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 <[email protected]>
  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: [email protected] and [email protected]
  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 #include <linux/byteorder/generic.h>
  93
  94 #include <net/net_namespace.h>
  95 #include <net/arp.h>
  96 #include <net/ip.h>
  97 #include <net/protocol.h>
  98 #include <net/route.h>
  99 #include <net/sock.h>
 100 #include <net/checksum.h>
 101 #include <net/inet_common.h>
 102 #include <linux/netfilter_ipv4.h>
 103 #ifdef CONFIG_IP_MROUTE
 104 #include <linux/mroute.h>
 105 #endif
 106 #ifdef CONFIG_PROC_FS
 107 #include <linux/proc_fs.h>
 108 #include <linux/seq_file.h>
 109 #endif
 110
 111 #ifdef CONFIG_IP_MULTICAST
 112 /* Parameter names and values are taken from igmp-v2-06 draft */
 113
 114 #define IGMP_V2_UNSOLICITED_REPORT_INTERVAL     (10*HZ)
 115 #define IGMP_V3_UNSOLICITED_REPORT_INTERVAL     (1*HZ)
 116 #define IGMP_QUERY_INTERVAL                     (125*HZ)
 117 #define IGMP_QUERY_RESPONSE_INTERVAL            (10*HZ)
 118
 119 #define IGMP_INITIAL_REPORT_DELAY               (1)
 120
 121 /* IGMP_INITIAL_REPORT_DELAY is not from IGMP specs!
 122  * IGMP specs require to report membership immediately after
 123  * joining a group, but we delay the first report by a
 124  * small interval. It seems more natural and still does not
 125  * contradict to specs provided this delay is small enough.
 126  */
 127
 128 #define IGMP_V1_SEEN(in_dev) \
 129         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 1 || \
 130          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 1 || \
 131          ((in_dev)->mr_v1_seen && \
 132           time_before(jiffies, (in_dev)->mr_v1_seen)))
 133 #define IGMP_V2_SEEN(in_dev) \
 134         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 2 || \
 135          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 2 || \
 136          ((in_dev)->mr_v2_seen && \
 137           time_before(jiffies, (in_dev)->mr_v2_seen)))
 138
 139 static int unsolicited_report_interval(struct in_device *in_dev)
 140 {
 141         int interval_ms, interval_jiffies;
 142
 143         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 144                 interval_ms = IN_DEV_CONF_GET(
 145                         in_dev,
 146                         IGMPV2_UNSOLICITED_REPORT_INTERVAL);
 147         else /* v3 */
 148                 interval_ms = IN_DEV_CONF_GET(
 149                         in_dev,
 150                         IGMPV3_UNSOLICITED_REPORT_INTERVAL);
 151
 152         interval_jiffies = msecs_to_jiffies(interval_ms);
 153
 154         /* _timer functions can't handle a delay of 0 jiffies so ensure
 155          *  we always return a positive value.
 156          */
 157         if (interval_jiffies <= 0)
 158                 interval_jiffies = 1;
 159         return interval_jiffies;
 160 }
 161
 162 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im);
 163 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im);
 164 static void igmpv3_clear_delrec(struct in_device *in_dev);
 165 static int sf_setstate(struct ip_mc_list *pmc);
 166 static void sf_markstate(struct ip_mc_list *pmc);
 167 #endif
 168 static void ip_mc_clear_src(struct ip_mc_list *pmc);
 169 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
 170                          int sfcount, __be32 *psfsrc, int delta);
 171
 172 static void ip_ma_put(struct ip_mc_list *im)
 173 {
 174         if (refcount_dec_and_test(&im->refcnt)) {
 175                 in_dev_put(im->interface);
 176                 kfree_rcu(im, rcu);
 177         }
 178 }
 179
 180 #define for_each_pmc_rcu(in_dev, pmc)                           \
 181         for (pmc = rcu_dereference(in_dev->mc_list);            \
 182              pmc != NULL;                                       \
 183              pmc = rcu_dereference(pmc->next_rcu))
 184
 185 #define for_each_pmc_rtnl(in_dev, pmc)                          \
 186         for (pmc = rtnl_dereference(in_dev->mc_list);           \
 187              pmc != NULL;                                       \
 188              pmc = rtnl_dereference(pmc->next_rcu))
 189
 190 #ifdef CONFIG_IP_MULTICAST
 191
 192 /*
 193  *      Timer management
 194  */
 195
 196 static void igmp_stop_timer(struct ip_mc_list *im)
 197 {
 198         spin_lock_bh(&im->lock);
 199         if (del_timer(&im->timer))
 200                 refcount_dec(&im->refcnt);
 201         im->tm_running = 0;
 202         im->reporter = 0;
 203         im->unsolicit_count = 0;
 204         spin_unlock_bh(&im->lock);
 205 }
 206
 207 /* It must be called with locked im->lock */
 208 static void igmp_start_timer(struct ip_mc_list *im, int max_delay)
 209 {
 210         int tv = prandom_u32() % max_delay;
 211
 212         im->tm_running = 1;
 213         if (!mod_timer(&im->timer, jiffies+tv+2))
 214                 refcount_inc(&im->refcnt);
 215 }
 216
 217 static void igmp_gq_start_timer(struct in_device *in_dev)
 218 {
 219         int tv = prandom_u32() % in_dev->mr_maxdelay;
 220         unsigned long exp = jiffies + tv + 2;
 221
 222         if (in_dev->mr_gq_running &&
 223             time_after_eq(exp, (in_dev->mr_gq_timer).expires))
 224                 return;
 225
 226         in_dev->mr_gq_running = 1;
 227         if (!mod_timer(&in_dev->mr_gq_timer, exp))
 228                 in_dev_hold(in_dev);
 229 }
 230
 231 static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
 232 {
 233         int tv = prandom_u32() % delay;
 234
 235         if (!mod_timer(&in_dev->mr_ifc_timer, jiffies+tv+2))
 236                 in_dev_hold(in_dev);
 237 }
 238
 239 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
 240 {
 241         spin_lock_bh(&im->lock);
 242         im->unsolicit_count = 0;
 243         if (del_timer(&im->timer)) {
 244                 if ((long)(im->timer.expires-jiffies) < max_delay) {
 245                         add_timer(&im->timer);
 246                         im->tm_running = 1;
 247                         spin_unlock_bh(&im->lock);
 248                         return;
 249                 }
 250                 refcount_dec(&im->refcnt);
 251         }
 252         igmp_start_timer(im, max_delay);
 253         spin_unlock_bh(&im->lock);
 254 }
 255
 256
 257 /*
 258  *      Send an IGMP report.
 259  */
 260
 261 #define IGMP_SIZE (sizeof(struct igmphdr)+sizeof(struct iphdr)+4)
 262
 263
 264 static int is_in(struct ip_mc_list *pmc, struct ip_sf_list *psf, int type,
 265         int gdeleted, int sdeleted)
 266 {
 267         switch (type) {
 268         case IGMPV3_MODE_IS_INCLUDE:
 269         case IGMPV3_MODE_IS_EXCLUDE:
 270                 if (gdeleted || sdeleted)
 271                         return 0;
 272                 if (!(pmc->gsquery && !psf->sf_gsresp)) {
 273                         if (pmc->sfmode == MCAST_INCLUDE)
 274                                 return 1;
 275                         /* don't include if this source is excluded
 276                          * in all filters
 277                          */
 278                         if (psf->sf_count[MCAST_INCLUDE])
 279                                 return type == IGMPV3_MODE_IS_INCLUDE;
 280                         return pmc->sfcount[MCAST_EXCLUDE] ==
 281                                 psf->sf_count[MCAST_EXCLUDE];
 282                 }
 283                 return 0;
 284         case IGMPV3_CHANGE_TO_INCLUDE:
 285                 if (gdeleted || sdeleted)
 286                         return 0;
 287                 return psf->sf_count[MCAST_INCLUDE] != 0;
 288         case IGMPV3_CHANGE_TO_EXCLUDE:
 289                 if (gdeleted || sdeleted)
 290                         return 0;
 291                 if (pmc->sfcount[MCAST_EXCLUDE] == 0 ||
 292                     psf->sf_count[MCAST_INCLUDE])
 293                         return 0;
 294                 return pmc->sfcount[MCAST_EXCLUDE] ==
 295                         psf->sf_count[MCAST_EXCLUDE];
 296         case IGMPV3_ALLOW_NEW_SOURCES:
 297                 if (gdeleted || !psf->sf_crcount)
 298                         return 0;
 299                 return (pmc->sfmode == MCAST_INCLUDE) ^ sdeleted;
 300         case IGMPV3_BLOCK_OLD_SOURCES:
 301                 if (pmc->sfmode == MCAST_INCLUDE)
 302                         return gdeleted || (psf->sf_crcount && sdeleted);
 303                 return psf->sf_crcount && !gdeleted && !sdeleted;
 304         }
 305         return 0;
 306 }
 307
 308 static int
 309 igmp_scount(struct ip_mc_list *pmc, int type, int gdeleted, int sdeleted)
 310 {
 311         struct ip_sf_list *psf;
 312         int scount = 0;
 313
 314         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 315                 if (!is_in(pmc, psf, type, gdeleted, sdeleted))
 316                         continue;
 317                 scount++;
 318         }
 319         return scount;
 320 }
 321
 322 /* source address selection per RFC 3376 section 4.2.13 */
 323 static __be32 igmpv3_get_srcaddr(struct net_device *dev,
 324                                  const struct flowi4 *fl4)
 325 {
 326         struct in_device *in_dev = __in_dev_get_rcu(dev);
 327
 328         if (!in_dev)
 329                 return htonl(INADDR_ANY);
 330
 331         for_ifa(in_dev) {
 332                 if (fl4->saddr == ifa->ifa_local)
 333                         return fl4->saddr;
 334         } endfor_ifa(in_dev);
 335
 336         return htonl(INADDR_ANY);
 337 }
 338
 339 static struct sk_buff *igmpv3_newpack(struct net_device *dev, unsigned int mtu)
 340 {
 341         struct sk_buff *skb;
 342         struct rtable *rt;
 343         struct iphdr *pip;
 344         struct igmpv3_report *pig;
 345         struct net *net = dev_net(dev);
 346         struct flowi4 fl4;
 347         int hlen = LL_RESERVED_SPACE(dev);
 348         int tlen = dev->needed_tailroom;
 349         unsigned int size = mtu;
 350
 351         while (1) {
 352                 skb = alloc_skb(size + hlen + tlen,
 353                                 GFP_ATOMIC | __GFP_NOWARN);
 354                 if (skb)
 355                         break;
 356                 size >>= 1;
 357                 if (size < 256)
 358                         return NULL;
 359         }
 360         skb->priority = TC_PRIO_CONTROL;
 361
 362         rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
 363                                    0, 0,
 364                                    IPPROTO_IGMP, 0, dev->ifindex);
 365         if (IS_ERR(rt)) {
 366                 kfree_skb(skb);
 367                 return NULL;
 368         }
 369
 370         skb_dst_set(skb, &rt->dst);
 371         skb->dev = dev;
 372
 373         skb_reserve(skb, hlen);
 374         skb_tailroom_reserve(skb, mtu, tlen);
 375
 376         skb_reset_network_header(skb);
 377         pip = ip_hdr(skb);
 378         skb_put(skb, sizeof(struct iphdr) + 4);
 379
 380         pip->version  = 4;
 381         pip->ihl      = (sizeof(struct iphdr)+4)>>2;
 382         pip->tos      = 0xc0;
 383         pip->frag_off = htons(IP_DF);
 384         pip->ttl      = 1;
 385         pip->daddr    = fl4.daddr;
 386
 387         rcu_read_lock();
 388         pip->saddr    = igmpv3_get_srcaddr(dev, &fl4);
 389         rcu_read_unlock();
 390
 391         pip->protocol = IPPROTO_IGMP;
 392         pip->tot_len  = 0;      /* filled in later */
 393         ip_select_ident(net, skb, NULL);
 394         ((u8 *)&pip[1])[0] = IPOPT_RA;
 395         ((u8 *)&pip[1])[1] = 4;
 396         ((u8 *)&pip[1])[2] = 0;
 397         ((u8 *)&pip[1])[3] = 0;
 398
 399         skb->transport_header = skb->network_header + sizeof(struct iphdr) + 4;
 400         skb_put(skb, sizeof(*pig));
 401         pig = igmpv3_report_hdr(skb);
 402         pig->type = IGMPV3_HOST_MEMBERSHIP_REPORT;
 403         pig->resv1 = 0;
 404         pig->csum = 0;
 405         pig->resv2 = 0;
 406         pig->ngrec = 0;
 407         return skb;
 408 }
 409
 410 static int igmpv3_sendpack(struct sk_buff *skb)
 411 {
 412         struct igmphdr *pig = igmp_hdr(skb);
 413         const int igmplen = skb_tail_pointer(skb) - skb_transport_header(skb);
 414
 415         pig->csum = ip_compute_csum(igmp_hdr(skb), igmplen);
 416
 417         return ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 418 }
 419
 420 static int grec_size(struct ip_mc_list *pmc, int type, int gdel, int sdel)
 421 {
 422         return sizeof(struct igmpv3_grec) + 4*igmp_scount(pmc, type, gdel, sdel);
 423 }
 424
 425 static struct sk_buff *add_grhead(struct sk_buff *skb, struct ip_mc_list *pmc,
 426         int type, struct igmpv3_grec **ppgr, unsigned int mtu)
 427 {
 428         struct net_device *dev = pmc->interface->dev;
 429         struct igmpv3_report *pih;
 430         struct igmpv3_grec *pgr;
 431
 432         if (!skb) {
 433                 skb = igmpv3_newpack(dev, mtu);
 434                 if (!skb)
 435                         return NULL;
 436         }
 437         pgr = skb_put(skb, sizeof(struct igmpv3_grec));
 438         pgr->grec_type = type;
 439         pgr->grec_auxwords = 0;
 440         pgr->grec_nsrcs = 0;
 441         pgr->grec_mca = pmc->multiaddr;
 442         pih = igmpv3_report_hdr(skb);
 443         pih->ngrec = htons(ntohs(pih->ngrec)+1);
 444         *ppgr = pgr;
 445         return skb;
 446 }
 447
 448 #define AVAILABLE(skb)  ((skb) ? skb_availroom(skb) : 0)
 449
 450 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
 451         int type, int gdeleted, int sdeleted)
 452 {
 453         struct net_device *dev = pmc->interface->dev;
 454         struct net *net = dev_net(dev);
 455         struct igmpv3_report *pih;
 456         struct igmpv3_grec *pgr = NULL;
 457         struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
 458         int scount, stotal, first, isquery, truncate;
 459         unsigned int mtu;
 460
 461         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 462                 return skb;
 463         if (ipv4_is_local_multicast(pmc->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
 464                 return skb;
 465
 466         mtu = READ_ONCE(dev->mtu);
 467         if (mtu < IPV4_MIN_MTU)
 468                 return skb;
 469
 470         isquery = type == IGMPV3_MODE_IS_INCLUDE ||
 471                   type == IGMPV3_MODE_IS_EXCLUDE;
 472         truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
 473                     type == IGMPV3_CHANGE_TO_EXCLUDE;
 474
 475         stotal = scount = 0;
 476
 477         psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
 478
 479         if (!*psf_list)
 480                 goto empty_source;
 481
 482         pih = skb ? igmpv3_report_hdr(skb) : NULL;
 483
 484         /* EX and TO_EX get a fresh packet, if needed */
 485         if (truncate) {
 486                 if (pih && pih->ngrec &&
 487                     AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
 488                         if (skb)
 489                                 igmpv3_sendpack(skb);
 490                         skb = igmpv3_newpack(dev, mtu);
 491                 }
 492         }
 493         first = 1;
 494         psf_prev = NULL;
 495         for (psf = *psf_list; psf; psf = psf_next) {
 496                 __be32 *psrc;
 497
 498                 psf_next = psf->sf_next;
 499
 500                 if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
 501                         psf_prev = psf;
 502                         continue;
 503                 }
 504
 505                 /* Based on RFC3376 5.1. Should not send source-list change
 506                  * records when there is a filter mode change.
 507                  */
 508                 if (((gdeleted && pmc->sfmode == MCAST_EXCLUDE) ||
 509                      (!gdeleted && pmc->crcount)) &&
 510                     (type == IGMPV3_ALLOW_NEW_SOURCES ||
 511                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount)
 512                         goto decrease_sf_crcount;
 513
 514                 /* clear marks on query responses */
 515                 if (isquery)
 516                         psf->sf_gsresp = 0;
 517
 518                 if (AVAILABLE(skb) < sizeof(__be32) +
 519                     first*sizeof(struct igmpv3_grec)) {
 520                         if (truncate && !first)
 521                                 break;   /* truncate these */
 522                         if (pgr)
 523                                 pgr->grec_nsrcs = htons(scount);
 524                         if (skb)
 525                                 igmpv3_sendpack(skb);
 526                         skb = igmpv3_newpack(dev, mtu);
 527                         first = 1;
 528                         scount = 0;
 529                 }
 530                 if (first) {
 531                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 532                         first = 0;
 533                 }
 534                 if (!skb)
 535                         return NULL;
 536                 psrc = skb_put(skb, sizeof(__be32));
 537                 *psrc = psf->sf_inaddr;
 538                 scount++; stotal++;
 539                 if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
 540                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
 541 decrease_sf_crcount:
 542                         psf->sf_crcount--;
 543                         if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
 544                                 if (psf_prev)
 545                                         psf_prev->sf_next = psf->sf_next;
 546                                 else
 547                                         *psf_list = psf->sf_next;
 548                                 kfree(psf);
 549                                 continue;
 550                         }
 551                 }
 552                 psf_prev = psf;
 553         }
 554
 555 empty_source:
 556         if (!stotal) {
 557                 if (type == IGMPV3_ALLOW_NEW_SOURCES ||
 558                     type == IGMPV3_BLOCK_OLD_SOURCES)
 559                         return skb;
 560                 if (pmc->crcount || isquery) {
 561                         /* make sure we have room for group header */
 562                         if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
 563                                 igmpv3_sendpack(skb);
 564                                 skb = NULL; /* add_grhead will get a new one */
 565                         }
 566                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 567                 }
 568         }
 569         if (pgr)
 570                 pgr->grec_nsrcs = htons(scount);
 571
 572         if (isquery)
 573                 pmc->gsquery = 0;       /* clear query state on report */
 574         return skb;
 575 }
 576
 577 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
 578 {
 579         struct sk_buff *skb = NULL;
 580         struct net *net = dev_net(in_dev->dev);
 581         int type;
 582
 583         if (!pmc) {
 584                 rcu_read_lock();
 585                 for_each_pmc_rcu(in_dev, pmc) {
 586                         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 587                                 continue;
 588                         if (ipv4_is_local_multicast(pmc->multiaddr) &&
 589                              !net->ipv4.sysctl_igmp_llm_reports)
 590                                 continue;
 591                         spin_lock_bh(&pmc->lock);
 592                         if (pmc->sfcount[MCAST_EXCLUDE])
 593                                 type = IGMPV3_MODE_IS_EXCLUDE;
 594                         else
 595                                 type = IGMPV3_MODE_IS_INCLUDE;
 596                         skb = add_grec(skb, pmc, type, 0, 0);
 597                         spin_unlock_bh(&pmc->lock);
 598                 }
 599                 rcu_read_unlock();
 600         } else {
 601                 spin_lock_bh(&pmc->lock);
 602                 if (pmc->sfcount[MCAST_EXCLUDE])
 603                         type = IGMPV3_MODE_IS_EXCLUDE;
 604                 else
 605                         type = IGMPV3_MODE_IS_INCLUDE;
 606                 skb = add_grec(skb, pmc, type, 0, 0);
 607                 spin_unlock_bh(&pmc->lock);
 608         }
 609         if (!skb)
 610                 return 0;
 611         return igmpv3_sendpack(skb);
 612 }
 613
 614 /*
 615  * remove zero-count source records from a source filter list
 616  */
 617 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
 618 {
 619         struct ip_sf_list *psf_prev, *psf_next, *psf;
 620
 621         psf_prev = NULL;
 622         for (psf = *ppsf; psf; psf = psf_next) {
 623                 psf_next = psf->sf_next;
 624                 if (psf->sf_crcount == 0) {
 625                         if (psf_prev)
 626                                 psf_prev->sf_next = psf->sf_next;
 627                         else
 628                                 *ppsf = psf->sf_next;
 629                         kfree(psf);
 630                 } else
 631                         psf_prev = psf;
 632         }
 633 }
 634
 635 static void igmpv3_send_cr(struct in_device *in_dev)
 636 {
 637         struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
 638         struct sk_buff *skb = NULL;
 639         int type, dtype;
 640
 641         rcu_read_lock();
 642         spin_lock_bh(&in_dev->mc_tomb_lock);
 643
 644         /* deleted MCA's */
 645         pmc_prev = NULL;
 646         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
 647                 pmc_next = pmc->next;
 648                 if (pmc->sfmode == MCAST_INCLUDE) {
 649                         type = IGMPV3_BLOCK_OLD_SOURCES;
 650                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 651                         skb = add_grec(skb, pmc, type, 1, 0);
 652                         skb = add_grec(skb, pmc, dtype, 1, 1);
 653                 }
 654                 if (pmc->crcount) {
 655                         if (pmc->sfmode == MCAST_EXCLUDE) {
 656                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 657                                 skb = add_grec(skb, pmc, type, 1, 0);
 658                         }
 659                         pmc->crcount--;
 660                         if (pmc->crcount == 0) {
 661                                 igmpv3_clear_zeros(&pmc->tomb);
 662                                 igmpv3_clear_zeros(&pmc->sources);
 663                         }
 664                 }
 665                 if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
 666                         if (pmc_prev)
 667                                 pmc_prev->next = pmc_next;
 668                         else
 669                                 in_dev->mc_tomb = pmc_next;
 670                         in_dev_put(pmc->interface);
 671                         kfree(pmc);
 672                 } else
 673                         pmc_prev = pmc;
 674         }
 675         spin_unlock_bh(&in_dev->mc_tomb_lock);
 676
 677         /* change recs */
 678         for_each_pmc_rcu(in_dev, pmc) {
 679                 spin_lock_bh(&pmc->lock);
 680                 if (pmc->sfcount[MCAST_EXCLUDE]) {
 681                         type = IGMPV3_BLOCK_OLD_SOURCES;
 682                         dtype = IGMPV3_ALLOW_NEW_SOURCES;
 683                 } else {
 684                         type = IGMPV3_ALLOW_NEW_SOURCES;
 685                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 686                 }
 687                 skb = add_grec(skb, pmc, type, 0, 0);
 688                 skb = add_grec(skb, pmc, dtype, 0, 1);  /* deleted sources */
 689
 690                 /* filter mode changes */
 691                 if (pmc->crcount) {
 692                         if (pmc->sfmode == MCAST_EXCLUDE)
 693                                 type = IGMPV3_CHANGE_TO_EXCLUDE;
 694                         else
 695                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 696                         skb = add_grec(skb, pmc, type, 0, 0);
 697                         pmc->crcount--;
 698                 }
 699                 spin_unlock_bh(&pmc->lock);
 700         }
 701         rcu_read_unlock();
 702
 703         if (!skb)
 704                 return;
 705         (void) igmpv3_sendpack(skb);
 706 }
 707
 708 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
 709         int type)
 710 {
 711         struct sk_buff *skb;
 712         struct iphdr *iph;
 713         struct igmphdr *ih;
 714         struct rtable *rt;
 715         struct net_device *dev = in_dev->dev;
 716         struct net *net = dev_net(dev);
 717         __be32  group = pmc ? pmc->multiaddr : 0;
 718         struct flowi4 fl4;
 719         __be32  dst;
 720         int hlen, tlen;
 721
 722         if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
 723                 return igmpv3_send_report(in_dev, pmc);
 724
 725         if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
 726                 return 0;
 727
 728         if (type == IGMP_HOST_LEAVE_MESSAGE)
 729                 dst = IGMP_ALL_ROUTER;
 730         else
 731                 dst = group;
 732
 733         rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
 734                                    0, 0,
 735                                    IPPROTO_IGMP, 0, dev->ifindex);
 736         if (IS_ERR(rt))
 737                 return -1;
 738
 739         hlen = LL_RESERVED_SPACE(dev);
 740         tlen = dev->needed_tailroom;
 741         skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
 742         if (!skb) {
 743                 ip_rt_put(rt);
 744                 return -1;
 745         }
 746         skb->priority = TC_PRIO_CONTROL;
 747
 748         skb_dst_set(skb, &rt->dst);
 749
 750         skb_reserve(skb, hlen);
 751
 752         skb_reset_network_header(skb);
 753         iph = ip_hdr(skb);
 754         skb_put(skb, sizeof(struct iphdr) + 4);
 755
 756         iph->version  = 4;
 757         iph->ihl      = (sizeof(struct iphdr)+4)>>2;
 758         iph->tos      = 0xc0;
 759         iph->frag_off = htons(IP_DF);
 760         iph->ttl      = 1;
 761         iph->daddr    = dst;
 762         iph->saddr    = fl4.saddr;
 763         iph->protocol = IPPROTO_IGMP;
 764         ip_select_ident(net, skb, NULL);
 765         ((u8 *)&iph[1])[0] = IPOPT_RA;
 766         ((u8 *)&iph[1])[1] = 4;
 767         ((u8 *)&iph[1])[2] = 0;
 768         ((u8 *)&iph[1])[3] = 0;
 769
 770         ih = skb_put(skb, sizeof(struct igmphdr));
 771         ih->type = type;
 772         ih->code = 0;
 773         ih->csum = 0;
 774         ih->group = group;
 775         ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
 776
 777         return ip_local_out(net, skb->sk, skb);
 778 }
 779
 780 static void igmp_gq_timer_expire(struct timer_list *t)
 781 {
 782         struct in_device *in_dev = from_timer(in_dev, t, mr_gq_timer);
 783
 784         in_dev->mr_gq_running = 0;
 785         igmpv3_send_report(in_dev, NULL);
 786         in_dev_put(in_dev);
 787 }
 788
 789 static void igmp_ifc_timer_expire(struct timer_list *t)
 790 {
 791         struct in_device *in_dev = from_timer(in_dev, t, mr_ifc_timer);
 792
 793         igmpv3_send_cr(in_dev);
 794         if (in_dev->mr_ifc_count) {
 795                 in_dev->mr_ifc_count--;
 796                 igmp_ifc_start_timer(in_dev,
 797                                      unsolicited_report_interval(in_dev));
 798         }
 799         in_dev_put(in_dev);
 800 }
 801
 802 static void igmp_ifc_event(struct in_device *in_dev)
 803 {
 804         struct net *net = dev_net(in_dev->dev);
 805         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 806                 return;
 807         in_dev->mr_ifc_count = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
 808         igmp_ifc_start_timer(in_dev, 1);
 809 }
 810
 811
 812 static void igmp_timer_expire(struct timer_list *t)
 813 {
 814         struct ip_mc_list *im = from_timer(im, t, timer);
 815         struct in_device *in_dev = im->interface;
 816
 817         spin_lock(&im->lock);
 818         im->tm_running = 0;
 819
 820         if (im->unsolicit_count && --im->unsolicit_count)
 821                 igmp_start_timer(im, unsolicited_report_interval(in_dev));
 822
 823         im->reporter = 1;
 824         spin_unlock(&im->lock);
 825
 826         if (IGMP_V1_SEEN(in_dev))
 827                 igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
 828         else if (IGMP_V2_SEEN(in_dev))
 829                 igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
 830         else
 831                 igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
 832
 833         ip_ma_put(im);
 834 }
 835
 836 /* mark EXCLUDE-mode sources */
 837 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 838 {
 839         struct ip_sf_list *psf;
 840         int i, scount;
 841
 842         scount = 0;
 843         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 844                 if (scount == nsrcs)
 845                         break;
 846                 for (i = 0; i < nsrcs; i++) {
 847                         /* skip inactive filters */
 848                         if (psf->sf_count[MCAST_INCLUDE] ||
 849                             pmc->sfcount[MCAST_EXCLUDE] !=
 850                             psf->sf_count[MCAST_EXCLUDE])
 851                                 break;
 852                         if (srcs[i] == psf->sf_inaddr) {
 853                                 scount++;
 854                                 break;
 855                         }
 856                 }
 857         }
 858         pmc->gsquery = 0;
 859         if (scount == nsrcs)    /* all sources excluded */
 860                 return 0;
 861         return 1;
 862 }
 863
 864 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 865 {
 866         struct ip_sf_list *psf;
 867         int i, scount;
 868
 869         if (pmc->sfmode == MCAST_EXCLUDE)
 870                 return igmp_xmarksources(pmc, nsrcs, srcs);
 871
 872         /* mark INCLUDE-mode sources */
 873         scount = 0;
 874         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 875                 if (scount == nsrcs)
 876                         break;
 877                 for (i = 0; i < nsrcs; i++)
 878                         if (srcs[i] == psf->sf_inaddr) {
 879                                 psf->sf_gsresp = 1;
 880                                 scount++;
 881                                 break;
 882                         }
 883         }
 884         if (!scount) {
 885                 pmc->gsquery = 0;
 886                 return 0;
 887         }
 888         pmc->gsquery = 1;
 889         return 1;
 890 }
 891
 892 /* return true if packet was dropped */
 893 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
 894 {
 895         struct ip_mc_list *im;
 896         struct net *net = dev_net(in_dev->dev);
 897
 898         /* Timers are only set for non-local groups */
 899
 900         if (group == IGMP_ALL_HOSTS)
 901                 return false;
 902         if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
 903                 return false;
 904
 905         rcu_read_lock();
 906         for_each_pmc_rcu(in_dev, im) {
 907                 if (im->multiaddr == group) {
 908                         igmp_stop_timer(im);
 909                         break;
 910                 }
 911         }
 912         rcu_read_unlock();
 913         return false;
 914 }
 915
 916 /* return true if packet was dropped */
 917 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
 918         int len)
 919 {
 920         struct igmphdr          *ih = igmp_hdr(skb);
 921         struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
 922         struct ip_mc_list       *im;
 923         __be32                  group = ih->group;
 924         int                     max_delay;
 925         int                     mark = 0;
 926         struct net              *net = dev_net(in_dev->dev);
 927
 928
 929         if (len == 8) {
 930                 if (ih->code == 0) {
 931                         /* Alas, old v1 router presents here. */
 932
 933                         max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 934                         in_dev->mr_v1_seen = jiffies +
 935                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 936                                 in_dev->mr_qri;
 937                         group = 0;
 938                 } else {
 939                         /* v2 router present */
 940                         max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
 941                         in_dev->mr_v2_seen = jiffies +
 942                                 (in_dev->mr_qrv * in_dev->mr_qi) +
 943                                 in_dev->mr_qri;
 944                 }
 945                 /* cancel the interface change timer */
 946                 in_dev->mr_ifc_count = 0;
 947                 if (del_timer(&in_dev->mr_ifc_timer))
 948                         __in_dev_put(in_dev);
 949                 /* clear deleted report items */
 950                 igmpv3_clear_delrec(in_dev);
 951         } else if (len < 12) {
 952                 return true;    /* ignore bogus packet; freed by caller */
 953         } else if (IGMP_V1_SEEN(in_dev)) {
 954                 /* This is a v3 query with v1 queriers present */
 955                 max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 956                 group = 0;
 957         } else if (IGMP_V2_SEEN(in_dev)) {
 958                 /* this is a v3 query with v2 queriers present;
 959                  * Interpretation of the max_delay code is problematic here.
 960                  * A real v2 host would use ih_code directly, while v3 has a
 961                  * different encoding. We use the v3 encoding as more likely
 962                  * to be intended in a v3 query.
 963                  */
 964                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 965                 if (!max_delay)
 966                         max_delay = 1;  /* can't mod w/ 0 */
 967         } else { /* v3 */
 968                 if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
 969                         return true;
 970
 971                 ih3 = igmpv3_query_hdr(skb);
 972                 if (ih3->nsrcs) {
 973                         if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
 974                                            + ntohs(ih3->nsrcs)*sizeof(__be32)))
 975                                 return true;
 976                         ih3 = igmpv3_query_hdr(skb);
 977                 }
 978
 979                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 980                 if (!max_delay)
 981                         max_delay = 1;  /* can't mod w/ 0 */
 982                 in_dev->mr_maxdelay = max_delay;
 983
 984                 /* RFC3376, 4.1.6. QRV and 4.1.7. QQIC, when the most recently
 985                  * received value was zero, use the default or statically
 986                  * configured value.
 987                  */
 988                 in_dev->mr_qrv = ih3->qrv ?: net->ipv4.sysctl_igmp_qrv;
 989                 in_dev->mr_qi = IGMPV3_QQIC(ih3->qqic)*HZ ?: IGMP_QUERY_INTERVAL;
 990
 991                 /* RFC3376, 8.3. Query Response Interval:
 992                  * The number of seconds represented by the [Query Response
 993                  * Interval] must be less than the [Query Interval].
 994                  */
 995                 if (in_dev->mr_qri >= in_dev->mr_qi)
 996                         in_dev->mr_qri = (in_dev->mr_qi/HZ - 1)*HZ;
 997
 998                 if (!group) { /* general query */
 999                         if (ih3->nsrcs)
1000                                 return true;    /* no sources allowed */
1001                         igmp_gq_start_timer(in_dev);
1002                         return false;
1003                 }
1004                 /* mark sources to include, if group & source-specific */
1005                 mark = ih3->nsrcs != 0;
1006         }
1007
1008         /*
1009          * - Start the timers in all of our membership records
1010          *   that the query applies to for the interface on
1011          *   which the query arrived excl. those that belong
1012          *   to a "local" group (224.0.0.X)
1013          * - For timers already running check if they need to
1014          *   be reset.
1015          * - Use the igmp->igmp_code field as the maximum
1016          *   delay possible
1017          */
1018         rcu_read_lock();
1019         for_each_pmc_rcu(in_dev, im) {
1020                 int changed;
1021
1022                 if (group && group != im->multiaddr)
1023                         continue;
1024                 if (im->multiaddr == IGMP_ALL_HOSTS)
1025                         continue;
1026                 if (ipv4_is_local_multicast(im->multiaddr) &&
1027                     !net->ipv4.sysctl_igmp_llm_reports)
1028                         continue;
1029                 spin_lock_bh(&im->lock);
1030                 if (im->tm_running)
1031                         im->gsquery = im->gsquery && mark;
1032                 else
1033                         im->gsquery = mark;
1034                 changed = !im->gsquery ||
1035                         igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
1036                 spin_unlock_bh(&im->lock);
1037                 if (changed)
1038                         igmp_mod_timer(im, max_delay);
1039         }
1040         rcu_read_unlock();
1041         return false;
1042 }
1043
1044 /* called in rcu_read_lock() section */
1045 int igmp_rcv(struct sk_buff *skb)
1046 {
1047         /* This basically follows the spec line by line -- see RFC1112 */
1048         struct igmphdr *ih;
1049         struct net_device *dev = skb->dev;
1050         struct in_device *in_dev;
1051         int len = skb->len;
1052         bool dropped = true;
1053
1054         if (netif_is_l3_master(dev)) {
1055                 dev = dev_get_by_index_rcu(dev_net(dev), IPCB(skb)->iif);
1056                 if (!dev)
1057                         goto drop;
1058         }
1059
1060         in_dev = __in_dev_get_rcu(dev);
1061         if (!in_dev)
1062                 goto drop;
1063
1064         if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
1065                 goto drop;
1066
1067         if (skb_checksum_simple_validate(skb))
1068                 goto drop;
1069
1070         ih = igmp_hdr(skb);
1071         switch (ih->type) {
1072         case IGMP_HOST_MEMBERSHIP_QUERY:
1073                 dropped = igmp_heard_query(in_dev, skb, len);
1074                 break;
1075         case IGMP_HOST_MEMBERSHIP_REPORT:
1076         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1077                 /* Is it our report looped back? */
1078                 if (rt_is_output_route(skb_rtable(skb)))
1079                         break;
1080                 /* don't rely on MC router hearing unicast reports */
1081                 if (skb->pkt_type == PACKET_MULTICAST ||
1082                     skb->pkt_type == PACKET_BROADCAST)
1083                         dropped = igmp_heard_report(in_dev, ih->group);
1084                 break;
1085         case IGMP_PIM:
1086 #ifdef CONFIG_IP_PIMSM_V1
1087                 return pim_rcv_v1(skb);
1088 #endif
1089         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1090         case IGMP_DVMRP:
1091         case IGMP_TRACE:
1092         case IGMP_HOST_LEAVE_MESSAGE:
1093         case IGMP_MTRACE:
1094         case IGMP_MTRACE_RESP:
1095                 break;
1096         default:
1097                 break;
1098         }
1099
1100 drop:
1101         if (dropped)
1102                 kfree_skb(skb);
1103         else
1104                 consume_skb(skb);
1105         return 0;
1106 }
1107
1108 #endif
1109
1110
1111 /*
1112  *      Add a filter to a device
1113  */
1114
1115 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1116 {
1117         char buf[MAX_ADDR_LEN];
1118         struct net_device *dev = in_dev->dev;
1119
1120         /* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1121            We will get multicast token leakage, when IFF_MULTICAST
1122            is changed. This check should be done in ndo_set_rx_mode
1123            routine. Something sort of:
1124            if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1125            --ANK
1126            */
1127         if (arp_mc_map(addr, buf, dev, 0) == 0)
1128                 dev_mc_add(dev, buf);
1129 }
1130
1131 /*
1132  *      Remove a filter from a device
1133  */
1134
1135 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1136 {
1137         char buf[MAX_ADDR_LEN];
1138         struct net_device *dev = in_dev->dev;
1139
1140         if (arp_mc_map(addr, buf, dev, 0) == 0)
1141                 dev_mc_del(dev, buf);
1142 }
1143
1144 #ifdef CONFIG_IP_MULTICAST
1145 /*
1146  * deleted ip_mc_list manipulation
1147  */
1148 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1149 {
1150         struct ip_mc_list *pmc;
1151         struct net *net = dev_net(in_dev->dev);
1152
1153         /* this is an "ip_mc_list" for convenience; only the fields below
1154          * are actually used. In particular, the refcnt and users are not
1155          * used for management of the delete list. Using the same structure
1156          * for deleted items allows change reports to use common code with
1157          * non-deleted or query-response MCA's.
1158          */
1159         pmc = kzalloc(sizeof(*pmc), GFP_KERNEL);
1160         if (!pmc)
1161                 return;
1162         spin_lock_init(&pmc->lock);
1163         spin_lock_bh(&im->lock);
1164         pmc->interface = im->interface;
1165         in_dev_hold(in_dev);
1166         pmc->multiaddr = im->multiaddr;
1167         pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1168         pmc->sfmode = im->sfmode;
1169         if (pmc->sfmode == MCAST_INCLUDE) {
1170                 struct ip_sf_list *psf;
1171
1172                 pmc->tomb = im->tomb;
1173                 pmc->sources = im->sources;
1174                 im->tomb = im->sources = NULL;
1175                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1176                         psf->sf_crcount = pmc->crcount;
1177         }
1178         spin_unlock_bh(&im->lock);
1179
1180         spin_lock_bh(&in_dev->mc_tomb_lock);
1181         pmc->next = in_dev->mc_tomb;
1182         in_dev->mc_tomb = pmc;
1183         spin_unlock_bh(&in_dev->mc_tomb_lock);
1184 }
1185
1186 /*
1187  * restore ip_mc_list deleted records
1188  */
1189 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1190 {
1191         struct ip_mc_list *pmc, *pmc_prev;
1192         struct ip_sf_list *psf;
1193         struct net *net = dev_net(in_dev->dev);
1194         __be32 multiaddr = im->multiaddr;
1195
1196         spin_lock_bh(&in_dev->mc_tomb_lock);
1197         pmc_prev = NULL;
1198         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1199                 if (pmc->multiaddr == multiaddr)
1200                         break;
1201                 pmc_prev = pmc;
1202         }
1203         if (pmc) {
1204                 if (pmc_prev)
1205                         pmc_prev->next = pmc->next;
1206                 else
1207                         in_dev->mc_tomb = pmc->next;
1208         }
1209         spin_unlock_bh(&in_dev->mc_tomb_lock);
1210
1211         spin_lock_bh(&im->lock);
1212         if (pmc) {
1213                 im->interface = pmc->interface;
1214                 if (im->sfmode == MCAST_INCLUDE) {
1215                         im->tomb = pmc->tomb;
1216                         im->sources = pmc->sources;
1217                         for (psf = im->sources; psf; psf = psf->sf_next)
1218                                 psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1219                 } else {
1220                         im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1221                 }
1222                 in_dev_put(pmc->interface);
1223                 kfree(pmc);
1224         }
1225         spin_unlock_bh(&im->lock);
1226 }
1227
1228 /*
1229  * flush ip_mc_list deleted records
1230  */
1231 static void igmpv3_clear_delrec(struct in_device *in_dev)
1232 {
1233         struct ip_mc_list *pmc, *nextpmc;
1234
1235         spin_lock_bh(&in_dev->mc_tomb_lock);
1236         pmc = in_dev->mc_tomb;
1237         in_dev->mc_tomb = NULL;
1238         spin_unlock_bh(&in_dev->mc_tomb_lock);
1239
1240         for (; pmc; pmc = nextpmc) {
1241                 nextpmc = pmc->next;
1242                 ip_mc_clear_src(pmc);
1243                 in_dev_put(pmc->interface);
1244                 kfree(pmc);
1245         }
1246         /* clear dead sources, too */
1247         rcu_read_lock();
1248         for_each_pmc_rcu(in_dev, pmc) {
1249                 struct ip_sf_list *psf, *psf_next;
1250
1251                 spin_lock_bh(&pmc->lock);
1252                 psf = pmc->tomb;
1253                 pmc->tomb = NULL;
1254                 spin_unlock_bh(&pmc->lock);
1255                 for (; psf; psf = psf_next) {
1256                         psf_next = psf->sf_next;
1257                         kfree(psf);
1258                 }
1259         }
1260         rcu_read_unlock();
1261 }
1262 #endif
1263
1264 static void igmp_group_dropped(struct ip_mc_list *im)
1265 {
1266         struct in_device *in_dev = im->interface;
1267 #ifdef CONFIG_IP_MULTICAST
1268         struct net *net = dev_net(in_dev->dev);
1269         int reporter;
1270 #endif
1271
1272         if (im->loaded) {
1273                 im->loaded = 0;
1274                 ip_mc_filter_del(in_dev, im->multiaddr);
1275         }
1276
1277 #ifdef CONFIG_IP_MULTICAST
1278         if (im->multiaddr == IGMP_ALL_HOSTS)
1279                 return;
1280         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1281                 return;
1282
1283         reporter = im->reporter;
1284         igmp_stop_timer(im);
1285
1286         if (!in_dev->dead) {
1287                 if (IGMP_V1_SEEN(in_dev))
1288                         return;
1289                 if (IGMP_V2_SEEN(in_dev)) {
1290                         if (reporter)
1291                                 igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1292                         return;
1293                 }
1294                 /* IGMPv3 */
1295                 igmpv3_add_delrec(in_dev, im);
1296
1297                 igmp_ifc_event(in_dev);
1298         }
1299 #endif
1300 }
1301
1302 static void igmp_group_added(struct ip_mc_list *im)
1303 {
1304         struct in_device *in_dev = im->interface;
1305 #ifdef CONFIG_IP_MULTICAST
1306         struct net *net = dev_net(in_dev->dev);
1307 #endif
1308
1309         if (im->loaded == 0) {
1310                 im->loaded = 1;
1311                 ip_mc_filter_add(in_dev, im->multiaddr);
1312         }
1313
1314 #ifdef CONFIG_IP_MULTICAST
1315         if (im->multiaddr == IGMP_ALL_HOSTS)
1316                 return;
1317         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1318                 return;
1319
1320         if (in_dev->dead)
1321                 return;
1322
1323         im->unsolicit_count = net->ipv4.sysctl_igmp_qrv;
1324         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1325                 spin_lock_bh(&im->lock);
1326                 igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1327                 spin_unlock_bh(&im->lock);
1328                 return;
1329         }
1330         /* else, v3 */
1331
1332         /* Based on RFC3376 5.1, for newly added INCLUDE SSM, we should
1333          * not send filter-mode change record as the mode should be from
1334          * IN() to IN(A).
1335          */
1336         if (im->sfmode == MCAST_EXCLUDE)
1337                 im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1338
1339         igmp_ifc_event(in_dev);
1340 #endif
1341 }
1342
1343
1344 /*
1345  *      Multicast list managers
1346  */
1347
1348 static u32 ip_mc_hash(const struct ip_mc_list *im)
1349 {
1350         return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1351 }
1352
1353 static void ip_mc_hash_add(struct in_device *in_dev,
1354                            struct ip_mc_list *im)
1355 {
1356         struct ip_mc_list __rcu **mc_hash;
1357         u32 hash;
1358
1359         mc_hash = rtnl_dereference(in_dev->mc_hash);
1360         if (mc_hash) {
1361                 hash = ip_mc_hash(im);
1362                 im->next_hash = mc_hash[hash];
1363                 rcu_assign_pointer(mc_hash[hash], im);
1364                 return;
1365         }
1366
1367         /* do not use a hash table for small number of items */
1368         if (in_dev->mc_count < 4)
1369                 return;
1370
1371         mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1372                           GFP_KERNEL);
1373         if (!mc_hash)
1374                 return;
1375
1376         for_each_pmc_rtnl(in_dev, im) {
1377                 hash = ip_mc_hash(im);
1378                 im->next_hash = mc_hash[hash];
1379                 RCU_INIT_POINTER(mc_hash[hash], im);
1380         }
1381
1382         rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1383 }
1384
1385 static void ip_mc_hash_remove(struct in_device *in_dev,
1386                               struct ip_mc_list *im)
1387 {
1388         struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1389         struct ip_mc_list *aux;
1390
1391         if (!mc_hash)
1392                 return;
1393         mc_hash += ip_mc_hash(im);
1394         while ((aux = rtnl_dereference(*mc_hash)) != im)
1395                 mc_hash = &aux->next_hash;
1396         *mc_hash = im->next_hash;
1397 }
1398
1399
1400 /*
1401  *      A socket has joined a multicast group on device dev.
1402  */
1403 static void __ip_mc_inc_group(struct in_device *in_dev, __be32 addr,
1404                               unsigned int mode)
1405 {
1406         struct ip_mc_list *im;
1407
1408         ASSERT_RTNL();
1409
1410         for_each_pmc_rtnl(in_dev, im) {
1411                 if (im->multiaddr == addr) {
1412                         im->users++;
1413                         ip_mc_add_src(in_dev, &addr, mode, 0, NULL, 0);
1414                         goto out;
1415                 }
1416         }
1417
1418         im = kzalloc(sizeof(*im), GFP_KERNEL);
1419         if (!im)
1420                 goto out;
1421
1422         im->users = 1;
1423         im->interface = in_dev;
1424         in_dev_hold(in_dev);
1425         im->multiaddr = addr;
1426         /* initial mode is (EX, empty) */
1427         im->sfmode = mode;
1428         im->sfcount[mode] = 1;
1429         refcount_set(&im->refcnt, 1);
1430         spin_lock_init(&im->lock);
1431 #ifdef CONFIG_IP_MULTICAST
1432         timer_setup(&im->timer, igmp_timer_expire, 0);
1433 #endif
1434
1435         im->next_rcu = in_dev->mc_list;
1436         in_dev->mc_count++;
1437         rcu_assign_pointer(in_dev->mc_list, im);
1438
1439         ip_mc_hash_add(in_dev, im);
1440
1441 #ifdef CONFIG_IP_MULTICAST
1442         igmpv3_del_delrec(in_dev, im);
1443 #endif
1444         igmp_group_added(im);
1445         if (!in_dev->dead)
1446                 ip_rt_multicast_event(in_dev);
1447 out:
1448         return;
1449 }
1450
1451 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1452 {
1453         __ip_mc_inc_group(in_dev, addr, MCAST_EXCLUDE);
1454 }
1455 EXPORT_SYMBOL(ip_mc_inc_group);
1456
1457 static int ip_mc_check_iphdr(struct sk_buff *skb)
1458 {
1459         const struct iphdr *iph;
1460         unsigned int len;
1461         unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1462
1463         if (!pskb_may_pull(skb, offset))
1464                 return -EINVAL;
1465
1466         iph = ip_hdr(skb);
1467
1468         if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1469                 return -EINVAL;
1470
1471         offset += ip_hdrlen(skb) - sizeof(*iph);
1472
1473         if (!pskb_may_pull(skb, offset))
1474                 return -EINVAL;
1475
1476         iph = ip_hdr(skb);
1477
1478         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1479                 return -EINVAL;
1480
1481         len = skb_network_offset(skb) + ntohs(iph->tot_len);
1482         if (skb->len < len || len < offset)
1483                 return -EINVAL;
1484
1485         skb_set_transport_header(skb, offset);
1486
1487         return 0;
1488 }
1489
1490 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1491 {
1492         unsigned int len = skb_transport_offset(skb);
1493
1494         len += sizeof(struct igmpv3_report);
1495
1496         return pskb_may_pull(skb, len) ? 0 : -EINVAL;
1497 }
1498
1499 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1500 {
1501         unsigned int len = skb_transport_offset(skb);
1502
1503         len += sizeof(struct igmphdr);
1504         if (skb->len < len)
1505                 return -EINVAL;
1506
1507         /* IGMPv{1,2}? */
1508         if (skb->len != len) {
1509                 /* or IGMPv3? */
1510                 len += sizeof(struct igmpv3_query) - sizeof(struct igmphdr);
1511                 if (skb->len < len || !pskb_may_pull(skb, len))
1512                         return -EINVAL;
1513         }
1514
1515         /* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1516          * all-systems destination addresses (224.0.0.1) for general queries
1517          */
1518         if (!igmp_hdr(skb)->group &&
1519             ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1520                 return -EINVAL;
1521
1522         return 0;
1523 }
1524
1525 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1526 {
1527         switch (igmp_hdr(skb)->type) {
1528         case IGMP_HOST_LEAVE_MESSAGE:
1529         case IGMP_HOST_MEMBERSHIP_REPORT:
1530         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1531                 /* fall through */
1532                 return 0;
1533         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1534                 return ip_mc_check_igmp_reportv3(skb);
1535         case IGMP_HOST_MEMBERSHIP_QUERY:
1536                 return ip_mc_check_igmp_query(skb);
1537         default:
1538                 return -ENOMSG;
1539         }
1540 }
1541
1542 static inline __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1543 {
1544         return skb_checksum_simple_validate(skb);
1545 }
1546
1547 static int __ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1548
1549 {
1550         struct sk_buff *skb_chk;
1551         unsigned int transport_len;
1552         unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1553         int ret = -EINVAL;
1554
1555         transport_len = ntohs(ip_hdr(skb)->tot_len) - ip_hdrlen(skb);
1556
1557         skb_chk = skb_checksum_trimmed(skb, transport_len,
1558                                        ip_mc_validate_checksum);
1559         if (!skb_chk)
1560                 goto err;
1561
1562         if (!pskb_may_pull(skb_chk, len))
1563                 goto err;
1564
1565         ret = ip_mc_check_igmp_msg(skb_chk);
1566         if (ret)
1567                 goto err;
1568
1569         if (skb_trimmed)
1570                 *skb_trimmed = skb_chk;
1571         /* free now unneeded clone */
1572         else if (skb_chk != skb)
1573                 kfree_skb(skb_chk);
1574
1575         ret = 0;
1576
1577 err:
1578         if (ret && skb_chk && skb_chk != skb)
1579                 kfree_skb(skb_chk);
1580
1581         return ret;
1582 }
1583
1584 /**
1585  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1586  * @skb: the skb to validate
1587  * @skb_trimmed: to store an skb pointer trimmed to IPv4 packet tail (optional)
1588  *
1589  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1590  * skb transport header accordingly and returns zero.
1591  *
1592  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1593  *  standard
1594  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1595  * -ENOMEM: A memory allocation failure happened.
1596  *
1597  * Optionally, an skb pointer might be provided via skb_trimmed (or set it
1598  * to NULL): After parsing an IGMP packet successfully it will point to
1599  * an skb which has its tail aligned to the IP packet end. This might
1600  * either be the originally provided skb or a trimmed, cloned version if
1601  * the skb frame had data beyond the IP packet. A cloned skb allows us
1602  * to leave the original skb and its full frame unchanged (which might be
1603  * desirable for layer 2 frame jugglers).
1604  *
1605  * Caller needs to set the skb network header and free any returned skb if it
1606  * differs from the provided skb.
1607  */
1608 int ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1609 {
1610         int ret = ip_mc_check_iphdr(skb);
1611
1612         if (ret < 0)
1613                 return ret;
1614
1615         if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1616                 return -ENOMSG;
1617
1618         return __ip_mc_check_igmp(skb, skb_trimmed);
1619 }
1620 EXPORT_SYMBOL(ip_mc_check_igmp);
1621
1622 /*
1623  *      Resend IGMP JOIN report; used by netdev notifier.
1624  */
1625 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1626 {
1627 #ifdef CONFIG_IP_MULTICAST
1628         struct ip_mc_list *im;
1629         int type;
1630         struct net *net = dev_net(in_dev->dev);
1631
1632         ASSERT_RTNL();
1633
1634         for_each_pmc_rtnl(in_dev, im) {
1635                 if (im->multiaddr == IGMP_ALL_HOSTS)
1636                         continue;
1637                 if (ipv4_is_local_multicast(im->multiaddr) &&
1638                     !net->ipv4.sysctl_igmp_llm_reports)
1639                         continue;
1640
1641                 /* a failover is happening and switches
1642                  * must be notified immediately
1643                  */
1644                 if (IGMP_V1_SEEN(in_dev))
1645                         type = IGMP_HOST_MEMBERSHIP_REPORT;
1646                 else if (IGMP_V2_SEEN(in_dev))
1647                         type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1648                 else
1649                         type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1650                 igmp_send_report(in_dev, im, type);
1651         }
1652 #endif
1653 }
1654
1655 /*
1656  *      A socket has left a multicast group on device dev
1657  */
1658
1659 void ip_mc_dec_group(struct in_device *in_dev, __be32 addr)
1660 {
1661         struct ip_mc_list *i;
1662         struct ip_mc_list __rcu **ip;
1663
1664         ASSERT_RTNL();
1665
1666         for (ip = &in_dev->mc_list;
1667              (i = rtnl_dereference(*ip)) != NULL;
1668              ip = &i->next_rcu) {
1669                 if (i->multiaddr == addr) {
1670                         if (--i->users == 0) {
1671                                 ip_mc_hash_remove(in_dev, i);
1672                                 *ip = i->next_rcu;
1673                                 in_dev->mc_count--;
1674                                 igmp_group_dropped(i);
1675                                 ip_mc_clear_src(i);
1676
1677                                 if (!in_dev->dead)
1678                                         ip_rt_multicast_event(in_dev);
1679
1680                                 ip_ma_put(i);
1681                                 return;
1682                         }
1683                         break;
1684                 }
1685         }
1686 }
1687 EXPORT_SYMBOL(ip_mc_dec_group);
1688
1689 /* Device changing type */
1690
1691 void ip_mc_unmap(struct in_device *in_dev)
1692 {
1693         struct ip_mc_list *pmc;
1694
1695         ASSERT_RTNL();
1696
1697         for_each_pmc_rtnl(in_dev, pmc)
1698                 igmp_group_dropped(pmc);
1699 }
1700
1701 void ip_mc_remap(struct in_device *in_dev)
1702 {
1703         struct ip_mc_list *pmc;
1704
1705         ASSERT_RTNL();
1706
1707         for_each_pmc_rtnl(in_dev, pmc) {
1708 #ifdef CONFIG_IP_MULTICAST
1709                 igmpv3_del_delrec(in_dev, pmc);
1710 #endif
1711                 igmp_group_added(pmc);
1712         }
1713 }
1714
1715 /* Device going down */
1716
1717 void ip_mc_down(struct in_device *in_dev)
1718 {
1719         struct ip_mc_list *pmc;
1720
1721         ASSERT_RTNL();
1722
1723         for_each_pmc_rtnl(in_dev, pmc)
1724                 igmp_group_dropped(pmc);
1725
1726 #ifdef CONFIG_IP_MULTICAST
1727         in_dev->mr_ifc_count = 0;
1728         if (del_timer(&in_dev->mr_ifc_timer))
1729                 __in_dev_put(in_dev);
1730         in_dev->mr_gq_running = 0;
1731         if (del_timer(&in_dev->mr_gq_timer))
1732                 __in_dev_put(in_dev);
1733 #endif
1734
1735         ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1736 }
1737
1738 #ifdef CONFIG_IP_MULTICAST
1739 static void ip_mc_reset(struct in_device *in_dev)
1740 {
1741         struct net *net = dev_net(in_dev->dev);
1742
1743         in_dev->mr_qi = IGMP_QUERY_INTERVAL;
1744         in_dev->mr_qri = IGMP_QUERY_RESPONSE_INTERVAL;
1745         in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1746 }
1747 #else
1748 static void ip_mc_reset(struct in_device *in_dev)
1749 {
1750 }
1751 #endif
1752
1753 void ip_mc_init_dev(struct in_device *in_dev)
1754 {
1755         ASSERT_RTNL();
1756
1757 #ifdef CONFIG_IP_MULTICAST
1758         timer_setup(&in_dev->mr_gq_timer, igmp_gq_timer_expire, 0);
1759         timer_setup(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire, 0);
1760 #endif
1761         ip_mc_reset(in_dev);
1762
1763         spin_lock_init(&in_dev->mc_tomb_lock);
1764 }
1765
1766 /* Device going up */
1767
1768 void ip_mc_up(struct in_device *in_dev)
1769 {
1770         struct ip_mc_list *pmc;
1771
1772         ASSERT_RTNL();
1773
1774         ip_mc_reset(in_dev);
1775         ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1776
1777         for_each_pmc_rtnl(in_dev, pmc) {
1778 #ifdef CONFIG_IP_MULTICAST
1779                 igmpv3_del_delrec(in_dev, pmc);
1780 #endif
1781                 igmp_group_added(pmc);
1782         }
1783 }
1784
1785 /*
1786  *      Device is about to be destroyed: clean up.
1787  */
1788
1789 void ip_mc_destroy_dev(struct in_device *in_dev)
1790 {
1791         struct ip_mc_list *i;
1792
1793         ASSERT_RTNL();
1794
1795         /* Deactivate timers */
1796         ip_mc_down(in_dev);
1797 #ifdef CONFIG_IP_MULTICAST
1798         igmpv3_clear_delrec(in_dev);
1799 #endif
1800
1801         while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1802                 in_dev->mc_list = i->next_rcu;
1803                 in_dev->mc_count--;
1804                 ip_ma_put(i);
1805         }
1806 }
1807
1808 /* RTNL is locked */
1809 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1810 {
1811         struct net_device *dev = NULL;
1812         struct in_device *idev = NULL;
1813
1814         if (imr->imr_ifindex) {
1815                 idev = inetdev_by_index(net, imr->imr_ifindex);
1816                 return idev;
1817         }
1818         if (imr->imr_address.s_addr) {
1819                 dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1820                 if (!dev)
1821                         return NULL;
1822         }
1823
1824         if (!dev) {
1825                 struct rtable *rt = ip_route_output(net,
1826                                                     imr->imr_multiaddr.s_addr,
1827                                                     0, 0, 0);
1828                 if (!IS_ERR(rt)) {
1829                         dev = rt->dst.dev;
1830                         ip_rt_put(rt);
1831                 }
1832         }
1833         if (dev) {
1834                 imr->imr_ifindex = dev->ifindex;
1835                 idev = __in_dev_get_rtnl(dev);
1836         }
1837         return idev;
1838 }
1839
1840 /*
1841  *      Join a socket to a group
1842  */
1843
1844 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1845         __be32 *psfsrc)
1846 {
1847         struct ip_sf_list *psf, *psf_prev;
1848         int rv = 0;
1849
1850         psf_prev = NULL;
1851         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1852                 if (psf->sf_inaddr == *psfsrc)
1853                         break;
1854                 psf_prev = psf;
1855         }
1856         if (!psf || psf->sf_count[sfmode] == 0) {
1857                 /* source filter not found, or count wrong =>  bug */
1858                 return -ESRCH;
1859         }
1860         psf->sf_count[sfmode]--;
1861         if (psf->sf_count[sfmode] == 0) {
1862                 ip_rt_multicast_event(pmc->interface);
1863         }
1864         if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1865 #ifdef CONFIG_IP_MULTICAST
1866                 struct in_device *in_dev = pmc->interface;
1867                 struct net *net = dev_net(in_dev->dev);
1868 #endif
1869
1870                 /* no more filters for this source */
1871                 if (psf_prev)
1872                         psf_prev->sf_next = psf->sf_next;
1873                 else
1874                         pmc->sources = psf->sf_next;
1875 #ifdef CONFIG_IP_MULTICAST
1876                 if (psf->sf_oldin &&
1877                     !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1878                         psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1879                         psf->sf_next = pmc->tomb;
1880                         pmc->tomb = psf;
1881                         rv = 1;
1882                 } else
1883 #endif
1884                         kfree(psf);
1885         }
1886         return rv;
1887 }
1888
1889 #ifndef CONFIG_IP_MULTICAST
1890 #define igmp_ifc_event(x)       do { } while (0)
1891 #endif
1892
1893 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1894                          int sfcount, __be32 *psfsrc, int delta)
1895 {
1896         struct ip_mc_list *pmc;
1897         int     changerec = 0;
1898         int     i, err;
1899
1900         if (!in_dev)
1901                 return -ENODEV;
1902         rcu_read_lock();
1903         for_each_pmc_rcu(in_dev, pmc) {
1904                 if (*pmca == pmc->multiaddr)
1905                         break;
1906         }
1907         if (!pmc) {
1908                 /* MCA not found?? bug */
1909                 rcu_read_unlock();
1910                 return -ESRCH;
1911         }
1912         spin_lock_bh(&pmc->lock);
1913         rcu_read_unlock();
1914 #ifdef CONFIG_IP_MULTICAST
1915         sf_markstate(pmc);
1916 #endif
1917         if (!delta) {
1918                 err = -EINVAL;
1919                 if (!pmc->sfcount[sfmode])
1920                         goto out_unlock;
1921                 pmc->sfcount[sfmode]--;
1922         }
1923         err = 0;
1924         for (i = 0; i < sfcount; i++) {
1925                 int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1926
1927                 changerec |= rv > 0;
1928                 if (!err && rv < 0)
1929                         err = rv;
1930         }
1931         if (pmc->sfmode == MCAST_EXCLUDE &&
1932             pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1933             pmc->sfcount[MCAST_INCLUDE]) {
1934 #ifdef CONFIG_IP_MULTICAST
1935                 struct ip_sf_list *psf;
1936                 struct net *net = dev_net(in_dev->dev);
1937 #endif
1938
1939                 /* filter mode change */
1940                 pmc->sfmode = MCAST_INCLUDE;
1941 #ifdef CONFIG_IP_MULTICAST
1942                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1943                 in_dev->mr_ifc_count = pmc->crcount;
1944                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1945                         psf->sf_crcount = 0;
1946                 igmp_ifc_event(pmc->interface);
1947         } else if (sf_setstate(pmc) || changerec) {
1948                 igmp_ifc_event(pmc->interface);
1949 #endif
1950         }
1951 out_unlock:
1952         spin_unlock_bh(&pmc->lock);
1953         return err;
1954 }
1955
1956 /*
1957  * Add multicast single-source filter to the interface list
1958  */
1959 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1960         __be32 *psfsrc)
1961 {
1962         struct ip_sf_list *psf, *psf_prev;
1963
1964         psf_prev = NULL;
1965         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1966                 if (psf->sf_inaddr == *psfsrc)
1967                         break;
1968                 psf_prev = psf;
1969         }
1970         if (!psf) {
1971                 psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1972                 if (!psf)
1973                         return -ENOBUFS;
1974                 psf->sf_inaddr = *psfsrc;
1975                 if (psf_prev) {
1976                         psf_prev->sf_next = psf;
1977                 } else
1978                         pmc->sources = psf;
1979         }
1980         psf->sf_count[sfmode]++;
1981         if (psf->sf_count[sfmode] == 1) {
1982                 ip_rt_multicast_event(pmc->interface);
1983         }
1984         return 0;
1985 }
1986
1987 #ifdef CONFIG_IP_MULTICAST
1988 static void sf_markstate(struct ip_mc_list *pmc)
1989 {
1990         struct ip_sf_list *psf;
1991         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1992
1993         for (psf = pmc->sources; psf; psf = psf->sf_next)
1994                 if (pmc->sfcount[MCAST_EXCLUDE]) {
1995                         psf->sf_oldin = mca_xcount ==
1996                                 psf->sf_count[MCAST_EXCLUDE] &&
1997                                 !psf->sf_count[MCAST_INCLUDE];
1998                 } else
1999                         psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
2000 }
2001
2002 static int sf_setstate(struct ip_mc_list *pmc)
2003 {
2004         struct ip_sf_list *psf, *dpsf;
2005         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
2006         int qrv = pmc->interface->mr_qrv;
2007         int new_in, rv;
2008
2009         rv = 0;
2010         for (psf = pmc->sources; psf; psf = psf->sf_next) {
2011                 if (pmc->sfcount[MCAST_EXCLUDE]) {
2012                         new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
2013                                 !psf->sf_count[MCAST_INCLUDE];
2014                 } else
2015                         new_in = psf->sf_count[MCAST_INCLUDE] != 0;
2016                 if (new_in) {
2017                         if (!psf->sf_oldin) {
2018                                 struct ip_sf_list *prev = NULL;
2019
2020                                 for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
2021                                         if (dpsf->sf_inaddr == psf->sf_inaddr)
2022                                                 break;
2023                                         prev = dpsf;
2024                                 }
2025                                 if (dpsf) {
2026                                         if (prev)
2027                                                 prev->sf_next = dpsf->sf_next;
2028                                         else
2029                                                 pmc->tomb = dpsf->sf_next;
2030                                         kfree(dpsf);
2031                                 }
2032                                 psf->sf_crcount = qrv;
2033                                 rv++;
2034                         }
2035                 } else if (psf->sf_oldin) {
2036
2037                         psf->sf_crcount = 0;
2038                         /*
2039                          * add or update "delete" records if an active filter
2040                          * is now inactive
2041                          */
2042                         for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
2043                                 if (dpsf->sf_inaddr == psf->sf_inaddr)
2044                                         break;
2045                         if (!dpsf) {
2046                                 dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
2047                                 if (!dpsf)
2048                                         continue;
2049                                 *dpsf = *psf;
2050                                 /* pmc->lock held by callers */
2051                                 dpsf->sf_next = pmc->tomb;
2052                                 pmc->tomb = dpsf;
2053                         }
2054                         dpsf->sf_crcount = qrv;
2055                         rv++;
2056                 }
2057         }
2058         return rv;
2059 }
2060 #endif
2061
2062 /*
2063  * Add multicast source filter list to the interface list
2064  */
2065 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2066                          int sfcount, __be32 *psfsrc, int delta)
2067 {
2068         struct ip_mc_list *pmc;
2069         int     isexclude;
2070         int     i, err;
2071
2072         if (!in_dev)
2073                 return -ENODEV;
2074         rcu_read_lock();
2075         for_each_pmc_rcu(in_dev, pmc) {
2076                 if (*pmca == pmc->multiaddr)
2077                         break;
2078         }
2079         if (!pmc) {
2080                 /* MCA not found?? bug */
2081                 rcu_read_unlock();
2082                 return -ESRCH;
2083         }
2084         spin_lock_bh(&pmc->lock);
2085         rcu_read_unlock();
2086
2087 #ifdef CONFIG_IP_MULTICAST
2088         sf_markstate(pmc);
2089 #endif
2090         isexclude = pmc->sfmode == MCAST_EXCLUDE;
2091         if (!delta)
2092                 pmc->sfcount[sfmode]++;
2093         err = 0;
2094         for (i = 0; i < sfcount; i++) {
2095                 err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2096                 if (err)
2097                         break;
2098         }
2099         if (err) {
2100                 int j;
2101
2102                 if (!delta)
2103                         pmc->sfcount[sfmode]--;
2104                 for (j = 0; j < i; j++)
2105                         (void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2106         } else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2107 #ifdef CONFIG_IP_MULTICAST
2108                 struct ip_sf_list *psf;
2109                 struct net *net = dev_net(pmc->interface->dev);
2110                 in_dev = pmc->interface;
2111 #endif
2112
2113                 /* filter mode change */
2114                 if (pmc->sfcount[MCAST_EXCLUDE])
2115                         pmc->sfmode = MCAST_EXCLUDE;
2116                 else if (pmc->sfcount[MCAST_INCLUDE])
2117                         pmc->sfmode = MCAST_INCLUDE;
2118 #ifdef CONFIG_IP_MULTICAST
2119                 /* else no filters; keep old mode for reports */
2120
2121                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
2122                 in_dev->mr_ifc_count = pmc->crcount;
2123                 for (psf = pmc->sources; psf; psf = psf->sf_next)
2124                         psf->sf_crcount = 0;
2125                 igmp_ifc_event(in_dev);
2126         } else if (sf_setstate(pmc)) {
2127                 igmp_ifc_event(in_dev);
2128 #endif
2129         }
2130         spin_unlock_bh(&pmc->lock);
2131         return err;
2132 }
2133
2134 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2135 {
2136         struct ip_sf_list *psf, *nextpsf, *tomb, *sources;
2137
2138         spin_lock_bh(&pmc->lock);
2139         tomb = pmc->tomb;
2140         pmc->tomb = NULL;
2141         sources = pmc->sources;
2142         pmc->sources = NULL;
2143         pmc->sfmode = MCAST_EXCLUDE;
2144         pmc->sfcount[MCAST_INCLUDE] = 0;
2145         pmc->sfcount[MCAST_EXCLUDE] = 1;
2146         spin_unlock_bh(&pmc->lock);
2147
2148         for (psf = tomb; psf; psf = nextpsf) {
2149                 nextpsf = psf->sf_next;
2150                 kfree(psf);
2151         }
2152         for (psf = sources; psf; psf = nextpsf) {
2153                 nextpsf = psf->sf_next;
2154                 kfree(psf);
2155         }
2156 }
2157
2158 /* Join a multicast group
2159  */
2160 static int __ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr,
2161                               unsigned int mode)
2162 {
2163         __be32 addr = imr->imr_multiaddr.s_addr;
2164         struct ip_mc_socklist *iml, *i;
2165         struct in_device *in_dev;
2166         struct inet_sock *inet = inet_sk(sk);
2167         struct net *net = sock_net(sk);
2168         int ifindex;
2169         int count = 0;
2170         int err;
2171
2172         ASSERT_RTNL();
2173
2174         if (!ipv4_is_multicast(addr))
2175                 return -EINVAL;
2176
2177         in_dev = ip_mc_find_dev(net, imr);
2178
2179         if (!in_dev) {
2180                 err = -ENODEV;
2181                 goto done;
2182         }
2183
2184         err = -EADDRINUSE;
2185         ifindex = imr->imr_ifindex;
2186         for_each_pmc_rtnl(inet, i) {
2187                 if (i->multi.imr_multiaddr.s_addr == addr &&
2188                     i->multi.imr_ifindex == ifindex)
2189                         goto done;
2190                 count++;
2191         }
2192         err = -ENOBUFS;
2193         if (count >= net->ipv4.sysctl_igmp_max_memberships)
2194                 goto done;
2195         iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2196         if (!iml)
2197                 goto done;
2198
2199         memcpy(&iml->multi, imr, sizeof(*imr));
2200         iml->next_rcu = inet->mc_list;
2201         iml->sflist = NULL;
2202         iml->sfmode = mode;
2203         rcu_assign_pointer(inet->mc_list, iml);
2204         __ip_mc_inc_group(in_dev, addr, mode);
2205         err = 0;
2206 done:
2207         return err;
2208 }
2209
2210 /* Join ASM (Any-Source Multicast) group
2211  */
2212 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2213 {
2214         return __ip_mc_join_group(sk, imr, MCAST_EXCLUDE);
2215 }
2216 EXPORT_SYMBOL(ip_mc_join_group);
2217
2218 /* Join SSM (Source-Specific Multicast) group
2219  */
2220 int ip_mc_join_group_ssm(struct sock *sk, struct ip_mreqn *imr,
2221                          unsigned int mode)
2222 {
2223         return __ip_mc_join_group(sk, imr, mode);
2224 }
2225
2226 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2227                            struct in_device *in_dev)
2228 {
2229         struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2230         int err;
2231
2232         if (!psf) {
2233                 /* any-source empty exclude case */
2234                 return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2235                         iml->sfmode, 0, NULL, 0);
2236         }
2237         err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2238                         iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2239         RCU_INIT_POINTER(iml->sflist, NULL);
2240         /* decrease mem now to avoid the memleak warning */
2241         atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
2242         kfree_rcu(psf, rcu);
2243         return err;
2244 }
2245
2246 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2247 {
2248         struct inet_sock *inet = inet_sk(sk);
2249         struct ip_mc_socklist *iml;
2250         struct ip_mc_socklist __rcu **imlp;
2251         struct in_device *in_dev;
2252         struct net *net = sock_net(sk);
2253         __be32 group = imr->imr_multiaddr.s_addr;
2254         u32 ifindex;
2255         int ret = -EADDRNOTAVAIL;
2256
2257         ASSERT_RTNL();
2258
2259         in_dev = ip_mc_find_dev(net, imr);
2260         if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2261                 ret = -ENODEV;
2262                 goto out;
2263         }
2264         ifindex = imr->imr_ifindex;
2265         for (imlp = &inet->mc_list;
2266              (iml = rtnl_dereference(*imlp)) != NULL;
2267              imlp = &iml->next_rcu) {
2268                 if (iml->multi.imr_multiaddr.s_addr != group)
2269                         continue;
2270                 if (ifindex) {
2271                         if (iml->multi.imr_ifindex != ifindex)
2272                                 continue;
2273                 } else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2274                                 iml->multi.imr_address.s_addr)
2275                         continue;
2276
2277                 (void) ip_mc_leave_src(sk, iml, in_dev);
2278
2279                 *imlp = iml->next_rcu;
2280
2281                 if (in_dev)
2282                         ip_mc_dec_group(in_dev, group);
2283
2284                 /* decrease mem now to avoid the memleak warning */
2285                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2286                 kfree_rcu(iml, rcu);
2287                 return 0;
2288         }
2289 out:
2290         return ret;
2291 }
2292 EXPORT_SYMBOL(ip_mc_leave_group);
2293
2294 int ip_mc_source(int add, int omode, struct sock *sk, struct
2295         ip_mreq_source *mreqs, int ifindex)
2296 {
2297         int err;
2298         struct ip_mreqn imr;
2299         __be32 addr = mreqs->imr_multiaddr;
2300         struct ip_mc_socklist *pmc;
2301         struct in_device *in_dev = NULL;
2302         struct inet_sock *inet = inet_sk(sk);
2303         struct ip_sf_socklist *psl;
2304         struct net *net = sock_net(sk);
2305         int leavegroup = 0;
2306         int i, j, rv;
2307
2308         if (!ipv4_is_multicast(addr))
2309                 return -EINVAL;
2310
2311         ASSERT_RTNL();
2312
2313         imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2314         imr.imr_address.s_addr = mreqs->imr_interface;
2315         imr.imr_ifindex = ifindex;
2316         in_dev = ip_mc_find_dev(net, &imr);
2317
2318         if (!in_dev) {
2319                 err = -ENODEV;
2320                 goto done;
2321         }
2322         err = -EADDRNOTAVAIL;
2323
2324         for_each_pmc_rtnl(inet, pmc) {
2325                 if ((pmc->multi.imr_multiaddr.s_addr ==
2326                      imr.imr_multiaddr.s_addr) &&
2327                     (pmc->multi.imr_ifindex == imr.imr_ifindex))
2328                         break;
2329         }
2330         if (!pmc) {             /* must have a prior join */
2331                 err = -EINVAL;
2332                 goto done;
2333         }
2334         /* if a source filter was set, must be the same mode as before */
2335         if (pmc->sflist) {
2336                 if (pmc->sfmode != omode) {
2337                         err = -EINVAL;
2338                         goto done;
2339                 }
2340         } else if (pmc->sfmode != omode) {
2341                 /* allow mode switches for empty-set filters */
2342                 ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2343                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2344                         NULL, 0);
2345                 pmc->sfmode = omode;
2346         }
2347
2348         psl = rtnl_dereference(pmc->sflist);
2349         if (!add) {
2350                 if (!psl)
2351                         goto done;      /* err = -EADDRNOTAVAIL */
2352                 rv = !0;
2353                 for (i = 0; i < psl->sl_count; i++) {
2354                         rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2355                                 sizeof(__be32));
2356                         if (rv == 0)
2357                                 break;
2358                 }
2359                 if (rv)         /* source not found */
2360                         goto done;      /* err = -EADDRNOTAVAIL */
2361
2362                 /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2363                 if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2364                         leavegroup = 1;
2365                         goto done;
2366                 }
2367
2368                 /* update the interface filter */
2369                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2370                         &mreqs->imr_sourceaddr, 1);
2371
2372                 for (j = i+1; j < psl->sl_count; j++)
2373                         psl->sl_addr[j-1] = psl->sl_addr[j];
2374                 psl->sl_count--;
2375                 err = 0;
2376                 goto done;
2377         }
2378         /* else, add a new source to the filter */
2379
2380         if (psl && psl->sl_count >= net->ipv4.sysctl_igmp_max_msf) {
2381                 err = -ENOBUFS;
2382                 goto done;
2383         }
2384         if (!psl || psl->sl_count == psl->sl_max) {
2385                 struct ip_sf_socklist *newpsl;
2386                 int count = IP_SFBLOCK;
2387
2388                 if (psl)
2389                         count += psl->sl_max;
2390                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2391                 if (!newpsl) {
2392                         err = -ENOBUFS;
2393                         goto done;
2394                 }
2395                 newpsl->sl_max = count;
2396                 newpsl->sl_count = count - IP_SFBLOCK;
2397                 if (psl) {
2398                         for (i = 0; i < psl->sl_count; i++)
2399                                 newpsl->sl_addr[i] = psl->sl_addr[i];
2400                         /* decrease mem now to avoid the memleak warning */
2401                         atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2402                         kfree_rcu(psl, rcu);
2403                 }
2404                 rcu_assign_pointer(pmc->sflist, newpsl);
2405                 psl = newpsl;
2406         }
2407         rv = 1; /* > 0 for insert logic below if sl_count is 0 */
2408         for (i = 0; i < psl->sl_count; i++) {
2409                 rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2410                         sizeof(__be32));
2411                 if (rv == 0)
2412                         break;
2413         }
2414         if (rv == 0)            /* address already there is an error */
2415                 goto done;
2416         for (j = psl->sl_count-1; j >= i; j--)
2417                 psl->sl_addr[j+1] = psl->sl_addr[j];
2418         psl->sl_addr[i] = mreqs->imr_sourceaddr;
2419         psl->sl_count++;
2420         err = 0;
2421         /* update the interface list */
2422         ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2423                 &mreqs->imr_sourceaddr, 1);
2424 done:
2425         if (leavegroup)
2426                 err = ip_mc_leave_group(sk, &imr);
2427         return err;
2428 }
2429
2430 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2431 {
2432         int err = 0;
2433         struct ip_mreqn imr;
2434         __be32 addr = msf->imsf_multiaddr;
2435         struct ip_mc_socklist *pmc;
2436         struct in_device *in_dev;
2437         struct inet_sock *inet = inet_sk(sk);
2438         struct ip_sf_socklist *newpsl, *psl;
2439         struct net *net = sock_net(sk);
2440         int leavegroup = 0;
2441
2442         if (!ipv4_is_multicast(addr))
2443                 return -EINVAL;
2444         if (msf->imsf_fmode != MCAST_INCLUDE &&
2445             msf->imsf_fmode != MCAST_EXCLUDE)
2446                 return -EINVAL;
2447
2448         ASSERT_RTNL();
2449
2450         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2451         imr.imr_address.s_addr = msf->imsf_interface;
2452         imr.imr_ifindex = ifindex;
2453         in_dev = ip_mc_find_dev(net, &imr);
2454
2455         if (!in_dev) {
2456                 err = -ENODEV;
2457                 goto done;
2458         }
2459
2460         /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2461         if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2462                 leavegroup = 1;
2463                 goto done;
2464         }
2465
2466         for_each_pmc_rtnl(inet, pmc) {
2467                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2468                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2469                         break;
2470         }
2471         if (!pmc) {             /* must have a prior join */
2472                 err = -EINVAL;
2473                 goto done;
2474         }
2475         if (msf->imsf_numsrc) {
2476                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2477                                                            GFP_KERNEL);
2478                 if (!newpsl) {
2479                         err = -ENOBUFS;
2480                         goto done;
2481                 }
2482                 newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2483                 memcpy(newpsl->sl_addr, msf->imsf_slist,
2484                         msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2485                 err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2486                         msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2487                 if (err) {
2488                         sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2489                         goto done;
2490                 }
2491         } else {
2492                 newpsl = NULL;
2493                 (void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2494                                      msf->imsf_fmode, 0, NULL, 0);
2495         }
2496         psl = rtnl_dereference(pmc->sflist);
2497         if (psl) {
2498                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2499                         psl->sl_count, psl->sl_addr, 0);
2500                 /* decrease mem now to avoid the memleak warning */
2501                 atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2502                 kfree_rcu(psl, rcu);
2503         } else
2504                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2505                         0, NULL, 0);
2506         rcu_assign_pointer(pmc->sflist, newpsl);
2507         pmc->sfmode = msf->imsf_fmode;
2508         err = 0;
2509 done:
2510         if (leavegroup)
2511                 err = ip_mc_leave_group(sk, &imr);
2512         return err;
2513 }
2514
2515 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2516         struct ip_msfilter __user *optval, int __user *optlen)
2517 {
2518         int err, len, count, copycount;
2519         struct ip_mreqn imr;
2520         __be32 addr = msf->imsf_multiaddr;
2521         struct ip_mc_socklist *pmc;
2522         struct in_device *in_dev;
2523         struct inet_sock *inet = inet_sk(sk);
2524         struct ip_sf_socklist *psl;
2525         struct net *net = sock_net(sk);
2526
2527         ASSERT_RTNL();
2528
2529         if (!ipv4_is_multicast(addr))
2530                 return -EINVAL;
2531
2532         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2533         imr.imr_address.s_addr = msf->imsf_interface;
2534         imr.imr_ifindex = 0;
2535         in_dev = ip_mc_find_dev(net, &imr);
2536
2537         if (!in_dev) {
2538                 err = -ENODEV;
2539                 goto done;
2540         }
2541         err = -EADDRNOTAVAIL;
2542
2543         for_each_pmc_rtnl(inet, pmc) {
2544                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2545                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2546                         break;
2547         }
2548         if (!pmc)               /* must have a prior join */
2549                 goto done;
2550         msf->imsf_fmode = pmc->sfmode;
2551         psl = rtnl_dereference(pmc->sflist);
2552         if (!psl) {
2553                 len = 0;
2554                 count = 0;
2555         } else {
2556                 count = psl->sl_count;
2557         }
2558         copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2559         len = copycount * sizeof(psl->sl_addr[0]);
2560         msf->imsf_numsrc = count;
2561         if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2562             copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2563                 return -EFAULT;
2564         }
2565         if (len &&
2566             copy_to_user(&optval->imsf_slist[0], psl->sl_addr, len))
2567                 return -EFAULT;
2568         return 0;
2569 done:
2570         return err;
2571 }
2572
2573 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2574         struct group_filter __user *optval, int __user *optlen)
2575 {
2576         int err, i, count, copycount;
2577         struct sockaddr_in *psin;
2578         __be32 addr;
2579         struct ip_mc_socklist *pmc;
2580         struct inet_sock *inet = inet_sk(sk);
2581         struct ip_sf_socklist *psl;
2582
2583         ASSERT_RTNL();
2584
2585         psin = (struct sockaddr_in *)&gsf->gf_group;
2586         if (psin->sin_family != AF_INET)
2587                 return -EINVAL;
2588         addr = psin->sin_addr.s_addr;
2589         if (!ipv4_is_multicast(addr))
2590                 return -EINVAL;
2591
2592         err = -EADDRNOTAVAIL;
2593
2594         for_each_pmc_rtnl(inet, pmc) {
2595                 if (pmc->multi.imr_multiaddr.s_addr == addr &&
2596                     pmc->multi.imr_ifindex == gsf->gf_interface)
2597                         break;
2598         }
2599         if (!pmc)               /* must have a prior join */
2600                 goto done;
2601         gsf->gf_fmode = pmc->sfmode;
2602         psl = rtnl_dereference(pmc->sflist);
2603         count = psl ? psl->sl_count : 0;
2604         copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2605         gsf->gf_numsrc = count;
2606         if (put_user(GROUP_FILTER_SIZE(copycount), optlen) ||
2607             copy_to_user(optval, gsf, GROUP_FILTER_SIZE(0))) {
2608                 return -EFAULT;
2609         }
2610         for (i = 0; i < copycount; i++) {
2611                 struct sockaddr_storage ss;
2612
2613                 psin = (struct sockaddr_in *)&ss;
2614                 memset(&ss, 0, sizeof(ss));
2615                 psin->sin_family = AF_INET;
2616                 psin->sin_addr.s_addr = psl->sl_addr[i];
2617                 if (copy_to_user(&optval->gf_slist[i], &ss, sizeof(ss)))
2618                         return -EFAULT;
2619         }
2620         return 0;
2621 done:
2622         return err;
2623 }
2624
2625 /*
2626  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2627  */
2628 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr,
2629                    int dif, int sdif)
2630 {
2631         struct inet_sock *inet = inet_sk(sk);
2632         struct ip_mc_socklist *pmc;
2633         struct ip_sf_socklist *psl;
2634         int i;
2635         int ret;
2636
2637         ret = 1;
2638         if (!ipv4_is_multicast(loc_addr))
2639                 goto out;
2640
2641         rcu_read_lock();
2642         for_each_pmc_rcu(inet, pmc) {
2643                 if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2644                     (pmc->multi.imr_ifindex == dif ||
2645                      (sdif && pmc->multi.imr_ifindex == sdif)))
2646                         break;
2647         }
2648         ret = inet->mc_all;
2649         if (!pmc)
2650                 goto unlock;
2651         psl = rcu_dereference(pmc->sflist);
2652         ret = (pmc->sfmode == MCAST_EXCLUDE);
2653         if (!psl)
2654                 goto unlock;
2655
2656         for (i = 0; i < psl->sl_count; i++) {
2657                 if (psl->sl_addr[i] == rmt_addr)
2658                         break;
2659         }
2660         ret = 0;
2661         if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2662                 goto unlock;
2663         if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2664                 goto unlock;
2665         ret = 1;
2666 unlock:
2667         rcu_read_unlock();
2668 out:
2669         return ret;
2670 }
2671
2672 /*
2673  *      A socket is closing.
2674  */
2675
2676 void ip_mc_drop_socket(struct sock *sk)
2677 {
2678         struct inet_sock *inet = inet_sk(sk);
2679         struct ip_mc_socklist *iml;
2680         struct net *net = sock_net(sk);
2681
2682         if (!inet->mc_list)
2683                 return;
2684
2685         rtnl_lock();
2686         while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2687                 struct in_device *in_dev;
2688
2689                 inet->mc_list = iml->next_rcu;
2690                 in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2691                 (void) ip_mc_leave_src(sk, iml, in_dev);
2692                 if (in_dev)
2693                         ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2694                 /* decrease mem now to avoid the memleak warning */
2695                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2696                 kfree_rcu(iml, rcu);
2697         }
2698         rtnl_unlock();
2699 }
2700
2701 /* called with rcu_read_lock() */
2702 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2703 {
2704         struct ip_mc_list *im;
2705         struct ip_mc_list __rcu **mc_hash;
2706         struct ip_sf_list *psf;
2707         int rv = 0;
2708
2709         mc_hash = rcu_dereference(in_dev->mc_hash);
2710         if (mc_hash) {
2711                 u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2712
2713                 for (im = rcu_dereference(mc_hash[hash]);
2714                      im != NULL;
2715                      im = rcu_dereference(im->next_hash)) {
2716                         if (im->multiaddr == mc_addr)
2717                                 break;
2718                 }
2719         } else {
2720                 for_each_pmc_rcu(in_dev, im) {
2721                         if (im->multiaddr == mc_addr)
2722                                 break;
2723                 }
2724         }
2725         if (im && proto == IPPROTO_IGMP) {
2726                 rv = 1;
2727         } else if (im) {
2728                 if (src_addr) {
2729                         for (psf = im->sources; psf; psf = psf->sf_next) {
2730                                 if (psf->sf_inaddr == src_addr)
2731                                         break;
2732                         }
2733                         if (psf)
2734                                 rv = psf->sf_count[MCAST_INCLUDE] ||
2735                                         psf->sf_count[MCAST_EXCLUDE] !=
2736                                         im->sfcount[MCAST_EXCLUDE];
2737                         else
2738                                 rv = im->sfcount[MCAST_EXCLUDE] != 0;
2739                 } else
2740                         rv = 1; /* unspecified source; tentatively allow */
2741         }
2742         return rv;
2743 }
2744
2745 #if defined(CONFIG_PROC_FS)
2746 struct igmp_mc_iter_state {
2747         struct seq_net_private p;
2748         struct net_device *dev;
2749         struct in_device *in_dev;
2750 };
2751
2752 #define igmp_mc_seq_private(seq)        ((struct igmp_mc_iter_state *)(seq)->private)
2753
2754 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2755 {
2756         struct net *net = seq_file_net(seq);
2757         struct ip_mc_list *im = NULL;
2758         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2759
2760         state->in_dev = NULL;
2761         for_each_netdev_rcu(net, state->dev) {
2762                 struct in_device *in_dev;
2763
2764                 in_dev = __in_dev_get_rcu(state->dev);
2765                 if (!in_dev)
2766                         continue;
2767                 im = rcu_dereference(in_dev->mc_list);
2768                 if (im) {
2769                         state->in_dev = in_dev;
2770                         break;
2771                 }
2772         }
2773         return im;
2774 }
2775
2776 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2777 {
2778         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2779
2780         im = rcu_dereference(im->next_rcu);
2781         while (!im) {
2782                 state->dev = next_net_device_rcu(state->dev);
2783                 if (!state->dev) {
2784                         state->in_dev = NULL;
2785                         break;
2786                 }
2787                 state->in_dev = __in_dev_get_rcu(state->dev);
2788                 if (!state->in_dev)
2789                         continue;
2790                 im = rcu_dereference(state->in_dev->mc_list);
2791         }
2792         return im;
2793 }
2794
2795 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2796 {
2797         struct ip_mc_list *im = igmp_mc_get_first(seq);
2798         if (im)
2799                 while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2800                         --pos;
2801         return pos ? NULL : im;
2802 }
2803
2804 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2805         __acquires(rcu)
2806 {
2807         rcu_read_lock();
2808         return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2809 }
2810
2811 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2812 {
2813         struct ip_mc_list *im;
2814         if (v == SEQ_START_TOKEN)
2815                 im = igmp_mc_get_first(seq);
2816         else
2817                 im = igmp_mc_get_next(seq, v);
2818         ++*pos;
2819         return im;
2820 }
2821
2822 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2823         __releases(rcu)
2824 {
2825         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2826
2827         state->in_dev = NULL;
2828         state->dev = NULL;
2829         rcu_read_unlock();
2830 }
2831
2832 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2833 {
2834         if (v == SEQ_START_TOKEN)
2835                 seq_puts(seq,
2836                          "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2837         else {
2838                 struct ip_mc_list *im = (struct ip_mc_list *)v;
2839                 struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2840                 char   *querier;
2841                 long delta;
2842
2843 #ifdef CONFIG_IP_MULTICAST
2844                 querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2845                           IGMP_V2_SEEN(state->in_dev) ? "V2" :
2846                           "V3";
2847 #else
2848                 querier = "NONE";
2849 #endif
2850
2851                 if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2852                         seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2853                                    state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2854                 }
2855
2856                 delta = im->timer.expires - jiffies;
2857                 seq_printf(seq,
2858                            "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2859                            im->multiaddr, im->users,
2860                            im->tm_running,
2861                            im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2862                            im->reporter);
2863         }
2864         return 0;
2865 }
2866
2867 static const struct seq_operations igmp_mc_seq_ops = {
2868         .start  =       igmp_mc_seq_start,
2869         .next   =       igmp_mc_seq_next,
2870         .stop   =       igmp_mc_seq_stop,
2871         .show   =       igmp_mc_seq_show,
2872 };
2873
2874 struct igmp_mcf_iter_state {
2875         struct seq_net_private p;
2876         struct net_device *dev;
2877         struct in_device *idev;
2878         struct ip_mc_list *im;
2879 };
2880
2881 #define igmp_mcf_seq_private(seq)       ((struct igmp_mcf_iter_state *)(seq)->private)
2882
2883 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2884 {
2885         struct net *net = seq_file_net(seq);
2886         struct ip_sf_list *psf = NULL;
2887         struct ip_mc_list *im = NULL;
2888         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2889
2890         state->idev = NULL;
2891         state->im = NULL;
2892         for_each_netdev_rcu(net, state->dev) {
2893                 struct in_device *idev;
2894                 idev = __in_dev_get_rcu(state->dev);
2895                 if (unlikely(!idev))
2896                         continue;
2897                 im = rcu_dereference(idev->mc_list);
2898                 if (likely(im)) {
2899                         spin_lock_bh(&im->lock);
2900                         psf = im->sources;
2901                         if (likely(psf)) {
2902                                 state->im = im;
2903                                 state->idev = idev;
2904                                 break;
2905                         }
2906                         spin_unlock_bh(&im->lock);
2907                 }
2908         }
2909         return psf;
2910 }
2911
2912 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2913 {
2914         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2915
2916         psf = psf->sf_next;
2917         while (!psf) {
2918                 spin_unlock_bh(&state->im->lock);
2919                 state->im = state->im->next;
2920                 while (!state->im) {
2921                         state->dev = next_net_device_rcu(state->dev);
2922                         if (!state->dev) {
2923                                 state->idev = NULL;
2924                                 goto out;
2925                         }
2926                         state->idev = __in_dev_get_rcu(state->dev);
2927                         if (!state->idev)
2928                                 continue;
2929                         state->im = rcu_dereference(state->idev->mc_list);
2930                 }
2931                 if (!state->im)
2932                         break;
2933                 spin_lock_bh(&state->im->lock);
2934                 psf = state->im->sources;
2935         }
2936 out:
2937         return psf;
2938 }
2939
2940 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2941 {
2942         struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2943         if (psf)
2944                 while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2945                         --pos;
2946         return pos ? NULL : psf;
2947 }
2948
2949 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2950         __acquires(rcu)
2951 {
2952         rcu_read_lock();
2953         return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2954 }
2955
2956 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2957 {
2958         struct ip_sf_list *psf;
2959         if (v == SEQ_START_TOKEN)
2960                 psf = igmp_mcf_get_first(seq);
2961         else
2962                 psf = igmp_mcf_get_next(seq, v);
2963         ++*pos;
2964         return psf;
2965 }
2966
2967 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2968         __releases(rcu)
2969 {
2970         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2971         if (likely(state->im)) {
2972                 spin_unlock_bh(&state->im->lock);
2973                 state->im = NULL;
2974         }
2975         state->idev = NULL;
2976         state->dev = NULL;
2977         rcu_read_unlock();
2978 }
2979
2980 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2981 {
2982         struct ip_sf_list *psf = (struct ip_sf_list *)v;
2983         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2984
2985         if (v == SEQ_START_TOKEN) {
2986                 seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
2987         } else {
2988                 seq_printf(seq,
2989                            "%3d %6.6s 0x%08x "
2990                            "0x%08x %6lu %6lu\n",
2991                            state->dev->ifindex, state->dev->name,
2992                            ntohl(state->im->multiaddr),
2993                            ntohl(psf->sf_inaddr),
2994                            psf->sf_count[MCAST_INCLUDE],
2995                            psf->sf_count[MCAST_EXCLUDE]);
2996         }
2997         return 0;
2998 }
2999
3000 static const struct seq_operations igmp_mcf_seq_ops = {
3001         .start  =       igmp_mcf_seq_start,
3002         .next   =       igmp_mcf_seq_next,
3003         .stop   =       igmp_mcf_seq_stop,
3004         .show   =       igmp_mcf_seq_show,
3005 };
3006
3007 static int __net_init igmp_net_init(struct net *net)
3008 {
3009         struct proc_dir_entry *pde;
3010         int err;
3011
3012         pde = proc_create_net("igmp", 0444, net->proc_net, &igmp_mc_seq_ops,
3013                         sizeof(struct igmp_mc_iter_state));
3014         if (!pde)
3015                 goto out_igmp;
3016         pde = proc_create_net("mcfilter", 0444, net->proc_net,
3017                         &igmp_mcf_seq_ops, sizeof(struct igmp_mcf_iter_state));
3018         if (!pde)
3019                 goto out_mcfilter;
3020         err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
3021                                    SOCK_DGRAM, 0, net);
3022         if (err < 0) {
3023                 pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
3024                        err);
3025                 goto out_sock;
3026         }
3027
3028         return 0;
3029
3030 out_sock:
3031         remove_proc_entry("mcfilter", net->proc_net);
3032 out_mcfilter:
3033         remove_proc_entry("igmp", net->proc_net);
3034 out_igmp:
3035         return -ENOMEM;
3036 }
3037
3038 static void __net_exit igmp_net_exit(struct net *net)
3039 {
3040         remove_proc_entry("mcfilter", net->proc_net);
3041         remove_proc_entry("igmp", net->proc_net);
3042         inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
3043 }
3044
3045 static struct pernet_operations igmp_net_ops = {
3046         .init = igmp_net_init,
3047         .exit = igmp_net_exit,
3048 };
3049 #endif
3050
3051 static int igmp_netdev_event(struct notifier_block *this,
3052                              unsigned long event, void *ptr)
3053 {
3054         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3055         struct in_device *in_dev;
3056
3057         switch (event) {
3058         case NETDEV_RESEND_IGMP:
3059                 in_dev = __in_dev_get_rtnl(dev);
3060                 if (in_dev)
3061                         ip_mc_rejoin_groups(in_dev);
3062                 break;
3063         default:
3064                 break;
3065         }
3066         return NOTIFY_DONE;
3067 }
3068
3069 static struct notifier_block igmp_notifier = {
3070         .notifier_call = igmp_netdev_event,
3071 };
3072
3073 int __init igmp_mc_init(void)
3074 {
3075 #if defined(CONFIG_PROC_FS)
3076         int err;
3077
3078         err = register_pernet_subsys(&igmp_net_ops);
3079         if (err)
3080                 return err;
3081         err = register_netdevice_notifier(&igmp_notifier);
3082         if (err)
3083                 goto reg_notif_fail;
3084         return 0;
3085
3086 reg_notif_fail:
3087         unregister_pernet_subsys(&igmp_net_ops);
3088         return err;
3089 #else
3090         return register_netdevice_notifier(&igmp_notifier);
3091 #endif
3092 }