net/ipv4/igmp.c

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