net/ipv4/ipmr.c

   1 /*
   2  *      IP multicast routing support for mrouted 3.6/3.8
   3  *
   4  *              (c) 1995 Alan Cox, <alan@lxorguk.ukuu.org.uk>
   5  *        Linux Consultancy and Custom Driver Development
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  *
  12  *      Fixes:
  13  *      Michael Chastain        :       Incorrect size of copying.
  14  *      Alan Cox                :       Added the cache manager code
  15  *      Alan Cox                :       Fixed the clone/copy bug and device race.
  16  *      Mike McLagan            :       Routing by source
  17  *      Malcolm Beattie         :       Buffer handling fixes.
  18  *      Alexey Kuznetsov        :       Double buffer free and other fixes.
  19  *      SVR Anand               :       Fixed several multicast bugs and problems.
  20  *      Alexey Kuznetsov        :       Status, optimisations and more.
  21  *      Brad Parker             :       Better behaviour on mrouted upcall
  22  *                                      overflow.
  23  *      Carlos Picoto           :       PIMv1 Support
  24  *      Pavlin Ivanov Radoslavov:       PIMv2 Registers must checksum only PIM header
  25  *                                      Relax this requirement to work with older peers.
  26  *
  27  */
  28
  29 #include <asm/uaccess.h>
  30 #include <linux/types.h>
  31 #include <linux/capability.h>
  32 #include <linux/errno.h>
  33 #include <linux/timer.h>
  34 #include <linux/mm.h>
  35 #include <linux/kernel.h>
  36 #include <linux/fcntl.h>
  37 #include <linux/stat.h>
  38 #include <linux/socket.h>
  39 #include <linux/in.h>
  40 #include <linux/inet.h>
  41 #include <linux/netdevice.h>
  42 #include <linux/inetdevice.h>
  43 #include <linux/igmp.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/seq_file.h>
  46 #include <linux/mroute.h>
  47 #include <linux/init.h>
  48 #include <linux/if_ether.h>
  49 #include <linux/slab.h>
  50 #include <net/net_namespace.h>
  51 #include <net/ip.h>
  52 #include <net/protocol.h>
  53 #include <linux/skbuff.h>
  54 #include <net/route.h>
  55 #include <net/sock.h>
  56 #include <net/icmp.h>
  57 #include <net/udp.h>
  58 #include <net/raw.h>
  59 #include <linux/notifier.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/netfilter_ipv4.h>
  62 #include <linux/compat.h>
  63 #include <linux/export.h>
  64 #include <net/ip_tunnels.h>
  65 #include <net/checksum.h>
  66 #include <net/netlink.h>
  67 #include <net/fib_rules.h>
  68 #include <linux/netconf.h>
  69 #include <linux/nospec.h>
  70
  71 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
  72 #define CONFIG_IP_PIMSM 1
  73 #endif
  74
  75 struct mr_table {
  76         struct list_head        list;
  77         possible_net_t          net;
  78         u32                     id;
  79         struct sock __rcu       *mroute_sk;
  80         struct timer_list       ipmr_expire_timer;
  81         struct list_head        mfc_unres_queue;
  82         struct list_head        mfc_cache_array[MFC_LINES];
  83         struct vif_device       vif_table[MAXVIFS];
  84         int                     maxvif;
  85         atomic_t                cache_resolve_queue_len;
  86         bool                    mroute_do_assert;
  87         bool                    mroute_do_pim;
  88 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
  89         int                     mroute_reg_vif_num;
  90 #endif
  91 };
  92
  93 struct ipmr_rule {
  94         struct fib_rule         common;
  95 };
  96
  97 struct ipmr_result {
  98         struct mr_table         *mrt;
  99 };
 100
 101 /* Big lock, protecting vif table, mrt cache and mroute socket state.
 102  * Note that the changes are semaphored via rtnl_lock.
 103  */
 104
 105 static DEFINE_RWLOCK(mrt_lock);
 106
 107 /*
 108  *      Multicast router control variables
 109  */
 110
 111 #define VIF_EXISTS(_mrt, _idx) ((_mrt)->vif_table[_idx].dev != NULL)
 112
 113 /* Special spinlock for queue of unresolved entries */
 114 static DEFINE_SPINLOCK(mfc_unres_lock);
 115
 116 /* We return to original Alan's scheme. Hash table of resolved
 117  * entries is changed only in process context and protected
 118  * with weak lock mrt_lock. Queue of unresolved entries is protected
 119  * with strong spinlock mfc_unres_lock.
 120  *
 121  * In this case data path is free of exclusive locks at all.
 122  */
 123
 124 static struct kmem_cache *mrt_cachep __read_mostly;
 125
 126 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 127 static void ipmr_free_table(struct mr_table *mrt);
 128
 129 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
 130                           struct sk_buff *skb, struct mfc_cache *cache,
 131                           int local);
 132 static int ipmr_cache_report(struct mr_table *mrt,
 133                              struct sk_buff *pkt, vifi_t vifi, int assert);
 134 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
 135                               struct mfc_cache *c, struct rtmsg *rtm);
 136 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
 137                                  int cmd);
 138 static void mroute_clean_tables(struct mr_table *mrt, bool all);
 139 static void ipmr_expire_process(unsigned long arg);
 140
 141 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 142 #define ipmr_for_each_table(mrt, net) \
 143         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
 144
 145 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 146 {
 147         struct mr_table *mrt;
 148
 149         ipmr_for_each_table(mrt, net) {
 150                 if (mrt->id == id)
 151                         return mrt;
 152         }
 153         return NULL;
 154 }
 155
 156 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 157                            struct mr_table **mrt)
 158 {
 159         int err;
 160         struct ipmr_result res;
 161         struct fib_lookup_arg arg = {
 162                 .result = &res,
 163                 .flags = FIB_LOOKUP_NOREF,
 164         };
 165
 166         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 167                                flowi4_to_flowi(flp4), 0, &arg);
 168         if (err < 0)
 169                 return err;
 170         *mrt = res.mrt;
 171         return 0;
 172 }
 173
 174 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 175                             int flags, struct fib_lookup_arg *arg)
 176 {
 177         struct ipmr_result *res = arg->result;
 178         struct mr_table *mrt;
 179
 180         switch (rule->action) {
 181         case FR_ACT_TO_TBL:
 182                 break;
 183         case FR_ACT_UNREACHABLE:
 184                 return -ENETUNREACH;
 185         case FR_ACT_PROHIBIT:
 186                 return -EACCES;
 187         case FR_ACT_BLACKHOLE:
 188         default:
 189                 return -EINVAL;
 190         }
 191
 192         mrt = ipmr_get_table(rule->fr_net, rule->table);
 193         if (!mrt)
 194                 return -EAGAIN;
 195         res->mrt = mrt;
 196         return 0;
 197 }
 198
 199 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 200 {
 201         return 1;
 202 }
 203
 204 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 205         FRA_GENERIC_POLICY,
 206 };
 207
 208 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 209                                struct fib_rule_hdr *frh, struct nlattr **tb)
 210 {
 211         return 0;
 212 }
 213
 214 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 215                              struct nlattr **tb)
 216 {
 217         return 1;
 218 }
 219
 220 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 221                           struct fib_rule_hdr *frh)
 222 {
 223         frh->dst_len = 0;
 224         frh->src_len = 0;
 225         frh->tos     = 0;
 226         return 0;
 227 }
 228
 229 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
 230         .family         = RTNL_FAMILY_IPMR,
 231         .rule_size      = sizeof(struct ipmr_rule),
 232         .addr_size      = sizeof(u32),
 233         .action         = ipmr_rule_action,
 234         .match          = ipmr_rule_match,
 235         .configure      = ipmr_rule_configure,
 236         .compare        = ipmr_rule_compare,
 237         .fill           = ipmr_rule_fill,
 238         .nlgroup        = RTNLGRP_IPV4_RULE,
 239         .policy         = ipmr_rule_policy,
 240         .owner          = THIS_MODULE,
 241 };
 242
 243 static int __net_init ipmr_rules_init(struct net *net)
 244 {
 245         struct fib_rules_ops *ops;
 246         struct mr_table *mrt;
 247         int err;
 248
 249         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 250         if (IS_ERR(ops))
 251                 return PTR_ERR(ops);
 252
 253         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 254
 255         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 256         if (!mrt) {
 257                 err = -ENOMEM;
 258                 goto err1;
 259         }
 260
 261         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 262         if (err < 0)
 263                 goto err2;
 264
 265         net->ipv4.mr_rules_ops = ops;
 266         return 0;
 267
 268 err2:
 269         ipmr_free_table(mrt);
 270 err1:
 271         fib_rules_unregister(ops);
 272         return err;
 273 }
 274
 275 static void __net_exit ipmr_rules_exit(struct net *net)
 276 {
 277         struct mr_table *mrt, *next;
 278
 279         rtnl_lock();
 280         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 281                 list_del(&mrt->list);
 282                 ipmr_free_table(mrt);
 283         }
 284         fib_rules_unregister(net->ipv4.mr_rules_ops);
 285         rtnl_unlock();
 286 }
 287 #else
 288 #define ipmr_for_each_table(mrt, net) \
 289         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 290
 291 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 292 {
 293         return net->ipv4.mrt;
 294 }
 295
 296 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 297                            struct mr_table **mrt)
 298 {
 299         *mrt = net->ipv4.mrt;
 300         return 0;
 301 }
 302
 303 static int __net_init ipmr_rules_init(struct net *net)
 304 {
 305         net->ipv4.mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 306         return net->ipv4.mrt ? 0 : -ENOMEM;
 307 }
 308
 309 static void __net_exit ipmr_rules_exit(struct net *net)
 310 {
 311         rtnl_lock();
 312         ipmr_free_table(net->ipv4.mrt);
 313         net->ipv4.mrt = NULL;
 314         rtnl_unlock();
 315 }
 316 #endif
 317
 318 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 319 {
 320         struct mr_table *mrt;
 321         unsigned int i;
 322
 323         mrt = ipmr_get_table(net, id);
 324         if (mrt)
 325                 return mrt;
 326
 327         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
 328         if (!mrt)
 329                 return NULL;
 330         write_pnet(&mrt->net, net);
 331         mrt->id = id;
 332
 333         /* Forwarding cache */
 334         for (i = 0; i < MFC_LINES; i++)
 335                 INIT_LIST_HEAD(&mrt->mfc_cache_array[i]);
 336
 337         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
 338
 339         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
 340                     (unsigned long)mrt);
 341
 342 #ifdef CONFIG_IP_PIMSM
 343         mrt->mroute_reg_vif_num = -1;
 344 #endif
 345 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 346         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 347 #endif
 348         return mrt;
 349 }
 350
 351 static void ipmr_free_table(struct mr_table *mrt)
 352 {
 353         del_timer_sync(&mrt->ipmr_expire_timer);
 354         mroute_clean_tables(mrt, true);
 355         kfree(mrt);
 356 }
 357
 358 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 359
 360 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 361 {
 362         struct net *net = dev_net(dev);
 363
 364         dev_close(dev);
 365
 366         dev = __dev_get_by_name(net, "tunl0");
 367         if (dev) {
 368                 const struct net_device_ops *ops = dev->netdev_ops;
 369                 struct ifreq ifr;
 370                 struct ip_tunnel_parm p;
 371
 372                 memset(&p, 0, sizeof(p));
 373                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 374                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 375                 p.iph.version = 4;
 376                 p.iph.ihl = 5;
 377                 p.iph.protocol = IPPROTO_IPIP;
 378                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 379                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 380
 381                 if (ops->ndo_do_ioctl) {
 382                         mm_segment_t oldfs = get_fs();
 383
 384                         set_fs(KERNEL_DS);
 385                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 386                         set_fs(oldfs);
 387                 }
 388         }
 389 }
 390
 391 static
 392 struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 393 {
 394         struct net_device  *dev;
 395
 396         dev = __dev_get_by_name(net, "tunl0");
 397
 398         if (dev) {
 399                 const struct net_device_ops *ops = dev->netdev_ops;
 400                 int err;
 401                 struct ifreq ifr;
 402                 struct ip_tunnel_parm p;
 403                 struct in_device  *in_dev;
 404
 405                 memset(&p, 0, sizeof(p));
 406                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 407                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 408                 p.iph.version = 4;
 409                 p.iph.ihl = 5;
 410                 p.iph.protocol = IPPROTO_IPIP;
 411                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 412                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 413
 414                 if (ops->ndo_do_ioctl) {
 415                         mm_segment_t oldfs = get_fs();
 416
 417                         set_fs(KERNEL_DS);
 418                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 419                         set_fs(oldfs);
 420                 } else {
 421                         err = -EOPNOTSUPP;
 422                 }
 423                 dev = NULL;
 424
 425                 if (err == 0 &&
 426                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 427                         dev->flags |= IFF_MULTICAST;
 428
 429                         in_dev = __in_dev_get_rtnl(dev);
 430                         if (!in_dev)
 431                                 goto failure;
 432
 433                         ipv4_devconf_setall(in_dev);
 434                         neigh_parms_data_state_setall(in_dev->arp_parms);
 435                         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 436
 437                         if (dev_open(dev))
 438                                 goto failure;
 439                         dev_hold(dev);
 440                 }
 441         }
 442         return dev;
 443
 444 failure:
 445         unregister_netdevice(dev);
 446         return NULL;
 447 }
 448
 449 #ifdef CONFIG_IP_PIMSM
 450
 451 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 452 {
 453         struct net *net = dev_net(dev);
 454         struct mr_table *mrt;
 455         struct flowi4 fl4 = {
 456                 .flowi4_oif     = dev->ifindex,
 457                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 458                 .flowi4_mark    = skb->mark,
 459         };
 460         int err;
 461
 462         err = ipmr_fib_lookup(net, &fl4, &mrt);
 463         if (err < 0) {
 464                 kfree_skb(skb);
 465                 return err;
 466         }
 467
 468         read_lock(&mrt_lock);
 469         dev->stats.tx_bytes += skb->len;
 470         dev->stats.tx_packets++;
 471         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 472         read_unlock(&mrt_lock);
 473         kfree_skb(skb);
 474         return NETDEV_TX_OK;
 475 }
 476
 477 static int reg_vif_get_iflink(const struct net_device *dev)
 478 {
 479         return 0;
 480 }
 481
 482 static const struct net_device_ops reg_vif_netdev_ops = {
 483         .ndo_start_xmit = reg_vif_xmit,
 484         .ndo_get_iflink = reg_vif_get_iflink,
 485 };
 486
 487 static void reg_vif_setup(struct net_device *dev)
 488 {
 489         dev->type               = ARPHRD_PIMREG;
 490         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 491         dev->flags              = IFF_NOARP;
 492         dev->netdev_ops         = &reg_vif_netdev_ops;
 493         dev->destructor         = free_netdev;
 494         dev->features           |= NETIF_F_NETNS_LOCAL;
 495 }
 496
 497 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 498 {
 499         struct net_device *dev;
 500         struct in_device *in_dev;
 501         char name[IFNAMSIZ];
 502
 503         if (mrt->id == RT_TABLE_DEFAULT)
 504                 sprintf(name, "pimreg");
 505         else
 506                 sprintf(name, "pimreg%u", mrt->id);
 507
 508         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 509
 510         if (!dev)
 511                 return NULL;
 512
 513         dev_net_set(dev, net);
 514
 515         if (register_netdevice(dev)) {
 516                 free_netdev(dev);
 517                 return NULL;
 518         }
 519
 520         rcu_read_lock();
 521         in_dev = __in_dev_get_rcu(dev);
 522         if (!in_dev) {
 523                 rcu_read_unlock();
 524                 goto failure;
 525         }
 526
 527         ipv4_devconf_setall(in_dev);
 528         neigh_parms_data_state_setall(in_dev->arp_parms);
 529         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 530         rcu_read_unlock();
 531
 532         if (dev_open(dev))
 533                 goto failure;
 534
 535         dev_hold(dev);
 536
 537         return dev;
 538
 539 failure:
 540         unregister_netdevice(dev);
 541         return NULL;
 542 }
 543 #endif
 544
 545 /**
 546  *      vif_delete - Delete a VIF entry
 547  *      @notify: Set to 1, if the caller is a notifier_call
 548  */
 549
 550 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 551                       struct list_head *head)
 552 {
 553         struct vif_device *v;
 554         struct net_device *dev;
 555         struct in_device *in_dev;
 556
 557         if (vifi < 0 || vifi >= mrt->maxvif)
 558                 return -EADDRNOTAVAIL;
 559
 560         v = &mrt->vif_table[vifi];
 561
 562         write_lock_bh(&mrt_lock);
 563         dev = v->dev;
 564         v->dev = NULL;
 565
 566         if (!dev) {
 567                 write_unlock_bh(&mrt_lock);
 568                 return -EADDRNOTAVAIL;
 569         }
 570
 571 #ifdef CONFIG_IP_PIMSM
 572         if (vifi == mrt->mroute_reg_vif_num)
 573                 mrt->mroute_reg_vif_num = -1;
 574 #endif
 575
 576         if (vifi + 1 == mrt->maxvif) {
 577                 int tmp;
 578
 579                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 580                         if (VIF_EXISTS(mrt, tmp))
 581                                 break;
 582                 }
 583                 mrt->maxvif = tmp+1;
 584         }
 585
 586         write_unlock_bh(&mrt_lock);
 587
 588         dev_set_allmulti(dev, -1);
 589
 590         in_dev = __in_dev_get_rtnl(dev);
 591         if (in_dev) {
 592                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 593                 inet_netconf_notify_devconf(dev_net(dev),
 594                                             NETCONFA_MC_FORWARDING,
 595                                             dev->ifindex, &in_dev->cnf);
 596                 ip_rt_multicast_event(in_dev);
 597         }
 598
 599         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 600                 unregister_netdevice_queue(dev, head);
 601
 602         dev_put(dev);
 603         return 0;
 604 }
 605
 606 static void ipmr_cache_free_rcu(struct rcu_head *head)
 607 {
 608         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
 609
 610         kmem_cache_free(mrt_cachep, c);
 611 }
 612
 613 static inline void ipmr_cache_free(struct mfc_cache *c)
 614 {
 615         call_rcu(&c->rcu, ipmr_cache_free_rcu);
 616 }
 617
 618 /* Destroy an unresolved cache entry, killing queued skbs
 619  * and reporting error to netlink readers.
 620  */
 621
 622 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 623 {
 624         struct net *net = read_pnet(&mrt->net);
 625         struct sk_buff *skb;
 626         struct nlmsgerr *e;
 627
 628         atomic_dec(&mrt->cache_resolve_queue_len);
 629
 630         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
 631                 if (ip_hdr(skb)->version == 0) {
 632                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 633                         nlh->nlmsg_type = NLMSG_ERROR;
 634                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 635                         skb_trim(skb, nlh->nlmsg_len);
 636                         e = nlmsg_data(nlh);
 637                         e->error = -ETIMEDOUT;
 638                         memset(&e->msg, 0, sizeof(e->msg));
 639
 640                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 641                 } else {
 642                         kfree_skb(skb);
 643                 }
 644         }
 645
 646         ipmr_cache_free(c);
 647 }
 648
 649
 650 /* Timer process for the unresolved queue. */
 651
 652 static void ipmr_expire_process(unsigned long arg)
 653 {
 654         struct mr_table *mrt = (struct mr_table *)arg;
 655         unsigned long now;
 656         unsigned long expires;
 657         struct mfc_cache *c, *next;
 658
 659         if (!spin_trylock(&mfc_unres_lock)) {
 660                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 661                 return;
 662         }
 663
 664         if (list_empty(&mrt->mfc_unres_queue))
 665                 goto out;
 666
 667         now = jiffies;
 668         expires = 10*HZ;
 669
 670         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 671                 if (time_after(c->mfc_un.unres.expires, now)) {
 672                         unsigned long interval = c->mfc_un.unres.expires - now;
 673                         if (interval < expires)
 674                                 expires = interval;
 675                         continue;
 676                 }
 677
 678                 list_del(&c->list);
 679                 mroute_netlink_event(mrt, c, RTM_DELROUTE);
 680                 ipmr_destroy_unres(mrt, c);
 681         }
 682
 683         if (!list_empty(&mrt->mfc_unres_queue))
 684                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 685
 686 out:
 687         spin_unlock(&mfc_unres_lock);
 688 }
 689
 690 /* Fill oifs list. It is called under write locked mrt_lock. */
 691
 692 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
 693                                    unsigned char *ttls)
 694 {
 695         int vifi;
 696
 697         cache->mfc_un.res.minvif = MAXVIFS;
 698         cache->mfc_un.res.maxvif = 0;
 699         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 700
 701         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 702                 if (VIF_EXISTS(mrt, vifi) &&
 703                     ttls[vifi] && ttls[vifi] < 255) {
 704                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 705                         if (cache->mfc_un.res.minvif > vifi)
 706                                 cache->mfc_un.res.minvif = vifi;
 707                         if (cache->mfc_un.res.maxvif <= vifi)
 708                                 cache->mfc_un.res.maxvif = vifi + 1;
 709                 }
 710         }
 711 }
 712
 713 static int vif_add(struct net *net, struct mr_table *mrt,
 714                    struct vifctl *vifc, int mrtsock)
 715 {
 716         int vifi = vifc->vifc_vifi;
 717         struct vif_device *v = &mrt->vif_table[vifi];
 718         struct net_device *dev;
 719         struct in_device *in_dev;
 720         int err;
 721
 722         /* Is vif busy ? */
 723         if (VIF_EXISTS(mrt, vifi))
 724                 return -EADDRINUSE;
 725
 726         switch (vifc->vifc_flags) {
 727 #ifdef CONFIG_IP_PIMSM
 728         case VIFF_REGISTER:
 729                 /*
 730                  * Special Purpose VIF in PIM
 731                  * All the packets will be sent to the daemon
 732                  */
 733                 if (mrt->mroute_reg_vif_num >= 0)
 734                         return -EADDRINUSE;
 735                 dev = ipmr_reg_vif(net, mrt);
 736                 if (!dev)
 737                         return -ENOBUFS;
 738                 err = dev_set_allmulti(dev, 1);
 739                 if (err) {
 740                         unregister_netdevice(dev);
 741                         dev_put(dev);
 742                         return err;
 743                 }
 744                 break;
 745 #endif
 746         case VIFF_TUNNEL:
 747                 dev = ipmr_new_tunnel(net, vifc);
 748                 if (!dev)
 749                         return -ENOBUFS;
 750                 err = dev_set_allmulti(dev, 1);
 751                 if (err) {
 752                         ipmr_del_tunnel(dev, vifc);
 753                         dev_put(dev);
 754                         return err;
 755                 }
 756                 break;
 757
 758         case VIFF_USE_IFINDEX:
 759         case 0:
 760                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 761                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 762                         if (dev && !__in_dev_get_rtnl(dev)) {
 763                                 dev_put(dev);
 764                                 return -EADDRNOTAVAIL;
 765                         }
 766                 } else {
 767                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 768                 }
 769                 if (!dev)
 770                         return -EADDRNOTAVAIL;
 771                 err = dev_set_allmulti(dev, 1);
 772                 if (err) {
 773                         dev_put(dev);
 774                         return err;
 775                 }
 776                 break;
 777         default:
 778                 return -EINVAL;
 779         }
 780
 781         in_dev = __in_dev_get_rtnl(dev);
 782         if (!in_dev) {
 783                 dev_put(dev);
 784                 return -EADDRNOTAVAIL;
 785         }
 786         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 787         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING, dev->ifindex,
 788                                     &in_dev->cnf);
 789         ip_rt_multicast_event(in_dev);
 790
 791         /* Fill in the VIF structures */
 792
 793         v->rate_limit = vifc->vifc_rate_limit;
 794         v->local = vifc->vifc_lcl_addr.s_addr;
 795         v->remote = vifc->vifc_rmt_addr.s_addr;
 796         v->flags = vifc->vifc_flags;
 797         if (!mrtsock)
 798                 v->flags |= VIFF_STATIC;
 799         v->threshold = vifc->vifc_threshold;
 800         v->bytes_in = 0;
 801         v->bytes_out = 0;
 802         v->pkt_in = 0;
 803         v->pkt_out = 0;
 804         v->link = dev->ifindex;
 805         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
 806                 v->link = dev_get_iflink(dev);
 807
 808         /* And finish update writing critical data */
 809         write_lock_bh(&mrt_lock);
 810         v->dev = dev;
 811 #ifdef CONFIG_IP_PIMSM
 812         if (v->flags & VIFF_REGISTER)
 813                 mrt->mroute_reg_vif_num = vifi;
 814 #endif
 815         if (vifi+1 > mrt->maxvif)
 816                 mrt->maxvif = vifi+1;
 817         write_unlock_bh(&mrt_lock);
 818         return 0;
 819 }
 820
 821 /* called with rcu_read_lock() */
 822 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 823                                          __be32 origin,
 824                                          __be32 mcastgrp)
 825 {
 826         int line = MFC_HASH(mcastgrp, origin);
 827         struct mfc_cache *c;
 828
 829         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
 830                 if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
 831                         return c;
 832         }
 833         return NULL;
 834 }
 835
 836 /* Look for a (*,*,oif) entry */
 837 static struct mfc_cache *ipmr_cache_find_any_parent(struct mr_table *mrt,
 838                                                     int vifi)
 839 {
 840         int line = MFC_HASH(htonl(INADDR_ANY), htonl(INADDR_ANY));
 841         struct mfc_cache *c;
 842
 843         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list)
 844                 if (c->mfc_origin == htonl(INADDR_ANY) &&
 845                     c->mfc_mcastgrp == htonl(INADDR_ANY) &&
 846                     c->mfc_un.res.ttls[vifi] < 255)
 847                         return c;
 848
 849         return NULL;
 850 }
 851
 852 /* Look for a (*,G) entry */
 853 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 854                                              __be32 mcastgrp, int vifi)
 855 {
 856         int line = MFC_HASH(mcastgrp, htonl(INADDR_ANY));
 857         struct mfc_cache *c, *proxy;
 858
 859         if (mcastgrp == htonl(INADDR_ANY))
 860                 goto skip;
 861
 862         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list)
 863                 if (c->mfc_origin == htonl(INADDR_ANY) &&
 864                     c->mfc_mcastgrp == mcastgrp) {
 865                         if (c->mfc_un.res.ttls[vifi] < 255)
 866                                 return c;
 867
 868                         /* It's ok if the vifi is part of the static tree */
 869                         proxy = ipmr_cache_find_any_parent(mrt,
 870                                                            c->mfc_parent);
 871                         if (proxy && proxy->mfc_un.res.ttls[vifi] < 255)
 872                                 return c;
 873                 }
 874
 875 skip:
 876         return ipmr_cache_find_any_parent(mrt, vifi);
 877 }
 878
 879 /*
 880  *      Allocate a multicast cache entry
 881  */
 882 static struct mfc_cache *ipmr_cache_alloc(void)
 883 {
 884         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 885
 886         if (c) {
 887                 c->mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 888                 c->mfc_un.res.minvif = MAXVIFS;
 889         }
 890         return c;
 891 }
 892
 893 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 894 {
 895         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 896
 897         if (c) {
 898                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
 899                 c->mfc_un.unres.expires = jiffies + 10*HZ;
 900         }
 901         return c;
 902 }
 903
 904 /*
 905  *      A cache entry has gone into a resolved state from queued
 906  */
 907
 908 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 909                                struct mfc_cache *uc, struct mfc_cache *c)
 910 {
 911         struct sk_buff *skb;
 912         struct nlmsgerr *e;
 913
 914         /* Play the pending entries through our router */
 915
 916         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
 917                 if (ip_hdr(skb)->version == 0) {
 918                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 919
 920                         if (__ipmr_fill_mroute(mrt, skb, c, nlmsg_data(nlh)) > 0) {
 921                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 922                                                  (u8 *)nlh;
 923                         } else {
 924                                 nlh->nlmsg_type = NLMSG_ERROR;
 925                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 926                                 skb_trim(skb, nlh->nlmsg_len);
 927                                 e = nlmsg_data(nlh);
 928                                 e->error = -EMSGSIZE;
 929                                 memset(&e->msg, 0, sizeof(e->msg));
 930                         }
 931
 932                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 933                 } else {
 934                         ip_mr_forward(net, mrt, skb, c, 0);
 935                 }
 936         }
 937 }
 938
 939 /*
 940  *      Bounce a cache query up to mrouted. We could use netlink for this but mrouted
 941  *      expects the following bizarre scheme.
 942  *
 943  *      Called under mrt_lock.
 944  */
 945
 946 static int ipmr_cache_report(struct mr_table *mrt,
 947                              struct sk_buff *pkt, vifi_t vifi, int assert)
 948 {
 949         struct sk_buff *skb;
 950         const int ihl = ip_hdrlen(pkt);
 951         struct igmphdr *igmp;
 952         struct igmpmsg *msg;
 953         struct sock *mroute_sk;
 954         int ret;
 955
 956 #ifdef CONFIG_IP_PIMSM
 957         if (assert == IGMPMSG_WHOLEPKT)
 958                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
 959         else
 960 #endif
 961                 skb = alloc_skb(128, GFP_ATOMIC);
 962
 963         if (!skb)
 964                 return -ENOBUFS;
 965
 966 #ifdef CONFIG_IP_PIMSM
 967         if (assert == IGMPMSG_WHOLEPKT) {
 968                 /* Ugly, but we have no choice with this interface.
 969                  * Duplicate old header, fix ihl, length etc.
 970                  * And all this only to mangle msg->im_msgtype and
 971                  * to set msg->im_mbz to "mbz" :-)
 972                  */
 973                 skb_push(skb, sizeof(struct iphdr));
 974                 skb_reset_network_header(skb);
 975                 skb_reset_transport_header(skb);
 976                 msg = (struct igmpmsg *)skb_network_header(skb);
 977                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
 978                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
 979                 msg->im_mbz = 0;
 980                 msg->im_vif = mrt->mroute_reg_vif_num;
 981                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
 982                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
 983                                              sizeof(struct iphdr));
 984         } else
 985 #endif
 986         {
 987
 988         /* Copy the IP header */
 989
 990         skb_set_network_header(skb, skb->len);
 991         skb_put(skb, ihl);
 992         skb_copy_to_linear_data(skb, pkt->data, ihl);
 993         ip_hdr(skb)->protocol = 0;      /* Flag to the kernel this is a route add */
 994         msg = (struct igmpmsg *)skb_network_header(skb);
 995         msg->im_vif = vifi;
 996         skb_dst_set(skb, dst_clone(skb_dst(pkt)));
 997
 998         /* Add our header */
 999
1000         igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
1001         igmp->type      =
1002         msg->im_msgtype = assert;
1003         igmp->code      = 0;
1004         ip_hdr(skb)->tot_len = htons(skb->len);         /* Fix the length */
1005         skb->transport_header = skb->network_header;
1006         }
1007
1008         rcu_read_lock();
1009         mroute_sk = rcu_dereference(mrt->mroute_sk);
1010         if (!mroute_sk) {
1011                 rcu_read_unlock();
1012                 kfree_skb(skb);
1013                 return -EINVAL;
1014         }
1015
1016         /* Deliver to mrouted */
1017
1018         ret = sock_queue_rcv_skb(mroute_sk, skb);
1019         rcu_read_unlock();
1020         if (ret < 0) {
1021                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1022                 kfree_skb(skb);
1023         }
1024
1025         return ret;
1026 }
1027
1028 /*
1029  *      Queue a packet for resolution. It gets locked cache entry!
1030  */
1031
1032 static int
1033 ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi, struct sk_buff *skb)
1034 {
1035         bool found = false;
1036         int err;
1037         struct mfc_cache *c;
1038         const struct iphdr *iph = ip_hdr(skb);
1039
1040         spin_lock_bh(&mfc_unres_lock);
1041         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
1042                 if (c->mfc_mcastgrp == iph->daddr &&
1043                     c->mfc_origin == iph->saddr) {
1044                         found = true;
1045                         break;
1046                 }
1047         }
1048
1049         if (!found) {
1050                 /* Create a new entry if allowable */
1051
1052                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1053                     (c = ipmr_cache_alloc_unres()) == NULL) {
1054                         spin_unlock_bh(&mfc_unres_lock);
1055
1056                         kfree_skb(skb);
1057                         return -ENOBUFS;
1058                 }
1059
1060                 /* Fill in the new cache entry */
1061
1062                 c->mfc_parent   = -1;
1063                 c->mfc_origin   = iph->saddr;
1064                 c->mfc_mcastgrp = iph->daddr;
1065
1066                 /* Reflect first query at mrouted. */
1067
1068                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1069                 if (err < 0) {
1070                         /* If the report failed throw the cache entry
1071                            out - Brad Parker
1072                          */
1073                         spin_unlock_bh(&mfc_unres_lock);
1074
1075                         ipmr_cache_free(c);
1076                         kfree_skb(skb);
1077                         return err;
1078                 }
1079
1080                 atomic_inc(&mrt->cache_resolve_queue_len);
1081                 list_add(&c->list, &mrt->mfc_unres_queue);
1082                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1083
1084                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1085                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1086         }
1087
1088         /* See if we can append the packet */
1089
1090         if (c->mfc_un.unres.unresolved.qlen > 3) {
1091                 kfree_skb(skb);
1092                 err = -ENOBUFS;
1093         } else {
1094                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1095                 err = 0;
1096         }
1097
1098         spin_unlock_bh(&mfc_unres_lock);
1099         return err;
1100 }
1101
1102 /*
1103  *      MFC cache manipulation by user space mroute daemon
1104  */
1105
1106 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1107 {
1108         int line;
1109         struct mfc_cache *c, *next;
1110
1111         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1112
1113         list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1114                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1115                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr &&
1116                     (parent == -1 || parent == c->mfc_parent)) {
1117                         list_del_rcu(&c->list);
1118                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1119                         ipmr_cache_free(c);
1120                         return 0;
1121                 }
1122         }
1123         return -ENOENT;
1124 }
1125
1126 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1127                         struct mfcctl *mfc, int mrtsock, int parent)
1128 {
1129         bool found = false;
1130         int line;
1131         struct mfc_cache *uc, *c;
1132
1133         if (mfc->mfcc_parent >= MAXVIFS)
1134                 return -ENFILE;
1135
1136         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1137
1138         list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1139                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1140                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr &&
1141                     (parent == -1 || parent == c->mfc_parent)) {
1142                         found = true;
1143                         break;
1144                 }
1145         }
1146
1147         if (found) {
1148                 write_lock_bh(&mrt_lock);
1149                 c->mfc_parent = mfc->mfcc_parent;
1150                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1151                 if (!mrtsock)
1152                         c->mfc_flags |= MFC_STATIC;
1153                 write_unlock_bh(&mrt_lock);
1154                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1155                 return 0;
1156         }
1157
1158         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1159             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1160                 return -EINVAL;
1161
1162         c = ipmr_cache_alloc();
1163         if (!c)
1164                 return -ENOMEM;
1165
1166         c->mfc_origin = mfc->mfcc_origin.s_addr;
1167         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1168         c->mfc_parent = mfc->mfcc_parent;
1169         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1170         if (!mrtsock)
1171                 c->mfc_flags |= MFC_STATIC;
1172
1173         list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1174
1175         /*
1176          *      Check to see if we resolved a queued list. If so we
1177          *      need to send on the frames and tidy up.
1178          */
1179         found = false;
1180         spin_lock_bh(&mfc_unres_lock);
1181         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1182                 if (uc->mfc_origin == c->mfc_origin &&
1183                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1184                         list_del(&uc->list);
1185                         atomic_dec(&mrt->cache_resolve_queue_len);
1186                         found = true;
1187                         break;
1188                 }
1189         }
1190         if (list_empty(&mrt->mfc_unres_queue))
1191                 del_timer(&mrt->ipmr_expire_timer);
1192         spin_unlock_bh(&mfc_unres_lock);
1193
1194         if (found) {
1195                 ipmr_cache_resolve(net, mrt, uc, c);
1196                 ipmr_cache_free(uc);
1197         }
1198         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1199         return 0;
1200 }
1201
1202 /*
1203  *      Close the multicast socket, and clear the vif tables etc
1204  */
1205
1206 static void mroute_clean_tables(struct mr_table *mrt, bool all)
1207 {
1208         int i;
1209         LIST_HEAD(list);
1210         struct mfc_cache *c, *next;
1211
1212         /* Shut down all active vif entries */
1213
1214         for (i = 0; i < mrt->maxvif; i++) {
1215                 if (!all && (mrt->vif_table[i].flags & VIFF_STATIC))
1216                         continue;
1217                 vif_delete(mrt, i, 0, &list);
1218         }
1219         unregister_netdevice_many(&list);
1220
1221         /* Wipe the cache */
1222
1223         for (i = 0; i < MFC_LINES; i++) {
1224                 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1225                         if (!all && (c->mfc_flags & MFC_STATIC))
1226                                 continue;
1227                         list_del_rcu(&c->list);
1228                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1229                         ipmr_cache_free(c);
1230                 }
1231         }
1232
1233         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1234                 spin_lock_bh(&mfc_unres_lock);
1235                 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1236                         list_del(&c->list);
1237                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1238                         ipmr_destroy_unres(mrt, c);
1239                 }
1240                 spin_unlock_bh(&mfc_unres_lock);
1241         }
1242 }
1243
1244 /* called from ip_ra_control(), before an RCU grace period,
1245  * we dont need to call synchronize_rcu() here
1246  */
1247 static void mrtsock_destruct(struct sock *sk)
1248 {
1249         struct net *net = sock_net(sk);
1250         struct mr_table *mrt;
1251
1252         rtnl_lock();
1253         ipmr_for_each_table(mrt, net) {
1254                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1255                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1256                         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1257                                                     NETCONFA_IFINDEX_ALL,
1258                                                     net->ipv4.devconf_all);
1259                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1260                         mroute_clean_tables(mrt, false);
1261                 }
1262         }
1263         rtnl_unlock();
1264 }
1265
1266 /*
1267  *      Socket options and virtual interface manipulation. The whole
1268  *      virtual interface system is a complete heap, but unfortunately
1269  *      that's how BSD mrouted happens to think. Maybe one day with a proper
1270  *      MOSPF/PIM router set up we can clean this up.
1271  */
1272
1273 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval, unsigned int optlen)
1274 {
1275         int ret, parent = 0;
1276         struct vifctl vif;
1277         struct mfcctl mfc;
1278         struct net *net = sock_net(sk);
1279         struct mr_table *mrt;
1280
1281         if (sk->sk_type != SOCK_RAW ||
1282             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1283                 return -EOPNOTSUPP;
1284
1285         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1286         if (!mrt)
1287                 return -ENOENT;
1288
1289         if (optname != MRT_INIT) {
1290                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1291                     !ns_capable(net->user_ns, CAP_NET_ADMIN))
1292                         return -EACCES;
1293         }
1294
1295         switch (optname) {
1296         case MRT_INIT:
1297                 if (optlen != sizeof(int))
1298                         return -EINVAL;
1299
1300                 rtnl_lock();
1301                 if (rtnl_dereference(mrt->mroute_sk)) {
1302                         rtnl_unlock();
1303                         return -EADDRINUSE;
1304                 }
1305
1306                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1307                 if (ret == 0) {
1308                         rcu_assign_pointer(mrt->mroute_sk, sk);
1309                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1310                         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1311                                                     NETCONFA_IFINDEX_ALL,
1312                                                     net->ipv4.devconf_all);
1313                 }
1314                 rtnl_unlock();
1315                 return ret;
1316         case MRT_DONE:
1317                 if (sk != rcu_access_pointer(mrt->mroute_sk))
1318                         return -EACCES;
1319                 return ip_ra_control(sk, 0, NULL);
1320         case MRT_ADD_VIF:
1321         case MRT_DEL_VIF:
1322                 if (optlen != sizeof(vif))
1323                         return -EINVAL;
1324                 if (copy_from_user(&vif, optval, sizeof(vif)))
1325                         return -EFAULT;
1326                 if (vif.vifc_vifi >= MAXVIFS)
1327                         return -ENFILE;
1328                 rtnl_lock();
1329                 if (optname == MRT_ADD_VIF) {
1330                         ret = vif_add(net, mrt, &vif,
1331                                       sk == rtnl_dereference(mrt->mroute_sk));
1332                 } else {
1333                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1334                 }
1335                 rtnl_unlock();
1336                 return ret;
1337
1338                 /*
1339                  *      Manipulate the forwarding caches. These live
1340                  *      in a sort of kernel/user symbiosis.
1341                  */
1342         case MRT_ADD_MFC:
1343         case MRT_DEL_MFC:
1344                 parent = -1;
1345         case MRT_ADD_MFC_PROXY:
1346         case MRT_DEL_MFC_PROXY:
1347                 if (optlen != sizeof(mfc))
1348                         return -EINVAL;
1349                 if (copy_from_user(&mfc, optval, sizeof(mfc)))
1350                         return -EFAULT;
1351                 if (parent == 0)
1352                         parent = mfc.mfcc_parent;
1353                 rtnl_lock();
1354                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1355                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1356                 else
1357                         ret = ipmr_mfc_add(net, mrt, &mfc,
1358                                            sk == rtnl_dereference(mrt->mroute_sk),
1359                                            parent);
1360                 rtnl_unlock();
1361                 return ret;
1362                 /*
1363                  *      Control PIM assert.
1364                  */
1365         case MRT_ASSERT:
1366         {
1367                 int v;
1368                 if (optlen != sizeof(v))
1369                         return -EINVAL;
1370                 if (get_user(v, (int __user *)optval))
1371                         return -EFAULT;
1372                 mrt->mroute_do_assert = v;
1373                 return 0;
1374         }
1375 #ifdef CONFIG_IP_PIMSM
1376         case MRT_PIM:
1377         {
1378                 int v;
1379
1380                 if (optlen != sizeof(v))
1381                         return -EINVAL;
1382                 if (get_user(v, (int __user *)optval))
1383                         return -EFAULT;
1384                 v = !!v;
1385
1386                 rtnl_lock();
1387                 ret = 0;
1388                 if (v != mrt->mroute_do_pim) {
1389                         mrt->mroute_do_pim = v;
1390                         mrt->mroute_do_assert = v;
1391                 }
1392                 rtnl_unlock();
1393                 return ret;
1394         }
1395 #endif
1396 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
1397         case MRT_TABLE:
1398         {
1399                 u32 v;
1400
1401                 if (optlen != sizeof(u32))
1402                         return -EINVAL;
1403                 if (get_user(v, (u32 __user *)optval))
1404                         return -EFAULT;
1405
1406                 /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
1407                 if (v != RT_TABLE_DEFAULT && v >= 1000000000)
1408                         return -EINVAL;
1409
1410                 rtnl_lock();
1411                 ret = 0;
1412                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1413                         ret = -EBUSY;
1414                 } else {
1415                         if (!ipmr_new_table(net, v))
1416                                 ret = -ENOMEM;
1417                         else
1418                                 raw_sk(sk)->ipmr_table = v;
1419                 }
1420                 rtnl_unlock();
1421                 return ret;
1422         }
1423 #endif
1424         /*
1425          *      Spurious command, or MRT_VERSION which you cannot
1426          *      set.
1427          */
1428         default:
1429                 return -ENOPROTOOPT;
1430         }
1431 }
1432
1433 /*
1434  *      Getsock opt support for the multicast routing system.
1435  */
1436
1437 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1438 {
1439         int olr;
1440         int val;
1441         struct net *net = sock_net(sk);
1442         struct mr_table *mrt;
1443
1444         if (sk->sk_type != SOCK_RAW ||
1445             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1446                 return -EOPNOTSUPP;
1447
1448         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1449         if (!mrt)
1450                 return -ENOENT;
1451
1452         if (optname != MRT_VERSION &&
1453 #ifdef CONFIG_IP_PIMSM
1454            optname != MRT_PIM &&
1455 #endif
1456            optname != MRT_ASSERT)
1457                 return -ENOPROTOOPT;
1458
1459         if (get_user(olr, optlen))
1460                 return -EFAULT;
1461
1462         olr = min_t(unsigned int, olr, sizeof(int));
1463         if (olr < 0)
1464                 return -EINVAL;
1465
1466         if (put_user(olr, optlen))
1467                 return -EFAULT;
1468         if (optname == MRT_VERSION)
1469                 val = 0x0305;
1470 #ifdef CONFIG_IP_PIMSM
1471         else if (optname == MRT_PIM)
1472                 val = mrt->mroute_do_pim;
1473 #endif
1474         else
1475                 val = mrt->mroute_do_assert;
1476         if (copy_to_user(optval, &val, olr))
1477                 return -EFAULT;
1478         return 0;
1479 }
1480
1481 /*
1482  *      The IP multicast ioctl support routines.
1483  */
1484
1485 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1486 {
1487         struct sioc_sg_req sr;
1488         struct sioc_vif_req vr;
1489         struct vif_device *vif;
1490         struct mfc_cache *c;
1491         struct net *net = sock_net(sk);
1492         struct mr_table *mrt;
1493
1494         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1495         if (!mrt)
1496                 return -ENOENT;
1497
1498         switch (cmd) {
1499         case SIOCGETVIFCNT:
1500                 if (copy_from_user(&vr, arg, sizeof(vr)))
1501                         return -EFAULT;
1502                 if (vr.vifi >= mrt->maxvif)
1503                         return -EINVAL;
1504                 read_lock(&mrt_lock);
1505                 vif = &mrt->vif_table[vr.vifi];
1506                 if (VIF_EXISTS(mrt, vr.vifi)) {
1507                         vr.icount = vif->pkt_in;
1508                         vr.ocount = vif->pkt_out;
1509                         vr.ibytes = vif->bytes_in;
1510                         vr.obytes = vif->bytes_out;
1511                         read_unlock(&mrt_lock);
1512
1513                         if (copy_to_user(arg, &vr, sizeof(vr)))
1514                                 return -EFAULT;
1515                         return 0;
1516                 }
1517                 read_unlock(&mrt_lock);
1518                 return -EADDRNOTAVAIL;
1519         case SIOCGETSGCNT:
1520                 if (copy_from_user(&sr, arg, sizeof(sr)))
1521                         return -EFAULT;
1522
1523                 rcu_read_lock();
1524                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1525                 if (c) {
1526                         sr.pktcnt = c->mfc_un.res.pkt;
1527                         sr.bytecnt = c->mfc_un.res.bytes;
1528                         sr.wrong_if = c->mfc_un.res.wrong_if;
1529                         rcu_read_unlock();
1530
1531                         if (copy_to_user(arg, &sr, sizeof(sr)))
1532                                 return -EFAULT;
1533                         return 0;
1534                 }
1535                 rcu_read_unlock();
1536                 return -EADDRNOTAVAIL;
1537         default:
1538                 return -ENOIOCTLCMD;
1539         }
1540 }
1541
1542 #ifdef CONFIG_COMPAT
1543 struct compat_sioc_sg_req {
1544         struct in_addr src;
1545         struct in_addr grp;
1546         compat_ulong_t pktcnt;
1547         compat_ulong_t bytecnt;
1548         compat_ulong_t wrong_if;
1549 };
1550
1551 struct compat_sioc_vif_req {
1552         vifi_t  vifi;           /* Which iface */
1553         compat_ulong_t icount;
1554         compat_ulong_t ocount;
1555         compat_ulong_t ibytes;
1556         compat_ulong_t obytes;
1557 };
1558
1559 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1560 {
1561         struct compat_sioc_sg_req sr;
1562         struct compat_sioc_vif_req vr;
1563         struct vif_device *vif;
1564         struct mfc_cache *c;
1565         struct net *net = sock_net(sk);
1566         struct mr_table *mrt;
1567
1568         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1569         if (!mrt)
1570                 return -ENOENT;
1571
1572         switch (cmd) {
1573         case SIOCGETVIFCNT:
1574                 if (copy_from_user(&vr, arg, sizeof(vr)))
1575                         return -EFAULT;
1576                 if (vr.vifi >= mrt->maxvif)
1577                         return -EINVAL;
1578                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1579                 read_lock(&mrt_lock);
1580                 vif = &mrt->vif_table[vr.vifi];
1581                 if (VIF_EXISTS(mrt, vr.vifi)) {
1582                         vr.icount = vif->pkt_in;
1583                         vr.ocount = vif->pkt_out;
1584                         vr.ibytes = vif->bytes_in;
1585                         vr.obytes = vif->bytes_out;
1586                         read_unlock(&mrt_lock);
1587
1588                         if (copy_to_user(arg, &vr, sizeof(vr)))
1589                                 return -EFAULT;
1590                         return 0;
1591                 }
1592                 read_unlock(&mrt_lock);
1593                 return -EADDRNOTAVAIL;
1594         case SIOCGETSGCNT:
1595                 if (copy_from_user(&sr, arg, sizeof(sr)))
1596                         return -EFAULT;
1597
1598                 rcu_read_lock();
1599                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1600                 if (c) {
1601                         sr.pktcnt = c->mfc_un.res.pkt;
1602                         sr.bytecnt = c->mfc_un.res.bytes;
1603                         sr.wrong_if = c->mfc_un.res.wrong_if;
1604                         rcu_read_unlock();
1605
1606                         if (copy_to_user(arg, &sr, sizeof(sr)))
1607                                 return -EFAULT;
1608                         return 0;
1609                 }
1610                 rcu_read_unlock();
1611                 return -EADDRNOTAVAIL;
1612         default:
1613                 return -ENOIOCTLCMD;
1614         }
1615 }
1616 #endif
1617
1618
1619 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1620 {
1621         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1622         struct net *net = dev_net(dev);
1623         struct mr_table *mrt;
1624         struct vif_device *v;
1625         int ct;
1626
1627         if (event != NETDEV_UNREGISTER)
1628                 return NOTIFY_DONE;
1629
1630         ipmr_for_each_table(mrt, net) {
1631                 v = &mrt->vif_table[0];
1632                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1633                         if (v->dev == dev)
1634                                 vif_delete(mrt, ct, 1, NULL);
1635                 }
1636         }
1637         return NOTIFY_DONE;
1638 }
1639
1640
1641 static struct notifier_block ip_mr_notifier = {
1642         .notifier_call = ipmr_device_event,
1643 };
1644
1645 /*
1646  *      Encapsulate a packet by attaching a valid IPIP header to it.
1647  *      This avoids tunnel drivers and other mess and gives us the speed so
1648  *      important for multicast video.
1649  */
1650
1651 static void ip_encap(struct net *net, struct sk_buff *skb,
1652                      __be32 saddr, __be32 daddr)
1653 {
1654         struct iphdr *iph;
1655         const struct iphdr *old_iph = ip_hdr(skb);
1656
1657         skb_push(skb, sizeof(struct iphdr));
1658         skb->transport_header = skb->network_header;
1659         skb_reset_network_header(skb);
1660         iph = ip_hdr(skb);
1661
1662         iph->version    =       4;
1663         iph->tos        =       old_iph->tos;
1664         iph->ttl        =       old_iph->ttl;
1665         iph->frag_off   =       0;
1666         iph->daddr      =       daddr;
1667         iph->saddr      =       saddr;
1668         iph->protocol   =       IPPROTO_IPIP;
1669         iph->ihl        =       5;
1670         iph->tot_len    =       htons(skb->len);
1671         ip_select_ident(net, skb, NULL);
1672         ip_send_check(iph);
1673
1674         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1675         nf_reset(skb);
1676 }
1677
1678 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1679                                       struct sk_buff *skb)
1680 {
1681         struct ip_options *opt = &(IPCB(skb)->opt);
1682
1683         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1684         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1685
1686         if (unlikely(opt->optlen))
1687                 ip_forward_options(skb);
1688
1689         return dst_output(net, sk, skb);
1690 }
1691
1692 /*
1693  *      Processing handlers for ipmr_forward
1694  */
1695
1696 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1697                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1698 {
1699         const struct iphdr *iph = ip_hdr(skb);
1700         struct vif_device *vif = &mrt->vif_table[vifi];
1701         struct net_device *dev;
1702         struct rtable *rt;
1703         struct flowi4 fl4;
1704         int    encap = 0;
1705
1706         if (!vif->dev)
1707                 goto out_free;
1708
1709 #ifdef CONFIG_IP_PIMSM
1710         if (vif->flags & VIFF_REGISTER) {
1711                 vif->pkt_out++;
1712                 vif->bytes_out += skb->len;
1713                 vif->dev->stats.tx_bytes += skb->len;
1714                 vif->dev->stats.tx_packets++;
1715                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1716                 goto out_free;
1717         }
1718 #endif
1719
1720         if (vif->flags & VIFF_TUNNEL) {
1721                 rt = ip_route_output_ports(net, &fl4, NULL,
1722                                            vif->remote, vif->local,
1723                                            0, 0,
1724                                            IPPROTO_IPIP,
1725                                            RT_TOS(iph->tos), vif->link);
1726                 if (IS_ERR(rt))
1727                         goto out_free;
1728                 encap = sizeof(struct iphdr);
1729         } else {
1730                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1731                                            0, 0,
1732                                            IPPROTO_IPIP,
1733                                            RT_TOS(iph->tos), vif->link);
1734                 if (IS_ERR(rt))
1735                         goto out_free;
1736         }
1737
1738         dev = rt->dst.dev;
1739
1740         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1741                 /* Do not fragment multicasts. Alas, IPv4 does not
1742                  * allow to send ICMP, so that packets will disappear
1743                  * to blackhole.
1744                  */
1745
1746                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1747                 ip_rt_put(rt);
1748                 goto out_free;
1749         }
1750
1751         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1752
1753         if (skb_cow(skb, encap)) {
1754                 ip_rt_put(rt);
1755                 goto out_free;
1756         }
1757
1758         vif->pkt_out++;
1759         vif->bytes_out += skb->len;
1760
1761         skb_dst_drop(skb);
1762         skb_dst_set(skb, &rt->dst);
1763         ip_decrease_ttl(ip_hdr(skb));
1764
1765         /* FIXME: forward and output firewalls used to be called here.
1766          * What do we do with netfilter? -- RR
1767          */
1768         if (vif->flags & VIFF_TUNNEL) {
1769                 ip_encap(net, skb, vif->local, vif->remote);
1770                 /* FIXME: extra output firewall step used to be here. --RR */
1771                 vif->dev->stats.tx_packets++;
1772                 vif->dev->stats.tx_bytes += skb->len;
1773         }
1774
1775         IPCB(skb)->flags |= IPSKB_FORWARDED;
1776
1777         /*
1778          * RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1779          * not only before forwarding, but after forwarding on all output
1780          * interfaces. It is clear, if mrouter runs a multicasting
1781          * program, it should receive packets not depending to what interface
1782          * program is joined.
1783          * If we will not make it, the program will have to join on all
1784          * interfaces. On the other hand, multihoming host (or router, but
1785          * not mrouter) cannot join to more than one interface - it will
1786          * result in receiving multiple packets.
1787          */
1788         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1789                 net, NULL, skb, skb->dev, dev,
1790                 ipmr_forward_finish);
1791         return;
1792
1793 out_free:
1794         kfree_skb(skb);
1795 }
1796
1797 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1798 {
1799         int ct;
1800
1801         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1802                 if (mrt->vif_table[ct].dev == dev)
1803                         break;
1804         }
1805         return ct;
1806 }
1807
1808 /* "local" means that we should preserve one skb (for local delivery) */
1809
1810 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1811                           struct sk_buff *skb, struct mfc_cache *cache,
1812                           int local)
1813 {
1814         int psend = -1;
1815         int vif, ct;
1816         int true_vifi = ipmr_find_vif(mrt, skb->dev);
1817
1818         vif = cache->mfc_parent;
1819         cache->mfc_un.res.pkt++;
1820         cache->mfc_un.res.bytes += skb->len;
1821
1822         if (cache->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1823                 struct mfc_cache *cache_proxy;
1824
1825                 /* For an (*,G) entry, we only check that the incomming
1826                  * interface is part of the static tree.
1827                  */
1828                 cache_proxy = ipmr_cache_find_any_parent(mrt, vif);
1829                 if (cache_proxy &&
1830                     cache_proxy->mfc_un.res.ttls[true_vifi] < 255)
1831                         goto forward;
1832         }
1833
1834         /*
1835          * Wrong interface: drop packet and (maybe) send PIM assert.
1836          */
1837         if (mrt->vif_table[vif].dev != skb->dev) {
1838                 if (rt_is_output_route(skb_rtable(skb))) {
1839                         /* It is our own packet, looped back.
1840                          * Very complicated situation...
1841                          *
1842                          * The best workaround until routing daemons will be
1843                          * fixed is not to redistribute packet, if it was
1844                          * send through wrong interface. It means, that
1845                          * multicast applications WILL NOT work for
1846                          * (S,G), which have default multicast route pointing
1847                          * to wrong oif. In any case, it is not a good
1848                          * idea to use multicasting applications on router.
1849                          */
1850                         goto dont_forward;
1851                 }
1852
1853                 cache->mfc_un.res.wrong_if++;
1854
1855                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1856                     /* pimsm uses asserts, when switching from RPT to SPT,
1857                      * so that we cannot check that packet arrived on an oif.
1858                      * It is bad, but otherwise we would need to move pretty
1859                      * large chunk of pimd to kernel. Ough... --ANK
1860                      */
1861                     (mrt->mroute_do_pim ||
1862                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1863                     time_after(jiffies,
1864                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1865                         cache->mfc_un.res.last_assert = jiffies;
1866                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1867                 }
1868                 goto dont_forward;
1869         }
1870
1871 forward:
1872         mrt->vif_table[vif].pkt_in++;
1873         mrt->vif_table[vif].bytes_in += skb->len;
1874
1875         /*
1876          *      Forward the frame
1877          */
1878         if (cache->mfc_origin == htonl(INADDR_ANY) &&
1879             cache->mfc_mcastgrp == htonl(INADDR_ANY)) {
1880                 if (true_vifi >= 0 &&
1881                     true_vifi != cache->mfc_parent &&
1882                     ip_hdr(skb)->ttl >
1883                                 cache->mfc_un.res.ttls[cache->mfc_parent]) {
1884                         /* It's an (*,*) entry and the packet is not coming from
1885                          * the upstream: forward the packet to the upstream
1886                          * only.
1887                          */
1888                         psend = cache->mfc_parent;
1889                         goto last_forward;
1890                 }
1891                 goto dont_forward;
1892         }
1893         for (ct = cache->mfc_un.res.maxvif - 1;
1894              ct >= cache->mfc_un.res.minvif; ct--) {
1895                 /* For (*,G) entry, don't forward to the incoming interface */
1896                 if ((cache->mfc_origin != htonl(INADDR_ANY) ||
1897                      ct != true_vifi) &&
1898                     ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1899                         if (psend != -1) {
1900                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1901
1902                                 if (skb2)
1903                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1904                                                         psend);
1905                         }
1906                         psend = ct;
1907                 }
1908         }
1909 last_forward:
1910         if (psend != -1) {
1911                 if (local) {
1912                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1913
1914                         if (skb2)
1915                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1916                 } else {
1917                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1918                         return;
1919                 }
1920         }
1921
1922 dont_forward:
1923         if (!local)
1924                 kfree_skb(skb);
1925 }
1926
1927 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1928 {
1929         struct rtable *rt = skb_rtable(skb);
1930         struct iphdr *iph = ip_hdr(skb);
1931         struct flowi4 fl4 = {
1932                 .daddr = iph->daddr,
1933                 .saddr = iph->saddr,
1934                 .flowi4_tos = RT_TOS(iph->tos),
1935                 .flowi4_oif = (rt_is_output_route(rt) ?
1936                                skb->dev->ifindex : 0),
1937                 .flowi4_iif = (rt_is_output_route(rt) ?
1938                                LOOPBACK_IFINDEX :
1939                                skb->dev->ifindex),
1940                 .flowi4_mark = skb->mark,
1941         };
1942         struct mr_table *mrt;
1943         int err;
1944
1945         err = ipmr_fib_lookup(net, &fl4, &mrt);
1946         if (err)
1947                 return ERR_PTR(err);
1948         return mrt;
1949 }
1950
1951 /*
1952  *      Multicast packets for forwarding arrive here
1953  *      Called with rcu_read_lock();
1954  */
1955
1956 int ip_mr_input(struct sk_buff *skb)
1957 {
1958         struct mfc_cache *cache;
1959         struct net *net = dev_net(skb->dev);
1960         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1961         struct mr_table *mrt;
1962
1963         /* Packet is looped back after forward, it should not be
1964          * forwarded second time, but still can be delivered locally.
1965          */
1966         if (IPCB(skb)->flags & IPSKB_FORWARDED)
1967                 goto dont_forward;
1968
1969         mrt = ipmr_rt_fib_lookup(net, skb);
1970         if (IS_ERR(mrt)) {
1971                 kfree_skb(skb);
1972                 return PTR_ERR(mrt);
1973         }
1974         if (!local) {
1975                 if (IPCB(skb)->opt.router_alert) {
1976                         if (ip_call_ra_chain(skb))
1977                                 return 0;
1978                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1979                         /* IGMPv1 (and broken IGMPv2 implementations sort of
1980                          * Cisco IOS <= 11.2(8)) do not put router alert
1981                          * option to IGMP packets destined to routable
1982                          * groups. It is very bad, because it means
1983                          * that we can forward NO IGMP messages.
1984                          */
1985                         struct sock *mroute_sk;
1986
1987                         mroute_sk = rcu_dereference(mrt->mroute_sk);
1988                         if (mroute_sk) {
1989                                 nf_reset(skb);
1990                                 raw_rcv(mroute_sk, skb);
1991                                 return 0;
1992                         }
1993                     }
1994         }
1995
1996         /* already under rcu_read_lock() */
1997         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1998         if (!cache) {
1999                 int vif = ipmr_find_vif(mrt, skb->dev);
2000
2001                 if (vif >= 0)
2002                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2003                                                     vif);
2004         }
2005
2006         /*
2007          *      No usable cache entry
2008          */
2009         if (!cache) {
2010                 int vif;
2011
2012                 if (local) {
2013                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2014                         ip_local_deliver(skb);
2015                         if (!skb2)
2016                                 return -ENOBUFS;
2017                         skb = skb2;
2018                 }
2019
2020                 read_lock(&mrt_lock);
2021                 vif = ipmr_find_vif(mrt, skb->dev);
2022                 if (vif >= 0) {
2023                         int err2 = ipmr_cache_unresolved(mrt, vif, skb);
2024                         read_unlock(&mrt_lock);
2025
2026                         return err2;
2027                 }
2028                 read_unlock(&mrt_lock);
2029                 kfree_skb(skb);
2030                 return -ENODEV;
2031         }
2032
2033         read_lock(&mrt_lock);
2034         ip_mr_forward(net, mrt, skb, cache, local);
2035         read_unlock(&mrt_lock);
2036
2037         if (local)
2038                 return ip_local_deliver(skb);
2039
2040         return 0;
2041
2042 dont_forward:
2043         if (local)
2044                 return ip_local_deliver(skb);
2045         kfree_skb(skb);
2046         return 0;
2047 }
2048
2049 #ifdef CONFIG_IP_PIMSM
2050 /* called with rcu_read_lock() */
2051 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
2052                      unsigned int pimlen)
2053 {
2054         struct net_device *reg_dev = NULL;
2055         struct iphdr *encap;
2056
2057         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
2058         /*
2059          * Check that:
2060          * a. packet is really sent to a multicast group
2061          * b. packet is not a NULL-REGISTER
2062          * c. packet is not truncated
2063          */
2064         if (!ipv4_is_multicast(encap->daddr) ||
2065             encap->tot_len == 0 ||
2066             ntohs(encap->tot_len) + pimlen > skb->len)
2067                 return 1;
2068
2069         read_lock(&mrt_lock);
2070         if (mrt->mroute_reg_vif_num >= 0)
2071                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
2072         read_unlock(&mrt_lock);
2073
2074         if (!reg_dev)
2075                 return 1;
2076
2077         skb->mac_header = skb->network_header;
2078         skb_pull(skb, (u8 *)encap - skb->data);
2079         skb_reset_network_header(skb);
2080         skb->protocol = htons(ETH_P_IP);
2081         skb->ip_summed = CHECKSUM_NONE;
2082
2083         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
2084
2085         netif_rx(skb);
2086
2087         return NET_RX_SUCCESS;
2088 }
2089 #endif
2090
2091 #ifdef CONFIG_IP_PIMSM_V1
2092 /*
2093  * Handle IGMP messages of PIMv1
2094  */
2095
2096 int pim_rcv_v1(struct sk_buff *skb)
2097 {
2098         struct igmphdr *pim;
2099         struct net *net = dev_net(skb->dev);
2100         struct mr_table *mrt;
2101
2102         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2103                 goto drop;
2104
2105         pim = igmp_hdr(skb);
2106
2107         mrt = ipmr_rt_fib_lookup(net, skb);
2108         if (IS_ERR(mrt))
2109                 goto drop;
2110         if (!mrt->mroute_do_pim ||
2111             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2112                 goto drop;
2113
2114         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2115 drop:
2116                 kfree_skb(skb);
2117         }
2118         return 0;
2119 }
2120 #endif
2121
2122 #ifdef CONFIG_IP_PIMSM_V2
2123 static int pim_rcv(struct sk_buff *skb)
2124 {
2125         struct pimreghdr *pim;
2126         struct net *net = dev_net(skb->dev);
2127         struct mr_table *mrt;
2128
2129         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2130                 goto drop;
2131
2132         pim = (struct pimreghdr *)skb_transport_header(skb);
2133         if (pim->type != ((PIM_VERSION << 4) | (PIM_REGISTER)) ||
2134             (pim->flags & PIM_NULL_REGISTER) ||
2135             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2136              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2137                 goto drop;
2138
2139         mrt = ipmr_rt_fib_lookup(net, skb);
2140         if (IS_ERR(mrt))
2141                 goto drop;
2142         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2143 drop:
2144                 kfree_skb(skb);
2145         }
2146         return 0;
2147 }
2148 #endif
2149
2150 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2151                               struct mfc_cache *c, struct rtmsg *rtm)
2152 {
2153         int ct;
2154         struct rtnexthop *nhp;
2155         struct nlattr *mp_attr;
2156         struct rta_mfc_stats mfcs;
2157
2158         /* If cache is unresolved, don't try to parse IIF and OIF */
2159         if (c->mfc_parent >= MAXVIFS)
2160                 return -ENOENT;
2161
2162         if (VIF_EXISTS(mrt, c->mfc_parent) &&
2163             nla_put_u32(skb, RTA_IIF, mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
2164                 return -EMSGSIZE;
2165
2166         if (!(mp_attr = nla_nest_start(skb, RTA_MULTIPATH)))
2167                 return -EMSGSIZE;
2168
2169         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2170                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2171                         if (!(nhp = nla_reserve_nohdr(skb, sizeof(*nhp)))) {
2172                                 nla_nest_cancel(skb, mp_attr);
2173                                 return -EMSGSIZE;
2174                         }
2175
2176                         nhp->rtnh_flags = 0;
2177                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2178                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2179                         nhp->rtnh_len = sizeof(*nhp);
2180                 }
2181         }
2182
2183         nla_nest_end(skb, mp_attr);
2184
2185         mfcs.mfcs_packets = c->mfc_un.res.pkt;
2186         mfcs.mfcs_bytes = c->mfc_un.res.bytes;
2187         mfcs.mfcs_wrong_if = c->mfc_un.res.wrong_if;
2188         if (nla_put(skb, RTA_MFC_STATS, sizeof(mfcs), &mfcs) < 0)
2189                 return -EMSGSIZE;
2190
2191         rtm->rtm_type = RTN_MULTICAST;
2192         return 1;
2193 }
2194
2195 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2196                    __be32 saddr, __be32 daddr,
2197                    struct rtmsg *rtm, int nowait, u32 portid)
2198 {
2199         struct mfc_cache *cache;
2200         struct mr_table *mrt;
2201         int err;
2202
2203         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2204         if (!mrt)
2205                 return -ENOENT;
2206
2207         rcu_read_lock();
2208         cache = ipmr_cache_find(mrt, saddr, daddr);
2209         if (!cache && skb->dev) {
2210                 int vif = ipmr_find_vif(mrt, skb->dev);
2211
2212                 if (vif >= 0)
2213                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2214         }
2215         if (!cache) {
2216                 struct sk_buff *skb2;
2217                 struct iphdr *iph;
2218                 struct net_device *dev;
2219                 int vif = -1;
2220
2221                 if (nowait) {
2222                         rcu_read_unlock();
2223                         return -EAGAIN;
2224                 }
2225
2226                 dev = skb->dev;
2227                 read_lock(&mrt_lock);
2228                 if (dev)
2229                         vif = ipmr_find_vif(mrt, dev);
2230                 if (vif < 0) {
2231                         read_unlock(&mrt_lock);
2232                         rcu_read_unlock();
2233                         return -ENODEV;
2234                 }
2235                 skb2 = skb_clone(skb, GFP_ATOMIC);
2236                 if (!skb2) {
2237                         read_unlock(&mrt_lock);
2238                         rcu_read_unlock();
2239                         return -ENOMEM;
2240                 }
2241
2242                 NETLINK_CB(skb2).portid = portid;
2243                 skb_push(skb2, sizeof(struct iphdr));
2244                 skb_reset_network_header(skb2);
2245                 iph = ip_hdr(skb2);
2246                 iph->ihl = sizeof(struct iphdr) >> 2;
2247                 iph->saddr = saddr;
2248                 iph->daddr = daddr;
2249                 iph->version = 0;
2250                 err = ipmr_cache_unresolved(mrt, vif, skb2);
2251                 read_unlock(&mrt_lock);
2252                 rcu_read_unlock();
2253                 return err;
2254         }
2255
2256         read_lock(&mrt_lock);
2257         if (!nowait && (rtm->rtm_flags & RTM_F_NOTIFY))
2258                 cache->mfc_flags |= MFC_NOTIFY;
2259         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2260         read_unlock(&mrt_lock);
2261         rcu_read_unlock();
2262         return err;
2263 }
2264
2265 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2266                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2267                             int flags)
2268 {
2269         struct nlmsghdr *nlh;
2270         struct rtmsg *rtm;
2271         int err;
2272
2273         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2274         if (!nlh)
2275                 return -EMSGSIZE;
2276
2277         rtm = nlmsg_data(nlh);
2278         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2279         rtm->rtm_dst_len  = 32;
2280         rtm->rtm_src_len  = 32;
2281         rtm->rtm_tos      = 0;
2282         rtm->rtm_table    = mrt->id;
2283         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2284                 goto nla_put_failure;
2285         rtm->rtm_type     = RTN_MULTICAST;
2286         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2287         if (c->mfc_flags & MFC_STATIC)
2288                 rtm->rtm_protocol = RTPROT_STATIC;
2289         else
2290                 rtm->rtm_protocol = RTPROT_MROUTED;
2291         rtm->rtm_flags    = 0;
2292
2293         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2294             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2295                 goto nla_put_failure;
2296         err = __ipmr_fill_mroute(mrt, skb, c, rtm);
2297         /* do not break the dump if cache is unresolved */
2298         if (err < 0 && err != -ENOENT)
2299                 goto nla_put_failure;
2300
2301         nlmsg_end(skb, nlh);
2302         return 0;
2303
2304 nla_put_failure:
2305         nlmsg_cancel(skb, nlh);
2306         return -EMSGSIZE;
2307 }
2308
2309 static size_t mroute_msgsize(bool unresolved, int maxvif)
2310 {
2311         size_t len =
2312                 NLMSG_ALIGN(sizeof(struct rtmsg))
2313                 + nla_total_size(4)     /* RTA_TABLE */
2314                 + nla_total_size(4)     /* RTA_SRC */
2315                 + nla_total_size(4)     /* RTA_DST */
2316                 ;
2317
2318         if (!unresolved)
2319                 len = len
2320                       + nla_total_size(4)       /* RTA_IIF */
2321                       + nla_total_size(0)       /* RTA_MULTIPATH */
2322                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2323                                                 /* RTA_MFC_STATS */
2324                       + nla_total_size(sizeof(struct rta_mfc_stats))
2325                 ;
2326
2327         return len;
2328 }
2329
2330 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2331                                  int cmd)
2332 {
2333         struct net *net = read_pnet(&mrt->net);
2334         struct sk_buff *skb;
2335         int err = -ENOBUFS;
2336
2337         skb = nlmsg_new(mroute_msgsize(mfc->mfc_parent >= MAXVIFS, mrt->maxvif),
2338                         GFP_ATOMIC);
2339         if (!skb)
2340                 goto errout;
2341
2342         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2343         if (err < 0)
2344                 goto errout;
2345
2346         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2347         return;
2348
2349 errout:
2350         kfree_skb(skb);
2351         if (err < 0)
2352                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2353 }
2354
2355 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2356 {
2357         struct net *net = sock_net(skb->sk);
2358         struct mr_table *mrt;
2359         struct mfc_cache *mfc;
2360         unsigned int t = 0, s_t;
2361         unsigned int h = 0, s_h;
2362         unsigned int e = 0, s_e;
2363
2364         s_t = cb->args[0];
2365         s_h = cb->args[1];
2366         s_e = cb->args[2];
2367
2368         rcu_read_lock();
2369         ipmr_for_each_table(mrt, net) {
2370                 if (t < s_t)
2371                         goto next_table;
2372                 if (t > s_t)
2373                         s_h = 0;
2374                 for (h = s_h; h < MFC_LINES; h++) {
2375                         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2376                                 if (e < s_e)
2377                                         goto next_entry;
2378                                 if (ipmr_fill_mroute(mrt, skb,
2379                                                      NETLINK_CB(cb->skb).portid,
2380                                                      cb->nlh->nlmsg_seq,
2381                                                      mfc, RTM_NEWROUTE,
2382                                                      NLM_F_MULTI) < 0)
2383                                         goto done;
2384 next_entry:
2385                                 e++;
2386                         }
2387                         e = s_e = 0;
2388                 }
2389                 spin_lock_bh(&mfc_unres_lock);
2390                 list_for_each_entry(mfc, &mrt->mfc_unres_queue, list) {
2391                         if (e < s_e)
2392                                 goto next_entry2;
2393                         if (ipmr_fill_mroute(mrt, skb,
2394                                              NETLINK_CB(cb->skb).portid,
2395                                              cb->nlh->nlmsg_seq,
2396                                              mfc, RTM_NEWROUTE,
2397                                              NLM_F_MULTI) < 0) {
2398                                 spin_unlock_bh(&mfc_unres_lock);
2399                                 goto done;
2400                         }
2401 next_entry2:
2402                         e++;
2403                 }
2404                 spin_unlock_bh(&mfc_unres_lock);
2405                 e = s_e = 0;
2406                 s_h = 0;
2407 next_table:
2408                 t++;
2409         }
2410 done:
2411         rcu_read_unlock();
2412
2413         cb->args[2] = e;
2414         cb->args[1] = h;
2415         cb->args[0] = t;
2416
2417         return skb->len;
2418 }
2419
2420 #ifdef CONFIG_PROC_FS
2421 /*
2422  *      The /proc interfaces to multicast routing :
2423  *      /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2424  */
2425 struct ipmr_vif_iter {
2426         struct seq_net_private p;
2427         struct mr_table *mrt;
2428         int ct;
2429 };
2430
2431 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2432                                            struct ipmr_vif_iter *iter,
2433                                            loff_t pos)
2434 {
2435         struct mr_table *mrt = iter->mrt;
2436
2437         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2438                 if (!VIF_EXISTS(mrt, iter->ct))
2439                         continue;
2440                 if (pos-- == 0)
2441                         return &mrt->vif_table[iter->ct];
2442         }
2443         return NULL;
2444 }
2445
2446 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2447         __acquires(mrt_lock)
2448 {
2449         struct ipmr_vif_iter *iter = seq->private;
2450         struct net *net = seq_file_net(seq);
2451         struct mr_table *mrt;
2452
2453         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2454         if (!mrt)
2455                 return ERR_PTR(-ENOENT);
2456
2457         iter->mrt = mrt;
2458
2459         read_lock(&mrt_lock);
2460         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2461                 : SEQ_START_TOKEN;
2462 }
2463
2464 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2465 {
2466         struct ipmr_vif_iter *iter = seq->private;
2467         struct net *net = seq_file_net(seq);
2468         struct mr_table *mrt = iter->mrt;
2469
2470         ++*pos;
2471         if (v == SEQ_START_TOKEN)
2472                 return ipmr_vif_seq_idx(net, iter, 0);
2473
2474         while (++iter->ct < mrt->maxvif) {
2475                 if (!VIF_EXISTS(mrt, iter->ct))
2476                         continue;
2477                 return &mrt->vif_table[iter->ct];
2478         }
2479         return NULL;
2480 }
2481
2482 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2483         __releases(mrt_lock)
2484 {
2485         read_unlock(&mrt_lock);
2486 }
2487
2488 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2489 {
2490         struct ipmr_vif_iter *iter = seq->private;
2491         struct mr_table *mrt = iter->mrt;
2492
2493         if (v == SEQ_START_TOKEN) {
2494                 seq_puts(seq,
2495                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2496         } else {
2497                 const struct vif_device *vif = v;
2498                 const char *name =  vif->dev ? vif->dev->name : "none";
2499
2500                 seq_printf(seq,
2501                            "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2502                            vif - mrt->vif_table,
2503                            name, vif->bytes_in, vif->pkt_in,
2504                            vif->bytes_out, vif->pkt_out,
2505                            vif->flags, vif->local, vif->remote);
2506         }
2507         return 0;
2508 }
2509
2510 static const struct seq_operations ipmr_vif_seq_ops = {
2511         .start = ipmr_vif_seq_start,
2512         .next  = ipmr_vif_seq_next,
2513         .stop  = ipmr_vif_seq_stop,
2514         .show  = ipmr_vif_seq_show,
2515 };
2516
2517 static int ipmr_vif_open(struct inode *inode, struct file *file)
2518 {
2519         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2520                             sizeof(struct ipmr_vif_iter));
2521 }
2522
2523 static const struct file_operations ipmr_vif_fops = {
2524         .owner   = THIS_MODULE,
2525         .open    = ipmr_vif_open,
2526         .read    = seq_read,
2527         .llseek  = seq_lseek,
2528         .release = seq_release_net,
2529 };
2530
2531 struct ipmr_mfc_iter {
2532         struct seq_net_private p;
2533         struct mr_table *mrt;
2534         struct list_head *cache;
2535         int ct;
2536 };
2537
2538
2539 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2540                                           struct ipmr_mfc_iter *it, loff_t pos)
2541 {
2542         struct mr_table *mrt = it->mrt;
2543         struct mfc_cache *mfc;
2544
2545         rcu_read_lock();
2546         for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2547                 it->cache = &mrt->mfc_cache_array[it->ct];
2548                 list_for_each_entry_rcu(mfc, it->cache, list)
2549                         if (pos-- == 0)
2550                                 return mfc;
2551         }
2552         rcu_read_unlock();
2553
2554         spin_lock_bh(&mfc_unres_lock);
2555         it->cache = &mrt->mfc_unres_queue;
2556         list_for_each_entry(mfc, it->cache, list)
2557                 if (pos-- == 0)
2558                         return mfc;
2559         spin_unlock_bh(&mfc_unres_lock);
2560
2561         it->cache = NULL;
2562         return NULL;
2563 }
2564
2565
2566 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2567 {
2568         struct ipmr_mfc_iter *it = seq->private;
2569         struct net *net = seq_file_net(seq);
2570         struct mr_table *mrt;
2571
2572         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2573         if (!mrt)
2574                 return ERR_PTR(-ENOENT);
2575
2576         it->mrt = mrt;
2577         it->cache = NULL;
2578         it->ct = 0;
2579         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2580                 : SEQ_START_TOKEN;
2581 }
2582
2583 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2584 {
2585         struct mfc_cache *mfc = v;
2586         struct ipmr_mfc_iter *it = seq->private;
2587         struct net *net = seq_file_net(seq);
2588         struct mr_table *mrt = it->mrt;
2589
2590         ++*pos;
2591
2592         if (v == SEQ_START_TOKEN)
2593                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2594
2595         if (mfc->list.next != it->cache)
2596                 return list_entry(mfc->list.next, struct mfc_cache, list);
2597
2598         if (it->cache == &mrt->mfc_unres_queue)
2599                 goto end_of_list;
2600
2601         BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2602
2603         while (++it->ct < MFC_LINES) {
2604                 it->cache = &mrt->mfc_cache_array[it->ct];
2605                 if (list_empty(it->cache))
2606                         continue;
2607                 return list_first_entry(it->cache, struct mfc_cache, list);
2608         }
2609
2610         /* exhausted cache_array, show unresolved */
2611         rcu_read_unlock();
2612         it->cache = &mrt->mfc_unres_queue;
2613         it->ct = 0;
2614
2615         spin_lock_bh(&mfc_unres_lock);
2616         if (!list_empty(it->cache))
2617                 return list_first_entry(it->cache, struct mfc_cache, list);
2618
2619 end_of_list:
2620         spin_unlock_bh(&mfc_unres_lock);
2621         it->cache = NULL;
2622
2623         return NULL;
2624 }
2625
2626 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2627 {
2628         struct ipmr_mfc_iter *it = seq->private;
2629         struct mr_table *mrt = it->mrt;
2630
2631         if (it->cache == &mrt->mfc_unres_queue)
2632                 spin_unlock_bh(&mfc_unres_lock);
2633         else if (it->cache == &mrt->mfc_cache_array[it->ct])
2634                 rcu_read_unlock();
2635 }
2636
2637 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2638 {
2639         int n;
2640
2641         if (v == SEQ_START_TOKEN) {
2642                 seq_puts(seq,
2643                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2644         } else {
2645                 const struct mfc_cache *mfc = v;
2646                 const struct ipmr_mfc_iter *it = seq->private;
2647                 const struct mr_table *mrt = it->mrt;
2648
2649                 seq_printf(seq, "%08X %08X %-3hd",
2650                            (__force u32) mfc->mfc_mcastgrp,
2651                            (__force u32) mfc->mfc_origin,
2652                            mfc->mfc_parent);
2653
2654                 if (it->cache != &mrt->mfc_unres_queue) {
2655                         seq_printf(seq, " %8lu %8lu %8lu",
2656                                    mfc->mfc_un.res.pkt,
2657                                    mfc->mfc_un.res.bytes,
2658                                    mfc->mfc_un.res.wrong_if);
2659                         for (n = mfc->mfc_un.res.minvif;
2660                              n < mfc->mfc_un.res.maxvif; n++) {
2661                                 if (VIF_EXISTS(mrt, n) &&
2662                                     mfc->mfc_un.res.ttls[n] < 255)
2663                                         seq_printf(seq,
2664                                            " %2d:%-3d",
2665                                            n, mfc->mfc_un.res.ttls[n]);
2666                         }
2667                 } else {
2668                         /* unresolved mfc_caches don't contain
2669                          * pkt, bytes and wrong_if values
2670                          */
2671                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2672                 }
2673                 seq_putc(seq, '\n');
2674         }
2675         return 0;
2676 }
2677
2678 static const struct seq_operations ipmr_mfc_seq_ops = {
2679         .start = ipmr_mfc_seq_start,
2680         .next  = ipmr_mfc_seq_next,
2681         .stop  = ipmr_mfc_seq_stop,
2682         .show  = ipmr_mfc_seq_show,
2683 };
2684
2685 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2686 {
2687         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2688                             sizeof(struct ipmr_mfc_iter));
2689 }
2690
2691 static const struct file_operations ipmr_mfc_fops = {
2692         .owner   = THIS_MODULE,
2693         .open    = ipmr_mfc_open,
2694         .read    = seq_read,
2695         .llseek  = seq_lseek,
2696         .release = seq_release_net,
2697 };
2698 #endif
2699
2700 #ifdef CONFIG_IP_PIMSM_V2
2701 static const struct net_protocol pim_protocol = {
2702         .handler        =       pim_rcv,
2703         .netns_ok       =       1,
2704 };
2705 #endif
2706
2707
2708 /*
2709  *      Setup for IP multicast routing
2710  */
2711 static int __net_init ipmr_net_init(struct net *net)
2712 {
2713         int err;
2714
2715         err = ipmr_rules_init(net);
2716         if (err < 0)
2717                 goto fail;
2718
2719 #ifdef CONFIG_PROC_FS
2720         err = -ENOMEM;
2721         if (!proc_create("ip_mr_vif", 0, net->proc_net, &ipmr_vif_fops))
2722                 goto proc_vif_fail;
2723         if (!proc_create("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_fops))
2724                 goto proc_cache_fail;
2725 #endif
2726         return 0;
2727
2728 #ifdef CONFIG_PROC_FS
2729 proc_cache_fail:
2730         remove_proc_entry("ip_mr_vif", net->proc_net);
2731 proc_vif_fail:
2732         ipmr_rules_exit(net);
2733 #endif
2734 fail:
2735         return err;
2736 }
2737
2738 static void __net_exit ipmr_net_exit(struct net *net)
2739 {
2740 #ifdef CONFIG_PROC_FS
2741         remove_proc_entry("ip_mr_cache", net->proc_net);
2742         remove_proc_entry("ip_mr_vif", net->proc_net);
2743 #endif
2744         ipmr_rules_exit(net);
2745 }
2746
2747 static struct pernet_operations ipmr_net_ops = {
2748         .init = ipmr_net_init,
2749         .exit = ipmr_net_exit,
2750 };
2751
2752 int __init ip_mr_init(void)
2753 {
2754         int err;
2755
2756         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2757                                        sizeof(struct mfc_cache),
2758                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2759                                        NULL);
2760         if (!mrt_cachep)
2761                 return -ENOMEM;
2762
2763         err = register_pernet_subsys(&ipmr_net_ops);
2764         if (err)
2765                 goto reg_pernet_fail;
2766
2767         err = register_netdevice_notifier(&ip_mr_notifier);
2768         if (err)
2769                 goto reg_notif_fail;
2770 #ifdef CONFIG_IP_PIMSM_V2
2771         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2772                 pr_err("%s: can't add PIM protocol\n", __func__);
2773                 err = -EAGAIN;
2774                 goto add_proto_fail;
2775         }
2776 #endif
2777         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
2778                       NULL, ipmr_rtm_dumproute, NULL);
2779         return 0;
2780
2781 #ifdef CONFIG_IP_PIMSM_V2
2782 add_proto_fail:
2783         unregister_netdevice_notifier(&ip_mr_notifier);
2784 #endif
2785 reg_notif_fail:
2786         unregister_pernet_subsys(&ipmr_net_ops);
2787 reg_pernet_fail:
2788         kmem_cache_destroy(mrt_cachep);
2789         return err;
2790 }