net/ipv4/igmp.c

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