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                 if (im->sfmode == MCAST_INCLUDE) {
1204                         im->tomb = pmc->tomb;
1205                         im->sources = pmc->sources;
1206                         for (psf = im->sources; psf; psf = psf->sf_next)
1207                                 psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1208                 } else {
1209                         im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1210                 }
1211                 in_dev_put(pmc->interface);
1212                 kfree(pmc);
1213         }
1214         spin_unlock_bh(&im->lock);
1215 }
1216
1217 /*
1218  * flush ip_mc_list deleted records
1219  */
1220 static void igmpv3_clear_delrec(struct in_device *in_dev)
1221 {
1222         struct ip_mc_list *pmc, *nextpmc;
1223
1224         spin_lock_bh(&in_dev->mc_tomb_lock);
1225         pmc = in_dev->mc_tomb;
1226         in_dev->mc_tomb = NULL;
1227         spin_unlock_bh(&in_dev->mc_tomb_lock);
1228
1229         for (; pmc; pmc = nextpmc) {
1230                 nextpmc = pmc->next;
1231                 ip_mc_clear_src(pmc);
1232                 in_dev_put(pmc->interface);
1233                 kfree(pmc);
1234         }
1235         /* clear dead sources, too */
1236         rcu_read_lock();
1237         for_each_pmc_rcu(in_dev, pmc) {
1238                 struct ip_sf_list *psf, *psf_next;
1239
1240                 spin_lock_bh(&pmc->lock);
1241                 psf = pmc->tomb;
1242                 pmc->tomb = NULL;
1243                 spin_unlock_bh(&pmc->lock);
1244                 for (; psf; psf = psf_next) {
1245                         psf_next = psf->sf_next;
1246                         kfree(psf);
1247                 }
1248         }
1249         rcu_read_unlock();
1250 }
1251 #endif
1252
1253 static void igmp_group_dropped(struct ip_mc_list *im)
1254 {
1255         struct in_device *in_dev = im->interface;
1256 #ifdef CONFIG_IP_MULTICAST
1257         struct net *net = dev_net(in_dev->dev);
1258         int reporter;
1259 #endif
1260
1261         if (im->loaded) {
1262                 im->loaded = 0;
1263                 ip_mc_filter_del(in_dev, im->multiaddr);
1264         }
1265
1266 #ifdef CONFIG_IP_MULTICAST
1267         if (im->multiaddr == IGMP_ALL_HOSTS)
1268                 return;
1269         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1270                 return;
1271
1272         reporter = im->reporter;
1273         igmp_stop_timer(im);
1274
1275         if (!in_dev->dead) {
1276                 if (IGMP_V1_SEEN(in_dev))
1277                         return;
1278                 if (IGMP_V2_SEEN(in_dev)) {
1279                         if (reporter)
1280                                 igmp_send_report(in_dev, im, IGMP_HOST_LEAVE_MESSAGE);
1281                         return;
1282                 }
1283                 /* IGMPv3 */
1284                 igmpv3_add_delrec(in_dev, im);
1285
1286                 igmp_ifc_event(in_dev);
1287         }
1288 #endif
1289 }
1290
1291 static void igmp_group_added(struct ip_mc_list *im, unsigned int mode)
1292 {
1293         struct in_device *in_dev = im->interface;
1294 #ifdef CONFIG_IP_MULTICAST
1295         struct net *net = dev_net(in_dev->dev);
1296 #endif
1297
1298         if (im->loaded == 0) {
1299                 im->loaded = 1;
1300                 ip_mc_filter_add(in_dev, im->multiaddr);
1301         }
1302
1303 #ifdef CONFIG_IP_MULTICAST
1304         if (im->multiaddr == IGMP_ALL_HOSTS)
1305                 return;
1306         if (ipv4_is_local_multicast(im->multiaddr) && !net->ipv4.sysctl_igmp_llm_reports)
1307                 return;
1308
1309         if (in_dev->dead)
1310                 return;
1311         if (IGMP_V1_SEEN(in_dev) || IGMP_V2_SEEN(in_dev)) {
1312                 spin_lock_bh(&im->lock);
1313                 igmp_start_timer(im, IGMP_INITIAL_REPORT_DELAY);
1314                 spin_unlock_bh(&im->lock);
1315                 return;
1316         }
1317         /* else, v3 */
1318
1319         /* Based on RFC3376 5.1, for newly added INCLUDE SSM, we should
1320          * not send filter-mode change record as the mode should be from
1321          * IN() to IN(A).
1322          */
1323         if (mode == MCAST_EXCLUDE)
1324                 im->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1325
1326         igmp_ifc_event(in_dev);
1327 #endif
1328 }
1329
1330
1331 /*
1332  *      Multicast list managers
1333  */
1334
1335 static u32 ip_mc_hash(const struct ip_mc_list *im)
1336 {
1337         return hash_32((__force u32)im->multiaddr, MC_HASH_SZ_LOG);
1338 }
1339
1340 static void ip_mc_hash_add(struct in_device *in_dev,
1341                            struct ip_mc_list *im)
1342 {
1343         struct ip_mc_list __rcu **mc_hash;
1344         u32 hash;
1345
1346         mc_hash = rtnl_dereference(in_dev->mc_hash);
1347         if (mc_hash) {
1348                 hash = ip_mc_hash(im);
1349                 im->next_hash = mc_hash[hash];
1350                 rcu_assign_pointer(mc_hash[hash], im);
1351                 return;
1352         }
1353
1354         /* do not use a hash table for small number of items */
1355         if (in_dev->mc_count < 4)
1356                 return;
1357
1358         mc_hash = kzalloc(sizeof(struct ip_mc_list *) << MC_HASH_SZ_LOG,
1359                           GFP_KERNEL);
1360         if (!mc_hash)
1361                 return;
1362
1363         for_each_pmc_rtnl(in_dev, im) {
1364                 hash = ip_mc_hash(im);
1365                 im->next_hash = mc_hash[hash];
1366                 RCU_INIT_POINTER(mc_hash[hash], im);
1367         }
1368
1369         rcu_assign_pointer(in_dev->mc_hash, mc_hash);
1370 }
1371
1372 static void ip_mc_hash_remove(struct in_device *in_dev,
1373                               struct ip_mc_list *im)
1374 {
1375         struct ip_mc_list __rcu **mc_hash = rtnl_dereference(in_dev->mc_hash);
1376         struct ip_mc_list *aux;
1377
1378         if (!mc_hash)
1379                 return;
1380         mc_hash += ip_mc_hash(im);
1381         while ((aux = rtnl_dereference(*mc_hash)) != im)
1382                 mc_hash = &aux->next_hash;
1383         *mc_hash = im->next_hash;
1384 }
1385
1386
1387 /*
1388  *      A socket has joined a multicast group on device dev.
1389  */
1390 void __ip_mc_inc_group(struct in_device *in_dev, __be32 addr, unsigned int mode)
1391 {
1392         struct ip_mc_list *im;
1393 #ifdef CONFIG_IP_MULTICAST
1394         struct net *net = dev_net(in_dev->dev);
1395 #endif
1396
1397         ASSERT_RTNL();
1398
1399         for_each_pmc_rtnl(in_dev, im) {
1400                 if (im->multiaddr == addr) {
1401                         im->users++;
1402                         ip_mc_add_src(in_dev, &addr, mode, 0, NULL, 0);
1403                         goto out;
1404                 }
1405         }
1406
1407         im = kzalloc(sizeof(*im), GFP_KERNEL);
1408         if (!im)
1409                 goto out;
1410
1411         im->users = 1;
1412         im->interface = in_dev;
1413         in_dev_hold(in_dev);
1414         im->multiaddr = addr;
1415         /* initial mode is (EX, empty) */
1416         im->sfmode = mode;
1417         im->sfcount[mode] = 1;
1418         refcount_set(&im->refcnt, 1);
1419         spin_lock_init(&im->lock);
1420 #ifdef CONFIG_IP_MULTICAST
1421         timer_setup(&im->timer, igmp_timer_expire, 0);
1422         im->unsolicit_count = net->ipv4.sysctl_igmp_qrv;
1423 #endif
1424
1425         im->next_rcu = in_dev->mc_list;
1426         in_dev->mc_count++;
1427         rcu_assign_pointer(in_dev->mc_list, im);
1428
1429         ip_mc_hash_add(in_dev, im);
1430
1431 #ifdef CONFIG_IP_MULTICAST
1432         igmpv3_del_delrec(in_dev, im);
1433 #endif
1434         igmp_group_added(im, mode);
1435         if (!in_dev->dead)
1436                 ip_rt_multicast_event(in_dev);
1437 out:
1438         return;
1439 }
1440
1441 void ip_mc_inc_group(struct in_device *in_dev, __be32 addr)
1442 {
1443         __ip_mc_inc_group(in_dev, addr, MCAST_EXCLUDE);
1444 }
1445 EXPORT_SYMBOL(ip_mc_inc_group);
1446
1447 static int ip_mc_check_iphdr(struct sk_buff *skb)
1448 {
1449         const struct iphdr *iph;
1450         unsigned int len;
1451         unsigned int offset = skb_network_offset(skb) + sizeof(*iph);
1452
1453         if (!pskb_may_pull(skb, offset))
1454                 return -EINVAL;
1455
1456         iph = ip_hdr(skb);
1457
1458         if (iph->version != 4 || ip_hdrlen(skb) < sizeof(*iph))
1459                 return -EINVAL;
1460
1461         offset += ip_hdrlen(skb) - sizeof(*iph);
1462
1463         if (!pskb_may_pull(skb, offset))
1464                 return -EINVAL;
1465
1466         iph = ip_hdr(skb);
1467
1468         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1469                 return -EINVAL;
1470
1471         len = skb_network_offset(skb) + ntohs(iph->tot_len);
1472         if (skb->len < len || len < offset)
1473                 return -EINVAL;
1474
1475         skb_set_transport_header(skb, offset);
1476
1477         return 0;
1478 }
1479
1480 static int ip_mc_check_igmp_reportv3(struct sk_buff *skb)
1481 {
1482         unsigned int len = skb_transport_offset(skb);
1483
1484         len += sizeof(struct igmpv3_report);
1485
1486         return pskb_may_pull(skb, len) ? 0 : -EINVAL;
1487 }
1488
1489 static int ip_mc_check_igmp_query(struct sk_buff *skb)
1490 {
1491         unsigned int len = skb_transport_offset(skb);
1492
1493         len += sizeof(struct igmphdr);
1494         if (skb->len < len)
1495                 return -EINVAL;
1496
1497         /* IGMPv{1,2}? */
1498         if (skb->len != len) {
1499                 /* or IGMPv3? */
1500                 len += sizeof(struct igmpv3_query) - sizeof(struct igmphdr);
1501                 if (skb->len < len || !pskb_may_pull(skb, len))
1502                         return -EINVAL;
1503         }
1504
1505         /* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
1506          * all-systems destination addresses (224.0.0.1) for general queries
1507          */
1508         if (!igmp_hdr(skb)->group &&
1509             ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
1510                 return -EINVAL;
1511
1512         return 0;
1513 }
1514
1515 static int ip_mc_check_igmp_msg(struct sk_buff *skb)
1516 {
1517         switch (igmp_hdr(skb)->type) {
1518         case IGMP_HOST_LEAVE_MESSAGE:
1519         case IGMP_HOST_MEMBERSHIP_REPORT:
1520         case IGMPV2_HOST_MEMBERSHIP_REPORT:
1521                 /* fall through */
1522                 return 0;
1523         case IGMPV3_HOST_MEMBERSHIP_REPORT:
1524                 return ip_mc_check_igmp_reportv3(skb);
1525         case IGMP_HOST_MEMBERSHIP_QUERY:
1526                 return ip_mc_check_igmp_query(skb);
1527         default:
1528                 return -ENOMSG;
1529         }
1530 }
1531
1532 static inline __sum16 ip_mc_validate_checksum(struct sk_buff *skb)
1533 {
1534         return skb_checksum_simple_validate(skb);
1535 }
1536
1537 static int __ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1538
1539 {
1540         struct sk_buff *skb_chk;
1541         unsigned int transport_len;
1542         unsigned int len = skb_transport_offset(skb) + sizeof(struct igmphdr);
1543         int ret = -EINVAL;
1544
1545         transport_len = ntohs(ip_hdr(skb)->tot_len) - ip_hdrlen(skb);
1546
1547         skb_chk = skb_checksum_trimmed(skb, transport_len,
1548                                        ip_mc_validate_checksum);
1549         if (!skb_chk)
1550                 goto err;
1551
1552         if (!pskb_may_pull(skb_chk, len))
1553                 goto err;
1554
1555         ret = ip_mc_check_igmp_msg(skb_chk);
1556         if (ret)
1557                 goto err;
1558
1559         if (skb_trimmed)
1560                 *skb_trimmed = skb_chk;
1561         /* free now unneeded clone */
1562         else if (skb_chk != skb)
1563                 kfree_skb(skb_chk);
1564
1565         ret = 0;
1566
1567 err:
1568         if (ret && skb_chk && skb_chk != skb)
1569                 kfree_skb(skb_chk);
1570
1571         return ret;
1572 }
1573
1574 /**
1575  * ip_mc_check_igmp - checks whether this is a sane IGMP packet
1576  * @skb: the skb to validate
1577  * @skb_trimmed: to store an skb pointer trimmed to IPv4 packet tail (optional)
1578  *
1579  * Checks whether an IPv4 packet is a valid IGMP packet. If so sets
1580  * skb transport header accordingly and returns zero.
1581  *
1582  * -EINVAL: A broken packet was detected, i.e. it violates some internet
1583  *  standard
1584  * -ENOMSG: IP header validation succeeded but it is not an IGMP packet.
1585  * -ENOMEM: A memory allocation failure happened.
1586  *
1587  * Optionally, an skb pointer might be provided via skb_trimmed (or set it
1588  * to NULL): After parsing an IGMP packet successfully it will point to
1589  * an skb which has its tail aligned to the IP packet end. This might
1590  * either be the originally provided skb or a trimmed, cloned version if
1591  * the skb frame had data beyond the IP packet. A cloned skb allows us
1592  * to leave the original skb and its full frame unchanged (which might be
1593  * desirable for layer 2 frame jugglers).
1594  *
1595  * Caller needs to set the skb network header and free any returned skb if it
1596  * differs from the provided skb.
1597  */
1598 int ip_mc_check_igmp(struct sk_buff *skb, struct sk_buff **skb_trimmed)
1599 {
1600         int ret = ip_mc_check_iphdr(skb);
1601
1602         if (ret < 0)
1603                 return ret;
1604
1605         if (ip_hdr(skb)->protocol != IPPROTO_IGMP)
1606                 return -ENOMSG;
1607
1608         return __ip_mc_check_igmp(skb, skb_trimmed);
1609 }
1610 EXPORT_SYMBOL(ip_mc_check_igmp);
1611
1612 /*
1613  *      Resend IGMP JOIN report; used by netdev notifier.
1614  */
1615 static void ip_mc_rejoin_groups(struct in_device *in_dev)
1616 {
1617 #ifdef CONFIG_IP_MULTICAST
1618         struct ip_mc_list *im;
1619         int type;
1620         struct net *net = dev_net(in_dev->dev);
1621
1622         ASSERT_RTNL();
1623
1624         for_each_pmc_rtnl(in_dev, im) {
1625                 if (im->multiaddr == IGMP_ALL_HOSTS)
1626                         continue;
1627                 if (ipv4_is_local_multicast(im->multiaddr) &&
1628                     !net->ipv4.sysctl_igmp_llm_reports)
1629                         continue;
1630
1631                 /* a failover is happening and switches
1632                  * must be notified immediately
1633                  */
1634                 if (IGMP_V1_SEEN(in_dev))
1635                         type = IGMP_HOST_MEMBERSHIP_REPORT;
1636                 else if (IGMP_V2_SEEN(in_dev))
1637                         type = IGMPV2_HOST_MEMBERSHIP_REPORT;
1638                 else
1639                         type = IGMPV3_HOST_MEMBERSHIP_REPORT;
1640                 igmp_send_report(in_dev, im, type);
1641         }
1642 #endif
1643 }
1644
1645 /*
1646  *      A socket has left a multicast group on device dev
1647  */
1648
1649 void ip_mc_dec_group(struct in_device *in_dev, __be32 addr)
1650 {
1651         struct ip_mc_list *i;
1652         struct ip_mc_list __rcu **ip;
1653
1654         ASSERT_RTNL();
1655
1656         for (ip = &in_dev->mc_list;
1657              (i = rtnl_dereference(*ip)) != NULL;
1658              ip = &i->next_rcu) {
1659                 if (i->multiaddr == addr) {
1660                         if (--i->users == 0) {
1661                                 ip_mc_hash_remove(in_dev, i);
1662                                 *ip = i->next_rcu;
1663                                 in_dev->mc_count--;
1664                                 igmp_group_dropped(i);
1665                                 ip_mc_clear_src(i);
1666
1667                                 if (!in_dev->dead)
1668                                         ip_rt_multicast_event(in_dev);
1669
1670                                 ip_ma_put(i);
1671                                 return;
1672                         }
1673                         break;
1674                 }
1675         }
1676 }
1677 EXPORT_SYMBOL(ip_mc_dec_group);
1678
1679 /* Device changing type */
1680
1681 void ip_mc_unmap(struct in_device *in_dev)
1682 {
1683         struct ip_mc_list *pmc;
1684
1685         ASSERT_RTNL();
1686
1687         for_each_pmc_rtnl(in_dev, pmc)
1688                 igmp_group_dropped(pmc);
1689 }
1690
1691 void ip_mc_remap(struct in_device *in_dev)
1692 {
1693         struct ip_mc_list *pmc;
1694
1695         ASSERT_RTNL();
1696
1697         for_each_pmc_rtnl(in_dev, pmc) {
1698 #ifdef CONFIG_IP_MULTICAST
1699                 igmpv3_del_delrec(in_dev, pmc);
1700 #endif
1701                 igmp_group_added(pmc, pmc->sfmode);
1702         }
1703 }
1704
1705 /* Device going down */
1706
1707 void ip_mc_down(struct in_device *in_dev)
1708 {
1709         struct ip_mc_list *pmc;
1710
1711         ASSERT_RTNL();
1712
1713         for_each_pmc_rtnl(in_dev, pmc)
1714                 igmp_group_dropped(pmc);
1715
1716 #ifdef CONFIG_IP_MULTICAST
1717         in_dev->mr_ifc_count = 0;
1718         if (del_timer(&in_dev->mr_ifc_timer))
1719                 __in_dev_put(in_dev);
1720         in_dev->mr_gq_running = 0;
1721         if (del_timer(&in_dev->mr_gq_timer))
1722                 __in_dev_put(in_dev);
1723 #endif
1724
1725         ip_mc_dec_group(in_dev, IGMP_ALL_HOSTS);
1726 }
1727
1728 void ip_mc_init_dev(struct in_device *in_dev)
1729 {
1730 #ifdef CONFIG_IP_MULTICAST
1731         struct net *net = dev_net(in_dev->dev);
1732 #endif
1733         ASSERT_RTNL();
1734
1735 #ifdef CONFIG_IP_MULTICAST
1736         timer_setup(&in_dev->mr_gq_timer, igmp_gq_timer_expire, 0);
1737         timer_setup(&in_dev->mr_ifc_timer, igmp_ifc_timer_expire, 0);
1738         in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1739 #endif
1740
1741         spin_lock_init(&in_dev->mc_tomb_lock);
1742 }
1743
1744 /* Device going up */
1745
1746 void ip_mc_up(struct in_device *in_dev)
1747 {
1748         struct ip_mc_list *pmc;
1749 #ifdef CONFIG_IP_MULTICAST
1750         struct net *net = dev_net(in_dev->dev);
1751 #endif
1752
1753         ASSERT_RTNL();
1754
1755 #ifdef CONFIG_IP_MULTICAST
1756         in_dev->mr_qrv = net->ipv4.sysctl_igmp_qrv;
1757 #endif
1758         ip_mc_inc_group(in_dev, IGMP_ALL_HOSTS);
1759
1760         for_each_pmc_rtnl(in_dev, pmc) {
1761 #ifdef CONFIG_IP_MULTICAST
1762                 igmpv3_del_delrec(in_dev, pmc);
1763 #endif
1764                 igmp_group_added(pmc, pmc->sfmode);
1765         }
1766 }
1767
1768 /*
1769  *      Device is about to be destroyed: clean up.
1770  */
1771
1772 void ip_mc_destroy_dev(struct in_device *in_dev)
1773 {
1774         struct ip_mc_list *i;
1775
1776         ASSERT_RTNL();
1777
1778         /* Deactivate timers */
1779         ip_mc_down(in_dev);
1780 #ifdef CONFIG_IP_MULTICAST
1781         igmpv3_clear_delrec(in_dev);
1782 #endif
1783
1784         while ((i = rtnl_dereference(in_dev->mc_list)) != NULL) {
1785                 in_dev->mc_list = i->next_rcu;
1786                 in_dev->mc_count--;
1787                 ip_ma_put(i);
1788         }
1789 }
1790
1791 /* RTNL is locked */
1792 static struct in_device *ip_mc_find_dev(struct net *net, struct ip_mreqn *imr)
1793 {
1794         struct net_device *dev = NULL;
1795         struct in_device *idev = NULL;
1796
1797         if (imr->imr_ifindex) {
1798                 idev = inetdev_by_index(net, imr->imr_ifindex);
1799                 return idev;
1800         }
1801         if (imr->imr_address.s_addr) {
1802                 dev = __ip_dev_find(net, imr->imr_address.s_addr, false);
1803                 if (!dev)
1804                         return NULL;
1805         }
1806
1807         if (!dev) {
1808                 struct rtable *rt = ip_route_output(net,
1809                                                     imr->imr_multiaddr.s_addr,
1810                                                     0, 0, 0);
1811                 if (!IS_ERR(rt)) {
1812                         dev = rt->dst.dev;
1813                         ip_rt_put(rt);
1814                 }
1815         }
1816         if (dev) {
1817                 imr->imr_ifindex = dev->ifindex;
1818                 idev = __in_dev_get_rtnl(dev);
1819         }
1820         return idev;
1821 }
1822
1823 /*
1824  *      Join a socket to a group
1825  */
1826
1827 static int ip_mc_del1_src(struct ip_mc_list *pmc, int sfmode,
1828         __be32 *psfsrc)
1829 {
1830         struct ip_sf_list *psf, *psf_prev;
1831         int rv = 0;
1832
1833         psf_prev = NULL;
1834         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1835                 if (psf->sf_inaddr == *psfsrc)
1836                         break;
1837                 psf_prev = psf;
1838         }
1839         if (!psf || psf->sf_count[sfmode] == 0) {
1840                 /* source filter not found, or count wrong =>  bug */
1841                 return -ESRCH;
1842         }
1843         psf->sf_count[sfmode]--;
1844         if (psf->sf_count[sfmode] == 0) {
1845                 ip_rt_multicast_event(pmc->interface);
1846         }
1847         if (!psf->sf_count[MCAST_INCLUDE] && !psf->sf_count[MCAST_EXCLUDE]) {
1848 #ifdef CONFIG_IP_MULTICAST
1849                 struct in_device *in_dev = pmc->interface;
1850                 struct net *net = dev_net(in_dev->dev);
1851 #endif
1852
1853                 /* no more filters for this source */
1854                 if (psf_prev)
1855                         psf_prev->sf_next = psf->sf_next;
1856                 else
1857                         pmc->sources = psf->sf_next;
1858 #ifdef CONFIG_IP_MULTICAST
1859                 if (psf->sf_oldin &&
1860                     !IGMP_V1_SEEN(in_dev) && !IGMP_V2_SEEN(in_dev)) {
1861                         psf->sf_crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1862                         psf->sf_next = pmc->tomb;
1863                         pmc->tomb = psf;
1864                         rv = 1;
1865                 } else
1866 #endif
1867                         kfree(psf);
1868         }
1869         return rv;
1870 }
1871
1872 #ifndef CONFIG_IP_MULTICAST
1873 #define igmp_ifc_event(x)       do { } while (0)
1874 #endif
1875
1876 static int ip_mc_del_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
1877                          int sfcount, __be32 *psfsrc, int delta)
1878 {
1879         struct ip_mc_list *pmc;
1880         int     changerec = 0;
1881         int     i, err;
1882
1883         if (!in_dev)
1884                 return -ENODEV;
1885         rcu_read_lock();
1886         for_each_pmc_rcu(in_dev, pmc) {
1887                 if (*pmca == pmc->multiaddr)
1888                         break;
1889         }
1890         if (!pmc) {
1891                 /* MCA not found?? bug */
1892                 rcu_read_unlock();
1893                 return -ESRCH;
1894         }
1895         spin_lock_bh(&pmc->lock);
1896         rcu_read_unlock();
1897 #ifdef CONFIG_IP_MULTICAST
1898         sf_markstate(pmc);
1899 #endif
1900         if (!delta) {
1901                 err = -EINVAL;
1902                 if (!pmc->sfcount[sfmode])
1903                         goto out_unlock;
1904                 pmc->sfcount[sfmode]--;
1905         }
1906         err = 0;
1907         for (i = 0; i < sfcount; i++) {
1908                 int rv = ip_mc_del1_src(pmc, sfmode, &psfsrc[i]);
1909
1910                 changerec |= rv > 0;
1911                 if (!err && rv < 0)
1912                         err = rv;
1913         }
1914         if (pmc->sfmode == MCAST_EXCLUDE &&
1915             pmc->sfcount[MCAST_EXCLUDE] == 0 &&
1916             pmc->sfcount[MCAST_INCLUDE]) {
1917 #ifdef CONFIG_IP_MULTICAST
1918                 struct ip_sf_list *psf;
1919                 struct net *net = dev_net(in_dev->dev);
1920 #endif
1921
1922                 /* filter mode change */
1923                 pmc->sfmode = MCAST_INCLUDE;
1924 #ifdef CONFIG_IP_MULTICAST
1925                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
1926                 in_dev->mr_ifc_count = pmc->crcount;
1927                 for (psf = pmc->sources; psf; psf = psf->sf_next)
1928                         psf->sf_crcount = 0;
1929                 igmp_ifc_event(pmc->interface);
1930         } else if (sf_setstate(pmc) || changerec) {
1931                 igmp_ifc_event(pmc->interface);
1932 #endif
1933         }
1934 out_unlock:
1935         spin_unlock_bh(&pmc->lock);
1936         return err;
1937 }
1938
1939 /*
1940  * Add multicast single-source filter to the interface list
1941  */
1942 static int ip_mc_add1_src(struct ip_mc_list *pmc, int sfmode,
1943         __be32 *psfsrc)
1944 {
1945         struct ip_sf_list *psf, *psf_prev;
1946
1947         psf_prev = NULL;
1948         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1949                 if (psf->sf_inaddr == *psfsrc)
1950                         break;
1951                 psf_prev = psf;
1952         }
1953         if (!psf) {
1954                 psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
1955                 if (!psf)
1956                         return -ENOBUFS;
1957                 psf->sf_inaddr = *psfsrc;
1958                 if (psf_prev) {
1959                         psf_prev->sf_next = psf;
1960                 } else
1961                         pmc->sources = psf;
1962         }
1963         psf->sf_count[sfmode]++;
1964         if (psf->sf_count[sfmode] == 1) {
1965                 ip_rt_multicast_event(pmc->interface);
1966         }
1967         return 0;
1968 }
1969
1970 #ifdef CONFIG_IP_MULTICAST
1971 static void sf_markstate(struct ip_mc_list *pmc)
1972 {
1973         struct ip_sf_list *psf;
1974         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1975
1976         for (psf = pmc->sources; psf; psf = psf->sf_next)
1977                 if (pmc->sfcount[MCAST_EXCLUDE]) {
1978                         psf->sf_oldin = mca_xcount ==
1979                                 psf->sf_count[MCAST_EXCLUDE] &&
1980                                 !psf->sf_count[MCAST_INCLUDE];
1981                 } else
1982                         psf->sf_oldin = psf->sf_count[MCAST_INCLUDE] != 0;
1983 }
1984
1985 static int sf_setstate(struct ip_mc_list *pmc)
1986 {
1987         struct ip_sf_list *psf, *dpsf;
1988         int mca_xcount = pmc->sfcount[MCAST_EXCLUDE];
1989         int qrv = pmc->interface->mr_qrv;
1990         int new_in, rv;
1991
1992         rv = 0;
1993         for (psf = pmc->sources; psf; psf = psf->sf_next) {
1994                 if (pmc->sfcount[MCAST_EXCLUDE]) {
1995                         new_in = mca_xcount == psf->sf_count[MCAST_EXCLUDE] &&
1996                                 !psf->sf_count[MCAST_INCLUDE];
1997                 } else
1998                         new_in = psf->sf_count[MCAST_INCLUDE] != 0;
1999                 if (new_in) {
2000                         if (!psf->sf_oldin) {
2001                                 struct ip_sf_list *prev = NULL;
2002
2003                                 for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next) {
2004                                         if (dpsf->sf_inaddr == psf->sf_inaddr)
2005                                                 break;
2006                                         prev = dpsf;
2007                                 }
2008                                 if (dpsf) {
2009                                         if (prev)
2010                                                 prev->sf_next = dpsf->sf_next;
2011                                         else
2012                                                 pmc->tomb = dpsf->sf_next;
2013                                         kfree(dpsf);
2014                                 }
2015                                 psf->sf_crcount = qrv;
2016                                 rv++;
2017                         }
2018                 } else if (psf->sf_oldin) {
2019
2020                         psf->sf_crcount = 0;
2021                         /*
2022                          * add or update "delete" records if an active filter
2023                          * is now inactive
2024                          */
2025                         for (dpsf = pmc->tomb; dpsf; dpsf = dpsf->sf_next)
2026                                 if (dpsf->sf_inaddr == psf->sf_inaddr)
2027                                         break;
2028                         if (!dpsf) {
2029                                 dpsf = kmalloc(sizeof(*dpsf), GFP_ATOMIC);
2030                                 if (!dpsf)
2031                                         continue;
2032                                 *dpsf = *psf;
2033                                 /* pmc->lock held by callers */
2034                                 dpsf->sf_next = pmc->tomb;
2035                                 pmc->tomb = dpsf;
2036                         }
2037                         dpsf->sf_crcount = qrv;
2038                         rv++;
2039                 }
2040         }
2041         return rv;
2042 }
2043 #endif
2044
2045 /*
2046  * Add multicast source filter list to the interface list
2047  */
2048 static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
2049                          int sfcount, __be32 *psfsrc, int delta)
2050 {
2051         struct ip_mc_list *pmc;
2052         int     isexclude;
2053         int     i, err;
2054
2055         if (!in_dev)
2056                 return -ENODEV;
2057         rcu_read_lock();
2058         for_each_pmc_rcu(in_dev, pmc) {
2059                 if (*pmca == pmc->multiaddr)
2060                         break;
2061         }
2062         if (!pmc) {
2063                 /* MCA not found?? bug */
2064                 rcu_read_unlock();
2065                 return -ESRCH;
2066         }
2067         spin_lock_bh(&pmc->lock);
2068         rcu_read_unlock();
2069
2070 #ifdef CONFIG_IP_MULTICAST
2071         sf_markstate(pmc);
2072 #endif
2073         isexclude = pmc->sfmode == MCAST_EXCLUDE;
2074         if (!delta)
2075                 pmc->sfcount[sfmode]++;
2076         err = 0;
2077         for (i = 0; i < sfcount; i++) {
2078                 err = ip_mc_add1_src(pmc, sfmode, &psfsrc[i]);
2079                 if (err)
2080                         break;
2081         }
2082         if (err) {
2083                 int j;
2084
2085                 if (!delta)
2086                         pmc->sfcount[sfmode]--;
2087                 for (j = 0; j < i; j++)
2088                         (void) ip_mc_del1_src(pmc, sfmode, &psfsrc[j]);
2089         } else if (isexclude != (pmc->sfcount[MCAST_EXCLUDE] != 0)) {
2090 #ifdef CONFIG_IP_MULTICAST
2091                 struct ip_sf_list *psf;
2092                 struct net *net = dev_net(pmc->interface->dev);
2093                 in_dev = pmc->interface;
2094 #endif
2095
2096                 /* filter mode change */
2097                 if (pmc->sfcount[MCAST_EXCLUDE])
2098                         pmc->sfmode = MCAST_EXCLUDE;
2099                 else if (pmc->sfcount[MCAST_INCLUDE])
2100                         pmc->sfmode = MCAST_INCLUDE;
2101 #ifdef CONFIG_IP_MULTICAST
2102                 /* else no filters; keep old mode for reports */
2103
2104                 pmc->crcount = in_dev->mr_qrv ?: net->ipv4.sysctl_igmp_qrv;
2105                 in_dev->mr_ifc_count = pmc->crcount;
2106                 for (psf = pmc->sources; psf; psf = psf->sf_next)
2107                         psf->sf_crcount = 0;
2108                 igmp_ifc_event(in_dev);
2109         } else if (sf_setstate(pmc)) {
2110                 igmp_ifc_event(in_dev);
2111 #endif
2112         }
2113         spin_unlock_bh(&pmc->lock);
2114         return err;
2115 }
2116
2117 static void ip_mc_clear_src(struct ip_mc_list *pmc)
2118 {
2119         struct ip_sf_list *psf, *nextpsf, *tomb, *sources;
2120
2121         spin_lock_bh(&pmc->lock);
2122         tomb = pmc->tomb;
2123         pmc->tomb = NULL;
2124         sources = pmc->sources;
2125         pmc->sources = NULL;
2126         pmc->sfmode = MCAST_EXCLUDE;
2127         pmc->sfcount[MCAST_INCLUDE] = 0;
2128         pmc->sfcount[MCAST_EXCLUDE] = 1;
2129         spin_unlock_bh(&pmc->lock);
2130
2131         for (psf = tomb; psf; psf = nextpsf) {
2132                 nextpsf = psf->sf_next;
2133                 kfree(psf);
2134         }
2135         for (psf = sources; psf; psf = nextpsf) {
2136                 nextpsf = psf->sf_next;
2137                 kfree(psf);
2138         }
2139 }
2140
2141 /* Join a multicast group
2142  */
2143 static int __ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr,
2144                               unsigned int mode)
2145 {
2146         __be32 addr = imr->imr_multiaddr.s_addr;
2147         struct ip_mc_socklist *iml, *i;
2148         struct in_device *in_dev;
2149         struct inet_sock *inet = inet_sk(sk);
2150         struct net *net = sock_net(sk);
2151         int ifindex;
2152         int count = 0;
2153         int err;
2154
2155         ASSERT_RTNL();
2156
2157         if (!ipv4_is_multicast(addr))
2158                 return -EINVAL;
2159
2160         in_dev = ip_mc_find_dev(net, imr);
2161
2162         if (!in_dev) {
2163                 err = -ENODEV;
2164                 goto done;
2165         }
2166
2167         err = -EADDRINUSE;
2168         ifindex = imr->imr_ifindex;
2169         for_each_pmc_rtnl(inet, i) {
2170                 if (i->multi.imr_multiaddr.s_addr == addr &&
2171                     i->multi.imr_ifindex == ifindex)
2172                         goto done;
2173                 count++;
2174         }
2175         err = -ENOBUFS;
2176         if (count >= net->ipv4.sysctl_igmp_max_memberships)
2177                 goto done;
2178         iml = sock_kmalloc(sk, sizeof(*iml), GFP_KERNEL);
2179         if (!iml)
2180                 goto done;
2181
2182         memcpy(&iml->multi, imr, sizeof(*imr));
2183         iml->next_rcu = inet->mc_list;
2184         iml->sflist = NULL;
2185         iml->sfmode = mode;
2186         rcu_assign_pointer(inet->mc_list, iml);
2187         __ip_mc_inc_group(in_dev, addr, mode);
2188         err = 0;
2189 done:
2190         return err;
2191 }
2192
2193 /* Join ASM (Any-Source Multicast) group
2194  */
2195 int ip_mc_join_group(struct sock *sk, struct ip_mreqn *imr)
2196 {
2197         return __ip_mc_join_group(sk, imr, MCAST_EXCLUDE);
2198 }
2199 EXPORT_SYMBOL(ip_mc_join_group);
2200
2201 /* Join SSM (Source-Specific Multicast) group
2202  */
2203 int ip_mc_join_group_ssm(struct sock *sk, struct ip_mreqn *imr,
2204                          unsigned int mode)
2205 {
2206         return __ip_mc_join_group(sk, imr, mode);
2207 }
2208
2209 static int ip_mc_leave_src(struct sock *sk, struct ip_mc_socklist *iml,
2210                            struct in_device *in_dev)
2211 {
2212         struct ip_sf_socklist *psf = rtnl_dereference(iml->sflist);
2213         int err;
2214
2215         if (!psf) {
2216                 /* any-source empty exclude case */
2217                 return ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2218                         iml->sfmode, 0, NULL, 0);
2219         }
2220         err = ip_mc_del_src(in_dev, &iml->multi.imr_multiaddr.s_addr,
2221                         iml->sfmode, psf->sl_count, psf->sl_addr, 0);
2222         RCU_INIT_POINTER(iml->sflist, NULL);
2223         /* decrease mem now to avoid the memleak warning */
2224         atomic_sub(IP_SFLSIZE(psf->sl_max), &sk->sk_omem_alloc);
2225         kfree_rcu(psf, rcu);
2226         return err;
2227 }
2228
2229 int ip_mc_leave_group(struct sock *sk, struct ip_mreqn *imr)
2230 {
2231         struct inet_sock *inet = inet_sk(sk);
2232         struct ip_mc_socklist *iml;
2233         struct ip_mc_socklist __rcu **imlp;
2234         struct in_device *in_dev;
2235         struct net *net = sock_net(sk);
2236         __be32 group = imr->imr_multiaddr.s_addr;
2237         u32 ifindex;
2238         int ret = -EADDRNOTAVAIL;
2239
2240         ASSERT_RTNL();
2241
2242         in_dev = ip_mc_find_dev(net, imr);
2243         if (!imr->imr_ifindex && !imr->imr_address.s_addr && !in_dev) {
2244                 ret = -ENODEV;
2245                 goto out;
2246         }
2247         ifindex = imr->imr_ifindex;
2248         for (imlp = &inet->mc_list;
2249              (iml = rtnl_dereference(*imlp)) != NULL;
2250              imlp = &iml->next_rcu) {
2251                 if (iml->multi.imr_multiaddr.s_addr != group)
2252                         continue;
2253                 if (ifindex) {
2254                         if (iml->multi.imr_ifindex != ifindex)
2255                                 continue;
2256                 } else if (imr->imr_address.s_addr && imr->imr_address.s_addr !=
2257                                 iml->multi.imr_address.s_addr)
2258                         continue;
2259
2260                 (void) ip_mc_leave_src(sk, iml, in_dev);
2261
2262                 *imlp = iml->next_rcu;
2263
2264                 if (in_dev)
2265                         ip_mc_dec_group(in_dev, group);
2266
2267                 /* decrease mem now to avoid the memleak warning */
2268                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2269                 kfree_rcu(iml, rcu);
2270                 return 0;
2271         }
2272 out:
2273         return ret;
2274 }
2275 EXPORT_SYMBOL(ip_mc_leave_group);
2276
2277 int ip_mc_source(int add, int omode, struct sock *sk, struct
2278         ip_mreq_source *mreqs, int ifindex)
2279 {
2280         int err;
2281         struct ip_mreqn imr;
2282         __be32 addr = mreqs->imr_multiaddr;
2283         struct ip_mc_socklist *pmc;
2284         struct in_device *in_dev = NULL;
2285         struct inet_sock *inet = inet_sk(sk);
2286         struct ip_sf_socklist *psl;
2287         struct net *net = sock_net(sk);
2288         int leavegroup = 0;
2289         int i, j, rv;
2290
2291         if (!ipv4_is_multicast(addr))
2292                 return -EINVAL;
2293
2294         ASSERT_RTNL();
2295
2296         imr.imr_multiaddr.s_addr = mreqs->imr_multiaddr;
2297         imr.imr_address.s_addr = mreqs->imr_interface;
2298         imr.imr_ifindex = ifindex;
2299         in_dev = ip_mc_find_dev(net, &imr);
2300
2301         if (!in_dev) {
2302                 err = -ENODEV;
2303                 goto done;
2304         }
2305         err = -EADDRNOTAVAIL;
2306
2307         for_each_pmc_rtnl(inet, pmc) {
2308                 if ((pmc->multi.imr_multiaddr.s_addr ==
2309                      imr.imr_multiaddr.s_addr) &&
2310                     (pmc->multi.imr_ifindex == imr.imr_ifindex))
2311                         break;
2312         }
2313         if (!pmc) {             /* must have a prior join */
2314                 err = -EINVAL;
2315                 goto done;
2316         }
2317         /* if a source filter was set, must be the same mode as before */
2318         if (pmc->sflist) {
2319                 if (pmc->sfmode != omode) {
2320                         err = -EINVAL;
2321                         goto done;
2322                 }
2323         } else if (pmc->sfmode != omode) {
2324                 /* allow mode switches for empty-set filters */
2325                 ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 0, NULL, 0);
2326                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, pmc->sfmode, 0,
2327                         NULL, 0);
2328                 pmc->sfmode = omode;
2329         }
2330
2331         psl = rtnl_dereference(pmc->sflist);
2332         if (!add) {
2333                 if (!psl)
2334                         goto done;      /* err = -EADDRNOTAVAIL */
2335                 rv = !0;
2336                 for (i = 0; i < psl->sl_count; i++) {
2337                         rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2338                                 sizeof(__be32));
2339                         if (rv == 0)
2340                                 break;
2341                 }
2342                 if (rv)         /* source not found */
2343                         goto done;      /* err = -EADDRNOTAVAIL */
2344
2345                 /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2346                 if (psl->sl_count == 1 && omode == MCAST_INCLUDE) {
2347                         leavegroup = 1;
2348                         goto done;
2349                 }
2350
2351                 /* update the interface filter */
2352                 ip_mc_del_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2353                         &mreqs->imr_sourceaddr, 1);
2354
2355                 for (j = i+1; j < psl->sl_count; j++)
2356                         psl->sl_addr[j-1] = psl->sl_addr[j];
2357                 psl->sl_count--;
2358                 err = 0;
2359                 goto done;
2360         }
2361         /* else, add a new source to the filter */
2362
2363         if (psl && psl->sl_count >= net->ipv4.sysctl_igmp_max_msf) {
2364                 err = -ENOBUFS;
2365                 goto done;
2366         }
2367         if (!psl || psl->sl_count == psl->sl_max) {
2368                 struct ip_sf_socklist *newpsl;
2369                 int count = IP_SFBLOCK;
2370
2371                 if (psl)
2372                         count += psl->sl_max;
2373                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(count), GFP_KERNEL);
2374                 if (!newpsl) {
2375                         err = -ENOBUFS;
2376                         goto done;
2377                 }
2378                 newpsl->sl_max = count;
2379                 newpsl->sl_count = count - IP_SFBLOCK;
2380                 if (psl) {
2381                         for (i = 0; i < psl->sl_count; i++)
2382                                 newpsl->sl_addr[i] = psl->sl_addr[i];
2383                         /* decrease mem now to avoid the memleak warning */
2384                         atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2385                         kfree_rcu(psl, rcu);
2386                 }
2387                 rcu_assign_pointer(pmc->sflist, newpsl);
2388                 psl = newpsl;
2389         }
2390         rv = 1; /* > 0 for insert logic below if sl_count is 0 */
2391         for (i = 0; i < psl->sl_count; i++) {
2392                 rv = memcmp(&psl->sl_addr[i], &mreqs->imr_sourceaddr,
2393                         sizeof(__be32));
2394                 if (rv == 0)
2395                         break;
2396         }
2397         if (rv == 0)            /* address already there is an error */
2398                 goto done;
2399         for (j = psl->sl_count-1; j >= i; j--)
2400                 psl->sl_addr[j+1] = psl->sl_addr[j];
2401         psl->sl_addr[i] = mreqs->imr_sourceaddr;
2402         psl->sl_count++;
2403         err = 0;
2404         /* update the interface list */
2405         ip_mc_add_src(in_dev, &mreqs->imr_multiaddr, omode, 1,
2406                 &mreqs->imr_sourceaddr, 1);
2407 done:
2408         if (leavegroup)
2409                 err = ip_mc_leave_group(sk, &imr);
2410         return err;
2411 }
2412
2413 int ip_mc_msfilter(struct sock *sk, struct ip_msfilter *msf, int ifindex)
2414 {
2415         int err = 0;
2416         struct ip_mreqn imr;
2417         __be32 addr = msf->imsf_multiaddr;
2418         struct ip_mc_socklist *pmc;
2419         struct in_device *in_dev;
2420         struct inet_sock *inet = inet_sk(sk);
2421         struct ip_sf_socklist *newpsl, *psl;
2422         struct net *net = sock_net(sk);
2423         int leavegroup = 0;
2424
2425         if (!ipv4_is_multicast(addr))
2426                 return -EINVAL;
2427         if (msf->imsf_fmode != MCAST_INCLUDE &&
2428             msf->imsf_fmode != MCAST_EXCLUDE)
2429                 return -EINVAL;
2430
2431         ASSERT_RTNL();
2432
2433         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2434         imr.imr_address.s_addr = msf->imsf_interface;
2435         imr.imr_ifindex = ifindex;
2436         in_dev = ip_mc_find_dev(net, &imr);
2437
2438         if (!in_dev) {
2439                 err = -ENODEV;
2440                 goto done;
2441         }
2442
2443         /* special case - (INCLUDE, empty) == LEAVE_GROUP */
2444         if (msf->imsf_fmode == MCAST_INCLUDE && msf->imsf_numsrc == 0) {
2445                 leavegroup = 1;
2446                 goto done;
2447         }
2448
2449         for_each_pmc_rtnl(inet, pmc) {
2450                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2451                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2452                         break;
2453         }
2454         if (!pmc) {             /* must have a prior join */
2455                 err = -EINVAL;
2456                 goto done;
2457         }
2458         if (msf->imsf_numsrc) {
2459                 newpsl = sock_kmalloc(sk, IP_SFLSIZE(msf->imsf_numsrc),
2460                                                            GFP_KERNEL);
2461                 if (!newpsl) {
2462                         err = -ENOBUFS;
2463                         goto done;
2464                 }
2465                 newpsl->sl_max = newpsl->sl_count = msf->imsf_numsrc;
2466                 memcpy(newpsl->sl_addr, msf->imsf_slist,
2467                         msf->imsf_numsrc * sizeof(msf->imsf_slist[0]));
2468                 err = ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2469                         msf->imsf_fmode, newpsl->sl_count, newpsl->sl_addr, 0);
2470                 if (err) {
2471                         sock_kfree_s(sk, newpsl, IP_SFLSIZE(newpsl->sl_max));
2472                         goto done;
2473                 }
2474         } else {
2475                 newpsl = NULL;
2476                 (void) ip_mc_add_src(in_dev, &msf->imsf_multiaddr,
2477                                      msf->imsf_fmode, 0, NULL, 0);
2478         }
2479         psl = rtnl_dereference(pmc->sflist);
2480         if (psl) {
2481                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2482                         psl->sl_count, psl->sl_addr, 0);
2483                 /* decrease mem now to avoid the memleak warning */
2484                 atomic_sub(IP_SFLSIZE(psl->sl_max), &sk->sk_omem_alloc);
2485                 kfree_rcu(psl, rcu);
2486         } else
2487                 (void) ip_mc_del_src(in_dev, &msf->imsf_multiaddr, pmc->sfmode,
2488                         0, NULL, 0);
2489         rcu_assign_pointer(pmc->sflist, newpsl);
2490         pmc->sfmode = msf->imsf_fmode;
2491         err = 0;
2492 done:
2493         if (leavegroup)
2494                 err = ip_mc_leave_group(sk, &imr);
2495         return err;
2496 }
2497
2498 int ip_mc_msfget(struct sock *sk, struct ip_msfilter *msf,
2499         struct ip_msfilter __user *optval, int __user *optlen)
2500 {
2501         int err, len, count, copycount;
2502         struct ip_mreqn imr;
2503         __be32 addr = msf->imsf_multiaddr;
2504         struct ip_mc_socklist *pmc;
2505         struct in_device *in_dev;
2506         struct inet_sock *inet = inet_sk(sk);
2507         struct ip_sf_socklist *psl;
2508         struct net *net = sock_net(sk);
2509
2510         ASSERT_RTNL();
2511
2512         if (!ipv4_is_multicast(addr))
2513                 return -EINVAL;
2514
2515         imr.imr_multiaddr.s_addr = msf->imsf_multiaddr;
2516         imr.imr_address.s_addr = msf->imsf_interface;
2517         imr.imr_ifindex = 0;
2518         in_dev = ip_mc_find_dev(net, &imr);
2519
2520         if (!in_dev) {
2521                 err = -ENODEV;
2522                 goto done;
2523         }
2524         err = -EADDRNOTAVAIL;
2525
2526         for_each_pmc_rtnl(inet, pmc) {
2527                 if (pmc->multi.imr_multiaddr.s_addr == msf->imsf_multiaddr &&
2528                     pmc->multi.imr_ifindex == imr.imr_ifindex)
2529                         break;
2530         }
2531         if (!pmc)               /* must have a prior join */
2532                 goto done;
2533         msf->imsf_fmode = pmc->sfmode;
2534         psl = rtnl_dereference(pmc->sflist);
2535         if (!psl) {
2536                 len = 0;
2537                 count = 0;
2538         } else {
2539                 count = psl->sl_count;
2540         }
2541         copycount = count < msf->imsf_numsrc ? count : msf->imsf_numsrc;
2542         len = copycount * sizeof(psl->sl_addr[0]);
2543         msf->imsf_numsrc = count;
2544         if (put_user(IP_MSFILTER_SIZE(copycount), optlen) ||
2545             copy_to_user(optval, msf, IP_MSFILTER_SIZE(0))) {
2546                 return -EFAULT;
2547         }
2548         if (len &&
2549             copy_to_user(&optval->imsf_slist[0], psl->sl_addr, len))
2550                 return -EFAULT;
2551         return 0;
2552 done:
2553         return err;
2554 }
2555
2556 int ip_mc_gsfget(struct sock *sk, struct group_filter *gsf,
2557         struct group_filter __user *optval, int __user *optlen)
2558 {
2559         int err, i, count, copycount;
2560         struct sockaddr_in *psin;
2561         __be32 addr;
2562         struct ip_mc_socklist *pmc;
2563         struct inet_sock *inet = inet_sk(sk);
2564         struct ip_sf_socklist *psl;
2565
2566         ASSERT_RTNL();
2567
2568         psin = (struct sockaddr_in *)&gsf->gf_group;
2569         if (psin->sin_family != AF_INET)
2570                 return -EINVAL;
2571         addr = psin->sin_addr.s_addr;
2572         if (!ipv4_is_multicast(addr))
2573                 return -EINVAL;
2574
2575         err = -EADDRNOTAVAIL;
2576
2577         for_each_pmc_rtnl(inet, pmc) {
2578                 if (pmc->multi.imr_multiaddr.s_addr == addr &&
2579                     pmc->multi.imr_ifindex == gsf->gf_interface)
2580                         break;
2581         }
2582         if (!pmc)               /* must have a prior join */
2583                 goto done;
2584         gsf->gf_fmode = pmc->sfmode;
2585         psl = rtnl_dereference(pmc->sflist);
2586         count = psl ? psl->sl_count : 0;
2587         copycount = count < gsf->gf_numsrc ? count : gsf->gf_numsrc;
2588         gsf->gf_numsrc = count;
2589         if (put_user(GROUP_FILTER_SIZE(copycount), optlen) ||
2590             copy_to_user(optval, gsf, GROUP_FILTER_SIZE(0))) {
2591                 return -EFAULT;
2592         }
2593         for (i = 0; i < copycount; i++) {
2594                 struct sockaddr_storage ss;
2595
2596                 psin = (struct sockaddr_in *)&ss;
2597                 memset(&ss, 0, sizeof(ss));
2598                 psin->sin_family = AF_INET;
2599                 psin->sin_addr.s_addr = psl->sl_addr[i];
2600                 if (copy_to_user(&optval->gf_slist[i], &ss, sizeof(ss)))
2601                         return -EFAULT;
2602         }
2603         return 0;
2604 done:
2605         return err;
2606 }
2607
2608 /*
2609  * check if a multicast source filter allows delivery for a given <src,dst,intf>
2610  */
2611 int ip_mc_sf_allow(struct sock *sk, __be32 loc_addr, __be32 rmt_addr,
2612                    int dif, int sdif)
2613 {
2614         struct inet_sock *inet = inet_sk(sk);
2615         struct ip_mc_socklist *pmc;
2616         struct ip_sf_socklist *psl;
2617         int i;
2618         int ret;
2619
2620         ret = 1;
2621         if (!ipv4_is_multicast(loc_addr))
2622                 goto out;
2623
2624         rcu_read_lock();
2625         for_each_pmc_rcu(inet, pmc) {
2626                 if (pmc->multi.imr_multiaddr.s_addr == loc_addr &&
2627                     (pmc->multi.imr_ifindex == dif ||
2628                      (sdif && pmc->multi.imr_ifindex == sdif)))
2629                         break;
2630         }
2631         ret = inet->mc_all;
2632         if (!pmc)
2633                 goto unlock;
2634         psl = rcu_dereference(pmc->sflist);
2635         ret = (pmc->sfmode == MCAST_EXCLUDE);
2636         if (!psl)
2637                 goto unlock;
2638
2639         for (i = 0; i < psl->sl_count; i++) {
2640                 if (psl->sl_addr[i] == rmt_addr)
2641                         break;
2642         }
2643         ret = 0;
2644         if (pmc->sfmode == MCAST_INCLUDE && i >= psl->sl_count)
2645                 goto unlock;
2646         if (pmc->sfmode == MCAST_EXCLUDE && i < psl->sl_count)
2647                 goto unlock;
2648         ret = 1;
2649 unlock:
2650         rcu_read_unlock();
2651 out:
2652         return ret;
2653 }
2654
2655 /*
2656  *      A socket is closing.
2657  */
2658
2659 void ip_mc_drop_socket(struct sock *sk)
2660 {
2661         struct inet_sock *inet = inet_sk(sk);
2662         struct ip_mc_socklist *iml;
2663         struct net *net = sock_net(sk);
2664
2665         if (!inet->mc_list)
2666                 return;
2667
2668         rtnl_lock();
2669         while ((iml = rtnl_dereference(inet->mc_list)) != NULL) {
2670                 struct in_device *in_dev;
2671
2672                 inet->mc_list = iml->next_rcu;
2673                 in_dev = inetdev_by_index(net, iml->multi.imr_ifindex);
2674                 (void) ip_mc_leave_src(sk, iml, in_dev);
2675                 if (in_dev)
2676                         ip_mc_dec_group(in_dev, iml->multi.imr_multiaddr.s_addr);
2677                 /* decrease mem now to avoid the memleak warning */
2678                 atomic_sub(sizeof(*iml), &sk->sk_omem_alloc);
2679                 kfree_rcu(iml, rcu);
2680         }
2681         rtnl_unlock();
2682 }
2683
2684 /* called with rcu_read_lock() */
2685 int ip_check_mc_rcu(struct in_device *in_dev, __be32 mc_addr, __be32 src_addr, u8 proto)
2686 {
2687         struct ip_mc_list *im;
2688         struct ip_mc_list __rcu **mc_hash;
2689         struct ip_sf_list *psf;
2690         int rv = 0;
2691
2692         mc_hash = rcu_dereference(in_dev->mc_hash);
2693         if (mc_hash) {
2694                 u32 hash = hash_32((__force u32)mc_addr, MC_HASH_SZ_LOG);
2695
2696                 for (im = rcu_dereference(mc_hash[hash]);
2697                      im != NULL;
2698                      im = rcu_dereference(im->next_hash)) {
2699                         if (im->multiaddr == mc_addr)
2700                                 break;
2701                 }
2702         } else {
2703                 for_each_pmc_rcu(in_dev, im) {
2704                         if (im->multiaddr == mc_addr)
2705                                 break;
2706                 }
2707         }
2708         if (im && proto == IPPROTO_IGMP) {
2709                 rv = 1;
2710         } else if (im) {
2711                 if (src_addr) {
2712                         for (psf = im->sources; psf; psf = psf->sf_next) {
2713                                 if (psf->sf_inaddr == src_addr)
2714                                         break;
2715                         }
2716                         if (psf)
2717                                 rv = psf->sf_count[MCAST_INCLUDE] ||
2718                                         psf->sf_count[MCAST_EXCLUDE] !=
2719                                         im->sfcount[MCAST_EXCLUDE];
2720                         else
2721                                 rv = im->sfcount[MCAST_EXCLUDE] != 0;
2722                 } else
2723                         rv = 1; /* unspecified source; tentatively allow */
2724         }
2725         return rv;
2726 }
2727
2728 #if defined(CONFIG_PROC_FS)
2729 struct igmp_mc_iter_state {
2730         struct seq_net_private p;
2731         struct net_device *dev;
2732         struct in_device *in_dev;
2733 };
2734
2735 #define igmp_mc_seq_private(seq)        ((struct igmp_mc_iter_state *)(seq)->private)
2736
2737 static inline struct ip_mc_list *igmp_mc_get_first(struct seq_file *seq)
2738 {
2739         struct net *net = seq_file_net(seq);
2740         struct ip_mc_list *im = NULL;
2741         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2742
2743         state->in_dev = NULL;
2744         for_each_netdev_rcu(net, state->dev) {
2745                 struct in_device *in_dev;
2746
2747                 in_dev = __in_dev_get_rcu(state->dev);
2748                 if (!in_dev)
2749                         continue;
2750                 im = rcu_dereference(in_dev->mc_list);
2751                 if (im) {
2752                         state->in_dev = in_dev;
2753                         break;
2754                 }
2755         }
2756         return im;
2757 }
2758
2759 static struct ip_mc_list *igmp_mc_get_next(struct seq_file *seq, struct ip_mc_list *im)
2760 {
2761         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2762
2763         im = rcu_dereference(im->next_rcu);
2764         while (!im) {
2765                 state->dev = next_net_device_rcu(state->dev);
2766                 if (!state->dev) {
2767                         state->in_dev = NULL;
2768                         break;
2769                 }
2770                 state->in_dev = __in_dev_get_rcu(state->dev);
2771                 if (!state->in_dev)
2772                         continue;
2773                 im = rcu_dereference(state->in_dev->mc_list);
2774         }
2775         return im;
2776 }
2777
2778 static struct ip_mc_list *igmp_mc_get_idx(struct seq_file *seq, loff_t pos)
2779 {
2780         struct ip_mc_list *im = igmp_mc_get_first(seq);
2781         if (im)
2782                 while (pos && (im = igmp_mc_get_next(seq, im)) != NULL)
2783                         --pos;
2784         return pos ? NULL : im;
2785 }
2786
2787 static void *igmp_mc_seq_start(struct seq_file *seq, loff_t *pos)
2788         __acquires(rcu)
2789 {
2790         rcu_read_lock();
2791         return *pos ? igmp_mc_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2792 }
2793
2794 static void *igmp_mc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2795 {
2796         struct ip_mc_list *im;
2797         if (v == SEQ_START_TOKEN)
2798                 im = igmp_mc_get_first(seq);
2799         else
2800                 im = igmp_mc_get_next(seq, v);
2801         ++*pos;
2802         return im;
2803 }
2804
2805 static void igmp_mc_seq_stop(struct seq_file *seq, void *v)
2806         __releases(rcu)
2807 {
2808         struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2809
2810         state->in_dev = NULL;
2811         state->dev = NULL;
2812         rcu_read_unlock();
2813 }
2814
2815 static int igmp_mc_seq_show(struct seq_file *seq, void *v)
2816 {
2817         if (v == SEQ_START_TOKEN)
2818                 seq_puts(seq,
2819                          "Idx\tDevice    : Count Querier\tGroup    Users Timer\tReporter\n");
2820         else {
2821                 struct ip_mc_list *im = (struct ip_mc_list *)v;
2822                 struct igmp_mc_iter_state *state = igmp_mc_seq_private(seq);
2823                 char   *querier;
2824                 long delta;
2825
2826 #ifdef CONFIG_IP_MULTICAST
2827                 querier = IGMP_V1_SEEN(state->in_dev) ? "V1" :
2828                           IGMP_V2_SEEN(state->in_dev) ? "V2" :
2829                           "V3";
2830 #else
2831                 querier = "NONE";
2832 #endif
2833
2834                 if (rcu_access_pointer(state->in_dev->mc_list) == im) {
2835                         seq_printf(seq, "%d\t%-10s: %5d %7s\n",
2836                                    state->dev->ifindex, state->dev->name, state->in_dev->mc_count, querier);
2837                 }
2838
2839                 delta = im->timer.expires - jiffies;
2840                 seq_printf(seq,
2841                            "\t\t\t\t%08X %5d %d:%08lX\t\t%d\n",
2842                            im->multiaddr, im->users,
2843                            im->tm_running,
2844                            im->tm_running ? jiffies_delta_to_clock_t(delta) : 0,
2845                            im->reporter);
2846         }
2847         return 0;
2848 }
2849
2850 static const struct seq_operations igmp_mc_seq_ops = {
2851         .start  =       igmp_mc_seq_start,
2852         .next   =       igmp_mc_seq_next,
2853         .stop   =       igmp_mc_seq_stop,
2854         .show   =       igmp_mc_seq_show,
2855 };
2856
2857 struct igmp_mcf_iter_state {
2858         struct seq_net_private p;
2859         struct net_device *dev;
2860         struct in_device *idev;
2861         struct ip_mc_list *im;
2862 };
2863
2864 #define igmp_mcf_seq_private(seq)       ((struct igmp_mcf_iter_state *)(seq)->private)
2865
2866 static inline struct ip_sf_list *igmp_mcf_get_first(struct seq_file *seq)
2867 {
2868         struct net *net = seq_file_net(seq);
2869         struct ip_sf_list *psf = NULL;
2870         struct ip_mc_list *im = NULL;
2871         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2872
2873         state->idev = NULL;
2874         state->im = NULL;
2875         for_each_netdev_rcu(net, state->dev) {
2876                 struct in_device *idev;
2877                 idev = __in_dev_get_rcu(state->dev);
2878                 if (unlikely(!idev))
2879                         continue;
2880                 im = rcu_dereference(idev->mc_list);
2881                 if (likely(im)) {
2882                         spin_lock_bh(&im->lock);
2883                         psf = im->sources;
2884                         if (likely(psf)) {
2885                                 state->im = im;
2886                                 state->idev = idev;
2887                                 break;
2888                         }
2889                         spin_unlock_bh(&im->lock);
2890                 }
2891         }
2892         return psf;
2893 }
2894
2895 static struct ip_sf_list *igmp_mcf_get_next(struct seq_file *seq, struct ip_sf_list *psf)
2896 {
2897         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2898
2899         psf = psf->sf_next;
2900         while (!psf) {
2901                 spin_unlock_bh(&state->im->lock);
2902                 state->im = state->im->next;
2903                 while (!state->im) {
2904                         state->dev = next_net_device_rcu(state->dev);
2905                         if (!state->dev) {
2906                                 state->idev = NULL;
2907                                 goto out;
2908                         }
2909                         state->idev = __in_dev_get_rcu(state->dev);
2910                         if (!state->idev)
2911                                 continue;
2912                         state->im = rcu_dereference(state->idev->mc_list);
2913                 }
2914                 if (!state->im)
2915                         break;
2916                 spin_lock_bh(&state->im->lock);
2917                 psf = state->im->sources;
2918         }
2919 out:
2920         return psf;
2921 }
2922
2923 static struct ip_sf_list *igmp_mcf_get_idx(struct seq_file *seq, loff_t pos)
2924 {
2925         struct ip_sf_list *psf = igmp_mcf_get_first(seq);
2926         if (psf)
2927                 while (pos && (psf = igmp_mcf_get_next(seq, psf)) != NULL)
2928                         --pos;
2929         return pos ? NULL : psf;
2930 }
2931
2932 static void *igmp_mcf_seq_start(struct seq_file *seq, loff_t *pos)
2933         __acquires(rcu)
2934 {
2935         rcu_read_lock();
2936         return *pos ? igmp_mcf_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2937 }
2938
2939 static void *igmp_mcf_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2940 {
2941         struct ip_sf_list *psf;
2942         if (v == SEQ_START_TOKEN)
2943                 psf = igmp_mcf_get_first(seq);
2944         else
2945                 psf = igmp_mcf_get_next(seq, v);
2946         ++*pos;
2947         return psf;
2948 }
2949
2950 static void igmp_mcf_seq_stop(struct seq_file *seq, void *v)
2951         __releases(rcu)
2952 {
2953         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2954         if (likely(state->im)) {
2955                 spin_unlock_bh(&state->im->lock);
2956                 state->im = NULL;
2957         }
2958         state->idev = NULL;
2959         state->dev = NULL;
2960         rcu_read_unlock();
2961 }
2962
2963 static int igmp_mcf_seq_show(struct seq_file *seq, void *v)
2964 {
2965         struct ip_sf_list *psf = (struct ip_sf_list *)v;
2966         struct igmp_mcf_iter_state *state = igmp_mcf_seq_private(seq);
2967
2968         if (v == SEQ_START_TOKEN) {
2969                 seq_puts(seq, "Idx Device        MCA        SRC    INC    EXC\n");
2970         } else {
2971                 seq_printf(seq,
2972                            "%3d %6.6s 0x%08x "
2973                            "0x%08x %6lu %6lu\n",
2974                            state->dev->ifindex, state->dev->name,
2975                            ntohl(state->im->multiaddr),
2976                            ntohl(psf->sf_inaddr),
2977                            psf->sf_count[MCAST_INCLUDE],
2978                            psf->sf_count[MCAST_EXCLUDE]);
2979         }
2980         return 0;
2981 }
2982
2983 static const struct seq_operations igmp_mcf_seq_ops = {
2984         .start  =       igmp_mcf_seq_start,
2985         .next   =       igmp_mcf_seq_next,
2986         .stop   =       igmp_mcf_seq_stop,
2987         .show   =       igmp_mcf_seq_show,
2988 };
2989
2990 static int __net_init igmp_net_init(struct net *net)
2991 {
2992         struct proc_dir_entry *pde;
2993         int err;
2994
2995         pde = proc_create_net("igmp", 0444, net->proc_net, &igmp_mc_seq_ops,
2996                         sizeof(struct igmp_mc_iter_state));
2997         if (!pde)
2998                 goto out_igmp;
2999         pde = proc_create_net("mcfilter", 0444, net->proc_net,
3000                         &igmp_mcf_seq_ops, sizeof(struct igmp_mcf_iter_state));
3001         if (!pde)
3002                 goto out_mcfilter;
3003         err = inet_ctl_sock_create(&net->ipv4.mc_autojoin_sk, AF_INET,
3004                                    SOCK_DGRAM, 0, net);
3005         if (err < 0) {
3006                 pr_err("Failed to initialize the IGMP autojoin socket (err %d)\n",
3007                        err);
3008                 goto out_sock;
3009         }
3010
3011         return 0;
3012
3013 out_sock:
3014         remove_proc_entry("mcfilter", net->proc_net);
3015 out_mcfilter:
3016         remove_proc_entry("igmp", net->proc_net);
3017 out_igmp:
3018         return -ENOMEM;
3019 }
3020
3021 static void __net_exit igmp_net_exit(struct net *net)
3022 {
3023         remove_proc_entry("mcfilter", net->proc_net);
3024         remove_proc_entry("igmp", net->proc_net);
3025         inet_ctl_sock_destroy(net->ipv4.mc_autojoin_sk);
3026 }
3027
3028 static struct pernet_operations igmp_net_ops = {
3029         .init = igmp_net_init,
3030         .exit = igmp_net_exit,
3031 };
3032 #endif
3033
3034 static int igmp_netdev_event(struct notifier_block *this,
3035                              unsigned long event, void *ptr)
3036 {
3037         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3038         struct in_device *in_dev;
3039
3040         switch (event) {
3041         case NETDEV_RESEND_IGMP:
3042                 in_dev = __in_dev_get_rtnl(dev);
3043                 if (in_dev)
3044                         ip_mc_rejoin_groups(in_dev);
3045                 break;
3046         default:
3047                 break;
3048         }
3049         return NOTIFY_DONE;
3050 }
3051
3052 static struct notifier_block igmp_notifier = {
3053         .notifier_call = igmp_netdev_event,
3054 };
3055
3056 int __init igmp_mc_init(void)
3057 {
3058 #if defined(CONFIG_PROC_FS)
3059         int err;
3060
3061         err = register_pernet_subsys(&igmp_net_ops);
3062         if (err)
3063                 return err;
3064         err = register_netdevice_notifier(&igmp_notifier);
3065         if (err)
3066                 goto reg_notif_fail;
3067         return 0;
3068
3069 reg_notif_fail:
3070         unregister_pernet_subsys(&igmp_net_ops);
3071         return err;
3072 #else
3073         return register_netdevice_notifier(&igmp_notifier);
3074 #endif
3075 }