Pileus Git - ~andy/linux/blob - 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/ipip.h>
  65 #include <net/checksum.h>
  66 #include <net/netlink.h>
  67 #include <net/fib_rules.h>
  68 #include <linux/netconf.h>
  69
  70 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
  71 #define CONFIG_IP_PIMSM 1
  72 #endif
  73
  74 struct mr_table {
  75         struct list_head        list;
  76 #ifdef CONFIG_NET_NS
  77         struct net              *net;
  78 #endif
  79         u32                     id;
  80         struct sock __rcu       *mroute_sk;
  81         struct timer_list       ipmr_expire_timer;
  82         struct list_head        mfc_unres_queue;
  83         struct list_head        mfc_cache_array[MFC_LINES];
  84         struct vif_device       vif_table[MAXVIFS];
  85         int                     maxvif;
  86         atomic_t                cache_resolve_queue_len;
  87         bool                    mroute_do_assert;
  88         bool                    mroute_do_pim;
  89 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
  90         int                     mroute_reg_vif_num;
  91 #endif
  92 };
  93
  94 struct ipmr_rule {
  95         struct fib_rule         common;
  96 };
  97
  98 struct ipmr_result {
  99         struct mr_table         *mrt;
 100 };
 101
 102 /* Big lock, protecting vif table, mrt cache and mroute socket state.
 103  * Note that the changes are semaphored via rtnl_lock.
 104  */
 105
 106 static DEFINE_RWLOCK(mrt_lock);
 107
 108 /*
 109  *      Multicast router control variables
 110  */
 111
 112 #define VIF_EXISTS(_mrt, _idx) ((_mrt)->vif_table[_idx].dev != NULL)
 113
 114 /* Special spinlock for queue of unresolved entries */
 115 static DEFINE_SPINLOCK(mfc_unres_lock);
 116
 117 /* We return to original Alan's scheme. Hash table of resolved
 118  * entries is changed only in process context and protected
 119  * with weak lock mrt_lock. Queue of unresolved entries is protected
 120  * with strong spinlock mfc_unres_lock.
 121  *
 122  * In this case data path is free of exclusive locks at all.
 123  */
 124
 125 static struct kmem_cache *mrt_cachep __read_mostly;
 126
 127 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 128 static void ipmr_free_table(struct mr_table *mrt);
 129
 130 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
 131                          struct sk_buff *skb, struct mfc_cache *cache,
 132                          int local);
 133 static int ipmr_cache_report(struct mr_table *mrt,
 134                              struct sk_buff *pkt, vifi_t vifi, int assert);
 135 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
 136                               struct mfc_cache *c, struct rtmsg *rtm);
 137 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
 138                                  int cmd);
 139 static void mroute_clean_tables(struct mr_table *mrt);
 140 static void ipmr_expire_process(unsigned long arg);
 141
 142 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 143 #define ipmr_for_each_table(mrt, net) \
 144         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
 145
 146 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 147 {
 148         struct mr_table *mrt;
 149
 150         ipmr_for_each_table(mrt, net) {
 151                 if (mrt->id == id)
 152                         return mrt;
 153         }
 154         return NULL;
 155 }
 156
 157 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 158                            struct mr_table **mrt)
 159 {
 160         struct ipmr_result res;
 161         struct fib_lookup_arg arg = { .result = &res, };
 162         int err;
 163
 164         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 165                                flowi4_to_flowi(flp4), 0, &arg);
 166         if (err < 0)
 167                 return err;
 168         *mrt = res.mrt;
 169         return 0;
 170 }
 171
 172 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 173                             int flags, struct fib_lookup_arg *arg)
 174 {
 175         struct ipmr_result *res = arg->result;
 176         struct mr_table *mrt;
 177
 178         switch (rule->action) {
 179         case FR_ACT_TO_TBL:
 180                 break;
 181         case FR_ACT_UNREACHABLE:
 182                 return -ENETUNREACH;
 183         case FR_ACT_PROHIBIT:
 184                 return -EACCES;
 185         case FR_ACT_BLACKHOLE:
 186         default:
 187                 return -EINVAL;
 188         }
 189
 190         mrt = ipmr_get_table(rule->fr_net, rule->table);
 191         if (mrt == NULL)
 192                 return -EAGAIN;
 193         res->mrt = mrt;
 194         return 0;
 195 }
 196
 197 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 198 {
 199         return 1;
 200 }
 201
 202 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 203         FRA_GENERIC_POLICY,
 204 };
 205
 206 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 207                                struct fib_rule_hdr *frh, struct nlattr **tb)
 208 {
 209         return 0;
 210 }
 211
 212 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 213                              struct nlattr **tb)
 214 {
 215         return 1;
 216 }
 217
 218 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 219                           struct fib_rule_hdr *frh)
 220 {
 221         frh->dst_len = 0;
 222         frh->src_len = 0;
 223         frh->tos     = 0;
 224         return 0;
 225 }
 226
 227 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
 228         .family         = RTNL_FAMILY_IPMR,
 229         .rule_size      = sizeof(struct ipmr_rule),
 230         .addr_size      = sizeof(u32),
 231         .action         = ipmr_rule_action,
 232         .match          = ipmr_rule_match,
 233         .configure      = ipmr_rule_configure,
 234         .compare        = ipmr_rule_compare,
 235         .default_pref   = fib_default_rule_pref,
 236         .fill           = ipmr_rule_fill,
 237         .nlgroup        = RTNLGRP_IPV4_RULE,
 238         .policy         = ipmr_rule_policy,
 239         .owner          = THIS_MODULE,
 240 };
 241
 242 static int __net_init ipmr_rules_init(struct net *net)
 243 {
 244         struct fib_rules_ops *ops;
 245         struct mr_table *mrt;
 246         int err;
 247
 248         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 249         if (IS_ERR(ops))
 250                 return PTR_ERR(ops);
 251
 252         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 253
 254         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 255         if (mrt == NULL) {
 256                 err = -ENOMEM;
 257                 goto err1;
 258         }
 259
 260         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 261         if (err < 0)
 262                 goto err2;
 263
 264         net->ipv4.mr_rules_ops = ops;
 265         return 0;
 266
 267 err2:
 268         kfree(mrt);
 269 err1:
 270         fib_rules_unregister(ops);
 271         return err;
 272 }
 273
 274 static void __net_exit ipmr_rules_exit(struct net *net)
 275 {
 276         struct mr_table *mrt, *next;
 277
 278         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 279                 list_del(&mrt->list);
 280                 ipmr_free_table(mrt);
 281         }
 282         fib_rules_unregister(net->ipv4.mr_rules_ops);
 283 }
 284 #else
 285 #define ipmr_for_each_table(mrt, net) \
 286         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 287
 288 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 289 {
 290         return net->ipv4.mrt;
 291 }
 292
 293 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 294                            struct mr_table **mrt)
 295 {
 296         *mrt = net->ipv4.mrt;
 297         return 0;
 298 }
 299
 300 static int __net_init ipmr_rules_init(struct net *net)
 301 {
 302         net->ipv4.mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 303         return net->ipv4.mrt ? 0 : -ENOMEM;
 304 }
 305
 306 static void __net_exit ipmr_rules_exit(struct net *net)
 307 {
 308         ipmr_free_table(net->ipv4.mrt);
 309 }
 310 #endif
 311
 312 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 313 {
 314         struct mr_table *mrt;
 315         unsigned int i;
 316
 317         mrt = ipmr_get_table(net, id);
 318         if (mrt != NULL)
 319                 return mrt;
 320
 321         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
 322         if (mrt == NULL)
 323                 return NULL;
 324         write_pnet(&mrt->net, net);
 325         mrt->id = id;
 326
 327         /* Forwarding cache */
 328         for (i = 0; i < MFC_LINES; i++)
 329                 INIT_LIST_HEAD(&mrt->mfc_cache_array[i]);
 330
 331         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
 332
 333         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
 334                     (unsigned long)mrt);
 335
 336 #ifdef CONFIG_IP_PIMSM
 337         mrt->mroute_reg_vif_num = -1;
 338 #endif
 339 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 340         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 341 #endif
 342         return mrt;
 343 }
 344
 345 static void ipmr_free_table(struct mr_table *mrt)
 346 {
 347         del_timer_sync(&mrt->ipmr_expire_timer);
 348         mroute_clean_tables(mrt);
 349         kfree(mrt);
 350 }
 351
 352 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 353
 354 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 355 {
 356         struct net *net = dev_net(dev);
 357
 358         dev_close(dev);
 359
 360         dev = __dev_get_by_name(net, "tunl0");
 361         if (dev) {
 362                 const struct net_device_ops *ops = dev->netdev_ops;
 363                 struct ifreq ifr;
 364                 struct ip_tunnel_parm p;
 365
 366                 memset(&p, 0, sizeof(p));
 367                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 368                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 369                 p.iph.version = 4;
 370                 p.iph.ihl = 5;
 371                 p.iph.protocol = IPPROTO_IPIP;
 372                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 373                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 374
 375                 if (ops->ndo_do_ioctl) {
 376                         mm_segment_t oldfs = get_fs();
 377
 378                         set_fs(KERNEL_DS);
 379                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 380                         set_fs(oldfs);
 381                 }
 382         }
 383 }
 384
 385 static
 386 struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 387 {
 388         struct net_device  *dev;
 389
 390         dev = __dev_get_by_name(net, "tunl0");
 391
 392         if (dev) {
 393                 const struct net_device_ops *ops = dev->netdev_ops;
 394                 int err;
 395                 struct ifreq ifr;
 396                 struct ip_tunnel_parm p;
 397                 struct in_device  *in_dev;
 398
 399                 memset(&p, 0, sizeof(p));
 400                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 401                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 402                 p.iph.version = 4;
 403                 p.iph.ihl = 5;
 404                 p.iph.protocol = IPPROTO_IPIP;
 405                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 406                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 407
 408                 if (ops->ndo_do_ioctl) {
 409                         mm_segment_t oldfs = get_fs();
 410
 411                         set_fs(KERNEL_DS);
 412                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 413                         set_fs(oldfs);
 414                 } else {
 415                         err = -EOPNOTSUPP;
 416                 }
 417                 dev = NULL;
 418
 419                 if (err == 0 &&
 420                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 421                         dev->flags |= IFF_MULTICAST;
 422
 423                         in_dev = __in_dev_get_rtnl(dev);
 424                         if (in_dev == NULL)
 425                                 goto failure;
 426
 427                         ipv4_devconf_setall(in_dev);
 428                         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 429
 430                         if (dev_open(dev))
 431                                 goto failure;
 432                         dev_hold(dev);
 433                 }
 434         }
 435         return dev;
 436
 437 failure:
 438         /* allow the register to be completed before unregistering. */
 439         rtnl_unlock();
 440         rtnl_lock();
 441
 442         unregister_netdevice(dev);
 443         return NULL;
 444 }
 445
 446 #ifdef CONFIG_IP_PIMSM
 447
 448 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 449 {
 450         struct net *net = dev_net(dev);
 451         struct mr_table *mrt;
 452         struct flowi4 fl4 = {
 453                 .flowi4_oif     = dev->ifindex,
 454                 .flowi4_iif     = skb->skb_iif,
 455                 .flowi4_mark    = skb->mark,
 456         };
 457         int err;
 458
 459         err = ipmr_fib_lookup(net, &fl4, &mrt);
 460         if (err < 0) {
 461                 kfree_skb(skb);
 462                 return err;
 463         }
 464
 465         read_lock(&mrt_lock);
 466         dev->stats.tx_bytes += skb->len;
 467         dev->stats.tx_packets++;
 468         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 469         read_unlock(&mrt_lock);
 470         kfree_skb(skb);
 471         return NETDEV_TX_OK;
 472 }
 473
 474 static const struct net_device_ops reg_vif_netdev_ops = {
 475         .ndo_start_xmit = reg_vif_xmit,
 476 };
 477
 478 static void reg_vif_setup(struct net_device *dev)
 479 {
 480         dev->type               = ARPHRD_PIMREG;
 481         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 482         dev->flags              = IFF_NOARP;
 483         dev->netdev_ops         = &reg_vif_netdev_ops,
 484         dev->destructor         = free_netdev;
 485         dev->features           |= NETIF_F_NETNS_LOCAL;
 486 }
 487
 488 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 489 {
 490         struct net_device *dev;
 491         struct in_device *in_dev;
 492         char name[IFNAMSIZ];
 493
 494         if (mrt->id == RT_TABLE_DEFAULT)
 495                 sprintf(name, "pimreg");
 496         else
 497                 sprintf(name, "pimreg%u", mrt->id);
 498
 499         dev = alloc_netdev(0, name, reg_vif_setup);
 500
 501         if (dev == NULL)
 502                 return NULL;
 503
 504         dev_net_set(dev, net);
 505
 506         if (register_netdevice(dev)) {
 507                 free_netdev(dev);
 508                 return NULL;
 509         }
 510         dev->iflink = 0;
 511
 512         rcu_read_lock();
 513         in_dev = __in_dev_get_rcu(dev);
 514         if (!in_dev) {
 515                 rcu_read_unlock();
 516                 goto failure;
 517         }
 518
 519         ipv4_devconf_setall(in_dev);
 520         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 521         rcu_read_unlock();
 522
 523         if (dev_open(dev))
 524                 goto failure;
 525
 526         dev_hold(dev);
 527
 528         return dev;
 529
 530 failure:
 531         /* allow the register to be completed before unregistering. */
 532         rtnl_unlock();
 533         rtnl_lock();
 534
 535         unregister_netdevice(dev);
 536         return NULL;
 537 }
 538 #endif
 539
 540 /**
 541  *      vif_delete - Delete a VIF entry
 542  *      @notify: Set to 1, if the caller is a notifier_call
 543  */
 544
 545 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 546                       struct list_head *head)
 547 {
 548         struct vif_device *v;
 549         struct net_device *dev;
 550         struct in_device *in_dev;
 551
 552         if (vifi < 0 || vifi >= mrt->maxvif)
 553                 return -EADDRNOTAVAIL;
 554
 555         v = &mrt->vif_table[vifi];
 556
 557         write_lock_bh(&mrt_lock);
 558         dev = v->dev;
 559         v->dev = NULL;
 560
 561         if (!dev) {
 562                 write_unlock_bh(&mrt_lock);
 563                 return -EADDRNOTAVAIL;
 564         }
 565
 566 #ifdef CONFIG_IP_PIMSM
 567         if (vifi == mrt->mroute_reg_vif_num)
 568                 mrt->mroute_reg_vif_num = -1;
 569 #endif
 570
 571         if (vifi + 1 == mrt->maxvif) {
 572                 int tmp;
 573
 574                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 575                         if (VIF_EXISTS(mrt, tmp))
 576                                 break;
 577                 }
 578                 mrt->maxvif = tmp+1;
 579         }
 580
 581         write_unlock_bh(&mrt_lock);
 582
 583         dev_set_allmulti(dev, -1);
 584
 585         in_dev = __in_dev_get_rtnl(dev);
 586         if (in_dev) {
 587                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 588                 inet_netconf_notify_devconf(dev_net(dev),
 589                                             NETCONFA_MC_FORWARDING,
 590                                             dev->ifindex, &in_dev->cnf);
 591                 ip_rt_multicast_event(in_dev);
 592         }
 593
 594         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 595                 unregister_netdevice_queue(dev, head);
 596
 597         dev_put(dev);
 598         return 0;
 599 }
 600
 601 static void ipmr_cache_free_rcu(struct rcu_head *head)
 602 {
 603         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
 604
 605         kmem_cache_free(mrt_cachep, c);
 606 }
 607
 608 static inline void ipmr_cache_free(struct mfc_cache *c)
 609 {
 610         call_rcu(&c->rcu, ipmr_cache_free_rcu);
 611 }
 612
 613 /* Destroy an unresolved cache entry, killing queued skbs
 614  * and reporting error to netlink readers.
 615  */
 616
 617 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 618 {
 619         struct net *net = read_pnet(&mrt->net);
 620         struct sk_buff *skb;
 621         struct nlmsgerr *e;
 622
 623         atomic_dec(&mrt->cache_resolve_queue_len);
 624
 625         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
 626                 if (ip_hdr(skb)->version == 0) {
 627                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 628                         nlh->nlmsg_type = NLMSG_ERROR;
 629                         nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
 630                         skb_trim(skb, nlh->nlmsg_len);
 631                         e = NLMSG_DATA(nlh);
 632                         e->error = -ETIMEDOUT;
 633                         memset(&e->msg, 0, sizeof(e->msg));
 634
 635                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 636                 } else {
 637                         kfree_skb(skb);
 638                 }
 639         }
 640
 641         ipmr_cache_free(c);
 642 }
 643
 644
 645 /* Timer process for the unresolved queue. */
 646
 647 static void ipmr_expire_process(unsigned long arg)
 648 {
 649         struct mr_table *mrt = (struct mr_table *)arg;
 650         unsigned long now;
 651         unsigned long expires;
 652         struct mfc_cache *c, *next;
 653
 654         if (!spin_trylock(&mfc_unres_lock)) {
 655                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 656                 return;
 657         }
 658
 659         if (list_empty(&mrt->mfc_unres_queue))
 660                 goto out;
 661
 662         now = jiffies;
 663         expires = 10*HZ;
 664
 665         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 666                 if (time_after(c->mfc_un.unres.expires, now)) {
 667                         unsigned long interval = c->mfc_un.unres.expires - now;
 668                         if (interval < expires)
 669                                 expires = interval;
 670                         continue;
 671                 }
 672
 673                 list_del(&c->list);
 674                 mroute_netlink_event(mrt, c, RTM_DELROUTE);
 675                 ipmr_destroy_unres(mrt, c);
 676         }
 677
 678         if (!list_empty(&mrt->mfc_unres_queue))
 679                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 680
 681 out:
 682         spin_unlock(&mfc_unres_lock);
 683 }
 684
 685 /* Fill oifs list. It is called under write locked mrt_lock. */
 686
 687 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
 688                                    unsigned char *ttls)
 689 {
 690         int vifi;
 691
 692         cache->mfc_un.res.minvif = MAXVIFS;
 693         cache->mfc_un.res.maxvif = 0;
 694         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 695
 696         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 697                 if (VIF_EXISTS(mrt, vifi) &&
 698                     ttls[vifi] && ttls[vifi] < 255) {
 699                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 700                         if (cache->mfc_un.res.minvif > vifi)
 701                                 cache->mfc_un.res.minvif = vifi;
 702                         if (cache->mfc_un.res.maxvif <= vifi)
 703                                 cache->mfc_un.res.maxvif = vifi + 1;
 704                 }
 705         }
 706 }
 707
 708 static int vif_add(struct net *net, struct mr_table *mrt,
 709                    struct vifctl *vifc, int mrtsock)
 710 {
 711         int vifi = vifc->vifc_vifi;
 712         struct vif_device *v = &mrt->vif_table[vifi];
 713         struct net_device *dev;
 714         struct in_device *in_dev;
 715         int err;
 716
 717         /* Is vif busy ? */
 718         if (VIF_EXISTS(mrt, vifi))
 719                 return -EADDRINUSE;
 720
 721         switch (vifc->vifc_flags) {
 722 #ifdef CONFIG_IP_PIMSM
 723         case VIFF_REGISTER:
 724                 /*
 725                  * Special Purpose VIF in PIM
 726                  * All the packets will be sent to the daemon
 727                  */
 728                 if (mrt->mroute_reg_vif_num >= 0)
 729                         return -EADDRINUSE;
 730                 dev = ipmr_reg_vif(net, mrt);
 731                 if (!dev)
 732                         return -ENOBUFS;
 733                 err = dev_set_allmulti(dev, 1);
 734                 if (err) {
 735                         unregister_netdevice(dev);
 736                         dev_put(dev);
 737                         return err;
 738                 }
 739                 break;
 740 #endif
 741         case VIFF_TUNNEL:
 742                 dev = ipmr_new_tunnel(net, vifc);
 743                 if (!dev)
 744                         return -ENOBUFS;
 745                 err = dev_set_allmulti(dev, 1);
 746                 if (err) {
 747                         ipmr_del_tunnel(dev, vifc);
 748                         dev_put(dev);
 749                         return err;
 750                 }
 751                 break;
 752
 753         case VIFF_USE_IFINDEX:
 754         case 0:
 755                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 756                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 757                         if (dev && __in_dev_get_rtnl(dev) == NULL) {
 758                                 dev_put(dev);
 759                                 return -EADDRNOTAVAIL;
 760                         }
 761                 } else {
 762                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 763                 }
 764                 if (!dev)
 765                         return -EADDRNOTAVAIL;
 766                 err = dev_set_allmulti(dev, 1);
 767                 if (err) {
 768                         dev_put(dev);
 769                         return err;
 770                 }
 771                 break;
 772         default:
 773                 return -EINVAL;
 774         }
 775
 776         in_dev = __in_dev_get_rtnl(dev);
 777         if (!in_dev) {
 778                 dev_put(dev);
 779                 return -EADDRNOTAVAIL;
 780         }
 781         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 782         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING, dev->ifindex,
 783                                     &in_dev->cnf);
 784         ip_rt_multicast_event(in_dev);
 785
 786         /* Fill in the VIF structures */
 787
 788         v->rate_limit = vifc->vifc_rate_limit;
 789         v->local = vifc->vifc_lcl_addr.s_addr;
 790         v->remote = vifc->vifc_rmt_addr.s_addr;
 791         v->flags = vifc->vifc_flags;
 792         if (!mrtsock)
 793                 v->flags |= VIFF_STATIC;
 794         v->threshold = vifc->vifc_threshold;
 795         v->bytes_in = 0;
 796         v->bytes_out = 0;
 797         v->pkt_in = 0;
 798         v->pkt_out = 0;
 799         v->link = dev->ifindex;
 800         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
 801                 v->link = dev->iflink;
 802
 803         /* And finish update writing critical data */
 804         write_lock_bh(&mrt_lock);
 805         v->dev = dev;
 806 #ifdef CONFIG_IP_PIMSM
 807         if (v->flags & VIFF_REGISTER)
 808                 mrt->mroute_reg_vif_num = vifi;
 809 #endif
 810         if (vifi+1 > mrt->maxvif)
 811                 mrt->maxvif = vifi+1;
 812         write_unlock_bh(&mrt_lock);
 813         return 0;
 814 }
 815
 816 /* called with rcu_read_lock() */
 817 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 818                                          __be32 origin,
 819                                          __be32 mcastgrp)
 820 {
 821         int line = MFC_HASH(mcastgrp, origin);
 822         struct mfc_cache *c;
 823
 824         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
 825                 if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
 826                         return c;
 827         }
 828         return NULL;
 829 }
 830
 831 /*
 832  *      Allocate a multicast cache entry
 833  */
 834 static struct mfc_cache *ipmr_cache_alloc(void)
 835 {
 836         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 837
 838         if (c)
 839                 c->mfc_un.res.minvif = MAXVIFS;
 840         return c;
 841 }
 842
 843 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 844 {
 845         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 846
 847         if (c) {
 848                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
 849                 c->mfc_un.unres.expires = jiffies + 10*HZ;
 850         }
 851         return c;
 852 }
 853
 854 /*
 855  *      A cache entry has gone into a resolved state from queued
 856  */
 857
 858 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 859                                struct mfc_cache *uc, struct mfc_cache *c)
 860 {
 861         struct sk_buff *skb;
 862         struct nlmsgerr *e;
 863
 864         /* Play the pending entries through our router */
 865
 866         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
 867                 if (ip_hdr(skb)->version == 0) {
 868                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 869
 870                         if (__ipmr_fill_mroute(mrt, skb, c, NLMSG_DATA(nlh)) > 0) {
 871                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 872                                                  (u8 *)nlh;
 873                         } else {
 874                                 nlh->nlmsg_type = NLMSG_ERROR;
 875                                 nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
 876                                 skb_trim(skb, nlh->nlmsg_len);
 877                                 e = NLMSG_DATA(nlh);
 878                                 e->error = -EMSGSIZE;
 879                                 memset(&e->msg, 0, sizeof(e->msg));
 880                         }
 881
 882                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 883                 } else {
 884                         ip_mr_forward(net, mrt, skb, c, 0);
 885                 }
 886         }
 887 }
 888
 889 /*
 890  *      Bounce a cache query up to mrouted. We could use netlink for this but mrouted
 891  *      expects the following bizarre scheme.
 892  *
 893  *      Called under mrt_lock.
 894  */
 895
 896 static int ipmr_cache_report(struct mr_table *mrt,
 897                              struct sk_buff *pkt, vifi_t vifi, int assert)
 898 {
 899         struct sk_buff *skb;
 900         const int ihl = ip_hdrlen(pkt);
 901         struct igmphdr *igmp;
 902         struct igmpmsg *msg;
 903         struct sock *mroute_sk;
 904         int ret;
 905
 906 #ifdef CONFIG_IP_PIMSM
 907         if (assert == IGMPMSG_WHOLEPKT)
 908                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
 909         else
 910 #endif
 911                 skb = alloc_skb(128, GFP_ATOMIC);
 912
 913         if (!skb)
 914                 return -ENOBUFS;
 915
 916 #ifdef CONFIG_IP_PIMSM
 917         if (assert == IGMPMSG_WHOLEPKT) {
 918                 /* Ugly, but we have no choice with this interface.
 919                  * Duplicate old header, fix ihl, length etc.
 920                  * And all this only to mangle msg->im_msgtype and
 921                  * to set msg->im_mbz to "mbz" :-)
 922                  */
 923                 skb_push(skb, sizeof(struct iphdr));
 924                 skb_reset_network_header(skb);
 925                 skb_reset_transport_header(skb);
 926                 msg = (struct igmpmsg *)skb_network_header(skb);
 927                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
 928                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
 929                 msg->im_mbz = 0;
 930                 msg->im_vif = mrt->mroute_reg_vif_num;
 931                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
 932                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
 933                                              sizeof(struct iphdr));
 934         } else
 935 #endif
 936         {
 937
 938         /* Copy the IP header */
 939
 940         skb->network_header = skb->tail;
 941         skb_put(skb, ihl);
 942         skb_copy_to_linear_data(skb, pkt->data, ihl);
 943         ip_hdr(skb)->protocol = 0;      /* Flag to the kernel this is a route add */
 944         msg = (struct igmpmsg *)skb_network_header(skb);
 945         msg->im_vif = vifi;
 946         skb_dst_set(skb, dst_clone(skb_dst(pkt)));
 947
 948         /* Add our header */
 949
 950         igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
 951         igmp->type      =
 952         msg->im_msgtype = assert;
 953         igmp->code      = 0;
 954         ip_hdr(skb)->tot_len = htons(skb->len);         /* Fix the length */
 955         skb->transport_header = skb->network_header;
 956         }
 957
 958         rcu_read_lock();
 959         mroute_sk = rcu_dereference(mrt->mroute_sk);
 960         if (mroute_sk == NULL) {
 961                 rcu_read_unlock();
 962                 kfree_skb(skb);
 963                 return -EINVAL;
 964         }
 965
 966         /* Deliver to mrouted */
 967
 968         ret = sock_queue_rcv_skb(mroute_sk, skb);
 969         rcu_read_unlock();
 970         if (ret < 0) {
 971                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
 972                 kfree_skb(skb);
 973         }
 974
 975         return ret;
 976 }
 977
 978 /*
 979  *      Queue a packet for resolution. It gets locked cache entry!
 980  */
 981
 982 static int
 983 ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi, struct sk_buff *skb)
 984 {
 985         bool found = false;
 986         int err;
 987         struct mfc_cache *c;
 988         const struct iphdr *iph = ip_hdr(skb);
 989
 990         spin_lock_bh(&mfc_unres_lock);
 991         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
 992                 if (c->mfc_mcastgrp == iph->daddr &&
 993                     c->mfc_origin == iph->saddr) {
 994                         found = true;
 995                         break;
 996                 }
 997         }
 998
 999         if (!found) {
1000                 /* Create a new entry if allowable */
1001
1002                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1003                     (c = ipmr_cache_alloc_unres()) == NULL) {
1004                         spin_unlock_bh(&mfc_unres_lock);
1005
1006                         kfree_skb(skb);
1007                         return -ENOBUFS;
1008                 }
1009
1010                 /* Fill in the new cache entry */
1011
1012                 c->mfc_parent   = -1;
1013                 c->mfc_origin   = iph->saddr;
1014                 c->mfc_mcastgrp = iph->daddr;
1015
1016                 /* Reflect first query at mrouted. */
1017
1018                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1019                 if (err < 0) {
1020                         /* If the report failed throw the cache entry
1021                            out - Brad Parker
1022                          */
1023                         spin_unlock_bh(&mfc_unres_lock);
1024
1025                         ipmr_cache_free(c);
1026                         kfree_skb(skb);
1027                         return err;
1028                 }
1029
1030                 atomic_inc(&mrt->cache_resolve_queue_len);
1031                 list_add(&c->list, &mrt->mfc_unres_queue);
1032                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1033
1034                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1035                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1036         }
1037
1038         /* See if we can append the packet */
1039
1040         if (c->mfc_un.unres.unresolved.qlen > 3) {
1041                 kfree_skb(skb);
1042                 err = -ENOBUFS;
1043         } else {
1044                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1045                 err = 0;
1046         }
1047
1048         spin_unlock_bh(&mfc_unres_lock);
1049         return err;
1050 }
1051
1052 /*
1053  *      MFC cache manipulation by user space mroute daemon
1054  */
1055
1056 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc)
1057 {
1058         int line;
1059         struct mfc_cache *c, *next;
1060
1061         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1062
1063         list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1064                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1065                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1066                         list_del_rcu(&c->list);
1067                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1068                         ipmr_cache_free(c);
1069                         return 0;
1070                 }
1071         }
1072         return -ENOENT;
1073 }
1074
1075 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1076                         struct mfcctl *mfc, int mrtsock)
1077 {
1078         bool found = false;
1079         int line;
1080         struct mfc_cache *uc, *c;
1081
1082         if (mfc->mfcc_parent >= MAXVIFS)
1083                 return -ENFILE;
1084
1085         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1086
1087         list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1088                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1089                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1090                         found = true;
1091                         break;
1092                 }
1093         }
1094
1095         if (found) {
1096                 write_lock_bh(&mrt_lock);
1097                 c->mfc_parent = mfc->mfcc_parent;
1098                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1099                 if (!mrtsock)
1100                         c->mfc_flags |= MFC_STATIC;
1101                 write_unlock_bh(&mrt_lock);
1102                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1103                 return 0;
1104         }
1105
1106         if (!ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1107                 return -EINVAL;
1108
1109         c = ipmr_cache_alloc();
1110         if (c == NULL)
1111                 return -ENOMEM;
1112
1113         c->mfc_origin = mfc->mfcc_origin.s_addr;
1114         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1115         c->mfc_parent = mfc->mfcc_parent;
1116         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1117         if (!mrtsock)
1118                 c->mfc_flags |= MFC_STATIC;
1119
1120         list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1121
1122         /*
1123          *      Check to see if we resolved a queued list. If so we
1124          *      need to send on the frames and tidy up.
1125          */
1126         found = false;
1127         spin_lock_bh(&mfc_unres_lock);
1128         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1129                 if (uc->mfc_origin == c->mfc_origin &&
1130                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1131                         list_del(&uc->list);
1132                         atomic_dec(&mrt->cache_resolve_queue_len);
1133                         found = true;
1134                         break;
1135                 }
1136         }
1137         if (list_empty(&mrt->mfc_unres_queue))
1138                 del_timer(&mrt->ipmr_expire_timer);
1139         spin_unlock_bh(&mfc_unres_lock);
1140
1141         if (found) {
1142                 ipmr_cache_resolve(net, mrt, uc, c);
1143                 ipmr_cache_free(uc);
1144         }
1145         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1146         return 0;
1147 }
1148
1149 /*
1150  *      Close the multicast socket, and clear the vif tables etc
1151  */
1152
1153 static void mroute_clean_tables(struct mr_table *mrt)
1154 {
1155         int i;
1156         LIST_HEAD(list);
1157         struct mfc_cache *c, *next;
1158
1159         /* Shut down all active vif entries */
1160
1161         for (i = 0; i < mrt->maxvif; i++) {
1162                 if (!(mrt->vif_table[i].flags & VIFF_STATIC))
1163                         vif_delete(mrt, i, 0, &list);
1164         }
1165         unregister_netdevice_many(&list);
1166
1167         /* Wipe the cache */
1168
1169         for (i = 0; i < MFC_LINES; i++) {
1170                 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1171                         if (c->mfc_flags & MFC_STATIC)
1172                                 continue;
1173                         list_del_rcu(&c->list);
1174                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1175                         ipmr_cache_free(c);
1176                 }
1177         }
1178
1179         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1180                 spin_lock_bh(&mfc_unres_lock);
1181                 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1182                         list_del(&c->list);
1183                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1184                         ipmr_destroy_unres(mrt, c);
1185                 }
1186                 spin_unlock_bh(&mfc_unres_lock);
1187         }
1188 }
1189
1190 /* called from ip_ra_control(), before an RCU grace period,
1191  * we dont need to call synchronize_rcu() here
1192  */
1193 static void mrtsock_destruct(struct sock *sk)
1194 {
1195         struct net *net = sock_net(sk);
1196         struct mr_table *mrt;
1197
1198         rtnl_lock();
1199         ipmr_for_each_table(mrt, net) {
1200                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1201                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1202                         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1203                                                     NETCONFA_IFINDEX_ALL,
1204                                                     net->ipv4.devconf_all);
1205                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1206                         mroute_clean_tables(mrt);
1207                 }
1208         }
1209         rtnl_unlock();
1210 }
1211
1212 /*
1213  *      Socket options and virtual interface manipulation. The whole
1214  *      virtual interface system is a complete heap, but unfortunately
1215  *      that's how BSD mrouted happens to think. Maybe one day with a proper
1216  *      MOSPF/PIM router set up we can clean this up.
1217  */
1218
1219 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval, unsigned int optlen)
1220 {
1221         int ret;
1222         struct vifctl vif;
1223         struct mfcctl mfc;
1224         struct net *net = sock_net(sk);
1225         struct mr_table *mrt;
1226
1227         if (sk->sk_type != SOCK_RAW ||
1228             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1229                 return -EOPNOTSUPP;
1230
1231         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1232         if (mrt == NULL)
1233                 return -ENOENT;
1234
1235         if (optname != MRT_INIT) {
1236                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1237                     !ns_capable(net->user_ns, CAP_NET_ADMIN))
1238                         return -EACCES;
1239         }
1240
1241         switch (optname) {
1242         case MRT_INIT:
1243                 if (optlen != sizeof(int))
1244                         return -EINVAL;
1245
1246                 rtnl_lock();
1247                 if (rtnl_dereference(mrt->mroute_sk)) {
1248                         rtnl_unlock();
1249                         return -EADDRINUSE;
1250                 }
1251
1252                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1253                 if (ret == 0) {
1254                         rcu_assign_pointer(mrt->mroute_sk, 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                 }
1260                 rtnl_unlock();
1261                 return ret;
1262         case MRT_DONE:
1263                 if (sk != rcu_access_pointer(mrt->mroute_sk))
1264                         return -EACCES;
1265                 return ip_ra_control(sk, 0, NULL);
1266         case MRT_ADD_VIF:
1267         case MRT_DEL_VIF:
1268                 if (optlen != sizeof(vif))
1269                         return -EINVAL;
1270                 if (copy_from_user(&vif, optval, sizeof(vif)))
1271                         return -EFAULT;
1272                 if (vif.vifc_vifi >= MAXVIFS)
1273                         return -ENFILE;
1274                 rtnl_lock();
1275                 if (optname == MRT_ADD_VIF) {
1276                         ret = vif_add(net, mrt, &vif,
1277                                       sk == rtnl_dereference(mrt->mroute_sk));
1278                 } else {
1279                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1280                 }
1281                 rtnl_unlock();
1282                 return ret;
1283
1284                 /*
1285                  *      Manipulate the forwarding caches. These live
1286                  *      in a sort of kernel/user symbiosis.
1287                  */
1288         case MRT_ADD_MFC:
1289         case MRT_DEL_MFC:
1290                 if (optlen != sizeof(mfc))
1291                         return -EINVAL;
1292                 if (copy_from_user(&mfc, optval, sizeof(mfc)))
1293                         return -EFAULT;
1294                 rtnl_lock();
1295                 if (optname == MRT_DEL_MFC)
1296                         ret = ipmr_mfc_delete(mrt, &mfc);
1297                 else
1298                         ret = ipmr_mfc_add(net, mrt, &mfc,
1299                                            sk == rtnl_dereference(mrt->mroute_sk));
1300                 rtnl_unlock();
1301                 return ret;
1302                 /*
1303                  *      Control PIM assert.
1304                  */
1305         case MRT_ASSERT:
1306         {
1307                 int v;
1308                 if (optlen != sizeof(v))
1309                         return -EINVAL;
1310                 if (get_user(v, (int __user *)optval))
1311                         return -EFAULT;
1312                 mrt->mroute_do_assert = v;
1313                 return 0;
1314         }
1315 #ifdef CONFIG_IP_PIMSM
1316         case MRT_PIM:
1317         {
1318                 int v;
1319
1320                 if (optlen != sizeof(v))
1321                         return -EINVAL;
1322                 if (get_user(v, (int __user *)optval))
1323                         return -EFAULT;
1324                 v = !!v;
1325
1326                 rtnl_lock();
1327                 ret = 0;
1328                 if (v != mrt->mroute_do_pim) {
1329                         mrt->mroute_do_pim = v;
1330                         mrt->mroute_do_assert = v;
1331                 }
1332                 rtnl_unlock();
1333                 return ret;
1334         }
1335 #endif
1336 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
1337         case MRT_TABLE:
1338         {
1339                 u32 v;
1340
1341                 if (optlen != sizeof(u32))
1342                         return -EINVAL;
1343                 if (get_user(v, (u32 __user *)optval))
1344                         return -EFAULT;
1345
1346                 /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
1347                 if (v != RT_TABLE_DEFAULT && v >= 1000000000)
1348                         return -EINVAL;
1349
1350                 rtnl_lock();
1351                 ret = 0;
1352                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1353                         ret = -EBUSY;
1354                 } else {
1355                         if (!ipmr_new_table(net, v))
1356                                 ret = -ENOMEM;
1357                         else
1358                                 raw_sk(sk)->ipmr_table = v;
1359                 }
1360                 rtnl_unlock();
1361                 return ret;
1362         }
1363 #endif
1364         /*
1365          *      Spurious command, or MRT_VERSION which you cannot
1366          *      set.
1367          */
1368         default:
1369                 return -ENOPROTOOPT;
1370         }
1371 }
1372
1373 /*
1374  *      Getsock opt support for the multicast routing system.
1375  */
1376
1377 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1378 {
1379         int olr;
1380         int val;
1381         struct net *net = sock_net(sk);
1382         struct mr_table *mrt;
1383
1384         if (sk->sk_type != SOCK_RAW ||
1385             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1386                 return -EOPNOTSUPP;
1387
1388         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1389         if (mrt == NULL)
1390                 return -ENOENT;
1391
1392         if (optname != MRT_VERSION &&
1393 #ifdef CONFIG_IP_PIMSM
1394            optname != MRT_PIM &&
1395 #endif
1396            optname != MRT_ASSERT)
1397                 return -ENOPROTOOPT;
1398
1399         if (get_user(olr, optlen))
1400                 return -EFAULT;
1401
1402         olr = min_t(unsigned int, olr, sizeof(int));
1403         if (olr < 0)
1404                 return -EINVAL;
1405
1406         if (put_user(olr, optlen))
1407                 return -EFAULT;
1408         if (optname == MRT_VERSION)
1409                 val = 0x0305;
1410 #ifdef CONFIG_IP_PIMSM
1411         else if (optname == MRT_PIM)
1412                 val = mrt->mroute_do_pim;
1413 #endif
1414         else
1415                 val = mrt->mroute_do_assert;
1416         if (copy_to_user(optval, &val, olr))
1417                 return -EFAULT;
1418         return 0;
1419 }
1420
1421 /*
1422  *      The IP multicast ioctl support routines.
1423  */
1424
1425 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1426 {
1427         struct sioc_sg_req sr;
1428         struct sioc_vif_req vr;
1429         struct vif_device *vif;
1430         struct mfc_cache *c;
1431         struct net *net = sock_net(sk);
1432         struct mr_table *mrt;
1433
1434         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1435         if (mrt == NULL)
1436                 return -ENOENT;
1437
1438         switch (cmd) {
1439         case SIOCGETVIFCNT:
1440                 if (copy_from_user(&vr, arg, sizeof(vr)))
1441                         return -EFAULT;
1442                 if (vr.vifi >= mrt->maxvif)
1443                         return -EINVAL;
1444                 read_lock(&mrt_lock);
1445                 vif = &mrt->vif_table[vr.vifi];
1446                 if (VIF_EXISTS(mrt, vr.vifi)) {
1447                         vr.icount = vif->pkt_in;
1448                         vr.ocount = vif->pkt_out;
1449                         vr.ibytes = vif->bytes_in;
1450                         vr.obytes = vif->bytes_out;
1451                         read_unlock(&mrt_lock);
1452
1453                         if (copy_to_user(arg, &vr, sizeof(vr)))
1454                                 return -EFAULT;
1455                         return 0;
1456                 }
1457                 read_unlock(&mrt_lock);
1458                 return -EADDRNOTAVAIL;
1459         case SIOCGETSGCNT:
1460                 if (copy_from_user(&sr, arg, sizeof(sr)))
1461                         return -EFAULT;
1462
1463                 rcu_read_lock();
1464                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1465                 if (c) {
1466                         sr.pktcnt = c->mfc_un.res.pkt;
1467                         sr.bytecnt = c->mfc_un.res.bytes;
1468                         sr.wrong_if = c->mfc_un.res.wrong_if;
1469                         rcu_read_unlock();
1470
1471                         if (copy_to_user(arg, &sr, sizeof(sr)))
1472                                 return -EFAULT;
1473                         return 0;
1474                 }
1475                 rcu_read_unlock();
1476                 return -EADDRNOTAVAIL;
1477         default:
1478                 return -ENOIOCTLCMD;
1479         }
1480 }
1481
1482 #ifdef CONFIG_COMPAT
1483 struct compat_sioc_sg_req {
1484         struct in_addr src;
1485         struct in_addr grp;
1486         compat_ulong_t pktcnt;
1487         compat_ulong_t bytecnt;
1488         compat_ulong_t wrong_if;
1489 };
1490
1491 struct compat_sioc_vif_req {
1492         vifi_t  vifi;           /* Which iface */
1493         compat_ulong_t icount;
1494         compat_ulong_t ocount;
1495         compat_ulong_t ibytes;
1496         compat_ulong_t obytes;
1497 };
1498
1499 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1500 {
1501         struct compat_sioc_sg_req sr;
1502         struct compat_sioc_vif_req vr;
1503         struct vif_device *vif;
1504         struct mfc_cache *c;
1505         struct net *net = sock_net(sk);
1506         struct mr_table *mrt;
1507
1508         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1509         if (mrt == NULL)
1510                 return -ENOENT;
1511
1512         switch (cmd) {
1513         case SIOCGETVIFCNT:
1514                 if (copy_from_user(&vr, arg, sizeof(vr)))
1515                         return -EFAULT;
1516                 if (vr.vifi >= mrt->maxvif)
1517                         return -EINVAL;
1518                 read_lock(&mrt_lock);
1519                 vif = &mrt->vif_table[vr.vifi];
1520                 if (VIF_EXISTS(mrt, vr.vifi)) {
1521                         vr.icount = vif->pkt_in;
1522                         vr.ocount = vif->pkt_out;
1523                         vr.ibytes = vif->bytes_in;
1524                         vr.obytes = vif->bytes_out;
1525                         read_unlock(&mrt_lock);
1526
1527                         if (copy_to_user(arg, &vr, sizeof(vr)))
1528                                 return -EFAULT;
1529                         return 0;
1530                 }
1531                 read_unlock(&mrt_lock);
1532                 return -EADDRNOTAVAIL;
1533         case SIOCGETSGCNT:
1534                 if (copy_from_user(&sr, arg, sizeof(sr)))
1535                         return -EFAULT;
1536
1537                 rcu_read_lock();
1538                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1539                 if (c) {
1540                         sr.pktcnt = c->mfc_un.res.pkt;
1541                         sr.bytecnt = c->mfc_un.res.bytes;
1542                         sr.wrong_if = c->mfc_un.res.wrong_if;
1543                         rcu_read_unlock();
1544
1545                         if (copy_to_user(arg, &sr, sizeof(sr)))
1546                                 return -EFAULT;
1547                         return 0;
1548                 }
1549                 rcu_read_unlock();
1550                 return -EADDRNOTAVAIL;
1551         default:
1552                 return -ENOIOCTLCMD;
1553         }
1554 }
1555 #endif
1556
1557
1558 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1559 {
1560         struct net_device *dev = ptr;
1561         struct net *net = dev_net(dev);
1562         struct mr_table *mrt;
1563         struct vif_device *v;
1564         int ct;
1565
1566         if (event != NETDEV_UNREGISTER)
1567                 return NOTIFY_DONE;
1568
1569         ipmr_for_each_table(mrt, net) {
1570                 v = &mrt->vif_table[0];
1571                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1572                         if (v->dev == dev)
1573                                 vif_delete(mrt, ct, 1, NULL);
1574                 }
1575         }
1576         return NOTIFY_DONE;
1577 }
1578
1579
1580 static struct notifier_block ip_mr_notifier = {
1581         .notifier_call = ipmr_device_event,
1582 };
1583
1584 /*
1585  *      Encapsulate a packet by attaching a valid IPIP header to it.
1586  *      This avoids tunnel drivers and other mess and gives us the speed so
1587  *      important for multicast video.
1588  */
1589
1590 static void ip_encap(struct sk_buff *skb, __be32 saddr, __be32 daddr)
1591 {
1592         struct iphdr *iph;
1593         const struct iphdr *old_iph = ip_hdr(skb);
1594
1595         skb_push(skb, sizeof(struct iphdr));
1596         skb->transport_header = skb->network_header;
1597         skb_reset_network_header(skb);
1598         iph = ip_hdr(skb);
1599
1600         iph->version    =       4;
1601         iph->tos        =       old_iph->tos;
1602         iph->ttl        =       old_iph->ttl;
1603         iph->frag_off   =       0;
1604         iph->daddr      =       daddr;
1605         iph->saddr      =       saddr;
1606         iph->protocol   =       IPPROTO_IPIP;
1607         iph->ihl        =       5;
1608         iph->tot_len    =       htons(skb->len);
1609         ip_select_ident(iph, skb_dst(skb), NULL);
1610         ip_send_check(iph);
1611
1612         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1613         nf_reset(skb);
1614 }
1615
1616 static inline int ipmr_forward_finish(struct sk_buff *skb)
1617 {
1618         struct ip_options *opt = &(IPCB(skb)->opt);
1619
1620         IP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTFORWDATAGRAMS);
1621         IP_ADD_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTOCTETS, skb->len);
1622
1623         if (unlikely(opt->optlen))
1624                 ip_forward_options(skb);
1625
1626         return dst_output(skb);
1627 }
1628
1629 /*
1630  *      Processing handlers for ipmr_forward
1631  */
1632
1633 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1634                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1635 {
1636         const struct iphdr *iph = ip_hdr(skb);
1637         struct vif_device *vif = &mrt->vif_table[vifi];
1638         struct net_device *dev;
1639         struct rtable *rt;
1640         struct flowi4 fl4;
1641         int    encap = 0;
1642
1643         if (vif->dev == NULL)
1644                 goto out_free;
1645
1646 #ifdef CONFIG_IP_PIMSM
1647         if (vif->flags & VIFF_REGISTER) {
1648                 vif->pkt_out++;
1649                 vif->bytes_out += skb->len;
1650                 vif->dev->stats.tx_bytes += skb->len;
1651                 vif->dev->stats.tx_packets++;
1652                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1653                 goto out_free;
1654         }
1655 #endif
1656
1657         if (vif->flags & VIFF_TUNNEL) {
1658                 rt = ip_route_output_ports(net, &fl4, NULL,
1659                                            vif->remote, vif->local,
1660                                            0, 0,
1661                                            IPPROTO_IPIP,
1662                                            RT_TOS(iph->tos), vif->link);
1663                 if (IS_ERR(rt))
1664                         goto out_free;
1665                 encap = sizeof(struct iphdr);
1666         } else {
1667                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1668                                            0, 0,
1669                                            IPPROTO_IPIP,
1670                                            RT_TOS(iph->tos), vif->link);
1671                 if (IS_ERR(rt))
1672                         goto out_free;
1673         }
1674
1675         dev = rt->dst.dev;
1676
1677         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1678                 /* Do not fragment multicasts. Alas, IPv4 does not
1679                  * allow to send ICMP, so that packets will disappear
1680                  * to blackhole.
1681                  */
1682
1683                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
1684                 ip_rt_put(rt);
1685                 goto out_free;
1686         }
1687
1688         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1689
1690         if (skb_cow(skb, encap)) {
1691                 ip_rt_put(rt);
1692                 goto out_free;
1693         }
1694
1695         vif->pkt_out++;
1696         vif->bytes_out += skb->len;
1697
1698         skb_dst_drop(skb);
1699         skb_dst_set(skb, &rt->dst);
1700         ip_decrease_ttl(ip_hdr(skb));
1701
1702         /* FIXME: forward and output firewalls used to be called here.
1703          * What do we do with netfilter? -- RR
1704          */
1705         if (vif->flags & VIFF_TUNNEL) {
1706                 ip_encap(skb, vif->local, vif->remote);
1707                 /* FIXME: extra output firewall step used to be here. --RR */
1708                 vif->dev->stats.tx_packets++;
1709                 vif->dev->stats.tx_bytes += skb->len;
1710         }
1711
1712         IPCB(skb)->flags |= IPSKB_FORWARDED;
1713
1714         /*
1715          * RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1716          * not only before forwarding, but after forwarding on all output
1717          * interfaces. It is clear, if mrouter runs a multicasting
1718          * program, it should receive packets not depending to what interface
1719          * program is joined.
1720          * If we will not make it, the program will have to join on all
1721          * interfaces. On the other hand, multihoming host (or router, but
1722          * not mrouter) cannot join to more than one interface - it will
1723          * result in receiving multiple packets.
1724          */
1725         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD, skb, skb->dev, dev,
1726                 ipmr_forward_finish);
1727         return;
1728
1729 out_free:
1730         kfree_skb(skb);
1731 }
1732
1733 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1734 {
1735         int ct;
1736
1737         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1738                 if (mrt->vif_table[ct].dev == dev)
1739                         break;
1740         }
1741         return ct;
1742 }
1743
1744 /* "local" means that we should preserve one skb (for local delivery) */
1745
1746 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
1747                          struct sk_buff *skb, struct mfc_cache *cache,
1748                          int local)
1749 {
1750         int psend = -1;
1751         int vif, ct;
1752
1753         vif = cache->mfc_parent;
1754         cache->mfc_un.res.pkt++;
1755         cache->mfc_un.res.bytes += skb->len;
1756
1757         /*
1758          * Wrong interface: drop packet and (maybe) send PIM assert.
1759          */
1760         if (mrt->vif_table[vif].dev != skb->dev) {
1761                 int true_vifi;
1762
1763                 if (rt_is_output_route(skb_rtable(skb))) {
1764                         /* It is our own packet, looped back.
1765                          * Very complicated situation...
1766                          *
1767                          * The best workaround until routing daemons will be
1768                          * fixed is not to redistribute packet, if it was
1769                          * send through wrong interface. It means, that
1770                          * multicast applications WILL NOT work for
1771                          * (S,G), which have default multicast route pointing
1772                          * to wrong oif. In any case, it is not a good
1773                          * idea to use multicasting applications on router.
1774                          */
1775                         goto dont_forward;
1776                 }
1777
1778                 cache->mfc_un.res.wrong_if++;
1779                 true_vifi = ipmr_find_vif(mrt, skb->dev);
1780
1781                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1782                     /* pimsm uses asserts, when switching from RPT to SPT,
1783                      * so that we cannot check that packet arrived on an oif.
1784                      * It is bad, but otherwise we would need to move pretty
1785                      * large chunk of pimd to kernel. Ough... --ANK
1786                      */
1787                     (mrt->mroute_do_pim ||
1788                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1789                     time_after(jiffies,
1790                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1791                         cache->mfc_un.res.last_assert = jiffies;
1792                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1793                 }
1794                 goto dont_forward;
1795         }
1796
1797         mrt->vif_table[vif].pkt_in++;
1798         mrt->vif_table[vif].bytes_in += skb->len;
1799
1800         /*
1801          *      Forward the frame
1802          */
1803         for (ct = cache->mfc_un.res.maxvif - 1;
1804              ct >= cache->mfc_un.res.minvif; ct--) {
1805                 if (ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1806                         if (psend != -1) {
1807                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1808
1809                                 if (skb2)
1810                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1811                                                         psend);
1812                         }
1813                         psend = ct;
1814                 }
1815         }
1816         if (psend != -1) {
1817                 if (local) {
1818                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1819
1820                         if (skb2)
1821                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1822                 } else {
1823                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1824                         return 0;
1825                 }
1826         }
1827
1828 dont_forward:
1829         if (!local)
1830                 kfree_skb(skb);
1831         return 0;
1832 }
1833
1834 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1835 {
1836         struct rtable *rt = skb_rtable(skb);
1837         struct iphdr *iph = ip_hdr(skb);
1838         struct flowi4 fl4 = {
1839                 .daddr = iph->daddr,
1840                 .saddr = iph->saddr,
1841                 .flowi4_tos = RT_TOS(iph->tos),
1842                 .flowi4_oif = (rt_is_output_route(rt) ?
1843                                skb->dev->ifindex : 0),
1844                 .flowi4_iif = (rt_is_output_route(rt) ?
1845                                LOOPBACK_IFINDEX :
1846                                skb->dev->ifindex),
1847                 .flowi4_mark = skb->mark,
1848         };
1849         struct mr_table *mrt;
1850         int err;
1851
1852         err = ipmr_fib_lookup(net, &fl4, &mrt);
1853         if (err)
1854                 return ERR_PTR(err);
1855         return mrt;
1856 }
1857
1858 /*
1859  *      Multicast packets for forwarding arrive here
1860  *      Called with rcu_read_lock();
1861  */
1862
1863 int ip_mr_input(struct sk_buff *skb)
1864 {
1865         struct mfc_cache *cache;
1866         struct net *net = dev_net(skb->dev);
1867         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1868         struct mr_table *mrt;
1869
1870         /* Packet is looped back after forward, it should not be
1871          * forwarded second time, but still can be delivered locally.
1872          */
1873         if (IPCB(skb)->flags & IPSKB_FORWARDED)
1874                 goto dont_forward;
1875
1876         mrt = ipmr_rt_fib_lookup(net, skb);
1877         if (IS_ERR(mrt)) {
1878                 kfree_skb(skb);
1879                 return PTR_ERR(mrt);
1880         }
1881         if (!local) {
1882                 if (IPCB(skb)->opt.router_alert) {
1883                         if (ip_call_ra_chain(skb))
1884                                 return 0;
1885                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1886                         /* IGMPv1 (and broken IGMPv2 implementations sort of
1887                          * Cisco IOS <= 11.2(8)) do not put router alert
1888                          * option to IGMP packets destined to routable
1889                          * groups. It is very bad, because it means
1890                          * that we can forward NO IGMP messages.
1891                          */
1892                         struct sock *mroute_sk;
1893
1894                         mroute_sk = rcu_dereference(mrt->mroute_sk);
1895                         if (mroute_sk) {
1896                                 nf_reset(skb);
1897                                 raw_rcv(mroute_sk, skb);
1898                                 return 0;
1899                         }
1900                     }
1901         }
1902
1903         /* already under rcu_read_lock() */
1904         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1905
1906         /*
1907          *      No usable cache entry
1908          */
1909         if (cache == NULL) {
1910                 int vif;
1911
1912                 if (local) {
1913                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1914                         ip_local_deliver(skb);
1915                         if (skb2 == NULL)
1916                                 return -ENOBUFS;
1917                         skb = skb2;
1918                 }
1919
1920                 read_lock(&mrt_lock);
1921                 vif = ipmr_find_vif(mrt, skb->dev);
1922                 if (vif >= 0) {
1923                         int err2 = ipmr_cache_unresolved(mrt, vif, skb);
1924                         read_unlock(&mrt_lock);
1925
1926                         return err2;
1927                 }
1928                 read_unlock(&mrt_lock);
1929                 kfree_skb(skb);
1930                 return -ENODEV;
1931         }
1932
1933         read_lock(&mrt_lock);
1934         ip_mr_forward(net, mrt, skb, cache, local);
1935         read_unlock(&mrt_lock);
1936
1937         if (local)
1938                 return ip_local_deliver(skb);
1939
1940         return 0;
1941
1942 dont_forward:
1943         if (local)
1944                 return ip_local_deliver(skb);
1945         kfree_skb(skb);
1946         return 0;
1947 }
1948
1949 #ifdef CONFIG_IP_PIMSM
1950 /* called with rcu_read_lock() */
1951 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
1952                      unsigned int pimlen)
1953 {
1954         struct net_device *reg_dev = NULL;
1955         struct iphdr *encap;
1956
1957         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
1958         /*
1959          * Check that:
1960          * a. packet is really sent to a multicast group
1961          * b. packet is not a NULL-REGISTER
1962          * c. packet is not truncated
1963          */
1964         if (!ipv4_is_multicast(encap->daddr) ||
1965             encap->tot_len == 0 ||
1966             ntohs(encap->tot_len) + pimlen > skb->len)
1967                 return 1;
1968
1969         read_lock(&mrt_lock);
1970         if (mrt->mroute_reg_vif_num >= 0)
1971                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
1972         read_unlock(&mrt_lock);
1973
1974         if (reg_dev == NULL)
1975                 return 1;
1976
1977         skb->mac_header = skb->network_header;
1978         skb_pull(skb, (u8 *)encap - skb->data);
1979         skb_reset_network_header(skb);
1980         skb->protocol = htons(ETH_P_IP);
1981         skb->ip_summed = CHECKSUM_NONE;
1982         skb->pkt_type = PACKET_HOST;
1983
1984         skb_tunnel_rx(skb, reg_dev);
1985
1986         netif_rx(skb);
1987
1988         return NET_RX_SUCCESS;
1989 }
1990 #endif
1991
1992 #ifdef CONFIG_IP_PIMSM_V1
1993 /*
1994  * Handle IGMP messages of PIMv1
1995  */
1996
1997 int pim_rcv_v1(struct sk_buff *skb)
1998 {
1999         struct igmphdr *pim;
2000         struct net *net = dev_net(skb->dev);
2001         struct mr_table *mrt;
2002
2003         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2004                 goto drop;
2005
2006         pim = igmp_hdr(skb);
2007
2008         mrt = ipmr_rt_fib_lookup(net, skb);
2009         if (IS_ERR(mrt))
2010                 goto drop;
2011         if (!mrt->mroute_do_pim ||
2012             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2013                 goto drop;
2014
2015         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2016 drop:
2017                 kfree_skb(skb);
2018         }
2019         return 0;
2020 }
2021 #endif
2022
2023 #ifdef CONFIG_IP_PIMSM_V2
2024 static int pim_rcv(struct sk_buff *skb)
2025 {
2026         struct pimreghdr *pim;
2027         struct net *net = dev_net(skb->dev);
2028         struct mr_table *mrt;
2029
2030         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2031                 goto drop;
2032
2033         pim = (struct pimreghdr *)skb_transport_header(skb);
2034         if (pim->type != ((PIM_VERSION << 4) | (PIM_REGISTER)) ||
2035             (pim->flags & PIM_NULL_REGISTER) ||
2036             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2037              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2038                 goto drop;
2039
2040         mrt = ipmr_rt_fib_lookup(net, skb);
2041         if (IS_ERR(mrt))
2042                 goto drop;
2043         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2044 drop:
2045                 kfree_skb(skb);
2046         }
2047         return 0;
2048 }
2049 #endif
2050
2051 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2052                               struct mfc_cache *c, struct rtmsg *rtm)
2053 {
2054         int ct;
2055         struct rtnexthop *nhp;
2056         struct nlattr *mp_attr;
2057         struct rta_mfc_stats mfcs;
2058
2059         /* If cache is unresolved, don't try to parse IIF and OIF */
2060         if (c->mfc_parent >= MAXVIFS)
2061                 return -ENOENT;
2062
2063         if (VIF_EXISTS(mrt, c->mfc_parent) &&
2064             nla_put_u32(skb, RTA_IIF, mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
2065                 return -EMSGSIZE;
2066
2067         if (!(mp_attr = nla_nest_start(skb, RTA_MULTIPATH)))
2068                 return -EMSGSIZE;
2069
2070         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2071                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2072                         if (!(nhp = nla_reserve_nohdr(skb, sizeof(*nhp)))) {
2073                                 nla_nest_cancel(skb, mp_attr);
2074                                 return -EMSGSIZE;
2075                         }
2076
2077                         nhp->rtnh_flags = 0;
2078                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2079                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2080                         nhp->rtnh_len = sizeof(*nhp);
2081                 }
2082         }
2083
2084         nla_nest_end(skb, mp_attr);
2085
2086         mfcs.mfcs_packets = c->mfc_un.res.pkt;
2087         mfcs.mfcs_bytes = c->mfc_un.res.bytes;
2088         mfcs.mfcs_wrong_if = c->mfc_un.res.wrong_if;
2089         if (nla_put(skb, RTA_MFC_STATS, sizeof(mfcs), &mfcs) < 0)
2090                 return -EMSGSIZE;
2091
2092         rtm->rtm_type = RTN_MULTICAST;
2093         return 1;
2094 }
2095
2096 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2097                    __be32 saddr, __be32 daddr,
2098                    struct rtmsg *rtm, int nowait)
2099 {
2100         struct mfc_cache *cache;
2101         struct mr_table *mrt;
2102         int err;
2103
2104         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2105         if (mrt == NULL)
2106                 return -ENOENT;
2107
2108         rcu_read_lock();
2109         cache = ipmr_cache_find(mrt, saddr, daddr);
2110
2111         if (cache == NULL) {
2112                 struct sk_buff *skb2;
2113                 struct iphdr *iph;
2114                 struct net_device *dev;
2115                 int vif = -1;
2116
2117                 if (nowait) {
2118                         rcu_read_unlock();
2119                         return -EAGAIN;
2120                 }
2121
2122                 dev = skb->dev;
2123                 read_lock(&mrt_lock);
2124                 if (dev)
2125                         vif = ipmr_find_vif(mrt, dev);
2126                 if (vif < 0) {
2127                         read_unlock(&mrt_lock);
2128                         rcu_read_unlock();
2129                         return -ENODEV;
2130                 }
2131                 skb2 = skb_clone(skb, GFP_ATOMIC);
2132                 if (!skb2) {
2133                         read_unlock(&mrt_lock);
2134                         rcu_read_unlock();
2135                         return -ENOMEM;
2136                 }
2137
2138                 skb_push(skb2, sizeof(struct iphdr));
2139                 skb_reset_network_header(skb2);
2140                 iph = ip_hdr(skb2);
2141                 iph->ihl = sizeof(struct iphdr) >> 2;
2142                 iph->saddr = saddr;
2143                 iph->daddr = daddr;
2144                 iph->version = 0;
2145                 err = ipmr_cache_unresolved(mrt, vif, skb2);
2146                 read_unlock(&mrt_lock);
2147                 rcu_read_unlock();
2148                 return err;
2149         }
2150
2151         read_lock(&mrt_lock);
2152         if (!nowait && (rtm->rtm_flags & RTM_F_NOTIFY))
2153                 cache->mfc_flags |= MFC_NOTIFY;
2154         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2155         read_unlock(&mrt_lock);
2156         rcu_read_unlock();
2157         return err;
2158 }
2159
2160 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2161                             u32 portid, u32 seq, struct mfc_cache *c, int cmd)
2162 {
2163         struct nlmsghdr *nlh;
2164         struct rtmsg *rtm;
2165         int err;
2166
2167         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), NLM_F_MULTI);
2168         if (nlh == NULL)
2169                 return -EMSGSIZE;
2170
2171         rtm = nlmsg_data(nlh);
2172         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2173         rtm->rtm_dst_len  = 32;
2174         rtm->rtm_src_len  = 32;
2175         rtm->rtm_tos      = 0;
2176         rtm->rtm_table    = mrt->id;
2177         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2178                 goto nla_put_failure;
2179         rtm->rtm_type     = RTN_MULTICAST;
2180         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2181         if (c->mfc_flags & MFC_STATIC)
2182                 rtm->rtm_protocol = RTPROT_STATIC;
2183         else
2184                 rtm->rtm_protocol = RTPROT_MROUTED;
2185         rtm->rtm_flags    = 0;
2186
2187         if (nla_put_be32(skb, RTA_SRC, c->mfc_origin) ||
2188             nla_put_be32(skb, RTA_DST, c->mfc_mcastgrp))
2189                 goto nla_put_failure;
2190         err = __ipmr_fill_mroute(mrt, skb, c, rtm);
2191         /* do not break the dump if cache is unresolved */
2192         if (err < 0 && err != -ENOENT)
2193                 goto nla_put_failure;
2194
2195         return nlmsg_end(skb, nlh);
2196
2197 nla_put_failure:
2198         nlmsg_cancel(skb, nlh);
2199         return -EMSGSIZE;
2200 }
2201
2202 static size_t mroute_msgsize(bool unresolved, int maxvif)
2203 {
2204         size_t len =
2205                 NLMSG_ALIGN(sizeof(struct rtmsg))
2206                 + nla_total_size(4)     /* RTA_TABLE */
2207                 + nla_total_size(4)     /* RTA_SRC */
2208                 + nla_total_size(4)     /* RTA_DST */
2209                 ;
2210
2211         if (!unresolved)
2212                 len = len
2213                       + nla_total_size(4)       /* RTA_IIF */
2214                       + nla_total_size(0)       /* RTA_MULTIPATH */
2215                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2216                                                 /* RTA_MFC_STATS */
2217                       + nla_total_size(sizeof(struct rta_mfc_stats))
2218                 ;
2219
2220         return len;
2221 }
2222
2223 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2224                                  int cmd)
2225 {
2226         struct net *net = read_pnet(&mrt->net);
2227         struct sk_buff *skb;
2228         int err = -ENOBUFS;
2229
2230         skb = nlmsg_new(mroute_msgsize(mfc->mfc_parent >= MAXVIFS, mrt->maxvif),
2231                         GFP_ATOMIC);
2232         if (skb == NULL)
2233                 goto errout;
2234
2235         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd);
2236         if (err < 0)
2237                 goto errout;
2238
2239         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2240         return;
2241
2242 errout:
2243         kfree_skb(skb);
2244         if (err < 0)
2245                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2246 }
2247
2248 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2249 {
2250         struct net *net = sock_net(skb->sk);
2251         struct mr_table *mrt;
2252         struct mfc_cache *mfc;
2253         unsigned int t = 0, s_t;
2254         unsigned int h = 0, s_h;
2255         unsigned int e = 0, s_e;
2256
2257         s_t = cb->args[0];
2258         s_h = cb->args[1];
2259         s_e = cb->args[2];
2260
2261         rcu_read_lock();
2262         ipmr_for_each_table(mrt, net) {
2263                 if (t < s_t)
2264                         goto next_table;
2265                 if (t > s_t)
2266                         s_h = 0;
2267                 for (h = s_h; h < MFC_LINES; h++) {
2268                         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2269                                 if (e < s_e)
2270                                         goto next_entry;
2271                                 if (ipmr_fill_mroute(mrt, skb,
2272                                                      NETLINK_CB(cb->skb).portid,
2273                                                      cb->nlh->nlmsg_seq,
2274                                                      mfc, RTM_NEWROUTE) < 0)
2275                                         goto done;
2276 next_entry:
2277                                 e++;
2278                         }
2279                         e = s_e = 0;
2280                 }
2281                 spin_lock_bh(&mfc_unres_lock);
2282                 list_for_each_entry(mfc, &mrt->mfc_unres_queue, list) {
2283                         if (e < s_e)
2284                                 goto next_entry2;
2285                         if (ipmr_fill_mroute(mrt, skb,
2286                                              NETLINK_CB(cb->skb).portid,
2287                                              cb->nlh->nlmsg_seq,
2288                                              mfc, RTM_NEWROUTE) < 0) {
2289                                 spin_unlock_bh(&mfc_unres_lock);
2290                                 goto done;
2291                         }
2292 next_entry2:
2293                         e++;
2294                 }
2295                 spin_unlock_bh(&mfc_unres_lock);
2296                 e = s_e = 0;
2297                 s_h = 0;
2298 next_table:
2299                 t++;
2300         }
2301 done:
2302         rcu_read_unlock();
2303
2304         cb->args[2] = e;
2305         cb->args[1] = h;
2306         cb->args[0] = t;
2307
2308         return skb->len;
2309 }
2310
2311 #ifdef CONFIG_PROC_FS
2312 /*
2313  *      The /proc interfaces to multicast routing :
2314  *      /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2315  */
2316 struct ipmr_vif_iter {
2317         struct seq_net_private p;
2318         struct mr_table *mrt;
2319         int ct;
2320 };
2321
2322 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2323                                            struct ipmr_vif_iter *iter,
2324                                            loff_t pos)
2325 {
2326         struct mr_table *mrt = iter->mrt;
2327
2328         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2329                 if (!VIF_EXISTS(mrt, iter->ct))
2330                         continue;
2331                 if (pos-- == 0)
2332                         return &mrt->vif_table[iter->ct];
2333         }
2334         return NULL;
2335 }
2336
2337 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2338         __acquires(mrt_lock)
2339 {
2340         struct ipmr_vif_iter *iter = seq->private;
2341         struct net *net = seq_file_net(seq);
2342         struct mr_table *mrt;
2343
2344         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2345         if (mrt == NULL)
2346                 return ERR_PTR(-ENOENT);
2347
2348         iter->mrt = mrt;
2349
2350         read_lock(&mrt_lock);
2351         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2352                 : SEQ_START_TOKEN;
2353 }
2354
2355 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2356 {
2357         struct ipmr_vif_iter *iter = seq->private;
2358         struct net *net = seq_file_net(seq);
2359         struct mr_table *mrt = iter->mrt;
2360
2361         ++*pos;
2362         if (v == SEQ_START_TOKEN)
2363                 return ipmr_vif_seq_idx(net, iter, 0);
2364
2365         while (++iter->ct < mrt->maxvif) {
2366                 if (!VIF_EXISTS(mrt, iter->ct))
2367                         continue;
2368                 return &mrt->vif_table[iter->ct];
2369         }
2370         return NULL;
2371 }
2372
2373 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2374         __releases(mrt_lock)
2375 {
2376         read_unlock(&mrt_lock);
2377 }
2378
2379 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2380 {
2381         struct ipmr_vif_iter *iter = seq->private;
2382         struct mr_table *mrt = iter->mrt;
2383
2384         if (v == SEQ_START_TOKEN) {
2385                 seq_puts(seq,
2386                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2387         } else {
2388                 const struct vif_device *vif = v;
2389                 const char *name =  vif->dev ? vif->dev->name : "none";
2390
2391                 seq_printf(seq,
2392                            "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2393                            vif - mrt->vif_table,
2394                            name, vif->bytes_in, vif->pkt_in,
2395                            vif->bytes_out, vif->pkt_out,
2396                            vif->flags, vif->local, vif->remote);
2397         }
2398         return 0;
2399 }
2400
2401 static const struct seq_operations ipmr_vif_seq_ops = {
2402         .start = ipmr_vif_seq_start,
2403         .next  = ipmr_vif_seq_next,
2404         .stop  = ipmr_vif_seq_stop,
2405         .show  = ipmr_vif_seq_show,
2406 };
2407
2408 static int ipmr_vif_open(struct inode *inode, struct file *file)
2409 {
2410         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2411                             sizeof(struct ipmr_vif_iter));
2412 }
2413
2414 static const struct file_operations ipmr_vif_fops = {
2415         .owner   = THIS_MODULE,
2416         .open    = ipmr_vif_open,
2417         .read    = seq_read,
2418         .llseek  = seq_lseek,
2419         .release = seq_release_net,
2420 };
2421
2422 struct ipmr_mfc_iter {
2423         struct seq_net_private p;
2424         struct mr_table *mrt;
2425         struct list_head *cache;
2426         int ct;
2427 };
2428
2429
2430 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2431                                           struct ipmr_mfc_iter *it, loff_t pos)
2432 {
2433         struct mr_table *mrt = it->mrt;
2434         struct mfc_cache *mfc;
2435
2436         rcu_read_lock();
2437         for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2438                 it->cache = &mrt->mfc_cache_array[it->ct];
2439                 list_for_each_entry_rcu(mfc, it->cache, list)
2440                         if (pos-- == 0)
2441                                 return mfc;
2442         }
2443         rcu_read_unlock();
2444
2445         spin_lock_bh(&mfc_unres_lock);
2446         it->cache = &mrt->mfc_unres_queue;
2447         list_for_each_entry(mfc, it->cache, list)
2448                 if (pos-- == 0)
2449                         return mfc;
2450         spin_unlock_bh(&mfc_unres_lock);
2451
2452         it->cache = NULL;
2453         return NULL;
2454 }
2455
2456
2457 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2458 {
2459         struct ipmr_mfc_iter *it = seq->private;
2460         struct net *net = seq_file_net(seq);
2461         struct mr_table *mrt;
2462
2463         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2464         if (mrt == NULL)
2465                 return ERR_PTR(-ENOENT);
2466
2467         it->mrt = mrt;
2468         it->cache = NULL;
2469         it->ct = 0;
2470         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2471                 : SEQ_START_TOKEN;
2472 }
2473
2474 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2475 {
2476         struct mfc_cache *mfc = v;
2477         struct ipmr_mfc_iter *it = seq->private;
2478         struct net *net = seq_file_net(seq);
2479         struct mr_table *mrt = it->mrt;
2480
2481         ++*pos;
2482
2483         if (v == SEQ_START_TOKEN)
2484                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2485
2486         if (mfc->list.next != it->cache)
2487                 return list_entry(mfc->list.next, struct mfc_cache, list);
2488
2489         if (it->cache == &mrt->mfc_unres_queue)
2490                 goto end_of_list;
2491
2492         BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2493
2494         while (++it->ct < MFC_LINES) {
2495                 it->cache = &mrt->mfc_cache_array[it->ct];
2496                 if (list_empty(it->cache))
2497                         continue;
2498                 return list_first_entry(it->cache, struct mfc_cache, list);
2499         }
2500
2501         /* exhausted cache_array, show unresolved */
2502         rcu_read_unlock();
2503         it->cache = &mrt->mfc_unres_queue;
2504         it->ct = 0;
2505
2506         spin_lock_bh(&mfc_unres_lock);
2507         if (!list_empty(it->cache))
2508                 return list_first_entry(it->cache, struct mfc_cache, list);
2509
2510 end_of_list:
2511         spin_unlock_bh(&mfc_unres_lock);
2512         it->cache = NULL;
2513
2514         return NULL;
2515 }
2516
2517 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2518 {
2519         struct ipmr_mfc_iter *it = seq->private;
2520         struct mr_table *mrt = it->mrt;
2521
2522         if (it->cache == &mrt->mfc_unres_queue)
2523                 spin_unlock_bh(&mfc_unres_lock);
2524         else if (it->cache == &mrt->mfc_cache_array[it->ct])
2525                 rcu_read_unlock();
2526 }
2527
2528 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2529 {
2530         int n;
2531
2532         if (v == SEQ_START_TOKEN) {
2533                 seq_puts(seq,
2534                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2535         } else {
2536                 const struct mfc_cache *mfc = v;
2537                 const struct ipmr_mfc_iter *it = seq->private;
2538                 const struct mr_table *mrt = it->mrt;
2539
2540                 seq_printf(seq, "%08X %08X %-3hd",
2541                            (__force u32) mfc->mfc_mcastgrp,
2542                            (__force u32) mfc->mfc_origin,
2543                            mfc->mfc_parent);
2544
2545                 if (it->cache != &mrt->mfc_unres_queue) {
2546                         seq_printf(seq, " %8lu %8lu %8lu",
2547                                    mfc->mfc_un.res.pkt,
2548                                    mfc->mfc_un.res.bytes,
2549                                    mfc->mfc_un.res.wrong_if);
2550                         for (n = mfc->mfc_un.res.minvif;
2551                              n < mfc->mfc_un.res.maxvif; n++) {
2552                                 if (VIF_EXISTS(mrt, n) &&
2553                                     mfc->mfc_un.res.ttls[n] < 255)
2554                                         seq_printf(seq,
2555                                            " %2d:%-3d",
2556                                            n, mfc->mfc_un.res.ttls[n]);
2557                         }
2558                 } else {
2559                         /* unresolved mfc_caches don't contain
2560                          * pkt, bytes and wrong_if values
2561                          */
2562                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2563                 }
2564                 seq_putc(seq, '\n');
2565         }
2566         return 0;
2567 }
2568
2569 static const struct seq_operations ipmr_mfc_seq_ops = {
2570         .start = ipmr_mfc_seq_start,
2571         .next  = ipmr_mfc_seq_next,
2572         .stop  = ipmr_mfc_seq_stop,
2573         .show  = ipmr_mfc_seq_show,
2574 };
2575
2576 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2577 {
2578         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2579                             sizeof(struct ipmr_mfc_iter));
2580 }
2581
2582 static const struct file_operations ipmr_mfc_fops = {
2583         .owner   = THIS_MODULE,
2584         .open    = ipmr_mfc_open,
2585         .read    = seq_read,
2586         .llseek  = seq_lseek,
2587         .release = seq_release_net,
2588 };
2589 #endif
2590
2591 #ifdef CONFIG_IP_PIMSM_V2
2592 static const struct net_protocol pim_protocol = {
2593         .handler        =       pim_rcv,
2594         .netns_ok       =       1,
2595 };
2596 #endif
2597
2598
2599 /*
2600  *      Setup for IP multicast routing
2601  */
2602 static int __net_init ipmr_net_init(struct net *net)
2603 {
2604         int err;
2605
2606         err = ipmr_rules_init(net);
2607         if (err < 0)
2608                 goto fail;
2609
2610 #ifdef CONFIG_PROC_FS
2611         err = -ENOMEM;
2612         if (!proc_net_fops_create(net, "ip_mr_vif", 0, &ipmr_vif_fops))
2613                 goto proc_vif_fail;
2614         if (!proc_net_fops_create(net, "ip_mr_cache", 0, &ipmr_mfc_fops))
2615                 goto proc_cache_fail;
2616 #endif
2617         return 0;
2618
2619 #ifdef CONFIG_PROC_FS
2620 proc_cache_fail:
2621         proc_net_remove(net, "ip_mr_vif");
2622 proc_vif_fail:
2623         ipmr_rules_exit(net);
2624 #endif
2625 fail:
2626         return err;
2627 }
2628
2629 static void __net_exit ipmr_net_exit(struct net *net)
2630 {
2631 #ifdef CONFIG_PROC_FS
2632         proc_net_remove(net, "ip_mr_cache");
2633         proc_net_remove(net, "ip_mr_vif");
2634 #endif
2635         ipmr_rules_exit(net);
2636 }
2637
2638 static struct pernet_operations ipmr_net_ops = {
2639         .init = ipmr_net_init,
2640         .exit = ipmr_net_exit,
2641 };
2642
2643 int __init ip_mr_init(void)
2644 {
2645         int err;
2646
2647         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2648                                        sizeof(struct mfc_cache),
2649                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2650                                        NULL);
2651         if (!mrt_cachep)
2652                 return -ENOMEM;
2653
2654         err = register_pernet_subsys(&ipmr_net_ops);
2655         if (err)
2656                 goto reg_pernet_fail;
2657
2658         err = register_netdevice_notifier(&ip_mr_notifier);
2659         if (err)
2660                 goto reg_notif_fail;
2661 #ifdef CONFIG_IP_PIMSM_V2
2662         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2663                 pr_err("%s: can't add PIM protocol\n", __func__);
2664                 err = -EAGAIN;
2665                 goto add_proto_fail;
2666         }
2667 #endif
2668         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
2669                       NULL, ipmr_rtm_dumproute, NULL);
2670         return 0;
2671
2672 #ifdef CONFIG_IP_PIMSM_V2
2673 add_proto_fail:
2674         unregister_netdevice_notifier(&ip_mr_notifier);
2675 #endif
2676 reg_notif_fail:
2677         unregister_pernet_subsys(&ipmr_net_ops);
2678 reg_pernet_fail:
2679         kmem_cache_destroy(mrt_cachep);
2680         return err;
2681 }