net/ipv4/igmp.c

   1 /*
   2  *      Linux NET3:     Internet Group Management Protocol  [IGMP]
   3  *
   4  *      This code implements the IGMP protocol as defined in RFC1112. There has
   5  *      been a further revision of this protocol since which is now supported.
   6  *
   7  *      If you have trouble with this module be careful what gcc you have used,
   8  *      the older version didn't come out right using gcc 2.5.8, the newer one
   9  *      seems to fall out with gcc 2.6.2.
  10  *
  11  *      Authors:
  12  *              Alan Cox <alan@lxorguk.ukuu.org.uk>
  13  *
  14  *      This program is free software; you can redistribute it and/or
  15  *      modify it under the terms of the GNU General Public License
  16  *      as published by the Free Software Foundation; either version
  17  *      2 of the License, or (at your option) any later version.
  18  *
  19  *      Fixes:
  20  *
  21  *              Alan Cox        :       Added lots of __inline__ to optimise
  22  *                                      the memory usage of all the tiny little
  23  *                                      functions.
  24  *              Alan Cox        :       Dumped the header building experiment.
  25  *              Alan Cox        :       Minor tweaks ready for multicast routing
  26  *                                      and extended IGMP protocol.
  27  *              Alan Cox        :       Removed a load of inline directives. Gcc 2.5.8
  28  *                                      writes utterly bogus code otherwise (sigh)
  29  *                                      fixed IGMP loopback to behave in the manner
  30  *                                      desired by mrouted, fixed the fact it has been
  31  *                                      broken since 1.3.6 and cleaned up a few minor
  32  *                                      points.
  33  *
  34  *              Chih-Jen Chang  :       Tried to revise IGMP to Version 2
  35  *              Tsu-Sheng Tsao          E-mail: chihjenc@scf.usc.edu and tsusheng@scf.usc.edu
  36  *                                      The enhancements are mainly based on Steve Deering's
  37  *                                      ipmulti-3.5 source code.
  38  *              Chih-Jen Chang  :       Added the igmp_get_mrouter_info and
  39  *              Tsu-Sheng Tsao          igmp_set_mrouter_info to keep track of
  40  *                                      the mrouted version on that device.
  41  *              Chih-Jen Chang  :       Added the max_resp_time parameter to
  42  *              Tsu-Sheng Tsao          igmp_heard_query(). Using this parameter
  43  *                                      to identify the multicast router version
  44  *                                      and do what the IGMP version 2 specified.
  45  *              Chih-Jen Chang  :       Added a timer to revert to IGMP V2 router
  46  *              Tsu-Sheng Tsao          if the specified time expired.
  47  *              Alan Cox        :       Stop IGMP from 0.0.0.0 being accepted.
  48  *              Alan Cox        :       Use GFP_ATOMIC in the right places.
  49  *              Christian Daudt :       igmp timer wasn't set for local group
  50  *                                      memberships but was being deleted,
  51  *                                      which caused a "del_timer() called
  52  *                                      from %p with timer not initialized\n"
  53  *                                      message (960131).
  54  *              Christian Daudt :       removed del_timer from
  55  *                                      igmp_timer_expire function (960205).
  56  *             Christian Daudt :       igmp_heard_report now only calls
  57  *                                     igmp_timer_expire if tm->running is
  58  *                                     true (960216).
  59  *              Malcolm Beattie :       ttl comparison wrong in igmp_rcv made
  60  *                                      igmp_heard_query never trigger. Expiry
  61  *                                      miscalculation fixed in igmp_heard_query
  62  *                                      and random() made to return unsigned to
  63  *                                      prevent negative expiry times.
  64  *              Alexey Kuznetsov:       Wrong group leaving behaviour, backport
  65  *                                      fix from pending 2.1.x patches.
  66  *              Alan Cox:               Forget to enable FDDI support earlier.
  67  *              Alexey Kuznetsov:       Fixed leaving groups on device down.
  68  *              Alexey Kuznetsov:       Accordance to igmp-v2-06 draft.
  69  *              David L Stevens:        IGMPv3 support, with help from
  70  *                                      Vinay Kulkarni
  71  */
  72
  73 #include <linux/module.h>
  74 #include <linux/slab.h>
  75 #include <linux/uaccess.h>
  76 #include <linux/types.h>
  77 #include <linux/kernel.h>
  78 #include <linux/jiffies.h>
  79 #include <linux/string.h>
  80 #include <linux/socket.h>
  81 #include <linux/sockios.h>
  82 #include <linux/in.h>
  83 #include <linux/inet.h>
  84 #include <linux/netdevice.h>
  85 #include <linux/skbuff.h>
  86 #include <linux/inetdevice.h>
  87 #include <linux/igmp.h>
  88 #include <linux/if_arp.h>
  89 #include <linux/rtnetlink.h>
  90 #include <linux/times.h>
  91 #include <linux/pkt_sched.h>
  92 #include <linux/byteorder/generic.h>
  93
  94 #include <net/net_namespace.h>
  95 #include <net/arp.h>
  96 #include <net/ip.h>
  97 #include <net/protocol.h>
  98 #include <net/route.h>
  99 #include <net/sock.h>
 100 #include <net/checksum.h>
 101 #include <net/inet_common.h>
 102 #include <linux/netfilter_ipv4.h>
 103 #ifdef CONFIG_IP_MROUTE
 104 #include <linux/mroute.h>
 105 #endif
 106 #ifdef CONFIG_PROC_FS
 107 #include <linux/proc_fs.h>
 108 #include <linux/seq_file.h>
 109 #endif
 110
 111 #ifdef CONFIG_IP_MULTICAST
 112 /* Parameter names and values are taken from igmp-v2-06 draft */
 113
 114 #define IGMP_V1_ROUTER_PRESENT_TIMEOUT          (400*HZ)
 115 #define IGMP_V2_ROUTER_PRESENT_TIMEOUT          (400*HZ)
 116 #define IGMP_V2_UNSOLICITED_REPORT_INTERVAL     (10*HZ)
 117 #define IGMP_V3_UNSOLICITED_REPORT_INTERVAL     (1*HZ)
 118 #define IGMP_QUERY_RESPONSE_INTERVAL            (10*HZ)
 119 #define IGMP_QUERY_ROBUSTNESS_VARIABLE          2
 120
 121
 122 #define IGMP_INITIAL_REPORT_DELAY               (1)
 123
 124 /* IGMP_INITIAL_REPORT_DELAY is not from IGMP specs!
 125  * IGMP specs require to report membership immediately after
 126  * joining a group, but we delay the first report by a
 127  * small interval. It seems more natural and still does not
 128  * contradict to specs provided this delay is small enough.
 129  */
 130
 131 #define IGMP_V1_SEEN(in_dev) \
 132         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 1 || \
 133          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 1 || \
 134          ((in_dev)->mr_v1_seen && \
 135           time_before(jiffies, (in_dev)->mr_v1_seen)))
 136 #define IGMP_V2_SEEN(in_dev) \
 137         (IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 2 || \
 138          IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 2 || \
 139          ((in_dev)->mr_v2_seen && \
 140           time_before(jiffies, (in_dev)->mr_v2_seen)))
 141
 142 static int unsolicited_report_interval(struct in_device *in_dev)
 143 {
 144         int interval_ms, interval_jiffies;
 145
 146         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 147                 interval_ms = IN_DEV_CONF_GET(
 148                         in_dev,
 149                         IGMPV2_UNSOLICITED_REPORT_INTERVAL);
 150         else /* v3 */
 151                 interval_ms = IN_DEV_CONF_GET(
 152                         in_dev,
 153                         IGMPV3_UNSOLICITED_REPORT_INTERVAL);
 154
 155         interval_jiffies = msecs_to_jiffies(interval_ms);
 156
 157         /* _timer functions can't handle a delay of 0 jiffies so ensure
 158          *  we always return a positive value.
 159          */
 160         if (interval_jiffies <= 0)
 161                 interval_jiffies = 1;
 162         return interval_jiffies;
 163 }
 164
 165 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im);
 166 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im);
 167 static void igmpv3_clear_delrec(struct in_device *in_dev);
 168 static int sf_setstate(struct ip_mc_list *pmc);
 169 static void sf_markstate(struct ip_mc_list *pmc);
 170 #endif
 171 static void ip_mc_clear_src(struct ip_mc_list *pmc);
 172 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
 173                          int sfcount, __be32 *psfsrc, int delta);
 174
 175 static void ip_ma_put(struct ip_mc_list *im)
 176 {
 177         if (refcount_dec_and_test(&im->refcnt)) {
 178                 in_dev_put(im->interface);
 179                 kfree_rcu(im, rcu);
 180         }
 181 }
 182
 183 #define for_each_pmc_rcu(in_dev, pmc)                           \
 184         for (pmc = rcu_dereference(in_dev->mc_list);            \
 185              pmc != NULL;                                       \
 186              pmc = rcu_dereference(pmc->next_rcu))
 187
 188 #define for_each_pmc_rtnl(in_dev, pmc)                          \
 189         for (pmc = rtnl_dereference(in_dev->mc_list);           \
 190              pmc != NULL;                                       \
 191              pmc = rtnl_dereference(pmc->next_rcu))
 192
 193 #ifdef CONFIG_IP_MULTICAST
 194
 195 /*
 196  *      Timer management
 197  */
 198
 199 static void igmp_stop_timer(struct ip_mc_list *im)
 200 {
 201         spin_lock_bh(&im->lock);
 202         if (del_timer(&im->timer))
 203                 refcount_dec(&im->refcnt);
 204         im->tm_running = 0;
 205         im->reporter = 0;
 206         im->unsolicit_count = 0;
 207         spin_unlock_bh(&im->lock);
 208 }
 209
 210 /* It must be called with locked im->lock */
 211 static void igmp_start_timer(struct ip_mc_list *im, int max_delay)
 212 {
 213         int tv = prandom_u32() % max_delay;
 214
 215         im->tm_running = 1;
 216         if (!mod_timer(&im->timer, jiffies+tv+2))
 217                 refcount_inc(&im->refcnt);
 218 }
 219
 220 static void igmp_gq_start_timer(struct in_device *in_dev)
 221 {
 222         int tv = prandom_u32() % in_dev->mr_maxdelay;
 223         unsigned long exp = jiffies + tv + 2;
 224
 225         if (in_dev->mr_gq_running &&
 226             time_after_eq(exp, (in_dev->mr_gq_timer).expires))
 227                 return;
 228
 229         in_dev->mr_gq_running = 1;
 230         if (!mod_timer(&in_dev->mr_gq_timer, exp))
 231                 in_dev_hold(in_dev);
 232 }
 233
 234 static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
 235 {
 236         int tv = prandom_u32() % delay;
 237
 238         if (!mod_timer(&in_dev->mr_ifc_timer, jiffies+tv+2))
 239                 in_dev_hold(in_dev);
 240 }
 241
 242 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
 243 {
 244         spin_lock_bh(&im->lock);
 245         im->unsolicit_count = 0;
 246         if (del_timer(&im->timer)) {
 247                 if ((long)(im->timer.expires-jiffies) < max_delay) {
 248                         add_timer(&im->timer);
 249                         im->tm_running = 1;
 250                         spin_unlock_bh(&im->lock);
 251                         return;
 252                 }
 253                 refcount_dec(&im->refcnt);
 254         }
 255         igmp_start_timer(im, max_delay);
 256         spin_unlock_bh(&im->lock);
 257 }
 258
 259
 260 /*
 261  *      Send an IGMP report.
 262  */
 263
 264 #define IGMP_SIZE (sizeof(struct igmphdr)+sizeof(struct iphdr)+4)
 265
 266
 267 static int is_in(struct ip_mc_list *pmc, struct ip_sf_list *psf, int type,
 268         int gdeleted, int sdeleted)
 269 {
 270         switch (type) {
 271         case IGMPV3_MODE_IS_INCLUDE:
 272         case IGMPV3_MODE_IS_EXCLUDE:
 273                 if (gdeleted || sdeleted)
 274                         return 0;
 275                 if (!(pmc->gsquery && !psf->sf_gsresp)) {
 276                         if (pmc->sfmode == MCAST_INCLUDE)
 277                                 return 1;
 278                         /* don't include if this source is excluded
 279                          * in all filters
 280                          */
 281                         if (psf->sf_count[MCAST_INCLUDE])
 282                                 return type == IGMPV3_MODE_IS_INCLUDE;
 283                         return pmc->sfcount[MCAST_EXCLUDE] ==
 284                                 psf->sf_count[MCAST_EXCLUDE];
 285                 }
 286                 return 0;
 287         case IGMPV3_CHANGE_TO_INCLUDE:
 288                 if (gdeleted || sdeleted)
 289                         return 0;
 290                 return psf->sf_count[MCAST_INCLUDE] != 0;
 291         case IGMPV3_CHANGE_TO_EXCLUDE:
 292                 if (gdeleted || sdeleted)
 293                         return 0;
 294                 if (pmc->sfcount[MCAST_EXCLUDE] == 0 ||
 295                     psf->sf_count[MCAST_INCLUDE])
 296                         return 0;
 297                 return pmc->sfcount[MCAST_EXCLUDE] ==
 298                         psf->sf_count[MCAST_EXCLUDE];
 299         case IGMPV3_ALLOW_NEW_SOURCES:
 300                 if (gdeleted || !psf->sf_crcount)
 301                         return 0;
 302                 return (pmc->sfmode == MCAST_INCLUDE) ^ sdeleted;
 303         case IGMPV3_BLOCK_OLD_SOURCES:
 304                 if (pmc->sfmode == MCAST_INCLUDE)
 305                         return gdeleted || (psf->sf_crcount && sdeleted);
 306                 return psf->sf_crcount && !gdeleted && !sdeleted;
 307         }
 308         return 0;
 309 }
 310
 311 static int
 312 igmp_scount(struct ip_mc_list *pmc, int type, int gdeleted, int sdeleted)
 313 {
 314         struct ip_sf_list *psf;
 315         int scount = 0;
 316
 317         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 318                 if (!is_in(pmc, psf, type, gdeleted, sdeleted))
 319                         continue;
 320                 scount++;
 321         }
 322         return scount;
 323 }
 324
 325 /* source address selection per RFC 3376 section 4.2.13 */
 326 static __be32 igmpv3_get_srcaddr(struct net_device *dev,
 327                                  const struct flowi4 *fl4)
 328 {
 329         struct in_device *in_dev = __in_dev_get_rcu(dev);
 330
 331         if (!in_dev)
 332                 return htonl(INADDR_ANY);
 333
 334         for_ifa(in_dev) {
 335                 if (fl4->saddr == ifa->ifa_local)
 336                         return fl4->saddr;
 337         } endfor_ifa(in_dev);
 338
 339         return htonl(INADDR_ANY);
 340 }
 341
 342 static struct sk_buff *igmpv3_newpack(struct net_device *dev, unsigned int mtu)
 343 {
 344         struct sk_buff *skb;
 345         struct rtable *rt;
 346         struct iphdr *pip;
 347         struct igmpv3_report *pig;
 348         struct net *net = dev_net(dev);
 349         struct flowi4 fl4;
 350         int hlen = LL_RESERVED_SPACE(dev);
 351         int tlen = dev->needed_tailroom;
 352         unsigned int size = mtu;
 353
 354         while (1) {
 355                 skb = alloc_skb(size + hlen + tlen,
 356                                 GFP_ATOMIC | __GFP_NOWARN);
 357                 if (skb)
 358                         break;
 359                 size >>= 1;
 360                 if (size < 256)
 361                         return NULL;
 362         }
 363         skb->priority = TC_PRIO_CONTROL;
 364
 365         rt = ip_route_output_ports(net, &fl4, NULL, IGMPV3_ALL_MCR, 0,
 366                                    0, 0,
 367                                    IPPROTO_IGMP, 0, dev->ifindex);
 368         if (IS_ERR(rt)) {
 369                 kfree_skb(skb);
 370                 return NULL;
 371         }
 372
 373         skb_dst_set(skb, &rt->dst);
 374         skb->dev = dev;
 375
 376         skb_reserve(skb, hlen);
 377         skb_tailroom_reserve(skb, mtu, tlen);
 378
 379         skb_reset_network_header(skb);
 380         pip = ip_hdr(skb);
 381         skb_put(skb, sizeof(struct iphdr) + 4);
 382
 383         pip->version  = 4;
 384         pip->ihl      = (sizeof(struct iphdr)+4)>>2;
 385         pip->tos      = 0xc0;
 386         pip->frag_off = htons(IP_DF);
 387         pip->ttl      = 1;
 388         pip->daddr    = fl4.daddr;
 389
 390         rcu_read_lock();
 391         pip->saddr    = igmpv3_get_srcaddr(dev, &fl4);
 392         rcu_read_unlock();
 393
 394         pip->protocol = IPPROTO_IGMP;
 395         pip->tot_len  = 0;      /* filled in later */
 396         ip_select_ident(net, skb, NULL);
 397         ((u8 *)&pip[1])[0] = IPOPT_RA;
 398         ((u8 *)&pip[1])[1] = 4;
 399         ((u8 *)&pip[1])[2] = 0;
 400         ((u8 *)&pip[1])[3] = 0;
 401
 402         skb->transport_header = skb->network_header + sizeof(struct iphdr) + 4;
 403         skb_put(skb, sizeof(*pig));
 404         pig = igmpv3_report_hdr(skb);
 405         pig->type = IGMPV3_HOST_MEMBERSHIP_REPORT;
 406         pig->resv1 = 0;
 407         pig->csum = 0;
 408         pig->resv2 = 0;
 409         pig->ngrec = 0;
 410         return skb;
 411 }
 412
 413 static int igmpv3_sendpack(struct sk_buff *skb)
 414 {
 415         struct igmphdr *pig = igmp_hdr(skb);
 416         const int igmplen = skb_tail_pointer(skb) - skb_transport_header(skb);
 417
 418         pig->csum = ip_compute_csum(igmp_hdr(skb), igmplen);
 419
 420         return ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 421 }
 422
 423 static int grec_size(struct ip_mc_list *pmc, int type, int gdel, int sdel)
 424 {
 425         return sizeof(struct igmpv3_grec) + 4*igmp_scount(pmc, type, gdel, sdel);
 426 }
 427
 428 static struct sk_buff *add_grhead(struct sk_buff *skb, struct ip_mc_list *pmc,
 429         int type, struct igmpv3_grec **ppgr, unsigned int mtu)
 430 {
 431         struct net_device *dev = pmc->interface->dev;
 432         struct igmpv3_report *pih;
 433         struct igmpv3_grec *pgr;
 434
 435         if (!skb) {
 436                 skb = igmpv3_newpack(dev, mtu);
 437                 if (!skb)
 438                         return NULL;
 439         }
 440         pgr = skb_put(skb, sizeof(struct igmpv3_grec));
 441         pgr->grec_type = type;
 442         pgr->grec_auxwords = 0;
 443         pgr->grec_nsrcs = 0;
 444         pgr->grec_mca = pmc->multiaddr;
 445         pih = igmpv3_report_hdr(skb);
 446         pih->ngrec = htons(ntohs(pih->ngrec)+1);
 447         *ppgr = pgr;
 448         return skb;
 449 }
 450
 451 #define AVAILABLE(skb)  ((skb) ? skb_availroom(skb) : 0)
 452
 453 static struct sk_buff *add_grec(struct sk_buff *skb, struct ip_mc_list *pmc,
 454         int type, int gdeleted, int sdeleted)
 455 {
 456         struct net_device *dev = pmc->interface->dev;
 457         struct net *net = dev_net(dev);
 458         struct igmpv3_report *pih;
 459         struct igmpv3_grec *pgr = NULL;
 460         struct ip_sf_list *psf, *psf_next, *psf_prev, **psf_list;
 461         int scount, stotal, first, isquery, truncate;
 462         unsigned int mtu;
 463
 464         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 465                 return skb;
 466         if (ipv4_is_local_multicast(pmc->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
 467                 return skb;
 468
 469         mtu = READ_ONCE(dev->mtu);
 470         if (mtu < IPV4_MIN_MTU)
 471                 return skb;
 472
 473         isquery = type == IGMPV3_MODE_IS_INCLUDE ||
 474                   type == IGMPV3_MODE_IS_EXCLUDE;
 475         truncate = type == IGMPV3_MODE_IS_EXCLUDE ||
 476                     type == IGMPV3_CHANGE_TO_EXCLUDE;
 477
 478         stotal = scount = 0;
 479
 480         psf_list = sdeleted ? &pmc->tomb : &pmc->sources;
 481
 482         if (!*psf_list)
 483                 goto empty_source;
 484
 485         pih = skb ? igmpv3_report_hdr(skb) : NULL;
 486
 487         /* EX and TO_EX get a fresh packet, if needed */
 488         if (truncate) {
 489                 if (pih && pih->ngrec &&
 490                     AVAILABLE(skb) < grec_size(pmc, type, gdeleted, sdeleted)) {
 491                         if (skb)
 492                                 igmpv3_sendpack(skb);
 493                         skb = igmpv3_newpack(dev, mtu);
 494                 }
 495         }
 496         first = 1;
 497         psf_prev = NULL;
 498         for (psf = *psf_list; psf; psf = psf_next) {
 499                 __be32 *psrc;
 500
 501                 psf_next = psf->sf_next;
 502
 503                 if (!is_in(pmc, psf, type, gdeleted, sdeleted)) {
 504                         psf_prev = psf;
 505                         continue;
 506                 }
 507
 508                 /* Based on RFC3376 5.1. Should not send source-list change
 509                  * records when there is a filter mode change.
 510                  */
 511                 if (((gdeleted && pmc->sfmode == MCAST_EXCLUDE) ||
 512                      (!gdeleted && pmc->crcount)) &&
 513                     (type == IGMPV3_ALLOW_NEW_SOURCES ||
 514                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount)
 515                         goto decrease_sf_crcount;
 516
 517                 /* clear marks on query responses */
 518                 if (isquery)
 519                         psf->sf_gsresp = 0;
 520
 521                 if (AVAILABLE(skb) < sizeof(__be32) +
 522                     first*sizeof(struct igmpv3_grec)) {
 523                         if (truncate && !first)
 524                                 break;   /* truncate these */
 525                         if (pgr)
 526                                 pgr->grec_nsrcs = htons(scount);
 527                         if (skb)
 528                                 igmpv3_sendpack(skb);
 529                         skb = igmpv3_newpack(dev, mtu);
 530                         first = 1;
 531                         scount = 0;
 532                 }
 533                 if (first) {
 534                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 535                         first = 0;
 536                 }
 537                 if (!skb)
 538                         return NULL;
 539                 psrc = skb_put(skb, sizeof(__be32));
 540                 *psrc = psf->sf_inaddr;
 541                 scount++; stotal++;
 542                 if ((type == IGMPV3_ALLOW_NEW_SOURCES ||
 543                      type == IGMPV3_BLOCK_OLD_SOURCES) && psf->sf_crcount) {
 544 decrease_sf_crcount:
 545                         psf->sf_crcount--;
 546                         if ((sdeleted || gdeleted) && psf->sf_crcount == 0) {
 547                                 if (psf_prev)
 548                                         psf_prev->sf_next = psf->sf_next;
 549                                 else
 550                                         *psf_list = psf->sf_next;
 551                                 kfree(psf);
 552                                 continue;
 553                         }
 554                 }
 555                 psf_prev = psf;
 556         }
 557
 558 empty_source:
 559         if (!stotal) {
 560                 if (type == IGMPV3_ALLOW_NEW_SOURCES ||
 561                     type == IGMPV3_BLOCK_OLD_SOURCES)
 562                         return skb;
 563                 if (pmc->crcount || isquery) {
 564                         /* make sure we have room for group header */
 565                         if (skb && AVAILABLE(skb) < sizeof(struct igmpv3_grec)) {
 566                                 igmpv3_sendpack(skb);
 567                                 skb = NULL; /* add_grhead will get a new one */
 568                         }
 569                         skb = add_grhead(skb, pmc, type, &pgr, mtu);
 570                 }
 571         }
 572         if (pgr)
 573                 pgr->grec_nsrcs = htons(scount);
 574
 575         if (isquery)
 576                 pmc->gsquery = 0;       /* clear query state on report */
 577         return skb;
 578 }
 579
 580 static int igmpv3_send_report(struct in_device *in_dev, struct ip_mc_list *pmc)
 581 {
 582         struct sk_buff *skb = NULL;
 583         struct net *net = dev_net(in_dev->dev);
 584         int type;
 585
 586         if (!pmc) {
 587                 rcu_read_lock();
 588                 for_each_pmc_rcu(in_dev, pmc) {
 589                         if (pmc->multiaddr == IGMP_ALL_HOSTS)
 590                                 continue;
 591                         if (ipv4_is_local_multicast(pmc->multiaddr) &&
 592                              !net->ipv4.sysctl_igmp_llm_reports)
 593                                 continue;
 594                         spin_lock_bh(&pmc->lock);
 595                         if (pmc->sfcount[MCAST_EXCLUDE])
 596                                 type = IGMPV3_MODE_IS_EXCLUDE;
 597                         else
 598                                 type = IGMPV3_MODE_IS_INCLUDE;
 599                         skb = add_grec(skb, pmc, type, 0, 0);
 600                         spin_unlock_bh(&pmc->lock);
 601                 }
 602                 rcu_read_unlock();
 603         } else {
 604                 spin_lock_bh(&pmc->lock);
 605                 if (pmc->sfcount[MCAST_EXCLUDE])
 606                         type = IGMPV3_MODE_IS_EXCLUDE;
 607                 else
 608                         type = IGMPV3_MODE_IS_INCLUDE;
 609                 skb = add_grec(skb, pmc, type, 0, 0);
 610                 spin_unlock_bh(&pmc->lock);
 611         }
 612         if (!skb)
 613                 return 0;
 614         return igmpv3_sendpack(skb);
 615 }
 616
 617 /*
 618  * remove zero-count source records from a source filter list
 619  */
 620 static void igmpv3_clear_zeros(struct ip_sf_list **ppsf)
 621 {
 622         struct ip_sf_list *psf_prev, *psf_next, *psf;
 623
 624         psf_prev = NULL;
 625         for (psf = *ppsf; psf; psf = psf_next) {
 626                 psf_next = psf->sf_next;
 627                 if (psf->sf_crcount == 0) {
 628                         if (psf_prev)
 629                                 psf_prev->sf_next = psf->sf_next;
 630                         else
 631                                 *ppsf = psf->sf_next;
 632                         kfree(psf);
 633                 } else
 634                         psf_prev = psf;
 635         }
 636 }
 637
 638 static void igmpv3_send_cr(struct in_device *in_dev)
 639 {
 640         struct ip_mc_list *pmc, *pmc_prev, *pmc_next;
 641         struct sk_buff *skb = NULL;
 642         int type, dtype;
 643
 644         rcu_read_lock();
 645         spin_lock_bh(&in_dev->mc_tomb_lock);
 646
 647         /* deleted MCA's */
 648         pmc_prev = NULL;
 649         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc_next) {
 650                 pmc_next = pmc->next;
 651                 if (pmc->sfmode == MCAST_INCLUDE) {
 652                         type = IGMPV3_BLOCK_OLD_SOURCES;
 653                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 654                         skb = add_grec(skb, pmc, type, 1, 0);
 655                         skb = add_grec(skb, pmc, dtype, 1, 1);
 656                 }
 657                 if (pmc->crcount) {
 658                         if (pmc->sfmode == MCAST_EXCLUDE) {
 659                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 660                                 skb = add_grec(skb, pmc, type, 1, 0);
 661                         }
 662                         pmc->crcount--;
 663                         if (pmc->crcount == 0) {
 664                                 igmpv3_clear_zeros(&pmc->tomb);
 665                                 igmpv3_clear_zeros(&pmc->sources);
 666                         }
 667                 }
 668                 if (pmc->crcount == 0 && !pmc->tomb && !pmc->sources) {
 669                         if (pmc_prev)
 670                                 pmc_prev->next = pmc_next;
 671                         else
 672                                 in_dev->mc_tomb = pmc_next;
 673                         in_dev_put(pmc->interface);
 674                         kfree(pmc);
 675                 } else
 676                         pmc_prev = pmc;
 677         }
 678         spin_unlock_bh(&in_dev->mc_tomb_lock);
 679
 680         /* change recs */
 681         for_each_pmc_rcu(in_dev, pmc) {
 682                 spin_lock_bh(&pmc->lock);
 683                 if (pmc->sfcount[MCAST_EXCLUDE]) {
 684                         type = IGMPV3_BLOCK_OLD_SOURCES;
 685                         dtype = IGMPV3_ALLOW_NEW_SOURCES;
 686                 } else {
 687                         type = IGMPV3_ALLOW_NEW_SOURCES;
 688                         dtype = IGMPV3_BLOCK_OLD_SOURCES;
 689                 }
 690                 skb = add_grec(skb, pmc, type, 0, 0);
 691                 skb = add_grec(skb, pmc, dtype, 0, 1);  /* deleted sources */
 692
 693                 /* filter mode changes */
 694                 if (pmc->crcount) {
 695                         if (pmc->sfmode == MCAST_EXCLUDE)
 696                                 type = IGMPV3_CHANGE_TO_EXCLUDE;
 697                         else
 698                                 type = IGMPV3_CHANGE_TO_INCLUDE;
 699                         skb = add_grec(skb, pmc, type, 0, 0);
 700                         pmc->crcount--;
 701                 }
 702                 spin_unlock_bh(&pmc->lock);
 703         }
 704         rcu_read_unlock();
 705
 706         if (!skb)
 707                 return;
 708         (void) igmpv3_sendpack(skb);
 709 }
 710
 711 static int igmp_send_report(struct in_device *in_dev, struct ip_mc_list *pmc,
 712         int type)
 713 {
 714         struct sk_buff *skb;
 715         struct iphdr *iph;
 716         struct igmphdr *ih;
 717         struct rtable *rt;
 718         struct net_device *dev = in_dev->dev;
 719         struct net *net = dev_net(dev);
 720         __be32  group = pmc ? pmc->multiaddr : 0;
 721         struct flowi4 fl4;
 722         __be32  dst;
 723         int hlen, tlen;
 724
 725         if (type == IGMPV3_HOST_MEMBERSHIP_REPORT)
 726                 return igmpv3_send_report(in_dev, pmc);
 727
 728         if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
 729                 return 0;
 730
 731         if (type == IGMP_HOST_LEAVE_MESSAGE)
 732                 dst = IGMP_ALL_ROUTER;
 733         else
 734                 dst = group;
 735
 736         rt = ip_route_output_ports(net, &fl4, NULL, dst, 0,
 737                                    0, 0,
 738                                    IPPROTO_IGMP, 0, dev->ifindex);
 739         if (IS_ERR(rt))
 740                 return -1;
 741
 742         hlen = LL_RESERVED_SPACE(dev);
 743         tlen = dev->needed_tailroom;
 744         skb = alloc_skb(IGMP_SIZE + hlen + tlen, GFP_ATOMIC);
 745         if (!skb) {
 746                 ip_rt_put(rt);
 747                 return -1;
 748         }
 749         skb->priority = TC_PRIO_CONTROL;
 750
 751         skb_dst_set(skb, &rt->dst);
 752
 753         skb_reserve(skb, hlen);
 754
 755         skb_reset_network_header(skb);
 756         iph = ip_hdr(skb);
 757         skb_put(skb, sizeof(struct iphdr) + 4);
 758
 759         iph->version  = 4;
 760         iph->ihl      = (sizeof(struct iphdr)+4)>>2;
 761         iph->tos      = 0xc0;
 762         iph->frag_off = htons(IP_DF);
 763         iph->ttl      = 1;
 764         iph->daddr    = dst;
 765         iph->saddr    = fl4.saddr;
 766         iph->protocol = IPPROTO_IGMP;
 767         ip_select_ident(net, skb, NULL);
 768         ((u8 *)&iph[1])[0] = IPOPT_RA;
 769         ((u8 *)&iph[1])[1] = 4;
 770         ((u8 *)&iph[1])[2] = 0;
 771         ((u8 *)&iph[1])[3] = 0;
 772
 773         ih = skb_put(skb, sizeof(struct igmphdr));
 774         ih->type = type;
 775         ih->code = 0;
 776         ih->csum = 0;
 777         ih->group = group;
 778         ih->csum = ip_compute_csum((void *)ih, sizeof(struct igmphdr));
 779
 780         return ip_local_out(net, skb->sk, skb);
 781 }
 782
 783 static void igmp_gq_timer_expire(struct timer_list *t)
 784 {
 785         struct in_device *in_dev = from_timer(in_dev, t, mr_gq_timer);
 786
 787         in_dev->mr_gq_running = 0;
 788         igmpv3_send_report(in_dev, NULL);
 789         in_dev_put(in_dev);
 790 }
 791
 792 static void igmp_ifc_timer_expire(struct timer_list *t)
 793 {
 794         struct in_device *in_dev = from_timer(in_dev, t, mr_ifc_timer);
 795
 796         igmpv3_send_cr(in_dev);
 797         if (in_dev->mr_ifc_count) {
 798                 in_dev->mr_ifc_count--;
 799                 igmp_ifc_start_timer(in_dev,
 800                                      unsolicited_report_interval(in_dev));
 801         }
 802         in_dev_put(in_dev);
 803 }
 804
 805 static void igmp_ifc_event(struct in_device *in_dev)
 806 {
 807         struct net *net = dev_net(in_dev->dev);
 808         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev))
 809                 return;
 810         in_dev->mr_ifc_count = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
 811         igmp_ifc_start_timer(in_dev, 1);
 812 }
 813
 814
 815 static void igmp_timer_expire(struct timer_list *t)
 816 {
 817         struct ip_mc_list *im = from_timer(im, t, timer);
 818         struct in_device *in_dev = im->interface;
 819
 820         spin_lock(&im->lock);
 821         im->tm_running = 0;
 822
 823         if (im->unsolicit_count) {
 824                 im->unsolicit_count--;
 825                 igmp_start_timer(im, unsolicited_report_interval(in_dev));
 826         }
 827         im->reporter = 1;
 828         spin_unlock(&im->lock);
 829
 830         if (IGMP_V1_SEEN(in_dev))
 831                 igmp_send_report(in_dev, im, IGMP_HOST_MEMBERSHIP_REPORT);
 832         else if (IGMP_V2_SEEN(in_dev))
 833                 igmp_send_report(in_dev, im, IGMPV2_HOST_MEMBERSHIP_REPORT);
 834         else
 835                 igmp_send_report(in_dev, im, IGMPV3_HOST_MEMBERSHIP_REPORT);
 836
 837         ip_ma_put(im);
 838 }
 839
 840 /* mark EXCLUDE-mode sources */
 841 static int igmp_xmarksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 842 {
 843         struct ip_sf_list *psf;
 844         int i, scount;
 845
 846         scount = 0;
 847         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 848                 if (scount == nsrcs)
 849                         break;
 850                 for (i = 0; i < nsrcs; i++) {
 851                         /* skip inactive filters */
 852                         if (psf->sf_count[MCAST_INCLUDE] ||
 853                             pmc->sfcount[MCAST_EXCLUDE] !=
 854                             psf->sf_count[MCAST_EXCLUDE])
 855                                 break;
 856                         if (srcs[i] == psf->sf_inaddr) {
 857                                 scount++;
 858                                 break;
 859                         }
 860                 }
 861         }
 862         pmc->gsquery = 0;
 863         if (scount == nsrcs)    /* all sources excluded */
 864                 return 0;
 865         return 1;
 866 }
 867
 868 static int igmp_marksources(struct ip_mc_list *pmc, int nsrcs, __be32 *srcs)
 869 {
 870         struct ip_sf_list *psf;
 871         int i, scount;
 872
 873         if (pmc->sfmode == MCAST_EXCLUDE)
 874                 return igmp_xmarksources(pmc, nsrcs, srcs);
 875
 876         /* mark INCLUDE-mode sources */
 877         scount = 0;
 878         for (psf = pmc->sources; psf; psf = psf->sf_next) {
 879                 if (scount == nsrcs)
 880                         break;
 881                 for (i = 0; i < nsrcs; i++)
 882                         if (srcs[i] == psf->sf_inaddr) {
 883                                 psf->sf_gsresp = 1;
 884                                 scount++;
 885                                 break;
 886                         }
 887         }
 888         if (!scount) {
 889                 pmc->gsquery = 0;
 890                 return 0;
 891         }
 892         pmc->gsquery = 1;
 893         return 1;
 894 }
 895
 896 /* return true if packet was dropped */
 897 static bool igmp_heard_report(struct in_device *in_dev, __be32 group)
 898 {
 899         struct ip_mc_list *im;
 900         struct net *net = dev_net(in_dev->dev);
 901
 902         /* Timers are only set for non-local groups */
 903
 904         if (group == IGMP_ALL_HOSTS)
 905                 return false;
 906         if (ipv4_is_local_multicast(group) && !net->ipv4.sysctl_igmp_llm_reports)
 907                 return false;
 908
 909         rcu_read_lock();
 910         for_each_pmc_rcu(in_dev, im) {
 911                 if (im->multiaddr == group) {
 912                         igmp_stop_timer(im);
 913                         break;
 914                 }
 915         }
 916         rcu_read_unlock();
 917         return false;
 918 }
 919
 920 /* return true if packet was dropped */
 921 static bool igmp_heard_query(struct in_device *in_dev, struct sk_buff *skb,
 922         int len)
 923 {
 924         struct igmphdr          *ih = igmp_hdr(skb);
 925         struct igmpv3_query *ih3 = igmpv3_query_hdr(skb);
 926         struct ip_mc_list       *im;
 927         __be32                  group = ih->group;
 928         int                     max_delay;
 929         int                     mark = 0;
 930         struct net              *net = dev_net(in_dev->dev);
 931
 932
 933         if (len == 8) {
 934                 if (ih->code == 0) {
 935                         /* Alas, old v1 router presents here. */
 936
 937                         max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 938                         in_dev->mr_v1_seen = jiffies +
 939                                 IGMP_V1_ROUTER_PRESENT_TIMEOUT;
 940                         group = 0;
 941                 } else {
 942                         /* v2 router present */
 943                         max_delay = ih->code*(HZ/IGMP_TIMER_SCALE);
 944                         in_dev->mr_v2_seen = jiffies +
 945                                 IGMP_V2_ROUTER_PRESENT_TIMEOUT;
 946                 }
 947                 /* cancel the interface change timer */
 948                 in_dev->mr_ifc_count = 0;
 949                 if (del_timer(&in_dev->mr_ifc_timer))
 950                         __in_dev_put(in_dev);
 951                 /* clear deleted report items */
 952                 igmpv3_clear_delrec(in_dev);
 953         } else if (len < 12) {
 954                 return true;    /* ignore bogus packet; freed by caller */
 955         } else if (IGMP_V1_SEEN(in_dev)) {
 956                 /* This is a v3 query with v1 queriers present */
 957                 max_delay = IGMP_QUERY_RESPONSE_INTERVAL;
 958                 group = 0;
 959         } else if (IGMP_V2_SEEN(in_dev)) {
 960                 /* this is a v3 query with v2 queriers present;
 961                  * Interpretation of the max_delay code is problematic here.
 962                  * A real v2 host would use ih_code directly, while v3 has a
 963                  * different encoding. We use the v3 encoding as more likely
 964                  * to be intended in a v3 query.
 965                  */
 966                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 967                 if (!max_delay)
 968                         max_delay = 1;  /* can't mod w/ 0 */
 969         } else { /* v3 */
 970                 if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)))
 971                         return true;
 972
 973                 ih3 = igmpv3_query_hdr(skb);
 974                 if (ih3->nsrcs) {
 975                         if (!pskb_may_pull(skb, sizeof(struct igmpv3_query)
 976                                            + ntohs(ih3->nsrcs)*sizeof(__be32)))
 977                                 return true;
 978                         ih3 = igmpv3_query_hdr(skb);
 979                 }
 980
 981                 max_delay = IGMPV3_MRC(ih3->code)*(HZ/IGMP_TIMER_SCALE);
 982                 if (!max_delay)
 983                         max_delay = 1;  /* can't mod w/ 0 */
 984                 in_dev->mr_maxdelay = max_delay;
 985                 if (ih3->qrv)
 986                         in_dev->mr_qrv = ih3->qrv;
 987                 if (!group) { /* general query */
 988                         if (ih3->nsrcs)
 989                                 return true;    /* no sources allowed */
 990                         igmp_gq_start_timer(in_dev);
 991                         return false;
 992                 }
 993                 /* mark sources to include, if group & source-specific */
 994                 mark = ih3->nsrcs != 0;
 995         }
 996
 997         /*
 998          * - Start the timers in all of our membership records
 999          *   that the query applies to for the interface on
1000          *   which the query arrived excl. those that belong
1001          *   to a "local" group (224.0.0.X)
1002          * - For timers already running check if they need to
1003          *   be reset.
1004          * - Use the igmp->igmp_code field as the maximum
1005          *   delay possible
1006          */
1007         rcu_read_lock();
1008         for_each_pmc_rcu(in_dev, im) {
1009                 int changed;
1010
1011                 if (group && group != im->multiaddr)
1012                         continue;
1013                 if (im->multiaddr == IGMP_ALL_HOSTS)
1014                         continue;
1015                 if (ipv4_is_local_multicast(im->multiaddr) &&
1016                     !net->ipv4.sysctl_igmp_llm_reports)
1017                         continue;
1018                 spin_lock_bh(&im->lock);
1019                 if (im->tm_running)
1020                         im->gsquery = im->gsquery && mark;
1021                 else
1022                         im->gsquery = mark;
1023                 changed = !im->gsquery ||
1024                         igmp_marksources(im, ntohs(ih3->nsrcs), ih3->srcs);
1025                 spin_unlock_bh(&im->lock);
1026                 if (changed)
1027                         igmp_mod_timer(im, max_delay);
1028         }
1029         rcu_read_unlock();
1030         return false;
1031 }
1032
1033 /* called in rcu_read_lock() section */
1034 int igmp_rcv(struct sk_buff *skb)
1035 {
1036         /* This basically follows the spec line by line -- see RFC1112 */
1037         struct igmphdr *ih;
1038         struct net_device *dev = skb->dev;
1039         struct in_device *in_dev;
1040         int len = skb->len;
1041         bool dropped = true;
1042
1043         if (netif_is_l3_master(dev)) {
1044                 dev = dev_get_by_index_rcu(dev_net(dev), IPCB(skb)->iif);
1045                 if (!dev)
1046                         goto drop;
1047         }
1048
1049         in_dev = __in_dev_get_rcu(dev);
1050         if (!in_dev)
1051                 goto drop;
1052
1053         if (!pskb_may_pull(skb, sizeof(struct igmphdr)))
1054                 goto drop;
1055
1056         if (skb_checksum_simple_validate(skb))
1057                 goto drop;
1058
1059         ih = igmp_hdr(skb);
1060         switch (ih->type) {
1061         case IGMP_HOST_MEMBERSHIP_QUERY:
1062                 dropped = igmp_heard_query(in_dev, skb, len);
1063                 break;
1064         case IGMP_HOST_MEMBERSHIP_REPORT:
1065         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1066                 /* Is it our report looped back? */
1067                 if (rt_is_output_route(skb_rtable(skb)))
1068                         break;
1069                 /* don't rely on MC router hearing unicast reports */
1070                 if (skb->pkt_type == PACKET_MULTICAST ||
1071                     skb->pkt_type == PACKET_BROADCAST)
1072                         dropped = igmp_heard_report(in_dev, ih->group);
1073                 break;
1074         case IGMP_PIM:
1075 #ifdef CONFIG_IP_PIMSM_V1
1076                 return pim_rcv_v1(skb);
1077 #endif
1078         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1079         case IGMP_DVMRP:
1080         case IGMP_TRACE:
1081         case IGMP_HOST_LEAVE_MESSAGE:
1082         case IGMP_MTRACE:
1083         case IGMP_MTRACE_RESP:
1084                 break;
1085         default:
1086                 break;
1087         }
1088
1089 drop:
1090         if (dropped)
1091                 kfree_skb(skb);
1092         else
1093                 consume_skb(skb);
1094         return 0;
1095 }
1096
1097 #endif
1098
1099
1100 /*
1101  *      Add a filter to a device
1102  */
1103
1104 static void ip_mc_filter_add(struct in_device *in_dev, __be32 addr)
1105 {
1106         char buf[MAX_ADDR_LEN];
1107         struct net_device *dev = in_dev->dev;
1108
1109         /* Checking for IFF_MULTICAST here is WRONG-WRONG-WRONG.
1110            We will get multicast token leakage, when IFF_MULTICAST
1111            is changed. This check should be done in ndo_set_rx_mode
1112            routine. Something sort of:
1113            if (dev->mc_list && dev->flags&IFF_MULTICAST) { do it; }
1114            --ANK
1115            */
1116         if (arp_mc_map(addr, buf, dev, 0) == 0)
1117                 dev_mc_add(dev, buf);
1118 }
1119
1120 /*
1121  *      Remove a filter from a device
1122  */
1123
1124 static void ip_mc_filter_del(struct in_device *in_dev, __be32 addr)
1125 {
1126         char buf[MAX_ADDR_LEN];
1127         struct net_device *dev = in_dev->dev;
1128
1129         if (arp_mc_map(addr, buf, dev, 0) == 0)
1130                 dev_mc_del(dev, buf);
1131 }
1132
1133 #ifdef CONFIG_IP_MULTICAST
1134 /*
1135  * deleted ip_mc_list manipulation
1136  */
1137 static void igmpv3_add_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1138 {
1139         struct ip_mc_list *pmc;
1140         struct net *net = dev_net(in_dev->dev);
1141
1142         /* this is an "ip_mc_list" for convenience; only the fields below
1143          * are actually used. In particular, the refcnt and users are not
1144          * used for management of the delete list. Using the same structure
1145          * for deleted items allows change reports to use common code with
1146          * non-deleted or query-response MCA's.
1147          */
1148         pmc = kzalloc(sizeof(*pmc), GFP_KERNEL);
1149         if (!pmc)
1150                 return;
1151         spin_lock_init(&pmc->lock);
1152         spin_lock_bh(&im->lock);
1153         pmc->interface = im->interface;
1154         in_dev_hold(in_dev);
1155         pmc->multiaddr = im->multiaddr;
1156         pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1157         pmc->sfmode = im->sfmode;
1158         if (pmc->sfmode == MCAST_INCLUDE) {
1159                 struct ip_sf_list *psf;
1160
1161                 pmc->tomb = im->tomb;
1162                 pmc->sources = im->sources;
1163                 im->tomb = im->sources = NULL;
1164                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1165                         psf->sf_crcount = pmc->crcount;
1166         }
1167         spin_unlock_bh(&im->lock);
1168
1169         spin_lock_bh(&in_dev->mc_tomb_lock);
1170         pmc->next = in_dev->mc_tomb;
1171         in_dev->mc_tomb = pmc;
1172         spin_unlock_bh(&in_dev->mc_tomb_lock);
1173 }
1174
1175 /*
1176  * restore ip_mc_list deleted records
1177  */
1178 static void igmpv3_del_delrec(struct in_device *in_dev, struct ip_mc_list *im)
1179 {
1180         struct ip_mc_list *pmc, *pmc_prev;
1181         struct ip_sf_list *psf;
1182         struct net *net = dev_net(in_dev->dev);
1183         __be32 multiaddr = im->multiaddr;
1184
1185         spin_lock_bh(&in_dev->mc_tomb_lock);
1186         pmc_prev = NULL;
1187         for (pmc = in_dev->mc_tomb; pmc; pmc = pmc->next) {
1188                 if (pmc->multiaddr == multiaddr)
1189                         break;
1190                 pmc_prev = pmc;
1191         }
1192         if (pmc) {
1193                 if (pmc_prev)
1194                         pmc_prev->next = pmc->next;
1195                 else
1196                         in_dev->mc_tomb = pmc->next;
1197         }
1198         spin_unlock_bh(&in_dev->mc_tomb_lock);
1199
1200         spin_lock_bh(&im->lock);
1201         if (pmc) {
1202                 im->interface = pmc->interface;
1203                 im->sfmode = pmc->sfmode;
1204                 if (pmc->sfmode == MCAST_INCLUDE) {
1205                         im->tomb = pmc->tomb;
1206                         im->sources = pmc->sources;
1207                         for (psf = im->sources; psf; psf = psf->sf_next)
1208                                 psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1209                 } else {
1210                         im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1211                 }
1212                 in_dev_put(pmc->interface);
1213                 kfree(pmc);
1214         }
1215         spin_unlock_bh(&im->lock);
1216 }
1217
1218 /*
1219  * flush ip_mc_list deleted records
1220  */
1221 static void igmpv3_clear_delrec(struct in_device *in_dev)
1222 {
1223         struct ip_mc_list *pmc, *nextpmc;
1224
1225         spin_lock_bh(&in_dev->mc_tomb_lock);
1226         pmc = in_dev->mc_tomb;
1227         in_dev->mc_tomb = NULL;
1228         spin_unlock_bh(&in_dev->mc_tomb_lock);
1229
1230         for (; pmc; pmc = nextpmc) {
1231                 nextpmc = pmc->next;
1232                 ip_mc_clear_src(pmc);
1233                 in_dev_put(pmc->interface);
1234                 kfree(pmc);
1235         }
1236         /* clear dead sources, too */
1237         rcu_read_lock();
1238         for_each_pmc_rcu(in_dev, pmc) {
1239                 struct ip_sf_list *psf, *psf_next;
1240
1241                 spin_lock_bh(&pmc->lock);
1242                 psf = pmc->tomb;
1243                 pmc->tomb = NULL;
1244                 spin_unlock_bh(&pmc->lock);
1245                 for (; psf; psf = psf_next) {
1246                         psf_next = psf->sf_next;
1247                         kfree(psf);
1248                 }
1249         }
1250         rcu_read_unlock();
1251 }
1252 #endif
1253
1254 static void igmp_group_dropped(struct ip_mc_list *im)
1255 {
1256         struct in_device *in_dev = im->interface;
1257 #ifdef CONFIG_IP_MULTICAST
1258         struct net *net = dev_net(in_dev->dev);
1259         int reporter;
1260 #endif
1261
1262         if (im->loaded) {
1263                 im->loaded = 0;
1264                 ip_mc_filter_del(in_dev, im->multiaddr);
1265         }
1266
1267 #ifdef CONFIG_IP_MULTICAST
1268         if (im->multiaddr == IGMP_ALL_HOSTS)
1269                 return;
1270         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1271                 return;
1272
1273         reporter = im->reporter;
1274         igmp_stop_timer(im);
1275
1276         if (!in_dev->dead) {
1277                 if (IGMP_V1_SEEN(in_dev))
1278                         return;
1279                 if (IGMP_V2_SEEN(in_dev)) {
1280                         if (reporter)
1281                                 igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1282                         return;
1283                 }
1284                 /* IGMPv3 */
1285                 igmpv3_add_delrec(in_dev, im);
1286
1287                 igmp_ifc_event(in_dev);
1288         }
1289 #endif
1290 }
1291
1292 static void igmp_group_added(struct ip_mc_list *im, unsigned int mode)
1293 {
1294         struct in_device *in_dev = im->interface;
1295 #ifdef CONFIG_IP_MULTICAST
1296         struct net *net = dev_net(in_dev->dev);
1297 #endif
1298
1299         if (im->loaded == 0) {
1300                 im->loaded = 1;
1301                 ip_mc_filter_add(in_dev, im->multiaddr);
1302         }
1303
1304 #ifdef CONFIG_IP_MULTICAST
1305         if (im->multiaddr == IGMP_ALL_HOSTS)
1306                 return;
1307         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1308                 return;
1309
1310         if (in_dev->dead)
1311                 return;
1312         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1313                 spin_lock_bh(&im->lock);
1314                 igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1315                 spin_unlock_bh(&im->lock);
1316                 return;
1317         }
1318         /* else, v3 */
1319
1320         /* Based on RFC3376 5.1, for newly added INCLUDE SSM, we should
1321          * not send filter-mode change record as the mode should be from
1322          * IN() to IN(A).
1323          */
1324         if (mode == MCAST_EXCLUDE)
1325                 im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1326
1327         igmp_ifc_event(in_dev);
1328 #endif
1329 }
1330
1331
1332 /*
1333  *      Multicast list managers
1334  */
1335
1336 static u32 ip_mc_hash(const struct ip_mc_list *im)
1337 {
1338         return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1339 }
1340
1341 static void ip_mc_hash_add(struct in_device *in_dev,
1342                            struct ip_mc_list *im)
1343 {
1344         struct ip_mc_list __rcu **mc_hash;
1345         u32 hash;
1346
1347         mc_hash = rtnl_dereference(in_dev->mc_hash);
1348         if (mc_hash) {
1349                 hash = ip_mc_hash(im);
1350                 im->next_hash = mc_hash[hash];
1351                 rcu_assign_pointer(mc_hash[hash], im);
1352                 return;
1353         }
1354
1355         /* do not use a hash table for small number of items */
1356         if (in_dev->mc_count < 4)
1357                 return;
1358
1359         mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1360                           GFP_KERNEL);
1361         if (!mc_hash)
1362                 return;
1363
1364         for_each_pmc_rtnl(in_dev, im) {
1365                 hash = ip_mc_hash(im);
1366                 im->next_hash = mc_hash[hash];
1367                 RCU_INIT_POINTER(mc_hash[hash], im);
1368         }
1369
1370         rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1371 }
1372
1373 static void ip_mc_hash_remove(struct in_device *in_dev,
1374                               struct ip_mc_list *im)
1375 {
1376         struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1377         struct ip_mc_list *aux;
1378
1379         if (!mc_hash)
1380                 return;
1381         mc_hash += ip_mc_hash(im);
1382         while ((aux = rtnl_dereference(*mc_hash)) != im)
1383                 mc_hash = &aux->next_hash;
1384         *mc_hash = im->next_hash;
1385 }
1386
1387
1388 /*
1389  *      A socket has joined a multicast group on device dev.
1390  */
1391 void __ip_mc_inc_group(struct in_device *in_dev, __be32 addr, unsigned int mode)
1392 {
1393         struct ip_mc_list *im;
1394 #ifdef CONFIG_IP_MULTICAST
1395         struct net *net = dev_net(in_dev->dev);
1396 #endif
1397
1398         ASSERT_RTNL();
1399
1400         for_each_pmc_rtnl(in_dev, im) {
1401                 if (im->multiaddr == addr) {
1402                         im->users++;
1403                         ip_mc_add_src(in_dev, &addr, mode, 0, NULL, 0);
1404                         goto out;
1405                 }
1406         }
1407
1408         im = kzalloc(sizeof(*im), GFP_KERNEL);
1409         if (!im)
1410                 goto out;
1411
1412         im->users = 1;
1413         im->interface = in_dev;
1414         in_dev_hold(in_dev);
1415         im->multiaddr = addr;
1416         /* initial mode is (EX, empty) */
1417         im->sfmode = mode;
1418         im->sfcount[mode] = 1;
1419         refcount_set(&im->refcnt, 1);
1420         spin_lock_init(&im->lock);
1421 #ifdef CONFIG_IP_MULTICAST
1422         timer_setup(&im->timer, igmp_timer_expire, 0);
1423         im->unsolicit_count = net->ipv4.sysctl_igmp_qrv;
1424 #endif
1425
1426         im->next_rcu = in_dev->mc_list;
1427         in_dev->mc_count++;
1428         rcu_assign_pointer(in_dev->mc_list, im);
1429
1430         ip_mc_hash_add(in_dev, im);
1431
1432 #ifdef CONFIG_IP_MULTICAST
1433         igmpv3_del_delrec(in_dev, im);
1434 #endif
1435         igmp_group_added(im, mode);
1436         if (!in_dev->dead)
1437                 ip_rt_multicast_event(in_dev);
1438 out:
1439         return;
1440 }
1441
1442 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1443 {
1444         __ip_mc_inc_group(in_dev, addr, MCAST_EXCLUDE);
1445 }
1446 EXPORT_SYMBOL(ip_mc_inc_group);
1447
1448 static int ip_mc_check_iphdr(struct sk_buff *skb)
1449 {
1450         const struct iphdr *iph;
1451         unsigned int len;
1452         unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1453
1454         if (!pskb_may_pull(skb, offset))
1455                 return -EINVAL;
1456
1457         iph = ip_hdr(skb);
1458
1459         if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1460                 return -EINVAL;
1461
1462         offset += ip_hdrlen(skb) - sizeof(*iph);
1463
1464         if (!pskb_may_pull(skb, offset))
1465                 return -EINVAL;
1466
1467         iph = ip_hdr(skb);
1468
1469         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1470                 return -EINVAL;
1471
1472         len = skb_network_offset(skb) + ntohs(iph->tot_len);
1473         if (skb->len < len || len < offset)
1474                 return -EINVAL;
1475
1476         skb_set_transport_header(skb, offset);
1477
1478         return 0;
1479 }
1480
1481 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1482 {
1483         unsigned int len = skb_transport_offset(skb);
1484
1485         len += sizeof(struct igmpv3_report);
1486
1487         return pskb_may_pull(skb, len) ? 0 : -EINVAL;
1488 }
1489
1490 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1491 {
1492         unsigned int len = skb_transport_offset(skb);
1493
1494         len += sizeof(struct igmphdr);
1495         if (skb->len < len)
1496                 return -EINVAL;
1497
1498         /* IGMPv{1,2}? */
1499         if (skb->len != len) {
1500                 /* or IGMPv3? */
1501                 len += sizeof(struct igmpv3_query) - sizeof(struct igmphdr);
1502                 if (skb->len < len || !pskb_may_pull(skb, len))
1503                         return -EINVAL;
1504         }
1505
1506         /* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1507          * all-systems destination addresses (224.0.0.1) for general queries
1508          */
1509         if (!igmp_hdr(skb)->group &&
1510             ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1511                 return -EINVAL;
1512
1513         return 0;
1514 }
1515
1516 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1517 {
1518         switch (igmp_hdr(skb)->type) {
1519         case IGMP_HOST_LEAVE_MESSAGE:
1520         case IGMP_HOST_MEMBERSHIP_REPORT:
1521         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1522                 /* fall through */
1523                 return 0;
1524         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1525                 return ip_mc_check_igmp_reportv3(skb);
1526         case IGMP_HOST_MEMBERSHIP_QUERY:
1527                 return ip_mc_check_igmp_query(skb);
1528         default:
1529                 return -ENOMSG;
1530         }
1531 }
1532
1533 static inline __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1534 {
1535         return skb_checksum_simple_validate(skb);
1536 }
1537
1538 static int __ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1539
1540 {
1541         struct sk_buff *skb_chk;
1542         unsigned int transport_len;
1543         unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1544         int ret = -EINVAL;
1545
1546         transport_len = ntohs(ip_hdr(skb)->tot_len) - ip_hdrlen(skb);
1547
1548         skb_chk = skb_checksum_trimmed(skb, transport_len,
1549                                        ip_mc_validate_checksum);
1550         if (!skb_chk)
1551                 goto err;
1552
1553         if (!pskb_may_pull(skb_chk, len))
1554                 goto err;
1555
1556         ret = ip_mc_check_igmp_msg(skb_chk);
1557         if (ret)
1558                 goto err;
1559
1560         if (skb_trimmed)
1561                 *skb_trimmed = skb_chk;
1562         /* free now unneeded clone */
1563         else if (skb_chk != skb)
1564                 kfree_skb(skb_chk);
1565
1566         ret = 0;
1567
1568 err:
1569         if (ret && skb_chk && skb_chk != skb)
1570                 kfree_skb(skb_chk);
1571
1572         return ret;
1573 }
1574
1575 /**
1576  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1577  * @skb: the skb to validate
1578  * @skb_trimmed: to store an skb pointer trimmed to IPv4 packet tail (optional)
1579  *
1580  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1581  * skb transport header accordingly and returns zero.
1582  *
1583  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1584  *  standard
1585  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1586  * -ENOMEM: A memory allocation failure happened.
1587  *
1588  * Optionally, an skb pointer might be provided via skb_trimmed (or set it
1589  * to NULL): After parsing an IGMP packet successfully it will point to
1590  * an skb which has its tail aligned to the IP packet end. This might
1591  * either be the originally provided skb or a trimmed, cloned version if
1592  * the skb frame had data beyond the IP packet. A cloned skb allows us
1593  * to leave the original skb and its full frame unchanged (which might be
1594  * desirable for layer 2 frame jugglers).
1595  *
1596  * Caller needs to set the skb network header and free any returned skb if it
1597  * differs from the provided skb.
1598  */
1599 int ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1600 {
1601         int ret = ip_mc_check_iphdr(skb);
1602
1603         if (ret < 0)
1604                 return ret;
1605
1606         if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1607                 return -ENOMSG;
1608
1609         return __ip_mc_check_igmp(skb, skb_trimmed);
1610 }
1611 EXPORT_SYMBOL(ip_mc_check_igmp);
1612
1613 /*
1614  *      Resend IGMP JOIN report; used by netdev notifier.
1615  */
1616 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1617 {
1618 #ifdef CONFIG_IP_MULTICAST
1619         struct ip_mc_list *im;
1620         int type;
1621         struct net *net = dev_net(in_dev->dev);
1622
1623         ASSERT_RTNL();
1624
1625         for_each_pmc_rtnl(in_dev, im) {
1626                 if (im->multiaddr == IGMP_ALL_HOSTS)
1627                         continue;
1628                 if (ipv4_is_local_multicast(im->multiaddr) &&
1629                     !net->ipv4.sysctl_igmp_llm_reports)
1630                         continue;
1631
1632                 /* a failover is happening and switches
1633                  * must be notified immediately
1634                  */
1635                 if (IGMP_V1_SEEN(in_dev))
1636                         type = IGMP_HOST_MEMBERSHIP_REPORT;
1637                 else if (IGMP_V2_SEEN(in_dev))
1638                         type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1639                 else
1640                         type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1641                 igmp_send_report(in_dev, im, type);
1642         }
1643 #endif
1644 }
1645
1646 /*
1647  *      A socket has left a multicast group on device dev
1648  */
1649
1650 void ip_mc_dec_group(struct in_device *in_dev, __be32 addr)
1651 {
1652         struct ip_mc_list *i;
1653         struct ip_mc_list __rcu **ip;
1654
1655         ASSERT_RTNL();
1656
1657         for (ip = &in_dev->mc_list;
1658              (i = rtnl_dereference(*ip)) != NULL;
1659              ip = &i->next_rcu) {
1660                 if (i->multiaddr == addr) {
1661                         if (--i->users == 0) {
1662                                 ip_mc_hash_remove(in_dev, i);
1663                                 *ip = i->next_rcu;
1664                                 in_dev->mc_count--;
1665                                 igmp_group_dropped(i);
1666                                 ip_mc_clear_src(i);
1667
1668                                 if (!in_dev->dead)
1669                                         ip_rt_multicast_event(in_dev);
1670
1671                                 ip_ma_put(i);
1672                                 return;
1673                         }
1674                         break;
1675                 }
1676         }
1677 }
1678 EXPORT_SYMBOL(ip_mc_dec_group);
1679
1680 /* Device changing type */
1681
1682 void ip_mc_unmap(struct in_device *in_dev)
1683 {
1684         struct ip_mc_list *pmc;
1685
1686         ASSERT_RTNL();
1687
1688         for_each_pmc_rtnl(in_dev, pmc)
1689                 igmp_group_dropped(pmc);
1690 }
1691
1692 void ip_mc_remap(struct in_device *in_dev)
1693 {
1694         struct ip_mc_list *pmc;
1695
1696         ASSERT_RTNL();
1697
1698         for_each_pmc_rtnl(in_dev, pmc) {
1699 #ifdef CONFIG_IP_MULTICAST
1700                 igmpv3_del_delrec(in_dev, pmc);
1701 #endif
1702                 igmp_group_added(pmc, pmc->sfmode);
1703         }
1704 }
1705
1706 /* Device going down */
1707
1708 void ip_mc_down(struct in_device *in_dev)
1709 {
1710         struct ip_mc_list *pmc;
1711
1712         ASSERT_RTNL();
1713
1714         for_each_pmc_rtnl(in_dev, pmc)
1715                 igmp_group_dropped(pmc);
1716
1717 #ifdef CONFIG_IP_MULTICAST
1718         in_dev->mr_ifc_count = 0;
1719         if (del_timer(&in_dev->mr_ifc_timer))
1720                 __in_dev_put(in_dev);
1721         in_dev->mr_gq_running = 0;
1722         if (del_timer(&in_dev->mr_gq_timer))
1723                 __in_dev_put(in_dev);
1724 #endif
1725
1726         ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1727 }
1728
1729 void ip_mc_init_dev(struct in_device *in_dev)
1730 {
1731 #ifdef CONFIG_IP_MULTICAST
1732         struct net *net = dev_net(in_dev->dev);
1733 #endif
1734         ASSERT_RTNL();
1735
1736 #ifdef CONFIG_IP_MULTICAST
1737         timer_setup(&in_dev->mr_gq_timer, igmp_gq_timer_expire, 0);
1738         timer_setup(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire, 0);
1739         in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1740 #endif
1741
1742         spin_lock_init(&in_dev->mc_tomb_lock);
1743 }
1744
1745 /* Device going up */
1746
1747 void ip_mc_up(struct in_device *in_dev)
1748 {
1749         struct ip_mc_list *pmc;
1750 #ifdef CONFIG_IP_MULTICAST
1751         struct net *net = dev_net(in_dev->dev);
1752 #endif
1753
1754         ASSERT_RTNL();
1755
1756 #ifdef CONFIG_IP_MULTICAST
1757         in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1758 #endif
1759         ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1760
1761         for_each_pmc_rtnl(in_dev, pmc) {
1762 #ifdef CONFIG_IP_MULTICAST
1763                 igmpv3_del_delrec(in_dev, pmc);
1764 #endif
1765                 igmp_group_added(pmc, pmc->sfmode);
1766         }
1767 }
1768
1769 /*
1770  *      Device is about to be destroyed: clean up.
1771  */
1772
1773 void ip_mc_destroy_dev(struct in_device *in_dev)
1774 {
1775         struct ip_mc_list *i;
1776
1777         ASSERT_RTNL();
1778
1779         /* Deactivate timers */
1780         ip_mc_down(in_dev);
1781 #ifdef CONFIG_IP_MULTICAST
1782         igmpv3_clear_delrec(in_dev);
1783 #endif
1784
1785         while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1786                 in_dev->mc_list = i->next_rcu;
1787                 in_dev->mc_count--;
1788                 ip_ma_put(i);
1789         }
1790 }
1791
1792 /* RTNL is locked */
1793 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1794 {
1795         struct net_device *dev = NULL;
1796         struct in_device *idev = NULL;
1797
1798         if (imr->imr_ifindex) {
1799                 idev = inetdev_by_index(net, imr->imr_ifindex);
1800                 return idev;
1801         }
1802         if (imr->imr_address.s_addr) {
1803                 dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1804                 if (!dev)
1805                         return NULL;
1806         }
1807
1808         if (!dev) {
1809                 struct rtable *rt = ip_route_output(net,
1810                                                     imr->imr_multiaddr.s_addr,
1811                                                     0, 0, 0);
1812                 if (!IS_ERR(rt)) {
1813                         dev = rt->dst.dev;
1814                         ip_rt_put(rt);
1815                 }
1816         }
1817         if (dev) {
1818                 imr->imr_ifindex = dev->ifindex;
1819                 idev = __in_dev_get_rtnl(dev);
1820         }
1821         return idev;
1822 }
1823
1824 /*
1825  *      Join a socket to a group
1826  */
1827
1828 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1829         __be32 *psfsrc)
1830 {
1831         struct ip_sf_list *psf, *psf_prev;
1832         int rv = 0;
1833
1834         psf_prev = NULL;
1835         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1836                 if (psf->sf_inaddr == *psfsrc)
1837                         break;
1838                 psf_prev = psf;
1839         }
1840         if (!psf || psf->sf_count[sfmode] == 0) {
1841                 /* source filter not found, or count wrong =>  bug */
1842                 return -ESRCH;
1843         }
1844         psf->sf_count[sfmode]--;
1845         if (psf->sf_count[sfmode] == 0) {
1846                 ip_rt_multicast_event(pmc->interface);
1847         }
1848         if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1849 #ifdef CONFIG_IP_MULTICAST
1850                 struct in_device *in_dev = pmc->interface;
1851                 struct net *net = dev_net(in_dev->dev);
1852 #endif
1853
1854                 /* no more filters for this source */
1855                 if (psf_prev)
1856                         psf_prev->sf_next = psf->sf_next;
1857                 else
1858                         pmc->sources = psf->sf_next;
1859 #ifdef CONFIG_IP_MULTICAST
1860                 if (psf->sf_oldin &&
1861                     !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1862                         psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1863                         psf->sf_next = pmc->tomb;
1864                         pmc->tomb = psf;
1865                         rv = 1;
1866                 } else
1867 #endif
1868                         kfree(psf);
1869         }
1870         return rv;
1871 }
1872
1873 #ifndef CONFIG_IP_MULTICAST
1874 #define igmp_ifc_event(x)       do { } while (0)
1875 #endif
1876
1877 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1878                          int sfcount, __be32 *psfsrc, int delta)
1879 {
1880         struct ip_mc_list *pmc;
1881         int     changerec = 0;
1882         int     i, err;
1883
1884         if (!in_dev)
1885                 return -ENODEV;
1886         rcu_read_lock();
1887         for_each_pmc_rcu(in_dev, pmc) {
1888                 if (*pmca == pmc->multiaddr)
1889                         break;
1890         }
1891         if (!pmc) {
1892                 /* MCA not found?? bug */
1893                 rcu_read_unlock();
1894                 return -ESRCH;
1895         }
1896         spin_lock_bh(&pmc->lock);
1897         rcu_read_unlock();
1898 #ifdef CONFIG_IP_MULTICAST
1899         sf_markstate(pmc);
1900 #endif
1901         if (!delta) {
1902                 err = -EINVAL;
1903                 if (!pmc->sfcount[sfmode])
1904                         goto out_unlock;
1905                 pmc->sfcount[sfmode]--;
1906         }
1907         err = 0;
1908         for (i = 0; i < sfcount; i++) {
1909                 int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1910
1911                 changerec |= rv > 0;
1912                 if (!err && rv < 0)
1913                         err = rv;
1914         }
1915         if (pmc->sfmode == MCAST_EXCLUDE &&
1916             pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1917             pmc->sfcount[MCAST_INCLUDE]) {
1918 #ifdef CONFIG_IP_MULTICAST
1919                 struct ip_sf_list *psf;
1920                 struct net *net = dev_net(in_dev->dev);
1921 #endif
1922
1923                 /* filter mode change */
1924                 pmc->sfmode = MCAST_INCLUDE;
1925 #ifdef CONFIG_IP_MULTICAST
1926                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1927                 in_dev->mr_ifc_count = pmc->crcount;
1928                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1929                         psf->sf_crcount = 0;
1930                 igmp_ifc_event(pmc->interface);
1931         } else if (sf_setstate(pmc) || changerec) {
1932                 igmp_ifc_event(pmc->interface);
1933 #endif
1934         }
1935 out_unlock:
1936         spin_unlock_bh(&pmc->lock);
1937         return err;
1938 }
1939
1940 /*
1941  * Add multicast single-source filter to the interface list
1942  */
1943 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1944         __be32 *psfsrc)
1945 {
1946         struct ip_sf_list *psf, *psf_prev;
1947
1948         psf_prev = NULL;
1949         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1950                 if (psf->sf_inaddr == *psfsrc)
1951                         break;
1952                 psf_prev = psf;
1953         }
1954         if (!psf) {
1955                 psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1956                 if (!psf)
1957                         return -ENOBUFS;
1958                 psf->sf_inaddr = *psfsrc;
1959                 if (psf_prev) {
1960                         psf_prev->sf_next = psf;
1961                 } else
1962                         pmc->sources = psf;
1963         }
1964         psf->sf_count[sfmode]++;
1965         if (psf->sf_count[sfmode] == 1) {
1966                 ip_rt_multicast_event(pmc->interface);
1967         }
1968         return 0;
1969 }
1970
1971 #ifdef CONFIG_IP_MULTICAST
1972 static void sf_markstate(struct ip_mc_list *pmc)
1973 {
1974         struct ip_sf_list *psf;
1975         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1976
1977         for (psf = pmc->sources; psf; psf = psf->sf_next)
1978                 if (pmc->sfcount[MCAST_EXCLUDE]) {
1979                         psf->sf_oldin = mca_xcount ==
1980                                 psf->sf_count[MCAST_EXCLUDE] &&
1981                                 !psf->sf_count[MCAST_INCLUDE];
1982                 } else
1983                         psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
1984 }
1985
1986 static int sf_setstate(struct ip_mc_list *pmc)
1987 {
1988         struct ip_sf_list *psf, *dpsf;
1989         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1990         int qrv = pmc->interface->mr_qrv;
1991         int new_in, rv;
1992
1993         rv = 0;
1994         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1995                 if (pmc->sfcount[MCAST_EXCLUDE]) {
1996                         new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
1997                                 !psf->sf_count[MCAST_INCLUDE];
1998                 } else
1999                         new_in = psf->sf_count[MCAST_INCLUDE] != 0;
2000                 if (new_in) {
2001                         if (!psf->sf_oldin) {
2002                                 struct ip_sf_list *prev = NULL;
2003
2004                                 for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
2005                                         if (dpsf->sf_inaddr == psf->sf_inaddr)
2006                                                 break;
2007                                         prev = dpsf;
2008                                 }
2009                                 if (dpsf) {
2010                                         if (prev)
2011                                                 prev->sf_next = dpsf->sf_next;
2012                                         else
2013                                                 pmc->tomb = dpsf->sf_next;
2014                                         kfree(dpsf);
2015                                 }
2016                                 psf->sf_crcount = qrv;
2017                                 rv++;
2018                         }
2019                 } else if (psf->sf_oldin) {
2020
2021                         psf->sf_crcount = 0;
2022                         /*
2023                          * add or update "delete" records if an active filter
2024                          * is now inactive
2025                          */
2026                         for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
2027                                 if (dpsf->sf_inaddr == psf->sf_inaddr)
2028                                         break;
2029                         if (!dpsf) {
2030                                 dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
2031                                 if (!dpsf)
2032                                         continue;
2033                                 *dpsf = *psf;
2034                                 /* pmc->lock held by callers */
2035                                 dpsf->sf_next = pmc->tomb;
2036                                 pmc->tomb = dpsf;
2037                         }
2038                         dpsf->sf_crcount = qrv;
2039                         rv++;
2040                 }
2041         }
2042         return rv;
2043 }
2044 #endif
2045
2046 /*
2047  * Add multicast source filter list to the interface list
2048  */
2049 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2050                          int sfcount, __be32 *psfsrc, int delta)
2051 {
2052         struct ip_mc_list *pmc;
2053         int     isexclude;
2054         int     i, err;
2055
2056         if (!in_dev)
2057                 return -ENODEV;
2058         rcu_read_lock();
2059         for_each_pmc_rcu(in_dev, pmc) {
2060                 if (*pmca == pmc->multiaddr)
2061                         break;
2062         }
2063         if (!pmc) {
2064                 /* MCA not found?? bug */
2065                 rcu_read_unlock();
2066                 return -ESRCH;
2067         }
2068         spin_lock_bh(&pmc->lock);
2069         rcu_read_unlock();
2070
2071 #ifdef CONFIG_IP_MULTICAST
2072         sf_markstate(pmc);
2073 #endif
2074         isexclude = pmc->sfmode == MCAST_EXCLUDE;
2075         if (!delta)
2076                 pmc->sfcount[sfmode]++;
2077         err = 0;
2078         for (i = 0; i < sfcount; i++) {
2079                 err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2080                 if (err)
2081                         break;
2082         }
2083         if (err) {
2084                 int j;
2085
2086                 if (!delta)
2087                         pmc->sfcount[sfmode]--;
2088                 for (j = 0; j < i; j++)
2089                         (void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2090         } else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2091 #ifdef CONFIG_IP_MULTICAST
2092                 struct ip_sf_list *psf;
2093                 struct net *net = dev_net(pmc->interface->dev);
2094                 in_dev = pmc->interface;
2095 #endif
2096
2097                 /* filter mode change */
2098                 if (pmc->sfcount[MCAST_EXCLUDE])
2099                         pmc->sfmode = MCAST_EXCLUDE;
2100                 else if (pmc->sfcount[MCAST_INCLUDE])
2101                         pmc->sfmode = MCAST_INCLUDE;
2102 #ifdef CONFIG_IP_MULTICAST
2103                 /* else no filters; keep old mode for reports */
2104
2105                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
2106                 in_dev->mr_ifc_count = pmc->crcount;
2107                 for (psf = pmc->sources; psf; psf = psf->sf_next)
2108                         psf->sf_crcount = 0;
2109                 igmp_ifc_event(in_dev);
2110         } else if (sf_setstate(pmc)) {
2111                 igmp_ifc_event(in_dev);
2112 #endif
2113         }
2114         spin_unlock_bh(&pmc->lock);
2115         return err;
2116 }
2117
2118 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2119 {
2120         struct ip_sf_list *psf, *nextpsf, *tomb, *sources;
2121
2122         spin_lock_bh(&pmc->lock);
2123         tomb = pmc->tomb;
2124         pmc->tomb = NULL;
2125         sources = pmc->sources;
2126         pmc->sources = NULL;
2127         pmc->sfmode = MCAST_EXCLUDE;
2128         pmc->sfcount[MCAST_INCLUDE] = 0;
2129         pmc->sfcount[MCAST_EXCLUDE] = 1;
2130         spin_unlock_bh(&pmc->lock);
2131
2132         for (psf = tomb; psf; psf = nextpsf) {
2133                 nextpsf = psf->sf_next;
2134                 kfree(psf);
2135         }
2136         for (psf = sources; psf; psf = nextpsf) {
2137                 nextpsf = psf->sf_next;
2138                 kfree(psf);
2139         }
2140 }
2141
2142 /* Join a multicast group
2143  */
2144 static int __ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr,
2145                               unsigned int mode)
2146 {
2147         __be32 addr = imr->imr_multiaddr.s_addr;
2148         struct ip_mc_socklist *iml, *i;
2149         struct in_device *in_dev;
2150         struct inet_sock *inet = inet_sk(sk);
2151         struct net *net = sock_net(sk);
2152         int ifindex;
2153         int count = 0;
2154         int err;
2155
2156         ASSERT_RTNL();
2157
2158         if (!ipv4_is_multicast(addr))
2159                 return -EINVAL;
2160
2161         in_dev = ip_mc_find_dev(net, imr);
2162
2163         if (!in_dev) {
2164                 err = -ENODEV;
2165                 goto done;
2166         }
2167
2168         err = -EADDRINUSE;
2169         ifindex = imr->imr_ifindex;
2170         for_each_pmc_rtnl(inet, i) {
2171                 if (i->multi.imr_multiaddr.s_addr == addr &&
2172                     i->multi.imr_ifindex == ifindex)
2173                         goto done;
2174                 count++;
2175         }
2176         err = -ENOBUFS;
2177         if (count >= net->ipv4.sysctl_igmp_max_memberships)
2178                 goto done;
2179         iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2180         if (!iml)
2181                 goto done;
2182
2183         memcpy(&iml->multi, imr, sizeof(*imr));
2184         iml->next_rcu = inet->mc_list;
2185         iml->sflist = NULL;
2186         iml->sfmode = mode;
2187         rcu_assign_pointer(inet->mc_list, iml);
2188         __ip_mc_inc_group(in_dev, addr, mode);
2189         err = 0;
2190 done:
2191         return err;
2192 }
2193
2194 /* Join ASM (Any-Source Multicast) group
2195  */
2196 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2197 {
2198         return __ip_mc_join_group(sk, imr, MCAST_EXCLUDE);
2199 }
2200 EXPORT_SYMBOL(ip_mc_join_group);
2201
2202 /* Join SSM (Source-Specific Multicast) group
2203  */
2204 int ip_mc_join_group_ssm(struct sock *sk, struct ip_mreqn *imr,
2205                          unsigned int mode)
2206 {
2207         return __ip_mc_join_group(sk, imr, mode);
2208 }
2209
2210 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2211                            struct in_device *in_dev)
2212 {
2213         struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2214         int err;
2215
2216         if (!psf) {
2217                 /* any-source empty exclude case */
2218                 return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2219                         iml->sfmode, 0, NULL, 0);
2220         }
2221         err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2222                         iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2223         RCU_INIT_POINTER(iml->sflist, NULL);
2224         /* decrease mem now to avoid the memleak warning */
2225         atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
2226         kfree_rcu(psf, rcu);
2227         return err;
2228 }
2229
2230 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2231 {
2232         struct inet_sock *inet = inet_sk(sk);
2233         struct ip_mc_socklist *iml;
2234         struct ip_mc_socklist __rcu **imlp;
2235         struct in_device *in_dev;
2236         struct net *net = sock_net(sk);
2237         __be32 group = imr->imr_multiaddr.s_addr;
2238         u32 ifindex;
2239         int ret = -EADDRNOTAVAIL;
2240
2241         ASSERT_RTNL();
2242
2243         in_dev = ip_mc_find_dev(net, imr);
2244         if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2245                 ret = -ENODEV;
2246                 goto out;
2247         }
2248         ifindex = imr->imr_ifindex;
2249         for (imlp = &inet->mc_list;
2250              (iml = rtnl_dereference(*imlp)) != NULL;
2251              imlp = &iml->next_rcu) {
2252                 if (iml->multi.imr_multiaddr.s_addr != group)
2253                         continue;
2254                 if (ifindex) {
2255                         if (iml->multi.imr_ifindex != ifindex)
2256                                 continue;
2257                 } else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2258                                 iml->multi.imr_address.s_addr)
2259                         continue;
2260
2261                 (void) ip_mc_leave_src(sk, iml, in_dev);
2262
2263                 *imlp = iml->next_rcu;
2264
2265                 if (in_dev)
2266                         ip_mc_dec_group(in_dev, group);
2267
2268                 /* decrease mem now to avoid the memleak warning */
2269                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2270                 kfree_rcu(iml, rcu);
2271                 return 0;
2272         }
2273 out:
2274         return ret;
2275 }
2276 EXPORT_SYMBOL(ip_mc_leave_group);
2277
2278 int ip_mc_source(int add, int omode, struct sock *sk, struct
2279         ip_mreq_source *mreqs, int ifindex)
2280 {
2281         int err;
2282         struct ip_mreqn imr;
2283         __be32 addr = mreqs->imr_multiaddr;
2284         struct ip_mc_socklist *pmc;
2285         struct in_device *in_dev = NULL;
2286         struct inet_sock *inet = inet_sk(sk);
2287         struct ip_sf_socklist *psl;
2288         struct net *net = sock_net(sk);
2289         int leavegroup = 0;
2290         int i, j, rv;
2291
2292         if (!ipv4_is_multicast(addr))
2293                 return -EINVAL;
2294
2295         ASSERT_RTNL();
2296
2297         imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2298         imr.imr_address.s_addr = mreqs->imr_interface;
2299         imr.imr_ifindex = ifindex;
2300         in_dev = ip_mc_find_dev(net, &imr);
2301
2302         if (!in_dev) {
2303                 err = -ENODEV;
2304                 goto done;
2305         }
2306         err = -EADDRNOTAVAIL;
2307
2308         for_each_pmc_rtnl(inet, pmc) {
2309                 if ((pmc->multi.imr_multiaddr.s_addr ==
2310                      imr.imr_multiaddr.s_addr) &&
2311                     (pmc->multi.imr_ifindex == imr.imr_ifindex))
2312                         break;
2313         }
2314         if (!pmc) {             /* must have a prior join */
2315                 err = -EINVAL;
2316                 goto done;
2317         }
2318         /* if a source filter was set, must be the same mode as before */
2319         if (pmc->sflist) {
2320                 if (pmc->sfmode != omode) {
2321                         err = -EINVAL;
2322                         goto done;
2323                 }
2324         } else if (pmc->sfmode != omode) {
2325                 /* allow mode switches for empty-set filters */
2326                 ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2327                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2328                         NULL, 0);
2329                 pmc->sfmode = omode;
2330         }
2331
2332         psl = rtnl_dereference(pmc->sflist);
2333         if (!add) {
2334                 if (!psl)
2335                         goto done;      /* err = -EADDRNOTAVAIL */
2336                 rv = !0;
2337                 for (i = 0; i < psl->sl_count; i++) {
2338                         rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2339                                 sizeof(__be32));
2340                         if (rv == 0)
2341                                 break;
2342                 }
2343                 if (rv)         /* source not found */
2344                         goto done;      /* err = -EADDRNOTAVAIL */
2345
2346                 /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2347                 if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2348                         leavegroup = 1;
2349                         goto done;
2350                 }
2351
2352                 /* update the interface filter */
2353                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2354                         &mreqs->imr_sourceaddr, 1);
2355
2356                 for (j = i+1; j < psl->sl_count; j++)
2357                         psl->sl_addr[j-1] = psl->sl_addr[j];
2358                 psl->sl_count--;
2359                 err = 0;
2360                 goto done;
2361         }
2362         /* else, add a new source to the filter */
2363
2364         if (psl && psl->sl_count >= net->ipv4.sysctl_igmp_max_msf) {
2365                 err = -ENOBUFS;
2366                 goto done;
2367         }
2368         if (!psl || psl->sl_count == psl->sl_max) {
2369                 struct ip_sf_socklist *newpsl;
2370                 int count = IP_SFBLOCK;
2371
2372                 if (psl)
2373                         count += psl->sl_max;
2374                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2375                 if (!newpsl) {
2376                         err = -ENOBUFS;
2377                         goto done;
2378                 }
2379                 newpsl->sl_max = count;
2380                 newpsl->sl_count = count - IP_SFBLOCK;
2381                 if (psl) {
2382                         for (i = 0; i < psl->sl_count; i++)
2383                                 newpsl->sl_addr[i] = psl->sl_addr[i];
2384                         /* decrease mem now to avoid the memleak warning */
2385                         atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2386                         kfree_rcu(psl, rcu);
2387                 }
2388                 rcu_assign_pointer(pmc->sflist, newpsl);
2389                 psl = newpsl;
2390         }
2391         rv = 1; /* > 0 for insert logic below if sl_count is 0 */
2392         for (i = 0; i < psl->sl_count; i++) {
2393                 rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2394                         sizeof(__be32));
2395                 if (rv == 0)
2396                         break;
2397         }
2398         if (rv == 0)            /* address already there is an error */
2399                 goto done;
2400         for (j = psl->sl_count-1; j >= i; j--)
2401                 psl->sl_addr[j+1] = psl->sl_addr[j];
2402         psl->sl_addr[i] = mreqs->imr_sourceaddr;
2403         psl->sl_count++;
2404         err = 0;
2405         /* update the interface list */
2406         ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2407                 &mreqs->imr_sourceaddr, 1);
2408 done:
2409         if (leavegroup)
2410                 err = ip_mc_leave_group(sk, &imr);
2411         return err;
2412 }
2413
2414 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2415 {
2416         int err = 0;
2417         struct ip_mreqn imr;
2418         __be32 addr = msf->imsf_multiaddr;
2419         struct ip_mc_socklist *pmc;
2420         struct in_device *in_dev;
2421         struct inet_sock *inet = inet_sk(sk);
2422         struct ip_sf_socklist *newpsl, *psl;
2423         struct net *net = sock_net(sk);
2424         int leavegroup = 0;
2425
2426         if (!ipv4_is_multicast(addr))
2427                 return -EINVAL;
2428         if (msf->imsf_fmode != MCAST_INCLUDE &&
2429             msf->imsf_fmode != MCAST_EXCLUDE)
2430                 return -EINVAL;
2431
2432         ASSERT_RTNL();
2433
2434         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2435         imr.imr_address.s_addr = msf->imsf_interface;
2436         imr.imr_ifindex = ifindex;
2437         in_dev = ip_mc_find_dev(net, &imr);
2438
2439         if (!in_dev) {
2440                 err = -ENODEV;
2441                 goto done;
2442         }
2443
2444         /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2445         if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2446                 leavegroup = 1;
2447                 goto done;
2448         }
2449
2450         for_each_pmc_rtnl(inet, pmc) {
2451                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2452                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2453                         break;
2454         }
2455         if (!pmc) {             /* must have a prior join */
2456                 err = -EINVAL;
2457                 goto done;
2458         }
2459         if (msf->imsf_numsrc) {
2460                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2461                                                            GFP_KERNEL);
2462                 if (!newpsl) {
2463                         err = -ENOBUFS;
2464                         goto done;
2465                 }
2466                 newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2467                 memcpy(newpsl->sl_addr, msf->imsf_slist,
2468                         msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2469                 err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2470                         msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2471                 if (err) {
2472                         sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2473                         goto done;
2474                 }
2475         } else {
2476                 newpsl = NULL;
2477                 (void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2478                                      msf->imsf_fmode, 0, NULL, 0);
2479         }
2480         psl = rtnl_dereference(pmc->sflist);
2481         if (psl) {
2482                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2483                         psl->sl_count, psl->sl_addr, 0);
2484                 /* decrease mem now to avoid the memleak warning */
2485                 atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2486                 kfree_rcu(psl, rcu);
2487         } else
2488                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2489                         0, NULL, 0);
2490         rcu_assign_pointer(pmc->sflist, newpsl);
2491         pmc->sfmode = msf->imsf_fmode;
2492         err = 0;
2493 done:
2494         if (leavegroup)
2495                 err = ip_mc_leave_group(sk, &imr);
2496         return err;
2497 }
2498
2499 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2500         struct ip_msfilter __user *optval, int __user *optlen)
2501 {
2502         int err, len, count, copycount;
2503         struct ip_mreqn imr;
2504         __be32 addr = msf->imsf_multiaddr;
2505         struct ip_mc_socklist *pmc;
2506         struct in_device *in_dev;
2507         struct inet_sock *inet = inet_sk(sk);
2508         struct ip_sf_socklist *psl;
2509         struct net *net = sock_net(sk);
2510
2511         ASSERT_RTNL();
2512
2513         if (!ipv4_is_multicast(addr))
2514                 return -EINVAL;
2515
2516         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2517         imr.imr_address.s_addr = msf->imsf_interface;
2518         imr.imr_ifindex = 0;
2519         in_dev = ip_mc_find_dev(net, &imr);
2520
2521         if (!in_dev) {
2522                 err = -ENODEV;
2523                 goto done;
2524         }
2525         err = -EADDRNOTAVAIL;
2526
2527         for_each_pmc_rtnl(inet, pmc) {
2528                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2529                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2530                         break;
2531         }
2532         if (!pmc)               /* must have a prior join */
2533                 goto done;
2534         msf->imsf_fmode = pmc->sfmode;
2535         psl = rtnl_dereference(pmc->sflist);
2536         if (!psl) {
2537                 len = 0;
2538                 count = 0;
2539         } else {
2540                 count = psl->sl_count;
2541         }
2542         copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2543         len = copycount * sizeof(psl->sl_addr[0]);
2544         msf->imsf_numsrc = count;
2545         if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2546             copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2547                 return -EFAULT;
2548         }
2549         if (len &&
2550             copy_to_user(&optval->imsf_slist[0], psl->sl_addr, len))
2551                 return -EFAULT;
2552         return 0;
2553 done:
2554         return err;
2555 }
2556
2557 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2558         struct group_filter __user *optval, int __user *optlen)
2559 {
2560         int err, i, count, copycount;
2561         struct sockaddr_in *psin;
2562         __be32 addr;
2563         struct ip_mc_socklist *pmc;
2564         struct inet_sock *inet = inet_sk(sk);
2565         struct ip_sf_socklist *psl;
2566
2567         ASSERT_RTNL();
2568
2569         psin = (struct sockaddr_in *)&gsf->gf_group;
2570         if (psin->sin_family != AF_INET)
2571                 return -EINVAL;
2572         addr = psin->sin_addr.s_addr;
2573         if (!ipv4_is_multicast(addr))
2574                 return -EINVAL;
2575
2576         err = -EADDRNOTAVAIL;
2577
2578         for_each_pmc_rtnl(inet, pmc) {
2579                 if (pmc->multi.imr_multiaddr.s_addr == addr &&
2580                     pmc->multi.imr_ifindex == gsf->gf_interface)
2581                         break;
2582         }
2583         if (!pmc)               /* must have a prior join */
2584                 goto done;
2585         gsf->gf_fmode = pmc->sfmode;
2586         psl = rtnl_dereference(pmc->sflist);
2587         count = psl ? psl->sl_count : 0;
2588         copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2589         gsf->gf_numsrc = count;
2590         if (put_user(GROUP_FILTER_SIZE(copycount), optlen) ||
2591             copy_to_user(optval, gsf, GROUP_FILTER_SIZE(0))) {
2592                 return -EFAULT;
2593         }
2594         for (i = 0; i < copycount; i++) {
2595                 struct sockaddr_storage ss;
2596
2597                 psin = (struct sockaddr_in *)&ss;
2598                 memset(&ss, 0, sizeof(ss));
2599                 psin->sin_family = AF_INET;
2600                 psin->sin_addr.s_addr = psl->sl_addr[i];
2601                 if (copy_to_user(&optval->gf_slist[i], &ss, sizeof(ss)))
2602                         return -EFAULT;
2603         }
2604         return 0;
2605 done:
2606         return err;
2607 }
2608
2609 /*
2610  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2611  */
2612 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr,
2613                    int dif, int sdif)
2614 {
2615         struct inet_sock *inet = inet_sk(sk);
2616         struct ip_mc_socklist *pmc;
2617         struct ip_sf_socklist *psl;
2618         int i;
2619         int ret;
2620
2621         ret = 1;
2622         if (!ipv4_is_multicast(loc_addr))
2623                 goto out;
2624
2625         rcu_read_lock();
2626         for_each_pmc_rcu(inet, pmc) {
2627                 if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2628                     (pmc->multi.imr_ifindex == dif ||
2629                      (sdif && pmc->multi.imr_ifindex == sdif)))
2630                         break;
2631         }
2632         ret = inet->mc_all;
2633         if (!pmc)
2634                 goto unlock;
2635         psl = rcu_dereference(pmc->sflist);
2636         ret = (pmc->sfmode == MCAST_EXCLUDE);
2637         if (!psl)
2638                 goto unlock;
2639
2640         for (i = 0; i < psl->sl_count; i++) {
2641                 if (psl->sl_addr[i] == rmt_addr)
2642                         break;
2643         }
2644         ret = 0;
2645         if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2646                 goto unlock;
2647         if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2648                 goto unlock;
2649         ret = 1;
2650 unlock:
2651         rcu_read_unlock();
2652 out:
2653         return ret;
2654 }
2655
2656 /*
2657  *      A socket is closing.
2658  */
2659
2660 void ip_mc_drop_socket(struct sock *sk)
2661 {
2662         struct inet_sock *inet = inet_sk(sk);
2663         struct ip_mc_socklist *iml;
2664         struct net *net = sock_net(sk);
2665
2666         if (!inet->mc_list)
2667                 return;
2668
2669         rtnl_lock();
2670         while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2671                 struct in_device *in_dev;
2672
2673                 inet->mc_list = iml->next_rcu;
2674                 in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2675                 (void) ip_mc_leave_src(sk, iml, in_dev);
2676                 if (in_dev)
2677                         ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2678                 /* decrease mem now to avoid the memleak warning */
2679                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2680                 kfree_rcu(iml, rcu);
2681         }
2682         rtnl_unlock();
2683 }
2684
2685 /* called with rcu_read_lock() */
2686 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2687 {
2688         struct ip_mc_list *im;
2689         struct ip_mc_list __rcu **mc_hash;
2690         struct ip_sf_list *psf;
2691         int rv = 0;
2692
2693         mc_hash = rcu_dereference(in_dev->mc_hash);
2694         if (mc_hash) {
2695                 u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2696
2697                 for (im = rcu_dereference(mc_hash[hash]);
2698                      im != NULL;
2699                      im = rcu_dereference(im->next_hash)) {
2700                         if (im->multiaddr == mc_addr)
2701                                 break;
2702                 }
2703         } else {
2704                 for_each_pmc_rcu(in_dev, im) {
2705                         if (im->multiaddr == mc_addr)
2706                                 break;
2707                 }
2708         }
2709         if (im && proto == IPPROTO_IGMP) {
2710                 rv = 1;
2711         } else if (im) {
2712                 if (src_addr) {
2713                         for (psf = im->sources; psf; psf = psf->sf_next) {
2714                                 if (psf->sf_inaddr == src_addr)
2715                                         break;
2716                         }
2717                         if (psf)
2718                                 rv = psf->sf_count[MCAST_INCLUDE] ||
2719                                         psf->sf_count[MCAST_EXCLUDE] !=
2720                                         im->sfcount[MCAST_EXCLUDE];
2721                         else
2722                                 rv = im->sfcount[MCAST_EXCLUDE] != 0;
2723                 } else
2724                         rv = 1; /* unspecified source; tentatively allow */
2725         }
2726         return rv;
2727 }
2728
2729 #if defined(CONFIG_PROC_FS)
2730 struct igmp_mc_iter_state {
2731         struct seq_net_private p;
2732         struct net_device *dev;
2733         struct in_device *in_dev;
2734 };
2735
2736 #define igmp_mc_seq_private(seq)        ((struct igmp_mc_iter_state *)(seq)->private)
2737
2738 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2739 {
2740         struct net *net = seq_file_net(seq);
2741         struct ip_mc_list *im = NULL;
2742         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2743
2744         state->in_dev = NULL;
2745         for_each_netdev_rcu(net, state->dev) {
2746                 struct in_device *in_dev;
2747
2748                 in_dev = __in_dev_get_rcu(state->dev);
2749                 if (!in_dev)
2750                         continue;
2751                 im = rcu_dereference(in_dev->mc_list);
2752                 if (im) {
2753                         state->in_dev = in_dev;
2754                         break;
2755                 }
2756         }
2757         return im;
2758 }
2759
2760 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2761 {
2762         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2763
2764         im = rcu_dereference(im->next_rcu);
2765         while (!im) {
2766                 state->dev = next_net_device_rcu(state->dev);
2767                 if (!state->dev) {
2768                         state->in_dev = NULL;
2769                         break;
2770                 }
2771                 state->in_dev = __in_dev_get_rcu(state->dev);
2772                 if (!state->in_dev)
2773                         continue;
2774                 im = rcu_dereference(state->in_dev->mc_list);
2775         }
2776         return im;
2777 }
2778
2779 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2780 {
2781         struct ip_mc_list *im = igmp_mc_get_first(seq);
2782         if (im)
2783                 while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2784                         --pos;
2785         return pos ? NULL : im;
2786 }
2787
2788 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2789         __acquires(rcu)
2790 {
2791         rcu_read_lock();
2792         return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2793 }
2794
2795 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2796 {
2797         struct ip_mc_list *im;
2798         if (v == SEQ_START_TOKEN)
2799                 im = igmp_mc_get_first(seq);
2800         else
2801                 im = igmp_mc_get_next(seq, v);
2802         ++*pos;
2803         return im;
2804 }
2805
2806 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2807         __releases(rcu)
2808 {
2809         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2810
2811         state->in_dev = NULL;
2812         state->dev = NULL;
2813         rcu_read_unlock();
2814 }
2815
2816 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2817 {
2818         if (v == SEQ_START_TOKEN)
2819                 seq_puts(seq,
2820                          "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2821         else {
2822                 struct ip_mc_list *im = (struct ip_mc_list *)v;
2823                 struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2824                 char   *querier;
2825                 long delta;
2826
2827 #ifdef CONFIG_IP_MULTICAST
2828                 querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2829                           IGMP_V2_SEEN(state->in_dev) ? "V2" :
2830                           "V3";
2831 #else
2832                 querier = "NONE";
2833 #endif
2834
2835                 if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2836                         seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2837                                    state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2838                 }
2839
2840                 delta = im->timer.expires - jiffies;
2841                 seq_printf(seq,
2842                            "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2843                            im->multiaddr, im->users,
2844                            im->tm_running,
2845                            im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2846                            im->reporter);
2847         }
2848         return 0;
2849 }
2850
2851 static const struct seq_operations igmp_mc_seq_ops = {
2852         .start  =       igmp_mc_seq_start,
2853         .next   =       igmp_mc_seq_next,
2854         .stop   =       igmp_mc_seq_stop,
2855         .show   =       igmp_mc_seq_show,
2856 };
2857
2858 struct igmp_mcf_iter_state {
2859         struct seq_net_private p;
2860         struct net_device *dev;
2861         struct in_device *idev;
2862         struct ip_mc_list *im;
2863 };
2864
2865 #define igmp_mcf_seq_private(seq)       ((struct igmp_mcf_iter_state *)(seq)->private)
2866
2867 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2868 {
2869         struct net *net = seq_file_net(seq);
2870         struct ip_sf_list *psf = NULL;
2871         struct ip_mc_list *im = NULL;
2872         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2873
2874         state->idev = NULL;
2875         state->im = NULL;
2876         for_each_netdev_rcu(net, state->dev) {
2877                 struct in_device *idev;
2878                 idev = __in_dev_get_rcu(state->dev);
2879                 if (unlikely(!idev))
2880                         continue;
2881                 im = rcu_dereference(idev->mc_list);
2882                 if (likely(im)) {
2883                         spin_lock_bh(&im->lock);
2884                         psf = im->sources;
2885                         if (likely(psf)) {
2886                                 state->im = im;
2887                                 state->idev = idev;
2888                                 break;
2889                         }
2890                         spin_unlock_bh(&im->lock);
2891                 }
2892         }
2893         return psf;
2894 }
2895
2896 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2897 {
2898         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2899
2900         psf = psf->sf_next;
2901         while (!psf) {
2902                 spin_unlock_bh(&state->im->lock);
2903                 state->im = state->im->next;
2904                 while (!state->im) {
2905                         state->dev = next_net_device_rcu(state->dev);
2906                         if (!state->dev) {
2907                                 state->idev = NULL;
2908                                 goto out;
2909                         }
2910                         state->idev = __in_dev_get_rcu(state->dev);
2911                         if (!state->idev)
2912                                 continue;
2913                         state->im = rcu_dereference(state->idev->mc_list);
2914                 }
2915                 if (!state->im)
2916                         break;
2917                 spin_lock_bh(&state->im->lock);
2918                 psf = state->im->sources;
2919         }
2920 out:
2921         return psf;
2922 }
2923
2924 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2925 {
2926         struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2927         if (psf)
2928                 while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2929                         --pos;
2930         return pos ? NULL : psf;
2931 }
2932
2933 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2934         __acquires(rcu)
2935 {
2936         rcu_read_lock();
2937         return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2938 }
2939
2940 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2941 {
2942         struct ip_sf_list *psf;
2943         if (v == SEQ_START_TOKEN)
2944                 psf = igmp_mcf_get_first(seq);
2945         else
2946                 psf = igmp_mcf_get_next(seq, v);
2947         ++*pos;
2948         return psf;
2949 }
2950
2951 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2952         __releases(rcu)
2953 {
2954         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2955         if (likely(state->im)) {
2956                 spin_unlock_bh(&state->im->lock);
2957                 state->im = NULL;
2958         }
2959         state->idev = NULL;
2960         state->dev = NULL;
2961         rcu_read_unlock();
2962 }
2963
2964 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2965 {
2966         struct ip_sf_list *psf = (struct ip_sf_list *)v;
2967         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2968
2969         if (v == SEQ_START_TOKEN) {
2970                 seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
2971         } else {
2972                 seq_printf(seq,
2973                            "%3d %6.6s 0x%08x "
2974                            "0x%08x %6lu %6lu\n",
2975                            state->dev->ifindex, state->dev->name,
2976                            ntohl(state->im->multiaddr),
2977                            ntohl(psf->sf_inaddr),
2978                            psf->sf_count[MCAST_INCLUDE],
2979                            psf->sf_count[MCAST_EXCLUDE]);
2980         }
2981         return 0;
2982 }
2983
2984 static const struct seq_operations igmp_mcf_seq_ops = {
2985         .start  =       igmp_mcf_seq_start,
2986         .next   =       igmp_mcf_seq_next,
2987         .stop   =       igmp_mcf_seq_stop,
2988         .show   =       igmp_mcf_seq_show,
2989 };
2990
2991 static int __net_init igmp_net_init(struct net *net)
2992 {
2993         struct proc_dir_entry *pde;
2994         int err;
2995
2996         pde = proc_create_net("igmp", 0444, net->proc_net, &igmp_mc_seq_ops,
2997                         sizeof(struct igmp_mc_iter_state));
2998         if (!pde)
2999                 goto out_igmp;
3000         pde = proc_create_net("mcfilter", 0444, net->proc_net,
3001                         &igmp_mcf_seq_ops, sizeof(struct igmp_mcf_iter_state));
3002         if (!pde)
3003                 goto out_mcfilter;
3004         err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
3005                                    SOCK_DGRAM, 0, net);
3006         if (err < 0) {
3007                 pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
3008                        err);
3009                 goto out_sock;
3010         }
3011
3012         return 0;
3013
3014 out_sock:
3015         remove_proc_entry("mcfilter", net->proc_net);
3016 out_mcfilter:
3017         remove_proc_entry("igmp", net->proc_net);
3018 out_igmp:
3019         return -ENOMEM;
3020 }
3021
3022 static void __net_exit igmp_net_exit(struct net *net)
3023 {
3024         remove_proc_entry("mcfilter", net->proc_net);
3025         remove_proc_entry("igmp", net->proc_net);
3026         inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
3027 }
3028
3029 static struct pernet_operations igmp_net_ops = {
3030         .init = igmp_net_init,
3031         .exit = igmp_net_exit,
3032 };
3033 #endif
3034
3035 static int igmp_netdev_event(struct notifier_block *this,
3036                              unsigned long event, void *ptr)
3037 {
3038         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3039         struct in_device *in_dev;
3040
3041         switch (event) {
3042         case NETDEV_RESEND_IGMP:
3043                 in_dev = __in_dev_get_rtnl(dev);
3044                 if (in_dev)
3045                         ip_mc_rejoin_groups(in_dev);
3046                 break;
3047         default:
3048                 break;
3049         }
3050         return NOTIFY_DONE;
3051 }
3052
3053 static struct notifier_block igmp_notifier = {
3054         .notifier_call = igmp_netdev_event,
3055 };
3056
3057 int __init igmp_mc_init(void)
3058 {
3059 #if defined(CONFIG_PROC_FS)
3060         int err;
3061
3062         err = register_pernet_subsys(&igmp_net_ops);
3063         if (err)
3064                 return err;
3065         err = register_netdevice_notifier(&igmp_notifier);
3066         if (err)
3067                 goto reg_notif_fail;
3068         return 0;
3069
3070 reg_notif_fail:
3071         unregister_pernet_subsys(&igmp_net_ops);
3072         return err;
3073 #else
3074         return register_netdevice_notifier(&igmp_notifier);
3075 #endif
3076 }