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