 in_dev->mc_count--;
1617                                 igmp_group_dropped(i);
1618                                 ip_mc_clear_src(i);
1619
1620                                 if (!in_dev->dead)
1621                                         ip_rt_multicast_event(in_dev);
1622
1623                                 ip_ma_put(i);
1624                                 return;
1625                         }
1626                         break;
1627                 }
1628         }
1629 }
1630 EXPORT_SYMBOL(ip_mc_dec_group);
1631
1632 /* Device changing type */
1633
1634 void ip_mc_unmap(struct in_device *in_dev)
1635 {
1636         struct ip_mc_list *pmc;
1637
1638         ASSERT_RTNL();
1639
1640         for_each_pmc_rtnl(in_dev, pmc)
1641                 igmp_group_dropped(pmc);
1642 }
1643
1644 void ip_mc_remap(struct in_device *in_dev)
1645 {
1646         struct ip_mc_list *pmc;
1647
1648         ASSERT_RTNL();
1649
1650         for_each_pmc_rtnl(in_dev, pmc) {
1651 #ifdef CONFIG_IP_MULTICAST
1652                 igmpv3_del_delrec(in_dev, pmc);
1653 #endif
1654                 igmp_group_added(pmc);
1655         }
1656 }
1657
1658 /* Device going down */
1659
1660 void ip_mc_down(struct in_device *in_dev)
1661 {
1662         struct ip_mc_list *pmc;
1663
1664         ASSERT_RTNL();
1665
1666         for_each_pmc_rtnl(in_dev, pmc)
1667                 igmp_group_dropped(pmc);
1668
1669 #ifdef CONFIG_IP_MULTICAST
1670         in_dev->mr_ifc_count = 0;
1671         if (del_timer(&in_dev->mr_ifc_timer))
1672                 __in_dev_put(in_dev);
1673         in_dev->mr_gq_running = 0;
1674         if (del_timer(&in_dev->mr_gq_timer))
1675                 __in_dev_put(in_dev);
1676 #endif
1677
1678         ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1679 }
1680
1681 void ip_mc_init_dev(struct in_device *in_dev)
1682 {
1683 #ifdef CONFIG_IP_MULTICAST
1684         struct net *net = dev_net(in_dev->dev);
1685 #endif
1686         ASSERT_RTNL();
1687
1688 #ifdef CONFIG_IP_MULTICAST
1689         setup_timer(&in_dev->mr_gq_timer, igmp_gq_timer_expire,
1690                         (unsigned long)in_dev);
1691         setup_timer(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire,
1692                         (unsigned long)in_dev);
1693         in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1694 #endif
1695
1696         spin_lock_init(&in_dev->mc_tomb_lock);
1697 }
1698
1699 /* Device going up */
1700
1701 void ip_mc_up(struct in_device *in_dev)
1702 {
1703         struct ip_mc_list *pmc;
1704 #ifdef CONFIG_IP_MULTICAST
1705         struct net *net = dev_net(in_dev->dev);
1706 #endif
1707
1708         ASSERT_RTNL();
1709
1710 #ifdef CONFIG_IP_MULTICAST
1711         in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1712 #endif
1713         ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1714
1715         for_each_pmc_rtnl(in_dev, pmc) {
1716 #ifdef CONFIG_IP_MULTICAST
1717                 igmpv3_del_delrec(in_dev, pmc);
1718 #endif
1719                 igmp_group_added(pmc);
1720         }
1721 }
1722
1723 /*
1724  *      Device is about to be destroyed: clean up.
1725  */
1726
1727 void ip_mc_destroy_dev(struct in_device *in_dev)
1728 {
1729         struct ip_mc_list *i;
1730
1731         ASSERT_RTNL();
1732
1733         /* Deactivate timers */
1734         ip_mc_down(in_dev);
1735 #ifdef CONFIG_IP_MULTICAST
1736         igmpv3_clear_delrec(in_dev);
1737 #endif
1738
1739         while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1740                 in_dev->mc_list = i->next_rcu;
1741                 in_dev->mc_count--;
1742                 ip_ma_put(i);
1743         }
1744 }
1745
1746 /* RTNL is locked */
1747 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1748 {
1749         struct net_device *dev = NULL;
1750         struct in_device *idev = NULL;
1751
1752         if (imr->imr_ifindex) {
1753                 idev = inetdev_by_index(net, imr->imr_ifindex);
1754                 return idev;
1755         }
1756         if (imr->imr_address.s_addr) {
1757                 dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1758                 if (!dev)
1759                         return NULL;
1760         }
1761
1762         if (!dev) {
1763                 struct rtable *rt = ip_route_output(net,
1764                                                     imr->imr_multiaddr.s_addr,
1765                                                     0, 0, 0);
1766                 if (!IS_ERR(rt)) {
1767                         dev = rt->dst.dev;
1768                         ip_rt_put(rt);
1769                 }
1770         }
1771         if (dev) {
1772                 imr->imr_ifindex = dev->ifindex;
1773                 idev = __in_dev_get_rtnl(dev);
1774         }
1775         return idev;
1776 }
1777
1778 /*
1779  *      Join a socket to a group
1780  */
1781
1782 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1783         __be32 *psfsrc)
1784 {
1785         struct ip_sf_list *psf, *psf_prev;
1786         int rv = 0;
1787
1788         psf_prev = NULL;
1789         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1790                 if (psf->sf_inaddr == *psfsrc)
1791                         break;
1792                 psf_prev = psf;
1793         }
1794         if (!psf || psf->sf_count[sfmode] == 0) {
1795                 /* source filter not found, or count wrong =>  bug */
1796                 return -ESRCH;
1797         }
1798         psf->sf_count[sfmode]--;
1799         if (psf->sf_count[sfmode] == 0) {
1800                 ip_rt_multicast_event(pmc->interface);
1801         }
1802         if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1803 #ifdef CONFIG_IP_MULTICAST
1804                 struct in_device *in_dev = pmc->interface;
1805                 struct net *net = dev_net(in_dev->dev);
1806 #endif
1807
1808                 /* no more filters for this source */
1809                 if (psf_prev)
1810                         psf_prev->sf_next = psf->sf_next;
1811                 else
1812                         pmc->sources = psf->sf_next;
1813 #ifdef CONFIG_IP_MULTICAST
1814                 if (psf->sf_oldin &&
1815                     !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1816                         psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1817                         psf->sf_next = pmc->tomb;
1818                         pmc->tomb = psf;
1819                         rv = 1;
1820                 } else
1821 #endif
1822                         kfree(psf);
1823         }
1824         return rv;
1825 }
1826
1827 #ifndef CONFIG_IP_MULTICAST
1828 #define igmp_ifc_event(x)       do { } while (0)
1829 #endif
1830
1831 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1832                          int sfcount, __be32 *psfsrc, int delta)
1833 {
1834         struct ip_mc_list *pmc;
1835         int     changerec = 0;
1836         int     i, err;
1837
1838         if (!in_dev)
1839                 return -ENODEV;
1840         rcu_read_lock();
1841         for_each_pmc_rcu(in_dev, pmc) {
1842                 if (*pmca == pmc->multiaddr)
1843                         break;
1844         }
1845         if (!pmc) {
1846                 /* MCA not found?? bug */
1847                 rcu_read_unlock();
1848                 return -ESRCH;
1849         }
1850         spin_lock_bh(&pmc->lock);
1851         rcu_read_unlock();
1852 #ifdef CONFIG_IP_MULTICAST
1853         sf_markstate(pmc);
1854 #endif
1855         if (!delta) {
1856                 err = -EINVAL;
1857                 if (!pmc->sfcount[sfmode])
1858                         goto out_unlock;
1859                 pmc->sfcount[sfmode]--;
1860         }
1861         err = 0;
1862         for (i = 0; i < sfcount; i++) {
1863                 int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1864
1865                 changerec |= rv > 0;
1866                 if (!err && rv < 0)
1867                         err = rv;
1868         }
1869         if (pmc->sfmode == MCAST_EXCLUDE &&
1870             pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1871             pmc->sfcount[MCAST_INCLUDE]) {
1872 #ifdef CONFIG_IP_MULTICAST
1873                 struct ip_sf_list *psf;
1874                 struct net *net = dev_net(in_dev->dev);
1875 #endif
1876
1877                 /* filter mode change */
1878                 pmc->sfmode = MCAST_INCLUDE;
1879 #ifdef CONFIG_IP_MULTICAST
1880                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1881                 in_dev->mr_ifc_count = pmc->crcount;
1882                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1883                         psf->sf_crcount = 0;
1884                 igmp_ifc_event(pmc->interface);
1885         } else if (sf_setstate(pmc) || changerec) {
1886                 igmp_ifc_event(pmc->interface);
1887 #endif
1888         }
1889 out_unlock:
1890         spin_unlock_bh(&pmc->lock);
1891         return err;
1892 }
1893
1894 /*
1895  * Add multicast single-source filter to the interface list
1896  */
1897 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1898         __be32 *psfsrc)
1899 {
1900         struct ip_sf_list *psf, *psf_prev;
1901
1902         psf_prev = NULL;
1903         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1904                 if (psf->sf_inaddr == *psfsrc)
1905                         break;
1906                 psf_prev = psf;
1907         }
1908         if (!psf) {
1909                 psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1910                 if (!psf)
1911                         return -ENOBUFS;
1912                 psf->sf_inaddr = *psfsrc;
1913                 if (psf_prev) {
1914                         psf_prev->sf_next = psf;
1915                 } else
1916                         pmc->sources = psf;
1917         }
1918         psf->sf_count[sfmode]++;
1919         if (psf->sf_count[sfmode] == 1) {
1920                 ip_rt_multicast_event(pmc->interface);
1921         }
1922         return 0;
1923 }
1924
1925 #ifdef CONFIG_IP_MULTICAST
1926 static void sf_markstate(struct ip_mc_list *pmc)
1927 {
1928         struct ip_sf_list *psf;
1929         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1930
1931         for (psf = pmc->sources; psf; psf = psf->sf_next)
1932                 if (pmc->sfcount[MCAST_EXCLUDE]) {
1933                         psf->sf_oldin = mca_xcount ==
1934                                 psf->sf_count[MCAST_EXCLUDE] &&
1935                                 !psf->sf_count[MCAST_INCLUDE];
1936                 } else
1937                         psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
1938 }
1939
1940 static int sf_setstate(struct ip_mc_list *pmc)
1941 {
1942         struct ip_sf_list *psf, *dpsf;
1943         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1944         int qrv = pmc->interface->mr_qrv;
1945         int new_in, rv;
1946
1947         rv = 0;
1948         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1949                 if (pmc->sfcount[MCAST_EXCLUDE]) {
1950                         new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
1951                                 !psf->sf_count[MCAST_INCLUDE];
1952                 } else
1953                         new_in = psf->sf_count[MCAST_INCLUDE] != 0;
1954                 if (new_in) {
1955                         if (!psf->sf_oldin) {
1956                                 struct ip_sf_list *prev = NULL;
1957
1958                                 for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
1959                                         if (dpsf->sf_inaddr == psf->sf_inaddr)
1960                                                 break;
1961                                         prev = dpsf;
1962                                 }
1963                                 if (dpsf) {
1964                                         if (prev)
1965                                                 prev->sf_next = dpsf->sf_next;
1966                                         else
1967                                                 pmc->tomb = dpsf->sf_next;
1968                                         kfree(dpsf);
1969                                 }
1970                                 psf->sf_crcount = qrv;
1971                                 rv++;
1972                         }
1973                 } else if (psf->sf_oldin) {
1974
1975                         psf->sf_crcount = 0;
1976                         /*
1977                          * add or update "delete" records if an active filter
1978                          * is now inactive
1979                          */
1980                         for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
1981                                 if (dpsf->sf_inaddr == psf->sf_inaddr)
1982                                         break;
1983                         if (!dpsf) {
1984                                 dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
1985                                 if (!dpsf)
1986                                         continue;
1987                                 *dpsf = *psf;
1988                                 /* pmc->lock held by callers */
1989                                 dpsf->sf_next = pmc->tomb;
1990                                 pmc->tomb = dpsf;
1991                         }
1992                         dpsf->sf_crcount = qrv;
1993                         rv++;
1994                 }
1995         }
1996         return rv;
1997 }
1998 #endif
1999
2000 /*
2001  * Add multicast source filter list to the interface list
2002  */
2003 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2004                          int sfcount, __be32 *psfsrc, int delta)
2005 {
2006         struct ip_mc_list *pmc;
2007         int     isexclude;
2008         int     i, err;
2009
2010         if (!in_dev)
2011                 return -ENODEV;
2012         rcu_read_lock();
2013         for_each_pmc_rcu(in_dev, pmc) {
2014                 if (*pmca == pmc->multiaddr)
2015                         break;
2016         }
2017         if (!pmc) {
2018                 /* MCA not found?? bug */
2019                 rcu_read_unlock();
2020                 return -ESRCH;
2021         }
2022         spin_lock_bh(&pmc->lock);
2023         rcu_read_unlock();
2024
2025 #ifdef CONFIG_IP_MULTICAST
2026         sf_markstate(pmc);
2027 #endif
2028         isexclude = pmc->sfmode == MCAST_EXCLUDE;
2029         if (!delta)
2030                 pmc->sfcount[sfmode]++;
2031         err = 0;
2032         for (i = 0; i < sfcount; i++) {
2033                 err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2034                 if (err)
2035                         break;
2036         }
2037         if (err) {
2038                 int j;
2039
2040                 if (!delta)
2041                         pmc->sfcount[sfmode]--;
2042                 for (j = 0; j < i; j++)
2043                         (void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2044         } else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2045 #ifdef CONFIG_IP_MULTICAST
2046                 struct ip_sf_list *psf;
2047                 struct net *net = dev_net(pmc->interface->dev);
2048                 in_dev = pmc->interface;
2049 #endif
2050
2051                 /* filter mode change */
2052                 if (pmc->sfcount[MCAST_EXCLUDE])
2053                         pmc->sfmode = MCAST_EXCLUDE;
2054                 else if (pmc->sfcount[MCAST_INCLUDE])
2055                         pmc->sfmode = MCAST_INCLUDE;
2056 #ifdef CONFIG_IP_MULTICAST
2057                 /* else no filters; keep old mode for reports */
2058
2059                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
2060                 in_dev->mr_ifc_count = pmc->crcount;
2061                 for (psf = pmc->sources; psf; psf = psf->sf_next)
2062                         psf->sf_crcount = 0;
2063                 igmp_ifc_event(in_dev);
2064         } else if (sf_setstate(pmc)) {
2065                 igmp_ifc_event(in_dev);
2066 #endif
2067         }
2068         spin_unlock_bh(&pmc->lock);
2069         return err;
2070 }
2071
2072 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2073 {
2074         struct ip_sf_list *psf, *nextpsf;
2075
2076         for (psf = pmc->tomb; psf; psf = nextpsf) {
2077                 nextpsf = psf->sf_next;
2078                 kfree(psf);
2079         }
2080         pmc->tomb = NULL;
2081         for (psf = pmc->sources; psf; psf = nextpsf) {
2082                 nextpsf = psf->sf_next;
2083                 kfree(psf);
2084         }
2085         pmc->sources = NULL;
2086         pmc->sfmode = MCAST_EXCLUDE;
2087         pmc->sfcount[MCAST_INCLUDE] = 0;
2088         pmc->sfcount[MCAST_EXCLUDE] = 1;
2089 }
2090
2091 /* Join a multicast group
2092  */
2093
2094 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2095 {
2096         __be32 addr = imr->imr_multiaddr.s_addr;
2097         struct ip_mc_socklist *iml, *i;
2098         struct in_device *in_dev;
2099         struct inet_sock *inet = inet_sk(sk);
2100         struct net *net = sock_net(sk);
2101         int ifindex;
2102         int count = 0;
2103         int err;
2104
2105         ASSERT_RTNL();
2106
2107         if (!ipv4_is_multicast(addr))
2108                 return -EINVAL;
2109
2110         in_dev = ip_mc_find_dev(net, imr);
2111
2112         if (!in_dev) {
2113                 err = -ENODEV;
2114                 goto done;
2115         }
2116
2117         err = -EADDRINUSE;
2118         ifindex = imr->imr_ifindex;
2119         for_each_pmc_rtnl(inet, i) {
2120                 if (i->multi.imr_multiaddr.s_addr == addr &&
2121                     i->multi.imr_ifindex == ifindex)
2122                         goto done;
2123                 count++;
2124         }
2125         err = -ENOBUFS;
2126         if (count >= net->ipv4.sysctl_igmp_max_memberships)
2127                 goto done;
2128         iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2129         if (!iml)
2130                 goto done;
2131
2132         memcpy(&iml->multi, imr, sizeof(*imr));
2133         iml->next_rcu = inet->mc_list;
2134         iml->sflist = NULL;
2135         iml->sfmode = MCAST_EXCLUDE;
2136         rcu_assign_pointer(inet->mc_list, iml);
2137         ip_mc_inc_group(in_dev, addr);
2138         err = 0;
2139 done:
2140         return err;
2141 }
2142 EXPORT_SYMBOL(ip_mc_join_group);
2143
2144 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2145                            struct in_device *in_dev)
2146 {
2147         struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2148         int err;
2149
2150         if (!psf) {
2151                 /* any-source empty exclude case */
2152                 return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2153                         iml->sfmode, 0, NULL, 0);
2154         }
2155         err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2156                         iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2157         RCU_INIT_POINTER(iml->sflist, NULL);
2158         /* decrease mem now to avoid the memleak warning */
2159         atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
2160         kfree_rcu(psf, rcu);
2161         return err;
2162 }
2163
2164 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2165 {
2166         struct inet_sock *inet = inet_sk(sk);
2167         struct ip_mc_socklist *iml;
2168         struct ip_mc_socklist __rcu **imlp;
2169         struct in_device *in_dev;
2170         struct net *net = sock_net(sk);
2171         __be32 group = imr->imr_multiaddr.s_addr;
2172         u32 ifindex;
2173         int ret = -EADDRNOTAVAIL;
2174
2175         ASSERT_RTNL();
2176
2177         in_dev = ip_mc_find_dev(net, imr);
2178         if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2179                 ret = -ENODEV;
2180                 goto out;
2181         }
2182         ifindex = imr->imr_ifindex;
2183         for (imlp = &inet->mc_list;
2184              (iml = rtnl_dereference(*imlp)) != NULL;
2185              imlp = &iml->next_rcu) {
2186                 if (iml->multi.imr_multiaddr.s_addr != group)
2187                         continue;
2188                 if (ifindex) {
2189                         if (iml->multi.imr_ifindex != ifindex)
2190                                 continue;
2191                 } else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2192                                 iml->multi.imr_address.s_addr)
2193                         continue;
2194
2195                 (void) ip_mc_leave_src(sk, iml, in_dev);
2196
2197                 *imlp = iml->next_rcu;
2198
2199                 if (in_dev)
2200                         ip_mc_dec_group(in_dev, group);
2201
2202                 /* decrease mem now to avoid the memleak warning */
2203                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2204                 kfree_rcu(iml, rcu);
2205                 return 0;
2206         }
2207 out:
2208         return ret;
2209 }
2210 EXPORT_SYMBOL(ip_mc_leave_group);
2211
2212 int ip_mc_source(int add, int omode, struct sock *sk, struct
2213         ip_mreq_source *mreqs, int ifindex)
2214 {
2215         int err;
2216         struct ip_mreqn imr;
2217         __be32 addr = mreqs->imr_multiaddr;
2218         struct ip_mc_socklist *pmc;
2219         struct in_device *in_dev = NULL;
2220         struct inet_sock *inet = inet_sk(sk);
2221         struct ip_sf_socklist *psl;
2222         struct net *net = sock_net(sk);
2223         int leavegroup = 0;
2224         int i, j, rv;
2225
2226         if (!ipv4_is_multicast(addr))
2227                 return -EINVAL;
2228
2229         ASSERT_RTNL();
2230
2231         imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2232         imr.imr_address.s_addr = mreqs->imr_interface;
2233         imr.imr_ifindex = ifindex;
2234         in_dev = ip_mc_find_dev(net, &imr);
2235
2236         if (!in_dev) {
2237                 err = -ENODEV;
2238                 goto done;
2239         }
2240         err = -EADDRNOTAVAIL;
2241
2242         for_each_pmc_rtnl(inet, pmc) {
2243                 if ((pmc->multi.imr_multiaddr.s_addr ==
2244                      imr.imr_multiaddr.s_addr) &&
2245                     (pmc->multi.imr_ifindex == imr.imr_ifindex))
2246                         break;
2247         }
2248         if (!pmc) {             /* must have a prior join */
2249                 err = -EINVAL;
2250                 goto done;
2251         }
2252         /* if a source filter was set, must be the same mode as before */
2253         if (pmc->sflist) {
2254                 if (pmc->sfmode != omode) {
2255                         err = -EINVAL;
2256                         goto done;
2257                 }
2258         } else if (pmc->sfmode != omode) {
2259                 /* allow mode switches for empty-set filters */
2260                 ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2261                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2262                         NULL, 0);
2263                 pmc->sfmode = omode;
2264         }
2265
2266         psl = rtnl_dereference(pmc->sflist);
2267         if (!add) {
2268                 if (!psl)
2269                         goto done;      /* err = -EADDRNOTAVAIL */
2270                 rv = !0;
2271                 for (i = 0; i < psl->sl_count; i++) {
2272                         rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2273                                 sizeof(__be32));
2274                         if (rv == 0)
2275                                 break;
2276                 }
2277                 if (rv)         /* source not found */
2278                         goto done;      /* err = -EADDRNOTAVAIL */
2279
2280                 /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2281                 if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2282                         leavegroup = 1;
2283                         goto done;
2284                 }
2285
2286                 /* update the interface filter */
2287                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2288                         &mreqs->imr_sourceaddr, 1);
2289
2290                 for (j = i+1; j < psl->sl_count; j++)
2291                         psl->sl_addr[j-1] = psl->sl_addr[j];
2292                 psl->sl_count--;
2293                 err = 0;
2294                 goto done;
2295         }
2296         /* else, add a new source to the filter */
2297
2298         if (psl && psl->sl_count >= net->ipv4.sysctl_igmp_max_msf) {
2299                 err = -ENOBUFS;
2300                 goto done;
2301         }
2302         if (!psl || psl->sl_count == psl->sl_max) {
2303                 struct ip_sf_socklist *newpsl;
2304                 int count = IP_SFBLOCK;
2305
2306                 if (psl)
2307                         count += psl->sl_max;
2308                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2309                 if (!newpsl) {
2310                         err = -ENOBUFS;
2311                         goto done;
2312                 }
2313                 newpsl->sl_max = count;
2314                 newpsl->sl_count = count - IP_SFBLOCK;
2315                 if (psl) {
2316                         for (i = 0; i < psl->sl_count; i++)
2317                                 newpsl->sl_addr[i] = psl->sl_addr[i];
2318                         /* decrease mem now to avoid the memleak warning */
2319                         atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2320                         kfree_rcu(psl, rcu);
2321                 }
2322                 rcu_assign_pointer(pmc->sflist, newpsl);
2323                 psl = newpsl;
2324         }
2325         rv = 1; /* > 0 for insert logic below if sl_count is 0 */
2326         for (i = 0; i < psl->sl_count; i++) {
2327                 rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2328                         sizeof(__be32));
2329                 if (rv == 0)
2330                         break;
2331         }
2332         if (rv == 0)            /* address already there is an error */
2333                 goto done;
2334         for (j = psl->sl_count-1; j >= i; j--)
2335                 psl->sl_addr[j+1] = psl->sl_addr[j];
2336         psl->sl_addr[i] = mreqs->imr_sourceaddr;
2337         psl->sl_count++;
2338         err = 0;
2339         /* update the interface list */
2340         ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2341                 &mreqs->imr_sourceaddr, 1);
2342 done:
2343         if (leavegroup)
2344                 err = ip_mc_leave_group(sk, &imr);
2345         return err;
2346 }
2347
2348 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2349 {
2350         int err = 0;
2351         struct ip_mreqn imr;
2352         __be32 addr = msf->imsf_multiaddr;
2353         struct ip_mc_socklist *pmc;
2354         struct in_device *in_dev;
2355         struct inet_sock *inet = inet_sk(sk);
2356         struct ip_sf_socklist *newpsl, *psl;
2357         struct net *net = sock_net(sk);
2358         int leavegroup = 0;
2359
2360         if (!ipv4_is_multicast(addr))
2361                 return -EINVAL;
2362         if (msf->imsf_fmode != MCAST_INCLUDE &&
2363             msf->imsf_fmode != MCAST_EXCLUDE)
2364                 return -EINVAL;
2365
2366         ASSERT_RTNL();
2367
2368         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2369         imr.imr_address.s_addr = msf->imsf_interface;
2370         imr.imr_ifindex = ifindex;
2371         in_dev = ip_mc_find_dev(net, &imr);
2372
2373         if (!in_dev) {
2374                 err = -ENODEV;
2375                 goto done;
2376         }
2377
2378         /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2379         if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2380                 leavegroup = 1;
2381                 goto done;
2382         }
2383
2384         for_each_pmc_rtnl(inet, pmc) {
2385                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2386                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2387                         break;
2388         }
2389         if (!pmc) {             /* must have a prior join */
2390                 err = -EINVAL;
2391                 goto done;
2392         }
2393         if (msf->imsf_numsrc) {
2394                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2395                                                            GFP_KERNEL);
2396                 if (!newpsl) {
2397                         err = -ENOBUFS;
2398                         goto done;
2399                 }
2400                 newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2401                 memcpy(newpsl->sl_addr, msf->imsf_slist,
2402                         msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2403                 err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2404                         msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2405                 if (err) {
2406                         sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2407                         goto done;
2408                 }
2409         } else {
2410                 newpsl = NULL;
2411                 (void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2412                                      msf->imsf_fmode, 0, NULL, 0);
2413         }
2414         psl = rtnl_dereference(pmc->sflist);
2415         if (psl) {
2416                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2417                         psl->sl_count, psl->sl_addr, 0);
2418                 /* decrease mem now to avoid the memleak warning */
2419                 atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2420                 kfree_rcu(psl, rcu);
2421         } else
2422                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2423                         0, NULL, 0);
2424         rcu_assign_pointer(pmc->sflist, newpsl);
2425         pmc->sfmode = msf->imsf_fmode;
2426         err = 0;
2427 done:
2428         if (leavegroup)
2429                 err = ip_mc_leave_group(sk, &imr);
2430         return err;
2431 }
2432
2433 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2434         struct ip_msfilter __user *optval, int __user *optlen)
2435 {
2436         int err, len, count, copycount;
2437         struct ip_mreqn imr;
2438         __be32 addr = msf->imsf_multiaddr;
2439         struct ip_mc_socklist *pmc;
2440         struct in_device *in_dev;
2441         struct inet_sock *inet = inet_sk(sk);
2442         struct ip_sf_socklist *psl;
2443         struct net *net = sock_net(sk);
2444
2445         ASSERT_RTNL();
2446
2447         if (!ipv4_is_multicast(addr))
2448                 return -EINVAL;
2449
2450         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2451         imr.imr_address.s_addr = msf->imsf_interface;
2452         imr.imr_ifindex = 0;
2453         in_dev = ip_mc_find_dev(net, &imr);
2454
2455         if (!in_dev) {
2456                 err = -ENODEV;
2457                 goto done;
2458         }
2459         err = -EADDRNOTAVAIL;
2460
2461         for_each_pmc_rtnl(inet, pmc) {
2462                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2463                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2464                         break;
2465         }
2466         if (!pmc)               /* must have a prior join */
2467                 goto done;
2468         msf->imsf_fmode = pmc->sfmode;
2469         psl = rtnl_dereference(pmc->sflist);
2470         if (!psl) {
2471                 len = 0;
2472                 count = 0;
2473         } else {
2474                 count = psl->sl_count;
2475         }
2476         copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2477         len = copycount * sizeof(psl->sl_addr[0]);
2478         msf->imsf_numsrc = count;
2479         if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2480             copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2481                 return -EFAULT;
2482         }
2483         if (len &&
2484             copy_to_user(&optval->imsf_slist[0], psl->sl_addr, len))
2485                 return -EFAULT;
2486         return 0;
2487 done:
2488         return err;
2489 }
2490
2491 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2492         struct group_filter __user *optval, int __user *optlen)
2493 {
2494         int err, i, count, copycount;
2495         struct sockaddr_in *psin;
2496         __be32 addr;
2497         struct ip_mc_socklist *pmc;
2498         struct inet_sock *inet = inet_sk(sk);
2499         struct ip_sf_socklist *psl;
2500
2501         ASSERT_RTNL();
2502
2503         psin = (struct sockaddr_in *)&gsf->gf_group;
2504         if (psin->sin_family != AF_INET)
2505                 return -EINVAL;
2506         addr = psin->sin_addr.s_addr;
2507         if (!ipv4_is_multicast(addr))
2508                 return -EINVAL;
2509
2510         err = -EADDRNOTAVAIL;
2511
2512         for_each_pmc_rtnl(inet, pmc) {
2513                 if (pmc->multi.imr_multiaddr.s_addr == addr &&
2514                     pmc->multi.imr_ifindex == gsf->gf_interface)
2515                         break;
2516         }
2517         if (!pmc)               /* must have a prior join */
2518                 goto done;
2519         gsf->gf_fmode = pmc->sfmode;
2520         psl = rtnl_dereference(pmc->sflist);
2521         count = psl ? psl->sl_count : 0;
2522         copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2523         gsf->gf_numsrc = count;
2524         if (put_user(GROUP_FILTER_SIZE(copycount), optlen) ||
2525             copy_to_user(optval, gsf, GROUP_FILTER_SIZE(0))) {
2526                 return -EFAULT;
2527         }
2528         for (i = 0; i < copycount; i++) {
2529                 struct sockaddr_storage ss;
2530
2531                 psin = (struct sockaddr_in *)&ss;
2532                 memset(&ss, 0, sizeof(ss));
2533                 psin->sin_family = AF_INET;
2534                 psin->sin_addr.s_addr = psl->sl_addr[i];
2535                 if (copy_to_user(&optval->gf_slist[i], &ss, sizeof(ss)))
2536                         return -EFAULT;
2537         }
2538         return 0;
2539 done:
2540         return err;
2541 }
2542
2543 /*
2544  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2545  */
2546 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr, int dif)
2547 {
2548         struct inet_sock *inet = inet_sk(sk);
2549         struct ip_mc_socklist *pmc;
2550         struct ip_sf_socklist *psl;
2551         int i;
2552         int ret;
2553
2554         ret = 1;
2555         if (!ipv4_is_multicast(loc_addr))
2556                 goto out;
2557
2558         rcu_read_lock();
2559         for_each_pmc_rcu(inet, pmc) {
2560                 if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2561                     pmc->multi.imr_ifindex == dif)
2562                         break;
2563         }
2564         ret = inet->mc_all;
2565         if (!pmc)
2566                 goto unlock;
2567         psl = rcu_dereference(pmc->sflist);
2568         ret = (pmc->sfmode == MCAST_EXCLUDE);
2569         if (!psl)
2570                 goto unlock;
2571
2572         for (i = 0; i < psl->sl_count; i++) {
2573                 if (psl->sl_addr[i] == rmt_addr)
2574                         break;
2575         }
2576         ret = 0;
2577         if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2578                 goto unlock;
2579         if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2580                 goto unlock;
2581         ret = 1;
2582 unlock:
2583         rcu_read_unlock();
2584 out:
2585         return ret;
2586 }
2587
2588 /*
2589  *      A socket is closing.
2590  */
2591
2592 void ip_mc_drop_socket(struct sock *sk)
2593 {
2594         struct inet_sock *inet = inet_sk(sk);
2595         struct ip_mc_socklist *iml;
2596         struct net *net = sock_net(sk);
2597
2598         if (!inet->mc_list)
2599                 return;
2600
2601         rtnl_lock();
2602         while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2603                 struct in_device *in_dev;
2604
2605                 inet->mc_list = iml->next_rcu;
2606                 in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2607                 (void) ip_mc_leave_src(sk, iml, in_dev);
2608                 if (in_dev)
2609                         ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2610                 /* decrease mem now to avoid the memleak warning */
2611                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2612                 kfree_rcu(iml, rcu);
2613         }
2614         rtnl_unlock();
2615 }
2616
2617 /* called with rcu_read_lock() */
2618 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2619 {
2620         struct ip_mc_list *im;
2621         struct ip_mc_list __rcu **mc_hash;
2622         struct ip_sf_list *psf;
2623         int rv = 0;
2624
2625         mc_hash = rcu_dereference(in_dev->mc_hash);
2626         if (mc_hash) {
2627                 u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2628
2629                 for (im = rcu_dereference(mc_hash[hash]);
2630                      im != NULL;
2631                      im = rcu_dereference(im->next_hash)) {
2632                         if (im->multiaddr == mc_addr)
2633                                 break;
2634                 }
2635         } else {
2636                 for_each_pmc_rcu(in_dev, im) {
2637                         if (im->multiaddr == mc_addr)
2638                                 break;
2639                 }
2640         }
2641         if (im && proto == IPPROTO_IGMP) {
2642                 rv = 1;
2643         } else if (im) {
2644                 if (src_addr) {
2645                         for (psf = im->sources; psf; psf = psf->sf_next) {
2646                                 if (psf->sf_inaddr == src_addr)
2647                                         break;
2648                         }
2649                         if (psf)
2650                                 rv = psf->sf_count[MCAST_INCLUDE] ||
2651                                         psf->sf_count[MCAST_EXCLUDE] !=
2652                                         im->sfcount[MCAST_EXCLUDE];
2653                         else
2654                                 rv = im->sfcount[MCAST_EXCLUDE] != 0;
2655                 } else
2656                         rv = 1; /* unspecified source; tentatively allow */
2657         }
2658         return rv;
2659 }
2660
2661 #if defined(CONFIG_PROC_FS)
2662 struct igmp_mc_iter_state {
2663         struct seq_net_private p;
2664         struct net_device *dev;
2665         struct in_device *in_dev;
2666 };
2667
2668 #define igmp_mc_seq_private(seq)        ((struct igmp_mc_iter_state *)(seq)->private)
2669
2670 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2671 {
2672         struct net *net = seq_file_net(seq);
2673         struct ip_mc_list *im = NULL;
2674         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2675
2676         state->in_dev = NULL;
2677         for_each_netdev_rcu(net, state->dev) {
2678                 struct in_device *in_dev;
2679
2680                 in_dev = __in_dev_get_rcu(state->dev);
2681                 if (!in_dev)
2682                         continue;
2683                 im = rcu_dereference(in_dev->mc_list);
2684                 if (im) {
2685                         state->in_dev = in_dev;
2686                         break;
2687                 }
2688         }
2689         return im;
2690 }
2691
2692 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2693 {
2694         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2695
2696         im = rcu_dereference(im->next_rcu);
2697         while (!im) {
2698                 state->dev = next_net_device_rcu(state->dev);
2699                 if (!state->dev) {
2700                         state->in_dev = NULL;
2701                         break;
2702                 }
2703                 state->in_dev = __in_dev_get_rcu(state->dev);
2704                 if (!state->in_dev)
2705                         continue;
2706                 im = rcu_dereference(state->in_dev->mc_list);
2707         }
2708         return im;
2709 }
2710
2711 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2712 {
2713         struct ip_mc_list *im = igmp_mc_get_first(seq);
2714         if (im)
2715                 while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2716                         --pos;
2717         return pos ? NULL : im;
2718 }
2719
2720 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2721         __acquires(rcu)
2722 {
2723         rcu_read_lock();
2724         return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2725 }
2726
2727 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2728 {
2729         struct ip_mc_list *im;
2730         if (v == SEQ_START_TOKEN)
2731                 im = igmp_mc_get_first(seq);
2732         else
2733                 im = igmp_mc_get_next(seq, v);
2734         ++*pos;
2735         return im;
2736 }
2737
2738 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2739         __releases(rcu)
2740 {
2741         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2742
2743         state->in_dev = NULL;
2744         state->dev = NULL;
2745         rcu_read_unlock();
2746 }
2747
2748 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2749 {
2750         if (v == SEQ_START_TOKEN)
2751                 seq_puts(seq,
2752                          "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2753         else {
2754                 struct ip_mc_list *im = (struct ip_mc_list *)v;
2755                 struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2756                 char   *querier;
2757                 long delta;
2758
2759 #ifdef CONFIG_IP_MULTICAST
2760                 querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2761                           IGMP_V2_SEEN(state->in_dev) ? "V2" :
2762                           "V3";
2763 #else
2764                 querier = "NONE";
2765 #endif
2766
2767                 if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2768                         seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2769                                    state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2770                 }
2771
2772                 delta = im->timer.expires - jiffies;
2773                 seq_printf(seq,
2774                            "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2775                            im->multiaddr, im->users,
2776                            im->tm_running,
2777                            im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2778                            im->reporter);
2779         }
2780         return 0;
2781 }
2782
2783 static const struct seq_operations igmp_mc_seq_ops = {
2784         .start  =       igmp_mc_seq_start,
2785         .next   =       igmp_mc_seq_next,
2786         .stop   =       igmp_mc_seq_stop,
2787         .show   =       igmp_mc_seq_show,
2788 };
2789
2790 static int igmp_mc_seq_open(struct inode *inode, struct file *file)
2791 {
2792         return seq_open_net(inode, file, &igmp_mc_seq_ops,
2793                         sizeof(struct igmp_mc_iter_state));
2794 }
2795
2796 static const struct file_operations igmp_mc_seq_fops = {
2797         .owner          =       THIS_MODULE,
2798         .open           =       igmp_mc_seq_open,
2799         .read           =       seq_read,
2800         .llseek         =       seq_lseek,
2801         .release        =       seq_release_net,
2802 };
2803
2804 struct igmp_mcf_iter_state {
2805         struct seq_net_private p;
2806         struct net_device *dev;
2807         struct in_device *idev;
2808         struct ip_mc_list *im;
2809 };
2810
2811 #define igmp_mcf_seq_private(seq)       ((struct igmp_mcf_iter_state *)(seq)->private)
2812
2813 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2814 {
2815         struct net *net = seq_file_net(seq);
2816         struct ip_sf_list *psf = NULL;
2817         struct ip_mc_list *im = NULL;
2818         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2819
2820         state->idev = NULL;
2821         state->im = NULL;
2822         for_each_netdev_rcu(net, state->dev) {
2823                 struct in_device *idev;
2824                 idev = __in_dev_get_rcu(state->dev);
2825                 if (unlikely(!idev))
2826                         continue;
2827                 im = rcu_dereference(idev->mc_list);
2828                 if (likely(im)) {
2829                         spin_lock_bh(&im->lock);
2830                         psf = im->sources;
2831                         if (likely(psf)) {
2832                                 state->im = im;
2833                                 state->idev = idev;
2834                                 break;
2835                         }
2836                         spin_unlock_bh(&im->lock);
2837                 }
2838         }
2839         return psf;
2840 }
2841
2842 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2843 {
2844         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2845
2846         psf = psf->sf_next;
2847         while (!psf) {
2848                 spin_unlock_bh(&state->im->lock);
2849                 state->im = state->im->next;
2850                 while (!state->im) {
2851                         state->dev = next_net_device_rcu(state->dev);
2852                         if (!state->dev) {
2853                                 state->idev = NULL;
2854                                 goto out;
2855                         }
2856                         state->idev = __in_dev_get_rcu(state->dev);
2857                         if (!state->idev)
2858                                 continue;
2859                         state->im = rcu_dereference(state->idev->mc_list);
2860                 }
2861                 if (!state->im)
2862                         break;
2863                 spin_lock_bh(&state->im->lock);
2864                 psf = state->im->sources;
2865         }
2866 out:
2867         return psf;
2868 }
2869
2870 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2871 {
2872         struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2873         if (psf)
2874                 while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2875                         --pos;
2876         return pos ? NULL : psf;
2877 }
2878
2879 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2880         __acquires(rcu)
2881 {
2882         rcu_read_lock();
2883         return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2884 }
2885
2886 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2887 {
2888         struct ip_sf_list *psf;
2889         if (v == SEQ_START_TOKEN)
2890                 psf = igmp_mcf_get_first(seq);
2891         else
2892                 psf = igmp_mcf_get_next(seq, v);
2893         ++*pos;
2894         return psf;
2895 }
2896
2897 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2898         __releases(rcu)
2899 {
2900         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2901         if (likely(state->im)) {
2902                 spin_unlock_bh(&state->im->lock);
2903                 state->im = NULL;
2904         }
2905         state->idev = NULL;
2906         state->dev = NULL;
2907         rcu_read_unlock();
2908 }
2909
2910 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2911 {
2912         struct ip_sf_list *psf = (struct ip_sf_list *)v;
2913         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2914
2915         if (v == SEQ_START_TOKEN) {
2916                 seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
2917         } else {
2918                 seq_printf(seq,
2919                            "%3d %6.6s 0x%08x "
2920                            "0x%08x %6lu %6lu\n",
2921                            state->dev->ifindex, state->dev->name,
2922                            ntohl(state->im->multiaddr),
2923                            ntohl(psf->sf_inaddr),
2924                            psf->sf_count[MCAST_INCLUDE],
2925                            psf->sf_count[MCAST_EXCLUDE]);
2926         }
2927         return 0;
2928 }
2929
2930 static const struct seq_operations igmp_mcf_seq_ops = {
2931         .start  =       igmp_mcf_seq_start,
2932         .next   =       igmp_mcf_seq_next,
2933         .stop   =       igmp_mcf_seq_stop,
2934         .show   =       igmp_mcf_seq_show,
2935 };
2936
2937 static int igmp_mcf_seq_open(struct inode *inode, struct file *file)
2938 {
2939         return seq_open_net(inode, file, &igmp_mcf_seq_ops,
2940                         sizeof(struct igmp_mcf_iter_state));
2941 }
2942
2943 static const struct file_operations igmp_mcf_seq_fops = {
2944         .owner          =       THIS_MODULE,
2945         .open           =       igmp_mcf_seq_open,
2946         .read           =       seq_read,
2947         .llseek         =       seq_lseek,
2948         .release        =       seq_release_net,
2949 };
2950
2951 static int __net_init igmp_net_init(struct net *net)
2952 {
2953         struct proc_dir_entry *pde;
2954         int err;
2955
2956         pde = proc_create("igmp", S_IRUGO, net->proc_net, &igmp_mc_seq_fops);
2957         if (!pde)
2958                 goto out_igmp;
2959         pde = proc_create("mcfilter", S_IRUGO, net->proc_net,
2960                           &igmp_mcf_seq_fops);
2961         if (!pde)
2962                 goto out_mcfilter;
2963         err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
2964                                    SOCK_DGRAM, 0, net);
2965         if (err < 0) {
2966                 pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
2967                        err);
2968                 goto out_sock;
2969         }
2970
2971         /* Sysctl initialization */
2972         net->ipv4.sysctl_igmp_max_memberships = 20;
2973         net->ipv4.sysctl_igmp_max_msf = 10;
2974         /* IGMP reports for link-local multicast groups are enabled by default */
2975         net->ipv4.sysctl_igmp_llm_reports = 1;
2976         net->ipv4.sysctl_igmp_qrv = 2;
2977         return 0;
2978
2979 out_sock:
2980         remove_proc_entry("mcfilter", net->proc_net);
2981 out_mcfilter:
2982         remove_proc_entry("igmp", net->proc_net);
2983 out_igmp:
2984         return -ENOMEM;
2985 }
2986
2987 static void __net_exit igmp_net_exit(struct net *net)
2988 {
2989         remove_proc_entry("mcfilter", net->proc_net);
2990         remove_proc_entry("igmp", net->proc_net);
2991         inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
2992 }
2993
2994 static struct pernet_operations igmp_net_ops = {
2995         .init = igmp_net_init,
2996         .exit = igmp_net_exit,
2997 };
2998 #endif
2999
3000 static int igmp_netdev_event(struct notifier_block *this,
3001                              unsigned long event, void *ptr)
3002 {
3003         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3004         struct in_device *in_dev;
3005
3006         switch (event) {
3007         case NETDEV_RESEND_IGMP:
3008                 in_dev = __in_dev_get_rtnl(dev);
3009                 if (in_dev)
3010                         ip_mc_rejoin_groups(in_dev);
3011                 break;
3012         default:
3013                 break;
3014         }
3015         return NOTIFY_DONE;
3016 }
3017
3018 static struct notifier_block igmp_notifier = {
3019         .notifier_call = igmp_netdev_event,
3020 };
3021
3022 int __init igmp_mc_init(void)
3023 {
3024 #if defined(CONFIG_PROC_FS)
3025         int err;
3026
3027         err = register_pernet_subsys(&igmp_net_ops);
3028         if (err)
3029                 return err;
3030         err = register_netdevice_notifier(&igmp_notifier);
3031         if (err)
3032                 goto reg_notif_fail;
3033         return 0;
3034
3035 reg_notif_fail:
3036         unregister_pernet_subsys(&igmp_net_ops);
3037         return err;
3038 #else
3039         return register_netdevice_notifier(&igmp_notifier);
3040 #endif
3041 }