net/ipv4/ipmr.c

   1 /*
   2  *      IP multicast routing support for mrouted 3.6/3.8
   3  *
   4  *              (c) 1995 Alan Cox, <alan@lxorguk.ukuu.org.uk>
   5  *        Linux Consultancy and Custom Driver Development
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  *
  12  *      Fixes:
  13  *      Michael Chastain        :       Incorrect size of copying.
  14  *      Alan Cox                :       Added the cache manager code
  15  *      Alan Cox                :       Fixed the clone/copy bug and device race.
  16  *      Mike McLagan            :       Routing by source
  17  *      Malcolm Beattie         :       Buffer handling fixes.
  18  *      Alexey Kuznetsov        :       Double buffer free and other fixes.
  19  *      SVR Anand               :       Fixed several multicast bugs and problems.
  20  *      Alexey Kuznetsov        :       Status, optimisations and more.
  21  *      Brad Parker             :       Better behaviour on mrouted upcall
  22  *                                      overflow.
  23  *      Carlos Picoto           :       PIMv1 Support
  24  *      Pavlin Ivanov Radoslavov:       PIMv2 Registers must checksum only PIM header
  25  *                                      Relax this requirement to work with older peers.
  26  *
  27  */
  28
  29 #include <asm/uaccess.h>
  30 #include <linux/types.h>
  31 #include <linux/capability.h>
  32 #include <linux/errno.h>
  33 #include <linux/timer.h>
  34 #include <linux/mm.h>
  35 #include <linux/kernel.h>
  36 #include <linux/fcntl.h>
  37 #include <linux/stat.h>
  38 #include <linux/socket.h>
  39 #include <linux/in.h>
  40 #include <linux/inet.h>
  41 #include <linux/netdevice.h>
  42 #include <linux/inetdevice.h>
  43 #include <linux/igmp.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/seq_file.h>
  46 #include <linux/mroute.h>
  47 #include <linux/init.h>
  48 #include <linux/if_ether.h>
  49 #include <linux/slab.h>
  50 #include <net/net_namespace.h>
  51 #include <net/ip.h>
  52 #include <net/protocol.h>
  53 #include <linux/skbuff.h>
  54 #include <net/route.h>
  55 #include <net/sock.h>
  56 #include <net/icmp.h>
  57 #include <net/udp.h>
  58 #include <net/raw.h>
  59 #include <linux/notifier.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/netfilter_ipv4.h>
  62 #include <linux/compat.h>
  63 #include <linux/export.h>
  64 #include <net/ip_tunnels.h>
  65 #include <net/checksum.h>
  66 #include <net/netlink.h>
  67 #include <net/fib_rules.h>
  68 #include <linux/netconf.h>
  69 #include <net/nexthop.h>
  70
  71 struct ipmr_rule {
  72         struct fib_rule         common;
  73 };
  74
  75 struct ipmr_result {
  76         struct mr_table         *mrt;
  77 };
  78
  79 /* Big lock, protecting vif table, mrt cache and mroute socket state.
  80  * Note that the changes are semaphored via rtnl_lock.
  81  */
  82
  83 static DEFINE_RWLOCK(mrt_lock);
  84
  85 /* Multicast router control variables */
  86
  87 /* Special spinlock for queue of unresolved entries */
  88 static DEFINE_SPINLOCK(mfc_unres_lock);
  89
  90 /* We return to original Alan's scheme. Hash table of resolved
  91  * entries is changed only in process context and protected
  92  * with weak lock mrt_lock. Queue of unresolved entries is protected
  93  * with strong spinlock mfc_unres_lock.
  94  *
  95  * In this case data path is free of exclusive locks at all.
  96  */
  97
  98 static struct kmem_cache *mrt_cachep __read_mostly;
  99
 100 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 101 static void ipmr_free_table(struct mr_table *mrt);
 102
 103 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
 104                           struct sk_buff *skb, struct mfc_cache *cache,
 105                           int local);
 106 static int ipmr_cache_report(struct mr_table *mrt,
 107                              struct sk_buff *pkt, vifi_t vifi, int assert);
 108 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
 109                               struct mfc_cache *c, struct rtmsg *rtm);
 110 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
 111                                  int cmd);
 112 static void mroute_clean_tables(struct mr_table *mrt, bool all);
 113 static void ipmr_expire_process(unsigned long arg);
 114
 115 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 116 #define ipmr_for_each_table(mrt, net) \
 117         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
 118
 119 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 120 {
 121         struct mr_table *mrt;
 122
 123         ipmr_for_each_table(mrt, net) {
 124                 if (mrt->id == id)
 125                         return mrt;
 126         }
 127         return NULL;
 128 }
 129
 130 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 131                            struct mr_table **mrt)
 132 {
 133         int err;
 134         struct ipmr_result res;
 135         struct fib_lookup_arg arg = {
 136                 .result = &res,
 137                 .flags = FIB_LOOKUP_NOREF,
 138         };
 139
 140         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 141                                flowi4_to_flowi(flp4), 0, &arg);
 142         if (err < 0)
 143                 return err;
 144         *mrt = res.mrt;
 145         return 0;
 146 }
 147
 148 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 149                             int flags, struct fib_lookup_arg *arg)
 150 {
 151         struct ipmr_result *res = arg->result;
 152         struct mr_table *mrt;
 153
 154         switch (rule->action) {
 155         case FR_ACT_TO_TBL:
 156                 break;
 157         case FR_ACT_UNREACHABLE:
 158                 return -ENETUNREACH;
 159         case FR_ACT_PROHIBIT:
 160                 return -EACCES;
 161         case FR_ACT_BLACKHOLE:
 162         default:
 163                 return -EINVAL;
 164         }
 165
 166         mrt = ipmr_get_table(rule->fr_net, rule->table);
 167         if (!mrt)
 168                 return -EAGAIN;
 169         res->mrt = mrt;
 170         return 0;
 171 }
 172
 173 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 174 {
 175         return 1;
 176 }
 177
 178 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 179         FRA_GENERIC_POLICY,
 180 };
 181
 182 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 183                                struct fib_rule_hdr *frh, struct nlattr **tb)
 184 {
 185         return 0;
 186 }
 187
 188 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 189                              struct nlattr **tb)
 190 {
 191         return 1;
 192 }
 193
 194 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 195                           struct fib_rule_hdr *frh)
 196 {
 197         frh->dst_len = 0;
 198         frh->src_len = 0;
 199         frh->tos     = 0;
 200         return 0;
 201 }
 202
 203 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
 204         .family         = RTNL_FAMILY_IPMR,
 205         .rule_size      = sizeof(struct ipmr_rule),
 206         .addr_size      = sizeof(u32),
 207         .action         = ipmr_rule_action,
 208         .match          = ipmr_rule_match,
 209         .configure      = ipmr_rule_configure,
 210         .compare        = ipmr_rule_compare,
 211         .fill           = ipmr_rule_fill,
 212         .nlgroup        = RTNLGRP_IPV4_RULE,
 213         .policy         = ipmr_rule_policy,
 214         .owner          = THIS_MODULE,
 215 };
 216
 217 static int __net_init ipmr_rules_init(struct net *net)
 218 {
 219         struct fib_rules_ops *ops;
 220         struct mr_table *mrt;
 221         int err;
 222
 223         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 224         if (IS_ERR(ops))
 225                 return PTR_ERR(ops);
 226
 227         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 228
 229         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 230         if (IS_ERR(mrt)) {
 231                 err = PTR_ERR(mrt);
 232                 goto err1;
 233         }
 234
 235         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 236         if (err < 0)
 237                 goto err2;
 238
 239         net->ipv4.mr_rules_ops = ops;
 240         return 0;
 241
 242 err2:
 243         ipmr_free_table(mrt);
 244 err1:
 245         fib_rules_unregister(ops);
 246         return err;
 247 }
 248
 249 static void __net_exit ipmr_rules_exit(struct net *net)
 250 {
 251         struct mr_table *mrt, *next;
 252
 253         rtnl_lock();
 254         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 255                 list_del(&mrt->list);
 256                 ipmr_free_table(mrt);
 257         }
 258         fib_rules_unregister(net->ipv4.mr_rules_ops);
 259         rtnl_unlock();
 260 }
 261 #else
 262 #define ipmr_for_each_table(mrt, net) \
 263         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 264
 265 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 266 {
 267         return net->ipv4.mrt;
 268 }
 269
 270 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 271                            struct mr_table **mrt)
 272 {
 273         *mrt = net->ipv4.mrt;
 274         return 0;
 275 }
 276
 277 static int __net_init ipmr_rules_init(struct net *net)
 278 {
 279         struct mr_table *mrt;
 280
 281         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 282         if (IS_ERR(mrt))
 283                 return PTR_ERR(mrt);
 284         net->ipv4.mrt = mrt;
 285         return 0;
 286 }
 287
 288 static void __net_exit ipmr_rules_exit(struct net *net)
 289 {
 290         rtnl_lock();
 291         ipmr_free_table(net->ipv4.mrt);
 292         net->ipv4.mrt = NULL;
 293         rtnl_unlock();
 294 }
 295 #endif
 296
 297 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 298 {
 299         struct mr_table *mrt;
 300         unsigned int i;
 301
 302         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 303         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 304                 return ERR_PTR(-EINVAL);
 305
 306         mrt = ipmr_get_table(net, id);
 307         if (mrt)
 308                 return mrt;
 309
 310         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
 311         if (!mrt)
 312                 return ERR_PTR(-ENOMEM);
 313         write_pnet(&mrt->net, net);
 314         mrt->id = id;
 315
 316         /* Forwarding cache */
 317         for (i = 0; i < MFC_LINES; i++)
 318                 INIT_LIST_HEAD(&mrt->mfc_cache_array[i]);
 319
 320         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
 321
 322         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
 323                     (unsigned long)mrt);
 324
 325         mrt->mroute_reg_vif_num = -1;
 326 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 327         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 328 #endif
 329         return mrt;
 330 }
 331
 332 static void ipmr_free_table(struct mr_table *mrt)
 333 {
 334         del_timer_sync(&mrt->ipmr_expire_timer);
 335         mroute_clean_tables(mrt, true);
 336         kfree(mrt);
 337 }
 338
 339 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 340
 341 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 342 {
 343         struct net *net = dev_net(dev);
 344
 345         dev_close(dev);
 346
 347         dev = __dev_get_by_name(net, "tunl0");
 348         if (dev) {
 349                 const struct net_device_ops *ops = dev->netdev_ops;
 350                 struct ifreq ifr;
 351                 struct ip_tunnel_parm p;
 352
 353                 memset(&p, 0, sizeof(p));
 354                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 355                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 356                 p.iph.version = 4;
 357                 p.iph.ihl = 5;
 358                 p.iph.protocol = IPPROTO_IPIP;
 359                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 360                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 361
 362                 if (ops->ndo_do_ioctl) {
 363                         mm_segment_t oldfs = get_fs();
 364
 365                         set_fs(KERNEL_DS);
 366                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 367                         set_fs(oldfs);
 368                 }
 369         }
 370 }
 371
 372 /* Initialize ipmr pimreg/tunnel in_device */
 373 static bool ipmr_init_vif_indev(const struct net_device *dev)
 374 {
 375         struct in_device *in_dev;
 376
 377         ASSERT_RTNL();
 378
 379         in_dev = __in_dev_get_rtnl(dev);
 380         if (!in_dev)
 381                 return false;
 382         ipv4_devconf_setall(in_dev);
 383         neigh_parms_data_state_setall(in_dev->arp_parms);
 384         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 385
 386         return true;
 387 }
 388
 389 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 390 {
 391         struct net_device  *dev;
 392
 393         dev = __dev_get_by_name(net, "tunl0");
 394
 395         if (dev) {
 396                 const struct net_device_ops *ops = dev->netdev_ops;
 397                 int err;
 398                 struct ifreq ifr;
 399                 struct ip_tunnel_parm p;
 400
 401                 memset(&p, 0, sizeof(p));
 402                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 403                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 404                 p.iph.version = 4;
 405                 p.iph.ihl = 5;
 406                 p.iph.protocol = IPPROTO_IPIP;
 407                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 408                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 409
 410                 if (ops->ndo_do_ioctl) {
 411                         mm_segment_t oldfs = get_fs();
 412
 413                         set_fs(KERNEL_DS);
 414                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 415                         set_fs(oldfs);
 416                 } else {
 417                         err = -EOPNOTSUPP;
 418                 }
 419                 dev = NULL;
 420
 421                 if (err == 0 &&
 422                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 423                         dev->flags |= IFF_MULTICAST;
 424                         if (!ipmr_init_vif_indev(dev))
 425                                 goto failure;
 426                         if (dev_open(dev))
 427                                 goto failure;
 428                         dev_hold(dev);
 429                 }
 430         }
 431         return dev;
 432
 433 failure:
 434         unregister_netdevice(dev);
 435         return NULL;
 436 }
 437
 438 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 439 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 440 {
 441         struct net *net = dev_net(dev);
 442         struct mr_table *mrt;
 443         struct flowi4 fl4 = {
 444                 .flowi4_oif     = dev->ifindex,
 445                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 446                 .flowi4_mark    = skb->mark,
 447         };
 448         int err;
 449
 450         err = ipmr_fib_lookup(net, &fl4, &mrt);
 451         if (err < 0) {
 452                 kfree_skb(skb);
 453                 return err;
 454         }
 455
 456         read_lock(&mrt_lock);
 457         dev->stats.tx_bytes += skb->len;
 458         dev->stats.tx_packets++;
 459         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 460         read_unlock(&mrt_lock);
 461         kfree_skb(skb);
 462         return NETDEV_TX_OK;
 463 }
 464
 465 static int reg_vif_get_iflink(const struct net_device *dev)
 466 {
 467         return 0;
 468 }
 469
 470 static const struct net_device_ops reg_vif_netdev_ops = {
 471         .ndo_start_xmit = reg_vif_xmit,
 472         .ndo_get_iflink = reg_vif_get_iflink,
 473 };
 474
 475 static void reg_vif_setup(struct net_device *dev)
 476 {
 477         dev->type               = ARPHRD_PIMREG;
 478         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 479         dev->flags              = IFF_NOARP;
 480         dev->netdev_ops         = &reg_vif_netdev_ops;
 481         dev->destructor         = free_netdev;
 482         dev->features           |= NETIF_F_NETNS_LOCAL;
 483 }
 484
 485 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 486 {
 487         struct net_device *dev;
 488         char name[IFNAMSIZ];
 489
 490         if (mrt->id == RT_TABLE_DEFAULT)
 491                 sprintf(name, "pimreg");
 492         else
 493                 sprintf(name, "pimreg%u", mrt->id);
 494
 495         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 496
 497         if (!dev)
 498                 return NULL;
 499
 500         dev_net_set(dev, net);
 501
 502         if (register_netdevice(dev)) {
 503                 free_netdev(dev);
 504                 return NULL;
 505         }
 506
 507         if (!ipmr_init_vif_indev(dev))
 508                 goto failure;
 509         if (dev_open(dev))
 510                 goto failure;
 511
 512         dev_hold(dev);
 513
 514         return dev;
 515
 516 failure:
 517         unregister_netdevice(dev);
 518         return NULL;
 519 }
 520
 521 /* called with rcu_read_lock() */
 522 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 523                      unsigned int pimlen)
 524 {
 525         struct net_device *reg_dev = NULL;
 526         struct iphdr *encap;
 527
 528         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 529         /* Check that:
 530          * a. packet is really sent to a multicast group
 531          * b. packet is not a NULL-REGISTER
 532          * c. packet is not truncated
 533          */
 534         if (!ipv4_is_multicast(encap->daddr) ||
 535             encap->tot_len == 0 ||
 536             ntohs(encap->tot_len) + pimlen > skb->len)
 537                 return 1;
 538
 539         read_lock(&mrt_lock);
 540         if (mrt->mroute_reg_vif_num >= 0)
 541                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 542         read_unlock(&mrt_lock);
 543
 544         if (!reg_dev)
 545                 return 1;
 546
 547         skb->mac_header = skb->network_header;
 548         skb_pull(skb, (u8 *)encap - skb->data);
 549         skb_reset_network_header(skb);
 550         skb->protocol = htons(ETH_P_IP);
 551         skb->ip_summed = CHECKSUM_NONE;
 552
 553         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 554
 555         netif_rx(skb);
 556
 557         return NET_RX_SUCCESS;
 558 }
 559 #else
 560 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 561 {
 562         return NULL;
 563 }
 564 #endif
 565
 566 /**
 567  *      vif_delete - Delete a VIF entry
 568  *      @notify: Set to 1, if the caller is a notifier_call
 569  */
 570 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 571                       struct list_head *head)
 572 {
 573         struct vif_device *v;
 574         struct net_device *dev;
 575         struct in_device *in_dev;
 576
 577         if (vifi < 0 || vifi >= mrt->maxvif)
 578                 return -EADDRNOTAVAIL;
 579
 580         v = &mrt->vif_table[vifi];
 581
 582         write_lock_bh(&mrt_lock);
 583         dev = v->dev;
 584         v->dev = NULL;
 585
 586         if (!dev) {
 587                 write_unlock_bh(&mrt_lock);
 588                 return -EADDRNOTAVAIL;
 589         }
 590
 591         if (vifi == mrt->mroute_reg_vif_num)
 592                 mrt->mroute_reg_vif_num = -1;
 593
 594         if (vifi + 1 == mrt->maxvif) {
 595                 int tmp;
 596
 597                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 598                         if (VIF_EXISTS(mrt, tmp))
 599                                 break;
 600                 }
 601                 mrt->maxvif = tmp+1;
 602         }
 603
 604         write_unlock_bh(&mrt_lock);
 605
 606         dev_set_allmulti(dev, -1);
 607
 608         in_dev = __in_dev_get_rtnl(dev);
 609         if (in_dev) {
 610                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 611                 inet_netconf_notify_devconf(dev_net(dev),
 612                                             NETCONFA_MC_FORWARDING,
 613                                             dev->ifindex, &in_dev->cnf);
 614                 ip_rt_multicast_event(in_dev);
 615         }
 616
 617         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 618                 unregister_netdevice_queue(dev, head);
 619
 620         dev_put(dev);
 621         return 0;
 622 }
 623
 624 static void ipmr_cache_free_rcu(struct rcu_head *head)
 625 {
 626         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
 627
 628         kmem_cache_free(mrt_cachep, c);
 629 }
 630
 631 static inline void ipmr_cache_free(struct mfc_cache *c)
 632 {
 633         call_rcu(&c->rcu, ipmr_cache_free_rcu);
 634 }
 635
 636 /* Destroy an unresolved cache entry, killing queued skbs
 637  * and reporting error to netlink readers.
 638  */
 639 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 640 {
 641         struct net *net = read_pnet(&mrt->net);
 642         struct sk_buff *skb;
 643         struct nlmsgerr *e;
 644
 645         atomic_dec(&mrt->cache_resolve_queue_len);
 646
 647         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
 648                 if (ip_hdr(skb)->version == 0) {
 649                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 650                         nlh->nlmsg_type = NLMSG_ERROR;
 651                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 652                         skb_trim(skb, nlh->nlmsg_len);
 653                         e = nlmsg_data(nlh);
 654                         e->error = -ETIMEDOUT;
 655                         memset(&e->msg, 0, sizeof(e->msg));
 656
 657                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 658                 } else {
 659                         kfree_skb(skb);
 660                 }
 661         }
 662
 663         ipmr_cache_free(c);
 664 }
 665
 666 /* Timer process for the unresolved queue. */
 667 static void ipmr_expire_process(unsigned long arg)
 668 {
 669         struct mr_table *mrt = (struct mr_table *)arg;
 670         unsigned long now;
 671         unsigned long expires;
 672         struct mfc_cache *c, *next;
 673
 674         if (!spin_trylock(&mfc_unres_lock)) {
 675                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 676                 return;
 677         }
 678
 679         if (list_empty(&mrt->mfc_unres_queue))
 680                 goto out;
 681
 682         now = jiffies;
 683         expires = 10*HZ;
 684
 685         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 686                 if (time_after(c->mfc_un.unres.expires, now)) {
 687                         unsigned long interval = c->mfc_un.unres.expires - now;
 688                         if (interval < expires)
 689                                 expires = interval;
 690                         continue;
 691                 }
 692
 693                 list_del(&c->list);
 694                 mroute_netlink_event(mrt, c, RTM_DELROUTE);
 695                 ipmr_destroy_unres(mrt, c);
 696         }
 697
 698         if (!list_empty(&mrt->mfc_unres_queue))
 699                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 700
 701 out:
 702         spin_unlock(&mfc_unres_lock);
 703 }
 704
 705 /* Fill oifs list. It is called under write locked mrt_lock. */
 706 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
 707                                    unsigned char *ttls)
 708 {
 709         int vifi;
 710
 711         cache->mfc_un.res.minvif = MAXVIFS;
 712         cache->mfc_un.res.maxvif = 0;
 713         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 714
 715         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 716                 if (VIF_EXISTS(mrt, vifi) &&
 717                     ttls[vifi] && ttls[vifi] < 255) {
 718                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 719                         if (cache->mfc_un.res.minvif > vifi)
 720                                 cache->mfc_un.res.minvif = vifi;
 721                         if (cache->mfc_un.res.maxvif <= vifi)
 722                                 cache->mfc_un.res.maxvif = vifi + 1;
 723                 }
 724         }
 725 }
 726
 727 static int vif_add(struct net *net, struct mr_table *mrt,
 728                    struct vifctl *vifc, int mrtsock)
 729 {
 730         int vifi = vifc->vifc_vifi;
 731         struct vif_device *v = &mrt->vif_table[vifi];
 732         struct net_device *dev;
 733         struct in_device *in_dev;
 734         int err;
 735
 736         /* Is vif busy ? */
 737         if (VIF_EXISTS(mrt, vifi))
 738                 return -EADDRINUSE;
 739
 740         switch (vifc->vifc_flags) {
 741         case VIFF_REGISTER:
 742                 if (!ipmr_pimsm_enabled())
 743                         return -EINVAL;
 744                 /* Special Purpose VIF in PIM
 745                  * All the packets will be sent to the daemon
 746                  */
 747                 if (mrt->mroute_reg_vif_num >= 0)
 748                         return -EADDRINUSE;
 749                 dev = ipmr_reg_vif(net, mrt);
 750                 if (!dev)
 751                         return -ENOBUFS;
 752                 err = dev_set_allmulti(dev, 1);
 753                 if (err) {
 754                         unregister_netdevice(dev);
 755                         dev_put(dev);
 756                         return err;
 757                 }
 758                 break;
 759         case VIFF_TUNNEL:
 760                 dev = ipmr_new_tunnel(net, vifc);
 761                 if (!dev)
 762                         return -ENOBUFS;
 763                 err = dev_set_allmulti(dev, 1);
 764                 if (err) {
 765                         ipmr_del_tunnel(dev, vifc);
 766                         dev_put(dev);
 767                         return err;
 768                 }
 769                 break;
 770         case VIFF_USE_IFINDEX:
 771         case 0:
 772                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 773                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 774                         if (dev && !__in_dev_get_rtnl(dev)) {
 775                                 dev_put(dev);
 776                                 return -EADDRNOTAVAIL;
 777                         }
 778                 } else {
 779                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 780                 }
 781                 if (!dev)
 782                         return -EADDRNOTAVAIL;
 783                 err = dev_set_allmulti(dev, 1);
 784                 if (err) {
 785                         dev_put(dev);
 786                         return err;
 787                 }
 788                 break;
 789         default:
 790                 return -EINVAL;
 791         }
 792
 793         in_dev = __in_dev_get_rtnl(dev);
 794         if (!in_dev) {
 795                 dev_put(dev);
 796                 return -EADDRNOTAVAIL;
 797         }
 798         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 799         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING, dev->ifindex,
 800                                     &in_dev->cnf);
 801         ip_rt_multicast_event(in_dev);
 802
 803         /* Fill in the VIF structures */
 804
 805         v->rate_limit = vifc->vifc_rate_limit;
 806         v->local = vifc->vifc_lcl_addr.s_addr;
 807         v->remote = vifc->vifc_rmt_addr.s_addr;
 808         v->flags = vifc->vifc_flags;
 809         if (!mrtsock)
 810                 v->flags |= VIFF_STATIC;
 811         v->threshold = vifc->vifc_threshold;
 812         v->bytes_in = 0;
 813         v->bytes_out = 0;
 814         v->pkt_in = 0;
 815         v->pkt_out = 0;
 816         v->link = dev->ifindex;
 817         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
 818                 v->link = dev_get_iflink(dev);
 819
 820         /* And finish update writing critical data */
 821         write_lock_bh(&mrt_lock);
 822         v->dev = dev;
 823         if (v->flags & VIFF_REGISTER)
 824                 mrt->mroute_reg_vif_num = vifi;
 825         if (vifi+1 > mrt->maxvif)
 826                 mrt->maxvif = vifi+1;
 827         write_unlock_bh(&mrt_lock);
 828         return 0;
 829 }
 830
 831 /* called with rcu_read_lock() */
 832 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 833                                          __be32 origin,
 834                                          __be32 mcastgrp)
 835 {
 836         int line = MFC_HASH(mcastgrp, origin);
 837         struct mfc_cache *c;
 838
 839         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
 840                 if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
 841                         return c;
 842         }
 843         return NULL;
 844 }
 845
 846 /* Look for a (*,*,oif) entry */
 847 static struct mfc_cache *ipmr_cache_find_any_parent(struct mr_table *mrt,
 848                                                     int vifi)
 849 {
 850         int line = MFC_HASH(htonl(INADDR_ANY), htonl(INADDR_ANY));
 851         struct mfc_cache *c;
 852
 853         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list)
 854                 if (c->mfc_origin == htonl(INADDR_ANY) &&
 855                     c->mfc_mcastgrp == htonl(INADDR_ANY) &&
 856                     c->mfc_un.res.ttls[vifi] < 255)
 857                         return c;
 858
 859         return NULL;
 860 }
 861
 862 /* Look for a (*,G) entry */
 863 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 864                                              __be32 mcastgrp, int vifi)
 865 {
 866         int line = MFC_HASH(mcastgrp, htonl(INADDR_ANY));
 867         struct mfc_cache *c, *proxy;
 868
 869         if (mcastgrp == htonl(INADDR_ANY))
 870                 goto skip;
 871
 872         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list)
 873                 if (c->mfc_origin == htonl(INADDR_ANY) &&
 874                     c->mfc_mcastgrp == mcastgrp) {
 875                         if (c->mfc_un.res.ttls[vifi] < 255)
 876                                 return c;
 877
 878                         /* It's ok if the vifi is part of the static tree */
 879                         proxy = ipmr_cache_find_any_parent(mrt,
 880                                                            c->mfc_parent);
 881                         if (proxy && proxy->mfc_un.res.ttls[vifi] < 255)
 882                                 return c;
 883                 }
 884
 885 skip:
 886         return ipmr_cache_find_any_parent(mrt, vifi);
 887 }
 888
 889 /* Allocate a multicast cache entry */
 890 static struct mfc_cache *ipmr_cache_alloc(void)
 891 {
 892         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 893
 894         if (c)
 895                 c->mfc_un.res.minvif = MAXVIFS;
 896         return c;
 897 }
 898
 899 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 900 {
 901         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 902
 903         if (c) {
 904                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
 905                 c->mfc_un.unres.expires = jiffies + 10*HZ;
 906         }
 907         return c;
 908 }
 909
 910 /* A cache entry has gone into a resolved state from queued */
 911 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 912                                struct mfc_cache *uc, struct mfc_cache *c)
 913 {
 914         struct sk_buff *skb;
 915         struct nlmsgerr *e;
 916
 917         /* Play the pending entries through our router */
 918         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
 919                 if (ip_hdr(skb)->version == 0) {
 920                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 921
 922                         if (__ipmr_fill_mroute(mrt, skb, c, nlmsg_data(nlh)) > 0) {
 923                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 924                                                  (u8 *)nlh;
 925                         } else {
 926                                 nlh->nlmsg_type = NLMSG_ERROR;
 927                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 928                                 skb_trim(skb, nlh->nlmsg_len);
 929                                 e = nlmsg_data(nlh);
 930                                 e->error = -EMSGSIZE;
 931                                 memset(&e->msg, 0, sizeof(e->msg));
 932                         }
 933
 934                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 935                 } else {
 936                         ip_mr_forward(net, mrt, skb, c, 0);
 937                 }
 938         }
 939 }
 940
 941 /* Bounce a cache query up to mrouted. We could use netlink for this but mrouted
 942  * expects the following bizarre scheme.
 943  *
 944  * Called under mrt_lock.
 945  */
 946 static int ipmr_cache_report(struct mr_table *mrt,
 947                              struct sk_buff *pkt, vifi_t vifi, int assert)
 948 {
 949         const int ihl = ip_hdrlen(pkt);
 950         struct sock *mroute_sk;
 951         struct igmphdr *igmp;
 952         struct igmpmsg *msg;
 953         struct sk_buff *skb;
 954         int ret;
 955
 956         if (assert == IGMPMSG_WHOLEPKT)
 957                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
 958         else
 959                 skb = alloc_skb(128, GFP_ATOMIC);
 960
 961         if (!skb)
 962                 return -ENOBUFS;
 963
 964         if (assert == IGMPMSG_WHOLEPKT) {
 965                 /* Ugly, but we have no choice with this interface.
 966                  * Duplicate old header, fix ihl, length etc.
 967                  * And all this only to mangle msg->im_msgtype and
 968                  * to set msg->im_mbz to "mbz" :-)
 969                  */
 970                 skb_push(skb, sizeof(struct iphdr));
 971                 skb_reset_network_header(skb);
 972                 skb_reset_transport_header(skb);
 973                 msg = (struct igmpmsg *)skb_network_header(skb);
 974                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
 975                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
 976                 msg->im_mbz = 0;
 977                 msg->im_vif = mrt->mroute_reg_vif_num;
 978                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
 979                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
 980                                              sizeof(struct iphdr));
 981         } else {
 982                 /* Copy the IP header */
 983                 skb_set_network_header(skb, skb->len);
 984                 skb_put(skb, ihl);
 985                 skb_copy_to_linear_data(skb, pkt->data, ihl);
 986                 /* Flag to the kernel this is a route add */
 987                 ip_hdr(skb)->protocol = 0;
 988                 msg = (struct igmpmsg *)skb_network_header(skb);
 989                 msg->im_vif = vifi;
 990                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
 991                 /* Add our header */
 992                 igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
 993                 igmp->type = assert;
 994                 msg->im_msgtype = assert;
 995                 igmp->code = 0;
 996                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
 997                 skb->transport_header = skb->network_header;
 998         }
 999
1000         rcu_read_lock();
1001         mroute_sk = rcu_dereference(mrt->mroute_sk);
1002         if (!mroute_sk) {
1003                 rcu_read_unlock();
1004                 kfree_skb(skb);
1005                 return -EINVAL;
1006         }
1007
1008         /* Deliver to mrouted */
1009         ret = sock_queue_rcv_skb(mroute_sk, skb);
1010         rcu_read_unlock();
1011         if (ret < 0) {
1012                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1013                 kfree_skb(skb);
1014         }
1015
1016         return ret;
1017 }
1018
1019 /* Queue a packet for resolution. It gets locked cache entry! */
1020 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1021                                  struct sk_buff *skb)
1022 {
1023         bool found = false;
1024         int err;
1025         struct mfc_cache *c;
1026         const struct iphdr *iph = ip_hdr(skb);
1027
1028         spin_lock_bh(&mfc_unres_lock);
1029         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
1030                 if (c->mfc_mcastgrp == iph->daddr &&
1031                     c->mfc_origin == iph->saddr) {
1032                         found = true;
1033                         break;
1034                 }
1035         }
1036
1037         if (!found) {
1038                 /* Create a new entry if allowable */
1039                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1040                     (c = ipmr_cache_alloc_unres()) == NULL) {
1041                         spin_unlock_bh(&mfc_unres_lock);
1042
1043                         kfree_skb(skb);
1044                         return -ENOBUFS;
1045                 }
1046
1047                 /* Fill in the new cache entry */
1048                 c->mfc_parent   = -1;
1049                 c->mfc_origin   = iph->saddr;
1050                 c->mfc_mcastgrp = iph->daddr;
1051
1052                 /* Reflect first query at mrouted. */
1053                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1054                 if (err < 0) {
1055                         /* If the report failed throw the cache entry
1056                            out - Brad Parker
1057                          */
1058                         spin_unlock_bh(&mfc_unres_lock);
1059
1060                         ipmr_cache_free(c);
1061                         kfree_skb(skb);
1062                         return err;
1063                 }
1064
1065                 atomic_inc(&mrt->cache_resolve_queue_len);
1066                 list_add(&c->list, &mrt->mfc_unres_queue);
1067                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1068
1069                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1070                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1071         }
1072
1073         /* See if we can append the packet */
1074         if (c->mfc_un.unres.unresolved.qlen > 3) {
1075                 kfree_skb(skb);
1076                 err = -ENOBUFS;
1077         } else {
1078                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1079                 err = 0;
1080         }
1081
1082         spin_unlock_bh(&mfc_unres_lock);
1083         return err;
1084 }
1085
1086 /* MFC cache manipulation by user space mroute daemon */
1087
1088 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1089 {
1090         int line;
1091         struct mfc_cache *c, *next;
1092
1093         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1094
1095         list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1096                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1097                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr &&
1098                     (parent == -1 || parent == c->mfc_parent)) {
1099                         list_del_rcu(&c->list);
1100                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1101                         ipmr_cache_free(c);
1102                         return 0;
1103                 }
1104         }
1105         return -ENOENT;
1106 }
1107
1108 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1109                         struct mfcctl *mfc, int mrtsock, int parent)
1110 {
1111         bool found = false;
1112         int line;
1113         struct mfc_cache *uc, *c;
1114
1115         if (mfc->mfcc_parent >= MAXVIFS)
1116                 return -ENFILE;
1117
1118         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1119
1120         list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1121                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1122                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr &&
1123                     (parent == -1 || parent == c->mfc_parent)) {
1124                         found = true;
1125                         break;
1126                 }
1127         }
1128
1129         if (found) {
1130                 write_lock_bh(&mrt_lock);
1131                 c->mfc_parent = mfc->mfcc_parent;
1132                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1133                 if (!mrtsock)
1134                         c->mfc_flags |= MFC_STATIC;
1135                 write_unlock_bh(&mrt_lock);
1136                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1137                 return 0;
1138         }
1139
1140         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1141             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1142                 return -EINVAL;
1143
1144         c = ipmr_cache_alloc();
1145         if (!c)
1146                 return -ENOMEM;
1147
1148         c->mfc_origin = mfc->mfcc_origin.s_addr;
1149         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1150         c->mfc_parent = mfc->mfcc_parent;
1151         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1152         if (!mrtsock)
1153                 c->mfc_flags |= MFC_STATIC;
1154
1155         list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1156
1157         /* Check to see if we resolved a queued list. If so we
1158          * need to send on the frames and tidy up.
1159          */
1160         found = false;
1161         spin_lock_bh(&mfc_unres_lock);
1162         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1163                 if (uc->mfc_origin == c->mfc_origin &&
1164                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1165                         list_del(&uc->list);
1166                         atomic_dec(&mrt->cache_resolve_queue_len);
1167                         found = true;
1168                         break;
1169                 }
1170         }
1171         if (list_empty(&mrt->mfc_unres_queue))
1172                 del_timer(&mrt->ipmr_expire_timer);
1173         spin_unlock_bh(&mfc_unres_lock);
1174
1175         if (found) {
1176                 ipmr_cache_resolve(net, mrt, uc, c);
1177                 ipmr_cache_free(uc);
1178         }
1179         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1180         return 0;
1181 }
1182
1183 /* Close the multicast socket, and clear the vif tables etc */
1184 static void mroute_clean_tables(struct mr_table *mrt, bool all)
1185 {
1186         int i;
1187         LIST_HEAD(list);
1188         struct mfc_cache *c, *next;
1189
1190         /* Shut down all active vif entries */
1191         for (i = 0; i < mrt->maxvif; i++) {
1192                 if (!all && (mrt->vif_table[i].flags & VIFF_STATIC))
1193                         continue;
1194                 vif_delete(mrt, i, 0, &list);
1195         }
1196         unregister_netdevice_many(&list);
1197
1198         /* Wipe the cache */
1199         for (i = 0; i < MFC_LINES; i++) {
1200                 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1201                         if (!all && (c->mfc_flags & MFC_STATIC))
1202                                 continue;
1203                         list_del_rcu(&c->list);
1204                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1205                         ipmr_cache_free(c);
1206                 }
1207         }
1208
1209         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1210                 spin_lock_bh(&mfc_unres_lock);
1211                 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1212                         list_del(&c->list);
1213                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1214                         ipmr_destroy_unres(mrt, c);
1215                 }
1216                 spin_unlock_bh(&mfc_unres_lock);
1217         }
1218 }
1219
1220 /* called from ip_ra_control(), before an RCU grace period,
1221  * we dont need to call synchronize_rcu() here
1222  */
1223 static void mrtsock_destruct(struct sock *sk)
1224 {
1225         struct net *net = sock_net(sk);
1226         struct mr_table *mrt;
1227
1228         rtnl_lock();
1229         ipmr_for_each_table(mrt, net) {
1230                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1231                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1232                         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1233                                                     NETCONFA_IFINDEX_ALL,
1234                                                     net->ipv4.devconf_all);
1235                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1236                         mroute_clean_tables(mrt, false);
1237                 }
1238         }
1239         rtnl_unlock();
1240 }
1241
1242 /* Socket options and virtual interface manipulation. The whole
1243  * virtual interface system is a complete heap, but unfortunately
1244  * that's how BSD mrouted happens to think. Maybe one day with a proper
1245  * MOSPF/PIM router set up we can clean this up.
1246  */
1247
1248 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval,
1249                          unsigned int optlen)
1250 {
1251         struct net *net = sock_net(sk);
1252         int val, ret = 0, parent = 0;
1253         struct mr_table *mrt;
1254         struct vifctl vif;
1255         struct mfcctl mfc;
1256         u32 uval;
1257
1258         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1259         rtnl_lock();
1260         if (sk->sk_type != SOCK_RAW ||
1261             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1262                 ret = -EOPNOTSUPP;
1263                 goto out_unlock;
1264         }
1265
1266         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1267         if (!mrt) {
1268                 ret = -ENOENT;
1269                 goto out_unlock;
1270         }
1271         if (optname != MRT_INIT) {
1272                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1273                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1274                         ret = -EACCES;
1275                         goto out_unlock;
1276                 }
1277         }
1278
1279         switch (optname) {
1280         case MRT_INIT:
1281                 if (optlen != sizeof(int)) {
1282                         ret = -EINVAL;
1283                         break;
1284                 }
1285                 if (rtnl_dereference(mrt->mroute_sk)) {
1286                         ret = -EADDRINUSE;
1287                         break;
1288                 }
1289
1290                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1291                 if (ret == 0) {
1292                         rcu_assign_pointer(mrt->mroute_sk, sk);
1293                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1294                         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1295                                                     NETCONFA_IFINDEX_ALL,
1296                                                     net->ipv4.devconf_all);
1297                 }
1298                 break;
1299         case MRT_DONE:
1300                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1301                         ret = -EACCES;
1302                 } else {
1303                         /* We need to unlock here because mrtsock_destruct takes
1304                          * care of rtnl itself and we can't change that due to
1305                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1306                          */
1307                         rtnl_unlock();
1308                         ret = ip_ra_control(sk, 0, NULL);
1309                         goto out;
1310                 }
1311                 break;
1312         case MRT_ADD_VIF:
1313         case MRT_DEL_VIF:
1314                 if (optlen != sizeof(vif)) {
1315                         ret = -EINVAL;
1316                         break;
1317                 }
1318                 if (copy_from_user(&vif, optval, sizeof(vif))) {
1319                         ret = -EFAULT;
1320                         break;
1321                 }
1322                 if (vif.vifc_vifi >= MAXVIFS) {
1323                         ret = -ENFILE;
1324                         break;
1325                 }
1326                 if (optname == MRT_ADD_VIF) {
1327                         ret = vif_add(net, mrt, &vif,
1328                                       sk == rtnl_dereference(mrt->mroute_sk));
1329                 } else {
1330                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1331                 }
1332                 break;
1333         /* Manipulate the forwarding caches. These live
1334          * in a sort of kernel/user symbiosis.
1335          */
1336         case MRT_ADD_MFC:
1337         case MRT_DEL_MFC:
1338                 parent = -1;
1339         case MRT_ADD_MFC_PROXY:
1340         case MRT_DEL_MFC_PROXY:
1341                 if (optlen != sizeof(mfc)) {
1342                         ret = -EINVAL;
1343                         break;
1344                 }
1345                 if (copy_from_user(&mfc, optval, sizeof(mfc))) {
1346                         ret = -EFAULT;
1347                         break;
1348                 }
1349                 if (parent == 0)
1350                         parent = mfc.mfcc_parent;
1351                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1352                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1353                 else
1354                         ret = ipmr_mfc_add(net, mrt, &mfc,
1355                                            sk == rtnl_dereference(mrt->mroute_sk),
1356                                            parent);
1357                 break;
1358         /* Control PIM assert. */
1359         case MRT_ASSERT:
1360                 if (optlen != sizeof(val)) {
1361                         ret = -EINVAL;
1362                         break;
1363                 }
1364                 if (get_user(val, (int __user *)optval)) {
1365                         ret = -EFAULT;
1366                         break;
1367                 }
1368                 mrt->mroute_do_assert = val;
1369                 break;
1370         case MRT_PIM:
1371                 if (!ipmr_pimsm_enabled()) {
1372                         ret = -ENOPROTOOPT;
1373                         break;
1374                 }
1375                 if (optlen != sizeof(val)) {
1376                         ret = -EINVAL;
1377                         break;
1378                 }
1379                 if (get_user(val, (int __user *)optval)) {
1380                         ret = -EFAULT;
1381                         break;
1382                 }
1383
1384                 val = !!val;
1385                 if (val != mrt->mroute_do_pim) {
1386                         mrt->mroute_do_pim = val;
1387                         mrt->mroute_do_assert = val;
1388                 }
1389                 break;
1390         case MRT_TABLE:
1391                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1392                         ret = -ENOPROTOOPT;
1393                         break;
1394                 }
1395                 if (optlen != sizeof(uval)) {
1396                         ret = -EINVAL;
1397                         break;
1398                 }
1399                 if (get_user(uval, (u32 __user *)optval)) {
1400                         ret = -EFAULT;
1401                         break;
1402                 }
1403
1404                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1405                         ret = -EBUSY;
1406                 } else {
1407                         mrt = ipmr_new_table(net, uval);
1408                         if (IS_ERR(mrt))
1409                                 ret = PTR_ERR(mrt);
1410                         else
1411                                 raw_sk(sk)->ipmr_table = uval;
1412                 }
1413                 break;
1414         /* Spurious command, or MRT_VERSION which you cannot set. */
1415         default:
1416                 ret = -ENOPROTOOPT;
1417         }
1418 out_unlock:
1419         rtnl_unlock();
1420 out:
1421         return ret;
1422 }
1423
1424 /* Getsock opt support for the multicast routing system. */
1425 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1426 {
1427         int olr;
1428         int val;
1429         struct net *net = sock_net(sk);
1430         struct mr_table *mrt;
1431
1432         if (sk->sk_type != SOCK_RAW ||
1433             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1434                 return -EOPNOTSUPP;
1435
1436         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1437         if (!mrt)
1438                 return -ENOENT;
1439
1440         switch (optname) {
1441         case MRT_VERSION:
1442                 val = 0x0305;
1443                 break;
1444         case MRT_PIM:
1445                 if (!ipmr_pimsm_enabled())
1446                         return -ENOPROTOOPT;
1447                 val = mrt->mroute_do_pim;
1448                 break;
1449         case MRT_ASSERT:
1450                 val = mrt->mroute_do_assert;
1451                 break;
1452         default:
1453                 return -ENOPROTOOPT;
1454         }
1455
1456         if (get_user(olr, optlen))
1457                 return -EFAULT;
1458         olr = min_t(unsigned int, olr, sizeof(int));
1459         if (olr < 0)
1460                 return -EINVAL;
1461         if (put_user(olr, optlen))
1462                 return -EFAULT;
1463         if (copy_to_user(optval, &val, olr))
1464                 return -EFAULT;
1465         return 0;
1466 }
1467
1468 /* The IP multicast ioctl support routines. */
1469 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1470 {
1471         struct sioc_sg_req sr;
1472         struct sioc_vif_req vr;
1473         struct vif_device *vif;
1474         struct mfc_cache *c;
1475         struct net *net = sock_net(sk);
1476         struct mr_table *mrt;
1477
1478         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1479         if (!mrt)
1480                 return -ENOENT;
1481
1482         switch (cmd) {
1483         case SIOCGETVIFCNT:
1484                 if (copy_from_user(&vr, arg, sizeof(vr)))
1485                         return -EFAULT;
1486                 if (vr.vifi >= mrt->maxvif)
1487                         return -EINVAL;
1488                 read_lock(&mrt_lock);
1489                 vif = &mrt->vif_table[vr.vifi];
1490                 if (VIF_EXISTS(mrt, vr.vifi)) {
1491                         vr.icount = vif->pkt_in;
1492                         vr.ocount = vif->pkt_out;
1493                         vr.ibytes = vif->bytes_in;
1494                         vr.obytes = vif->bytes_out;
1495                         read_unlock(&mrt_lock);
1496
1497                         if (copy_to_user(arg, &vr, sizeof(vr)))
1498                                 return -EFAULT;
1499                         return 0;
1500                 }
1501                 read_unlock(&mrt_lock);
1502                 return -EADDRNOTAVAIL;
1503         case SIOCGETSGCNT:
1504                 if (copy_from_user(&sr, arg, sizeof(sr)))
1505                         return -EFAULT;
1506
1507                 rcu_read_lock();
1508                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1509                 if (c) {
1510                         sr.pktcnt = c->mfc_un.res.pkt;
1511                         sr.bytecnt = c->mfc_un.res.bytes;
1512                         sr.wrong_if = c->mfc_un.res.wrong_if;
1513                         rcu_read_unlock();
1514
1515                         if (copy_to_user(arg, &sr, sizeof(sr)))
1516                                 return -EFAULT;
1517                         return 0;
1518                 }
1519                 rcu_read_unlock();
1520                 return -EADDRNOTAVAIL;
1521         default:
1522                 return -ENOIOCTLCMD;
1523         }
1524 }
1525
1526 #ifdef CONFIG_COMPAT
1527 struct compat_sioc_sg_req {
1528         struct in_addr src;
1529         struct in_addr grp;
1530         compat_ulong_t pktcnt;
1531         compat_ulong_t bytecnt;
1532         compat_ulong_t wrong_if;
1533 };
1534
1535 struct compat_sioc_vif_req {
1536         vifi_t  vifi;           /* Which iface */
1537         compat_ulong_t icount;
1538         compat_ulong_t ocount;
1539         compat_ulong_t ibytes;
1540         compat_ulong_t obytes;
1541 };
1542
1543 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1544 {
1545         struct compat_sioc_sg_req sr;
1546         struct compat_sioc_vif_req vr;
1547         struct vif_device *vif;
1548         struct mfc_cache *c;
1549         struct net *net = sock_net(sk);
1550         struct mr_table *mrt;
1551
1552         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1553         if (!mrt)
1554                 return -ENOENT;
1555
1556         switch (cmd) {
1557         case SIOCGETVIFCNT:
1558                 if (copy_from_user(&vr, arg, sizeof(vr)))
1559                         return -EFAULT;
1560                 if (vr.vifi >= mrt->maxvif)
1561                         return -EINVAL;
1562                 read_lock(&mrt_lock);
1563                 vif = &mrt->vif_table[vr.vifi];
1564                 if (VIF_EXISTS(mrt, vr.vifi)) {
1565                         vr.icount = vif->pkt_in;
1566                         vr.ocount = vif->pkt_out;
1567                         vr.ibytes = vif->bytes_in;
1568                         vr.obytes = vif->bytes_out;
1569                         read_unlock(&mrt_lock);
1570
1571                         if (copy_to_user(arg, &vr, sizeof(vr)))
1572                                 return -EFAULT;
1573                         return 0;
1574                 }
1575                 read_unlock(&mrt_lock);
1576                 return -EADDRNOTAVAIL;
1577         case SIOCGETSGCNT:
1578                 if (copy_from_user(&sr, arg, sizeof(sr)))
1579                         return -EFAULT;
1580
1581                 rcu_read_lock();
1582                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1583                 if (c) {
1584                         sr.pktcnt = c->mfc_un.res.pkt;
1585                         sr.bytecnt = c->mfc_un.res.bytes;
1586                         sr.wrong_if = c->mfc_un.res.wrong_if;
1587                         rcu_read_unlock();
1588
1589                         if (copy_to_user(arg, &sr, sizeof(sr)))
1590                                 return -EFAULT;
1591                         return 0;
1592                 }
1593                 rcu_read_unlock();
1594                 return -EADDRNOTAVAIL;
1595         default:
1596                 return -ENOIOCTLCMD;
1597         }
1598 }
1599 #endif
1600
1601 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1602 {
1603         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1604         struct net *net = dev_net(dev);
1605         struct mr_table *mrt;
1606         struct vif_device *v;
1607         int ct;
1608
1609         if (event != NETDEV_UNREGISTER)
1610                 return NOTIFY_DONE;
1611
1612         ipmr_for_each_table(mrt, net) {
1613                 v = &mrt->vif_table[0];
1614                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1615                         if (v->dev == dev)
1616                                 vif_delete(mrt, ct, 1, NULL);
1617                 }
1618         }
1619         return NOTIFY_DONE;
1620 }
1621
1622 static struct notifier_block ip_mr_notifier = {
1623         .notifier_call = ipmr_device_event,
1624 };
1625
1626 /* Encapsulate a packet by attaching a valid IPIP header to it.
1627  * This avoids tunnel drivers and other mess and gives us the speed so
1628  * important for multicast video.
1629  */
1630 static void ip_encap(struct net *net, struct sk_buff *skb,
1631                      __be32 saddr, __be32 daddr)
1632 {
1633         struct iphdr *iph;
1634         const struct iphdr *old_iph = ip_hdr(skb);
1635
1636         skb_push(skb, sizeof(struct iphdr));
1637         skb->transport_header = skb->network_header;
1638         skb_reset_network_header(skb);
1639         iph = ip_hdr(skb);
1640
1641         iph->version    =       4;
1642         iph->tos        =       old_iph->tos;
1643         iph->ttl        =       old_iph->ttl;
1644         iph->frag_off   =       0;
1645         iph->daddr      =       daddr;
1646         iph->saddr      =       saddr;
1647         iph->protocol   =       IPPROTO_IPIP;
1648         iph->ihl        =       5;
1649         iph->tot_len    =       htons(skb->len);
1650         ip_select_ident(net, skb, NULL);
1651         ip_send_check(iph);
1652
1653         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1654         nf_reset(skb);
1655 }
1656
1657 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1658                                       struct sk_buff *skb)
1659 {
1660         struct ip_options *opt = &(IPCB(skb)->opt);
1661
1662         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1663         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1664
1665         if (unlikely(opt->optlen))
1666                 ip_forward_options(skb);
1667
1668         return dst_output(net, sk, skb);
1669 }
1670
1671 /* Processing handlers for ipmr_forward */
1672
1673 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1674                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1675 {
1676         const struct iphdr *iph = ip_hdr(skb);
1677         struct vif_device *vif = &mrt->vif_table[vifi];
1678         struct net_device *dev;
1679         struct rtable *rt;
1680         struct flowi4 fl4;
1681         int    encap = 0;
1682
1683         if (!vif->dev)
1684                 goto out_free;
1685
1686         if (vif->flags & VIFF_REGISTER) {
1687                 vif->pkt_out++;
1688                 vif->bytes_out += skb->len;
1689                 vif->dev->stats.tx_bytes += skb->len;
1690                 vif->dev->stats.tx_packets++;
1691                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1692                 goto out_free;
1693         }
1694
1695         if (vif->flags & VIFF_TUNNEL) {
1696                 rt = ip_route_output_ports(net, &fl4, NULL,
1697                                            vif->remote, vif->local,
1698                                            0, 0,
1699                                            IPPROTO_IPIP,
1700                                            RT_TOS(iph->tos), vif->link);
1701                 if (IS_ERR(rt))
1702                         goto out_free;
1703                 encap = sizeof(struct iphdr);
1704         } else {
1705                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1706                                            0, 0,
1707                                            IPPROTO_IPIP,
1708                                            RT_TOS(iph->tos), vif->link);
1709                 if (IS_ERR(rt))
1710                         goto out_free;
1711         }
1712
1713         dev = rt->dst.dev;
1714
1715         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1716                 /* Do not fragment multicasts. Alas, IPv4 does not
1717                  * allow to send ICMP, so that packets will disappear
1718                  * to blackhole.
1719                  */
1720                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1721                 ip_rt_put(rt);
1722                 goto out_free;
1723         }
1724
1725         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1726
1727         if (skb_cow(skb, encap)) {
1728                 ip_rt_put(rt);
1729                 goto out_free;
1730         }
1731
1732         vif->pkt_out++;
1733         vif->bytes_out += skb->len;
1734
1735         skb_dst_drop(skb);
1736         skb_dst_set(skb, &rt->dst);
1737         ip_decrease_ttl(ip_hdr(skb));
1738
1739         /* FIXME: forward and output firewalls used to be called here.
1740          * What do we do with netfilter? -- RR
1741          */
1742         if (vif->flags & VIFF_TUNNEL) {
1743                 ip_encap(net, skb, vif->local, vif->remote);
1744                 /* FIXME: extra output firewall step used to be here. --RR */
1745                 vif->dev->stats.tx_packets++;
1746                 vif->dev->stats.tx_bytes += skb->len;
1747         }
1748
1749         IPCB(skb)->flags |= IPSKB_FORWARDED;
1750
1751         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1752          * not only before forwarding, but after forwarding on all output
1753          * interfaces. It is clear, if mrouter runs a multicasting
1754          * program, it should receive packets not depending to what interface
1755          * program is joined.
1756          * If we will not make it, the program will have to join on all
1757          * interfaces. On the other hand, multihoming host (or router, but
1758          * not mrouter) cannot join to more than one interface - it will
1759          * result in receiving multiple packets.
1760          */
1761         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1762                 net, NULL, skb, skb->dev, dev,
1763                 ipmr_forward_finish);
1764         return;
1765
1766 out_free:
1767         kfree_skb(skb);
1768 }
1769
1770 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1771 {
1772         int ct;
1773
1774         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1775                 if (mrt->vif_table[ct].dev == dev)
1776                         break;
1777         }
1778         return ct;
1779 }
1780
1781 /* "local" means that we should preserve one skb (for local delivery) */
1782 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1783                           struct sk_buff *skb, struct mfc_cache *cache,
1784                           int local)
1785 {
1786         int psend = -1;
1787         int vif, ct;
1788         int true_vifi = ipmr_find_vif(mrt, skb->dev);
1789
1790         vif = cache->mfc_parent;
1791         cache->mfc_un.res.pkt++;
1792         cache->mfc_un.res.bytes += skb->len;
1793
1794         if (cache->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1795                 struct mfc_cache *cache_proxy;
1796
1797                 /* For an (*,G) entry, we only check that the incomming
1798                  * interface is part of the static tree.
1799                  */
1800                 cache_proxy = ipmr_cache_find_any_parent(mrt, vif);
1801                 if (cache_proxy &&
1802                     cache_proxy->mfc_un.res.ttls[true_vifi] < 255)
1803                         goto forward;
1804         }
1805
1806         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1807         if (mrt->vif_table[vif].dev != skb->dev) {
1808                 if (rt_is_output_route(skb_rtable(skb))) {
1809                         /* It is our own packet, looped back.
1810                          * Very complicated situation...
1811                          *
1812                          * The best workaround until routing daemons will be
1813                          * fixed is not to redistribute packet, if it was
1814                          * send through wrong interface. It means, that
1815                          * multicast applications WILL NOT work for
1816                          * (S,G), which have default multicast route pointing
1817                          * to wrong oif. In any case, it is not a good
1818                          * idea to use multicasting applications on router.
1819                          */
1820                         goto dont_forward;
1821                 }
1822
1823                 cache->mfc_un.res.wrong_if++;
1824
1825                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1826                     /* pimsm uses asserts, when switching from RPT to SPT,
1827                      * so that we cannot check that packet arrived on an oif.
1828                      * It is bad, but otherwise we would need to move pretty
1829                      * large chunk of pimd to kernel. Ough... --ANK
1830                      */
1831                     (mrt->mroute_do_pim ||
1832                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1833                     time_after(jiffies,
1834                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1835                         cache->mfc_un.res.last_assert = jiffies;
1836                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1837                 }
1838                 goto dont_forward;
1839         }
1840
1841 forward:
1842         mrt->vif_table[vif].pkt_in++;
1843         mrt->vif_table[vif].bytes_in += skb->len;
1844
1845         /* Forward the frame */
1846         if (cache->mfc_origin == htonl(INADDR_ANY) &&
1847             cache->mfc_mcastgrp == htonl(INADDR_ANY)) {
1848                 if (true_vifi >= 0 &&
1849                     true_vifi != cache->mfc_parent &&
1850                     ip_hdr(skb)->ttl >
1851                                 cache->mfc_un.res.ttls[cache->mfc_parent]) {
1852                         /* It's an (*,*) entry and the packet is not coming from
1853                          * the upstream: forward the packet to the upstream
1854                          * only.
1855                          */
1856                         psend = cache->mfc_parent;
1857                         goto last_forward;
1858                 }
1859                 goto dont_forward;
1860         }
1861         for (ct = cache->mfc_un.res.maxvif - 1;
1862              ct >= cache->mfc_un.res.minvif; ct--) {
1863                 /* For (*,G) entry, don't forward to the incoming interface */
1864                 if ((cache->mfc_origin != htonl(INADDR_ANY) ||
1865                      ct != true_vifi) &&
1866                     ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1867                         if (psend != -1) {
1868                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1869
1870                                 if (skb2)
1871                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1872                                                         psend);
1873                         }
1874                         psend = ct;
1875                 }
1876         }
1877 last_forward:
1878         if (psend != -1) {
1879                 if (local) {
1880                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1881
1882                         if (skb2)
1883                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1884                 } else {
1885                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1886                         return;
1887                 }
1888         }
1889
1890 dont_forward:
1891         if (!local)
1892                 kfree_skb(skb);
1893 }
1894
1895 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1896 {
1897         struct rtable *rt = skb_rtable(skb);
1898         struct iphdr *iph = ip_hdr(skb);
1899         struct flowi4 fl4 = {
1900                 .daddr = iph->daddr,
1901                 .saddr = iph->saddr,
1902                 .flowi4_tos = RT_TOS(iph->tos),
1903                 .flowi4_oif = (rt_is_output_route(rt) ?
1904                                skb->dev->ifindex : 0),
1905                 .flowi4_iif = (rt_is_output_route(rt) ?
1906                                LOOPBACK_IFINDEX :
1907                                skb->dev->ifindex),
1908                 .flowi4_mark = skb->mark,
1909         };
1910         struct mr_table *mrt;
1911         int err;
1912
1913         err = ipmr_fib_lookup(net, &fl4, &mrt);
1914         if (err)
1915                 return ERR_PTR(err);
1916         return mrt;
1917 }
1918
1919 /* Multicast packets for forwarding arrive here
1920  * Called with rcu_read_lock();
1921  */
1922 int ip_mr_input(struct sk_buff *skb)
1923 {
1924         struct mfc_cache *cache;
1925         struct net *net = dev_net(skb->dev);
1926         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1927         struct mr_table *mrt;
1928
1929         /* Packet is looped back after forward, it should not be
1930          * forwarded second time, but still can be delivered locally.
1931          */
1932         if (IPCB(skb)->flags & IPSKB_FORWARDED)
1933                 goto dont_forward;
1934
1935         mrt = ipmr_rt_fib_lookup(net, skb);
1936         if (IS_ERR(mrt)) {
1937                 kfree_skb(skb);
1938                 return PTR_ERR(mrt);
1939         }
1940         if (!local) {
1941                 if (IPCB(skb)->opt.router_alert) {
1942                         if (ip_call_ra_chain(skb))
1943                                 return 0;
1944                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1945                         /* IGMPv1 (and broken IGMPv2 implementations sort of
1946                          * Cisco IOS <= 11.2(8)) do not put router alert
1947                          * option to IGMP packets destined to routable
1948                          * groups. It is very bad, because it means
1949                          * that we can forward NO IGMP messages.
1950                          */
1951                         struct sock *mroute_sk;
1952
1953                         mroute_sk = rcu_dereference(mrt->mroute_sk);
1954                         if (mroute_sk) {
1955                                 nf_reset(skb);
1956                                 raw_rcv(mroute_sk, skb);
1957                                 return 0;
1958                         }
1959                     }
1960         }
1961
1962         /* already under rcu_read_lock() */
1963         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1964         if (!cache) {
1965                 int vif = ipmr_find_vif(mrt, skb->dev);
1966
1967                 if (vif >= 0)
1968                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
1969                                                     vif);
1970         }
1971
1972         /* No usable cache entry */
1973         if (!cache) {
1974                 int vif;
1975
1976                 if (local) {
1977                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1978                         ip_local_deliver(skb);
1979                         if (!skb2)
1980                                 return -ENOBUFS;
1981                         skb = skb2;
1982                 }
1983
1984                 read_lock(&mrt_lock);
1985                 vif = ipmr_find_vif(mrt, skb->dev);
1986                 if (vif >= 0) {
1987                         int err2 = ipmr_cache_unresolved(mrt, vif, skb);
1988                         read_unlock(&mrt_lock);
1989
1990                         return err2;
1991                 }
1992                 read_unlock(&mrt_lock);
1993                 kfree_skb(skb);
1994                 return -ENODEV;
1995         }
1996
1997         read_lock(&mrt_lock);
1998         ip_mr_forward(net, mrt, skb, cache, local);
1999         read_unlock(&mrt_lock);
2000
2001         if (local)
2002                 return ip_local_deliver(skb);
2003
2004         return 0;
2005
2006 dont_forward:
2007         if (local)
2008                 return ip_local_deliver(skb);
2009         kfree_skb(skb);
2010         return 0;
2011 }
2012
2013 #ifdef CONFIG_IP_PIMSM_V1
2014 /* Handle IGMP messages of PIMv1 */
2015 int pim_rcv_v1(struct sk_buff *skb)
2016 {
2017         struct igmphdr *pim;
2018         struct net *net = dev_net(skb->dev);
2019         struct mr_table *mrt;
2020
2021         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2022                 goto drop;
2023
2024         pim = igmp_hdr(skb);
2025
2026         mrt = ipmr_rt_fib_lookup(net, skb);
2027         if (IS_ERR(mrt))
2028                 goto drop;
2029         if (!mrt->mroute_do_pim ||
2030             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2031                 goto drop;
2032
2033         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2034 drop:
2035                 kfree_skb(skb);
2036         }
2037         return 0;
2038 }
2039 #endif
2040
2041 #ifdef CONFIG_IP_PIMSM_V2
2042 static int pim_rcv(struct sk_buff *skb)
2043 {
2044         struct pimreghdr *pim;
2045         struct net *net = dev_net(skb->dev);
2046         struct mr_table *mrt;
2047
2048         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2049                 goto drop;
2050
2051         pim = (struct pimreghdr *)skb_transport_header(skb);
2052         if (pim->type != ((PIM_VERSION << 4) | (PIM_REGISTER)) ||
2053             (pim->flags & PIM_NULL_REGISTER) ||
2054             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2055              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2056                 goto drop;
2057
2058         mrt = ipmr_rt_fib_lookup(net, skb);
2059         if (IS_ERR(mrt))
2060                 goto drop;
2061         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2062 drop:
2063                 kfree_skb(skb);
2064         }
2065         return 0;
2066 }
2067 #endif
2068
2069 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2070                               struct mfc_cache *c, struct rtmsg *rtm)
2071 {
2072         int ct;
2073         struct rtnexthop *nhp;
2074         struct nlattr *mp_attr;
2075         struct rta_mfc_stats mfcs;
2076
2077         /* If cache is unresolved, don't try to parse IIF and OIF */
2078         if (c->mfc_parent >= MAXVIFS)
2079                 return -ENOENT;
2080
2081         if (VIF_EXISTS(mrt, c->mfc_parent) &&
2082             nla_put_u32(skb, RTA_IIF, mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
2083                 return -EMSGSIZE;
2084
2085         if (!(mp_attr = nla_nest_start(skb, RTA_MULTIPATH)))
2086                 return -EMSGSIZE;
2087
2088         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2089                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2090                         if (!(nhp = nla_reserve_nohdr(skb, sizeof(*nhp)))) {
2091                                 nla_nest_cancel(skb, mp_attr);
2092                                 return -EMSGSIZE;
2093                         }
2094
2095                         nhp->rtnh_flags = 0;
2096                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2097                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2098                         nhp->rtnh_len = sizeof(*nhp);
2099                 }
2100         }
2101
2102         nla_nest_end(skb, mp_attr);
2103
2104         mfcs.mfcs_packets = c->mfc_un.res.pkt;
2105         mfcs.mfcs_bytes = c->mfc_un.res.bytes;
2106         mfcs.mfcs_wrong_if = c->mfc_un.res.wrong_if;
2107         if (nla_put_64bit(skb, RTA_MFC_STATS, sizeof(mfcs), &mfcs, RTA_PAD) < 0)
2108                 return -EMSGSIZE;
2109
2110         rtm->rtm_type = RTN_MULTICAST;
2111         return 1;
2112 }
2113
2114 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2115                    __be32 saddr, __be32 daddr,
2116                    struct rtmsg *rtm, int nowait)
2117 {
2118         struct mfc_cache *cache;
2119         struct mr_table *mrt;
2120         int err;
2121
2122         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2123         if (!mrt)
2124                 return -ENOENT;
2125
2126         rcu_read_lock();
2127         cache = ipmr_cache_find(mrt, saddr, daddr);
2128         if (!cache && skb->dev) {
2129                 int vif = ipmr_find_vif(mrt, skb->dev);
2130
2131                 if (vif >= 0)
2132                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2133         }
2134         if (!cache) {
2135                 struct sk_buff *skb2;
2136                 struct iphdr *iph;
2137                 struct net_device *dev;
2138                 int vif = -1;
2139
2140                 if (nowait) {
2141                         rcu_read_unlock();
2142                         return -EAGAIN;
2143                 }
2144
2145                 dev = skb->dev;
2146                 read_lock(&mrt_lock);
2147                 if (dev)
2148                         vif = ipmr_find_vif(mrt, dev);
2149                 if (vif < 0) {
2150                         read_unlock(&mrt_lock);
2151                         rcu_read_unlock();
2152                         return -ENODEV;
2153                 }
2154                 skb2 = skb_clone(skb, GFP_ATOMIC);
2155                 if (!skb2) {
2156                         read_unlock(&mrt_lock);
2157                         rcu_read_unlock();
2158                         return -ENOMEM;
2159                 }
2160
2161                 skb_push(skb2, sizeof(struct iphdr));
2162                 skb_reset_network_header(skb2);
2163                 iph = ip_hdr(skb2);
2164                 iph->ihl = sizeof(struct iphdr) >> 2;
2165                 iph->saddr = saddr;
2166                 iph->daddr = daddr;
2167                 iph->version = 0;
2168                 err = ipmr_cache_unresolved(mrt, vif, skb2);
2169                 read_unlock(&mrt_lock);
2170                 rcu_read_unlock();
2171                 return err;
2172         }
2173
2174         read_lock(&mrt_lock);
2175         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2176         read_unlock(&mrt_lock);
2177         rcu_read_unlock();
2178         return err;
2179 }
2180
2181 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2182                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2183                             int flags)
2184 {
2185         struct nlmsghdr *nlh;
2186         struct rtmsg *rtm;
2187         int err;
2188
2189         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2190         if (!nlh)
2191                 return -EMSGSIZE;
2192
2193         rtm = nlmsg_data(nlh);
2194         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2195         rtm->rtm_dst_len  = 32;
2196         rtm->rtm_src_len  = 32;
2197         rtm->rtm_tos      = 0;
2198         rtm->rtm_table    = mrt->id;
2199         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2200                 goto nla_put_failure;
2201         rtm->rtm_type     = RTN_MULTICAST;
2202         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2203         if (c->mfc_flags & MFC_STATIC)
2204                 rtm->rtm_protocol = RTPROT_STATIC;
2205         else
2206                 rtm->rtm_protocol = RTPROT_MROUTED;
2207         rtm->rtm_flags    = 0;
2208
2209         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2210             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2211                 goto nla_put_failure;
2212         err = __ipmr_fill_mroute(mrt, skb, c, rtm);
2213         /* do not break the dump if cache is unresolved */
2214         if (err < 0 && err != -ENOENT)
2215                 goto nla_put_failure;
2216
2217         nlmsg_end(skb, nlh);
2218         return 0;
2219
2220 nla_put_failure:
2221         nlmsg_cancel(skb, nlh);
2222         return -EMSGSIZE;
2223 }
2224
2225 static size_t mroute_msgsize(bool unresolved, int maxvif)
2226 {
2227         size_t len =
2228                 NLMSG_ALIGN(sizeof(struct rtmsg))
2229                 + nla_total_size(4)     /* RTA_TABLE */
2230                 + nla_total_size(4)     /* RTA_SRC */
2231                 + nla_total_size(4)     /* RTA_DST */
2232                 ;
2233
2234         if (!unresolved)
2235                 len = len
2236                       + nla_total_size(4)       /* RTA_IIF */
2237                       + nla_total_size(0)       /* RTA_MULTIPATH */
2238                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2239                                                 /* RTA_MFC_STATS */
2240                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2241                 ;
2242
2243         return len;
2244 }
2245
2246 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2247                                  int cmd)
2248 {
2249         struct net *net = read_pnet(&mrt->net);
2250         struct sk_buff *skb;
2251         int err = -ENOBUFS;
2252
2253         skb = nlmsg_new(mroute_msgsize(mfc->mfc_parent >= MAXVIFS, mrt->maxvif),
2254                         GFP_ATOMIC);
2255         if (!skb)
2256                 goto errout;
2257
2258         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2259         if (err < 0)
2260                 goto errout;
2261
2262         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2263         return;
2264
2265 errout:
2266         kfree_skb(skb);
2267         if (err < 0)
2268                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2269 }
2270
2271 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2272 {
2273         struct net *net = sock_net(skb->sk);
2274         struct mr_table *mrt;
2275         struct mfc_cache *mfc;
2276         unsigned int t = 0, s_t;
2277         unsigned int h = 0, s_h;
2278         unsigned int e = 0, s_e;
2279
2280         s_t = cb->args[0];
2281         s_h = cb->args[1];
2282         s_e = cb->args[2];
2283
2284         rcu_read_lock();
2285         ipmr_for_each_table(mrt, net) {
2286                 if (t < s_t)
2287                         goto next_table;
2288                 if (t > s_t)
2289                         s_h = 0;
2290                 for (h = s_h; h < MFC_LINES; h++) {
2291                         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2292                                 if (e < s_e)
2293                                         goto next_entry;
2294                                 if (ipmr_fill_mroute(mrt, skb,
2295                                                      NETLINK_CB(cb->skb).portid,
2296                                                      cb->nlh->nlmsg_seq,
2297                                                      mfc, RTM_NEWROUTE,
2298                                                      NLM_F_MULTI) < 0)
2299                                         goto done;
2300 next_entry:
2301                                 e++;
2302                         }
2303                         e = s_e = 0;
2304                 }
2305                 spin_lock_bh(&mfc_unres_lock);
2306                 list_for_each_entry(mfc, &mrt->mfc_unres_queue, list) {
2307                         if (e < s_e)
2308                                 goto next_entry2;
2309                         if (ipmr_fill_mroute(mrt, skb,
2310                                              NETLINK_CB(cb->skb).portid,
2311                                              cb->nlh->nlmsg_seq,
2312                                              mfc, RTM_NEWROUTE,
2313                                              NLM_F_MULTI) < 0) {
2314                                 spin_unlock_bh(&mfc_unres_lock);
2315                                 goto done;
2316                         }
2317 next_entry2:
2318                         e++;
2319                 }
2320                 spin_unlock_bh(&mfc_unres_lock);
2321                 e = s_e = 0;
2322                 s_h = 0;
2323 next_table:
2324                 t++;
2325         }
2326 done:
2327         rcu_read_unlock();
2328
2329         cb->args[2] = e;
2330         cb->args[1] = h;
2331         cb->args[0] = t;
2332
2333         return skb->len;
2334 }
2335
2336 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2337         [RTA_SRC]       = { .type = NLA_U32 },
2338         [RTA_DST]       = { .type = NLA_U32 },
2339         [RTA_IIF]       = { .type = NLA_U32 },
2340         [RTA_TABLE]     = { .type = NLA_U32 },
2341         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2342 };
2343
2344 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2345 {
2346         switch (rtm_protocol) {
2347         case RTPROT_STATIC:
2348         case RTPROT_MROUTED:
2349                 return true;
2350         }
2351         return false;
2352 }
2353
2354 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2355 {
2356         struct rtnexthop *rtnh = nla_data(nla);
2357         int remaining = nla_len(nla), vifi = 0;
2358
2359         while (rtnh_ok(rtnh, remaining)) {
2360                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2361                 if (++vifi == MAXVIFS)
2362                         break;
2363                 rtnh = rtnh_next(rtnh, &remaining);
2364         }
2365
2366         return remaining > 0 ? -EINVAL : vifi;
2367 }
2368
2369 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2370 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2371                             struct mfcctl *mfcc, int *mrtsock,
2372                             struct mr_table **mrtret)
2373 {
2374         struct net_device *dev = NULL;
2375         u32 tblid = RT_TABLE_DEFAULT;
2376         struct mr_table *mrt;
2377         struct nlattr *attr;
2378         struct rtmsg *rtm;
2379         int ret, rem;
2380
2381         ret = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipmr_policy);
2382         if (ret < 0)
2383                 goto out;
2384         rtm = nlmsg_data(nlh);
2385
2386         ret = -EINVAL;
2387         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2388             rtm->rtm_type != RTN_MULTICAST ||
2389             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2390             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2391                 goto out;
2392
2393         memset(mfcc, 0, sizeof(*mfcc));
2394         mfcc->mfcc_parent = -1;
2395         ret = 0;
2396         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2397                 switch (nla_type(attr)) {
2398                 case RTA_SRC:
2399                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2400                         break;
2401                 case RTA_DST:
2402                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2403                         break;
2404                 case RTA_IIF:
2405                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2406                         if (!dev) {
2407                                 ret = -ENODEV;
2408                                 goto out;
2409                         }
2410                         break;
2411                 case RTA_MULTIPATH:
2412                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2413                                 ret = -EINVAL;
2414                                 goto out;
2415                         }
2416                         break;
2417                 case RTA_PREFSRC:
2418                         ret = 1;
2419                         break;
2420                 case RTA_TABLE:
2421                         tblid = nla_get_u32(attr);
2422                         break;
2423                 }
2424         }
2425         mrt = ipmr_get_table(net, tblid);
2426         if (!mrt) {
2427                 ret = -ENOENT;
2428                 goto out;
2429         }
2430         *mrtret = mrt;
2431         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2432         if (dev)
2433                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2434
2435 out:
2436         return ret;
2437 }
2438
2439 /* takes care of both newroute and delroute */
2440 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh)
2441 {
2442         struct net *net = sock_net(skb->sk);
2443         int ret, mrtsock, parent;
2444         struct mr_table *tbl;
2445         struct mfcctl mfcc;
2446
2447         mrtsock = 0;
2448         tbl = NULL;
2449         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl);
2450         if (ret < 0)
2451                 return ret;
2452
2453         parent = ret ? mfcc.mfcc_parent : -1;
2454         if (nlh->nlmsg_type == RTM_NEWROUTE)
2455                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2456         else
2457                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2458 }
2459
2460 #ifdef CONFIG_PROC_FS
2461 /* The /proc interfaces to multicast routing :
2462  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2463  */
2464 struct ipmr_vif_iter {
2465         struct seq_net_private p;
2466         struct mr_table *mrt;
2467         int ct;
2468 };
2469
2470 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2471                                            struct ipmr_vif_iter *iter,
2472                                            loff_t pos)
2473 {
2474         struct mr_table *mrt = iter->mrt;
2475
2476         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2477                 if (!VIF_EXISTS(mrt, iter->ct))
2478                         continue;
2479                 if (pos-- == 0)
2480                         return &mrt->vif_table[iter->ct];
2481         }
2482         return NULL;
2483 }
2484
2485 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2486         __acquires(mrt_lock)
2487 {
2488         struct ipmr_vif_iter *iter = seq->private;
2489         struct net *net = seq_file_net(seq);
2490         struct mr_table *mrt;
2491
2492         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2493         if (!mrt)
2494                 return ERR_PTR(-ENOENT);
2495
2496         iter->mrt = mrt;
2497
2498         read_lock(&mrt_lock);
2499         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2500                 : SEQ_START_TOKEN;
2501 }
2502
2503 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2504 {
2505         struct ipmr_vif_iter *iter = seq->private;
2506         struct net *net = seq_file_net(seq);
2507         struct mr_table *mrt = iter->mrt;
2508
2509         ++*pos;
2510         if (v == SEQ_START_TOKEN)
2511                 return ipmr_vif_seq_idx(net, iter, 0);
2512
2513         while (++iter->ct < mrt->maxvif) {
2514                 if (!VIF_EXISTS(mrt, iter->ct))
2515                         continue;
2516                 return &mrt->vif_table[iter->ct];
2517         }
2518         return NULL;
2519 }
2520
2521 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2522         __releases(mrt_lock)
2523 {
2524         read_unlock(&mrt_lock);
2525 }
2526
2527 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2528 {
2529         struct ipmr_vif_iter *iter = seq->private;
2530         struct mr_table *mrt = iter->mrt;
2531
2532         if (v == SEQ_START_TOKEN) {
2533                 seq_puts(seq,
2534                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2535         } else {
2536                 const struct vif_device *vif = v;
2537                 const char *name =  vif->dev ? vif->dev->name : "none";
2538
2539                 seq_printf(seq,
2540                            "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2541                            vif - mrt->vif_table,
2542                            name, vif->bytes_in, vif->pkt_in,
2543                            vif->bytes_out, vif->pkt_out,
2544                            vif->flags, vif->local, vif->remote);
2545         }
2546         return 0;
2547 }
2548
2549 static const struct seq_operations ipmr_vif_seq_ops = {
2550         .start = ipmr_vif_seq_start,
2551         .next  = ipmr_vif_seq_next,
2552         .stop  = ipmr_vif_seq_stop,
2553         .show  = ipmr_vif_seq_show,
2554 };
2555
2556 static int ipmr_vif_open(struct inode *inode, struct file *file)
2557 {
2558         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2559                             sizeof(struct ipmr_vif_iter));
2560 }
2561
2562 static const struct file_operations ipmr_vif_fops = {
2563         .owner   = THIS_MODULE,
2564         .open    = ipmr_vif_open,
2565         .read    = seq_read,
2566         .llseek  = seq_lseek,
2567         .release = seq_release_net,
2568 };
2569
2570 struct ipmr_mfc_iter {
2571         struct seq_net_private p;
2572         struct mr_table *mrt;
2573         struct list_head *cache;
2574         int ct;
2575 };
2576
2577
2578 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2579                                           struct ipmr_mfc_iter *it, loff_t pos)
2580 {
2581         struct mr_table *mrt = it->mrt;
2582         struct mfc_cache *mfc;
2583
2584         rcu_read_lock();
2585         for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2586                 it->cache = &mrt->mfc_cache_array[it->ct];
2587                 list_for_each_entry_rcu(mfc, it->cache, list)
2588                         if (pos-- == 0)
2589                                 return mfc;
2590         }
2591         rcu_read_unlock();
2592
2593         spin_lock_bh(&mfc_unres_lock);
2594         it->cache = &mrt->mfc_unres_queue;
2595         list_for_each_entry(mfc, it->cache, list)
2596                 if (pos-- == 0)
2597                         return mfc;
2598         spin_unlock_bh(&mfc_unres_lock);
2599
2600         it->cache = NULL;
2601         return NULL;
2602 }
2603
2604
2605 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2606 {
2607         struct ipmr_mfc_iter *it = seq->private;
2608         struct net *net = seq_file_net(seq);
2609         struct mr_table *mrt;
2610
2611         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2612         if (!mrt)
2613                 return ERR_PTR(-ENOENT);
2614
2615         it->mrt = mrt;
2616         it->cache = NULL;
2617         it->ct = 0;
2618         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2619                 : SEQ_START_TOKEN;
2620 }
2621
2622 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2623 {
2624         struct mfc_cache *mfc = v;
2625         struct ipmr_mfc_iter *it = seq->private;
2626         struct net *net = seq_file_net(seq);
2627         struct mr_table *mrt = it->mrt;
2628
2629         ++*pos;
2630
2631         if (v == SEQ_START_TOKEN)
2632                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2633
2634         if (mfc->list.next != it->cache)
2635                 return list_entry(mfc->list.next, struct mfc_cache, list);
2636
2637         if (it->cache == &mrt->mfc_unres_queue)
2638                 goto end_of_list;
2639
2640         BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2641
2642         while (++it->ct < MFC_LINES) {
2643                 it->cache = &mrt->mfc_cache_array[it->ct];
2644                 if (list_empty(it->cache))
2645                         continue;
2646                 return list_first_entry(it->cache, struct mfc_cache, list);
2647         }
2648
2649         /* exhausted cache_array, show unresolved */
2650         rcu_read_unlock();
2651         it->cache = &mrt->mfc_unres_queue;
2652         it->ct = 0;
2653
2654         spin_lock_bh(&mfc_unres_lock);
2655         if (!list_empty(it->cache))
2656                 return list_first_entry(it->cache, struct mfc_cache, list);
2657
2658 end_of_list:
2659         spin_unlock_bh(&mfc_unres_lock);
2660         it->cache = NULL;
2661
2662         return NULL;
2663 }
2664
2665 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2666 {
2667         struct ipmr_mfc_iter *it = seq->private;
2668         struct mr_table *mrt = it->mrt;
2669
2670         if (it->cache == &mrt->mfc_unres_queue)
2671                 spin_unlock_bh(&mfc_unres_lock);
2672         else if (it->cache == &mrt->mfc_cache_array[it->ct])
2673                 rcu_read_unlock();
2674 }
2675
2676 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2677 {
2678         int n;
2679
2680         if (v == SEQ_START_TOKEN) {
2681                 seq_puts(seq,
2682                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2683         } else {
2684                 const struct mfc_cache *mfc = v;
2685                 const struct ipmr_mfc_iter *it = seq->private;
2686                 const struct mr_table *mrt = it->mrt;
2687
2688                 seq_printf(seq, "%08X %08X %-3hd",
2689                            (__force u32) mfc->mfc_mcastgrp,
2690                            (__force u32) mfc->mfc_origin,
2691                            mfc->mfc_parent);
2692
2693                 if (it->cache != &mrt->mfc_unres_queue) {
2694                         seq_printf(seq, " %8lu %8lu %8lu",
2695                                    mfc->mfc_un.res.pkt,
2696                                    mfc->mfc_un.res.bytes,
2697                                    mfc->mfc_un.res.wrong_if);
2698                         for (n = mfc->mfc_un.res.minvif;
2699                              n < mfc->mfc_un.res.maxvif; n++) {
2700                                 if (VIF_EXISTS(mrt, n) &&
2701                                     mfc->mfc_un.res.ttls[n] < 255)
2702                                         seq_printf(seq,
2703                                            " %2d:%-3d",
2704                                            n, mfc->mfc_un.res.ttls[n]);
2705                         }
2706                 } else {
2707                         /* unresolved mfc_caches don't contain
2708                          * pkt, bytes and wrong_if values
2709                          */
2710                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2711                 }
2712                 seq_putc(seq, '\n');
2713         }
2714         return 0;
2715 }
2716
2717 static const struct seq_operations ipmr_mfc_seq_ops = {
2718         .start = ipmr_mfc_seq_start,
2719         .next  = ipmr_mfc_seq_next,
2720         .stop  = ipmr_mfc_seq_stop,
2721         .show  = ipmr_mfc_seq_show,
2722 };
2723
2724 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2725 {
2726         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2727                             sizeof(struct ipmr_mfc_iter));
2728 }
2729
2730 static const struct file_operations ipmr_mfc_fops = {
2731         .owner   = THIS_MODULE,
2732         .open    = ipmr_mfc_open,
2733         .read    = seq_read,
2734         .llseek  = seq_lseek,
2735         .release = seq_release_net,
2736 };
2737 #endif
2738
2739 #ifdef CONFIG_IP_PIMSM_V2
2740 static const struct net_protocol pim_protocol = {
2741         .handler        =       pim_rcv,
2742         .netns_ok       =       1,
2743 };
2744 #endif
2745
2746 /* Setup for IP multicast routing */
2747 static int __net_init ipmr_net_init(struct net *net)
2748 {
2749         int err;
2750
2751         err = ipmr_rules_init(net);
2752         if (err < 0)
2753                 goto fail;
2754
2755 #ifdef CONFIG_PROC_FS
2756         err = -ENOMEM;
2757         if (!proc_create("ip_mr_vif", 0, net->proc_net, &ipmr_vif_fops))
2758                 goto proc_vif_fail;
2759         if (!proc_create("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_fops))
2760                 goto proc_cache_fail;
2761 #endif
2762         return 0;
2763
2764 #ifdef CONFIG_PROC_FS
2765 proc_cache_fail:
2766         remove_proc_entry("ip_mr_vif", net->proc_net);
2767 proc_vif_fail:
2768         ipmr_rules_exit(net);
2769 #endif
2770 fail:
2771         return err;
2772 }
2773
2774 static void __net_exit ipmr_net_exit(struct net *net)
2775 {
2776 #ifdef CONFIG_PROC_FS
2777         remove_proc_entry("ip_mr_cache", net->proc_net);
2778         remove_proc_entry("ip_mr_vif", net->proc_net);
2779 #endif
2780         ipmr_rules_exit(net);
2781 }
2782
2783 static struct pernet_operations ipmr_net_ops = {
2784         .init = ipmr_net_init,
2785         .exit = ipmr_net_exit,
2786 };
2787
2788 int __init ip_mr_init(void)
2789 {
2790         int err;
2791
2792         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2793                                        sizeof(struct mfc_cache),
2794                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2795                                        NULL);
2796
2797         err = register_pernet_subsys(&ipmr_net_ops);
2798         if (err)
2799                 goto reg_pernet_fail;
2800
2801         err = register_netdevice_notifier(&ip_mr_notifier);
2802         if (err)
2803                 goto reg_notif_fail;
2804 #ifdef CONFIG_IP_PIMSM_V2
2805         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2806                 pr_err("%s: can't add PIM protocol\n", __func__);
2807                 err = -EAGAIN;
2808                 goto add_proto_fail;
2809         }
2810 #endif
2811         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
2812                       NULL, ipmr_rtm_dumproute, NULL);
2813         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
2814                       ipmr_rtm_route, NULL, NULL);
2815         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
2816                       ipmr_rtm_route, NULL, NULL);
2817         return 0;
2818
2819 #ifdef CONFIG_IP_PIMSM_V2
2820 add_proto_fail:
2821         unregister_netdevice_notifier(&ip_mr_notifier);
2822 #endif
2823 reg_notif_fail:
2824         unregister_pernet_subsys(&ipmr_net_ops);
2825 reg_pernet_fail:
2826         kmem_cache_destroy(mrt_cachep);
2827         return err;
2828 }