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         cache->mfc_un.res.lastuse = jiffies;
 726 }
 727
 728 static int vif_add(struct net *net, struct mr_table *mrt,
 729                    struct vifctl *vifc, int mrtsock)
 730 {
 731         int vifi = vifc->vifc_vifi;
 732         struct vif_device *v = &mrt->vif_table[vifi];
 733         struct net_device *dev;
 734         struct in_device *in_dev;
 735         int err;
 736
 737         /* Is vif busy ? */
 738         if (VIF_EXISTS(mrt, vifi))
 739                 return -EADDRINUSE;
 740
 741         switch (vifc->vifc_flags) {
 742         case VIFF_REGISTER:
 743                 if (!ipmr_pimsm_enabled())
 744                         return -EINVAL;
 745                 /* Special Purpose VIF in PIM
 746                  * All the packets will be sent to the daemon
 747                  */
 748                 if (mrt->mroute_reg_vif_num >= 0)
 749                         return -EADDRINUSE;
 750                 dev = ipmr_reg_vif(net, mrt);
 751                 if (!dev)
 752                         return -ENOBUFS;
 753                 err = dev_set_allmulti(dev, 1);
 754                 if (err) {
 755                         unregister_netdevice(dev);
 756                         dev_put(dev);
 757                         return err;
 758                 }
 759                 break;
 760         case VIFF_TUNNEL:
 761                 dev = ipmr_new_tunnel(net, vifc);
 762                 if (!dev)
 763                         return -ENOBUFS;
 764                 err = dev_set_allmulti(dev, 1);
 765                 if (err) {
 766                         ipmr_del_tunnel(dev, vifc);
 767                         dev_put(dev);
 768                         return err;
 769                 }
 770                 break;
 771         case VIFF_USE_IFINDEX:
 772         case 0:
 773                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 774                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 775                         if (dev && !__in_dev_get_rtnl(dev)) {
 776                                 dev_put(dev);
 777                                 return -EADDRNOTAVAIL;
 778                         }
 779                 } else {
 780                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 781                 }
 782                 if (!dev)
 783                         return -EADDRNOTAVAIL;
 784                 err = dev_set_allmulti(dev, 1);
 785                 if (err) {
 786                         dev_put(dev);
 787                         return err;
 788                 }
 789                 break;
 790         default:
 791                 return -EINVAL;
 792         }
 793
 794         in_dev = __in_dev_get_rtnl(dev);
 795         if (!in_dev) {
 796                 dev_put(dev);
 797                 return -EADDRNOTAVAIL;
 798         }
 799         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 800         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING, dev->ifindex,
 801                                     &in_dev->cnf);
 802         ip_rt_multicast_event(in_dev);
 803
 804         /* Fill in the VIF structures */
 805
 806         v->rate_limit = vifc->vifc_rate_limit;
 807         v->local = vifc->vifc_lcl_addr.s_addr;
 808         v->remote = vifc->vifc_rmt_addr.s_addr;
 809         v->flags = vifc->vifc_flags;
 810         if (!mrtsock)
 811                 v->flags |= VIFF_STATIC;
 812         v->threshold = vifc->vifc_threshold;
 813         v->bytes_in = 0;
 814         v->bytes_out = 0;
 815         v->pkt_in = 0;
 816         v->pkt_out = 0;
 817         v->link = dev->ifindex;
 818         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
 819                 v->link = dev_get_iflink(dev);
 820
 821         /* And finish update writing critical data */
 822         write_lock_bh(&mrt_lock);
 823         v->dev = dev;
 824         if (v->flags & VIFF_REGISTER)
 825                 mrt->mroute_reg_vif_num = vifi;
 826         if (vifi+1 > mrt->maxvif)
 827                 mrt->maxvif = vifi+1;
 828         write_unlock_bh(&mrt_lock);
 829         return 0;
 830 }
 831
 832 /* called with rcu_read_lock() */
 833 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 834                                          __be32 origin,
 835                                          __be32 mcastgrp)
 836 {
 837         int line = MFC_HASH(mcastgrp, origin);
 838         struct mfc_cache *c;
 839
 840         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
 841                 if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
 842                         return c;
 843         }
 844         return NULL;
 845 }
 846
 847 /* Look for a (*,*,oif) entry */
 848 static struct mfc_cache *ipmr_cache_find_any_parent(struct mr_table *mrt,
 849                                                     int vifi)
 850 {
 851         int line = MFC_HASH(htonl(INADDR_ANY), htonl(INADDR_ANY));
 852         struct mfc_cache *c;
 853
 854         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list)
 855                 if (c->mfc_origin == htonl(INADDR_ANY) &&
 856                     c->mfc_mcastgrp == htonl(INADDR_ANY) &&
 857                     c->mfc_un.res.ttls[vifi] < 255)
 858                         return c;
 859
 860         return NULL;
 861 }
 862
 863 /* Look for a (*,G) entry */
 864 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 865                                              __be32 mcastgrp, int vifi)
 866 {
 867         int line = MFC_HASH(mcastgrp, htonl(INADDR_ANY));
 868         struct mfc_cache *c, *proxy;
 869
 870         if (mcastgrp == htonl(INADDR_ANY))
 871                 goto skip;
 872
 873         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list)
 874                 if (c->mfc_origin == htonl(INADDR_ANY) &&
 875                     c->mfc_mcastgrp == mcastgrp) {
 876                         if (c->mfc_un.res.ttls[vifi] < 255)
 877                                 return c;
 878
 879                         /* It's ok if the vifi is part of the static tree */
 880                         proxy = ipmr_cache_find_any_parent(mrt,
 881                                                            c->mfc_parent);
 882                         if (proxy && proxy->mfc_un.res.ttls[vifi] < 255)
 883                                 return c;
 884                 }
 885
 886 skip:
 887         return ipmr_cache_find_any_parent(mrt, vifi);
 888 }
 889
 890 /* Allocate a multicast cache entry */
 891 static struct mfc_cache *ipmr_cache_alloc(void)
 892 {
 893         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 894
 895         if (c) {
 896                 c->mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 897                 c->mfc_un.res.minvif = MAXVIFS;
 898         }
 899         return c;
 900 }
 901
 902 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 903 {
 904         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 905
 906         if (c) {
 907                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
 908                 c->mfc_un.unres.expires = jiffies + 10*HZ;
 909         }
 910         return c;
 911 }
 912
 913 /* A cache entry has gone into a resolved state from queued */
 914 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 915                                struct mfc_cache *uc, struct mfc_cache *c)
 916 {
 917         struct sk_buff *skb;
 918         struct nlmsgerr *e;
 919
 920         /* Play the pending entries through our router */
 921         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
 922                 if (ip_hdr(skb)->version == 0) {
 923                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 924
 925                         if (__ipmr_fill_mroute(mrt, skb, c, nlmsg_data(nlh)) > 0) {
 926                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 927                                                  (u8 *)nlh;
 928                         } else {
 929                                 nlh->nlmsg_type = NLMSG_ERROR;
 930                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 931                                 skb_trim(skb, nlh->nlmsg_len);
 932                                 e = nlmsg_data(nlh);
 933                                 e->error = -EMSGSIZE;
 934                                 memset(&e->msg, 0, sizeof(e->msg));
 935                         }
 936
 937                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 938                 } else {
 939                         ip_mr_forward(net, mrt, skb, c, 0);
 940                 }
 941         }
 942 }
 943
 944 /* Bounce a cache query up to mrouted. We could use netlink for this but mrouted
 945  * expects the following bizarre scheme.
 946  *
 947  * Called under mrt_lock.
 948  */
 949 static int ipmr_cache_report(struct mr_table *mrt,
 950                              struct sk_buff *pkt, vifi_t vifi, int assert)
 951 {
 952         const int ihl = ip_hdrlen(pkt);
 953         struct sock *mroute_sk;
 954         struct igmphdr *igmp;
 955         struct igmpmsg *msg;
 956         struct sk_buff *skb;
 957         int ret;
 958
 959         if (assert == IGMPMSG_WHOLEPKT)
 960                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
 961         else
 962                 skb = alloc_skb(128, GFP_ATOMIC);
 963
 964         if (!skb)
 965                 return -ENOBUFS;
 966
 967         if (assert == IGMPMSG_WHOLEPKT) {
 968                 /* Ugly, but we have no choice with this interface.
 969                  * Duplicate old header, fix ihl, length etc.
 970                  * And all this only to mangle msg->im_msgtype and
 971                  * to set msg->im_mbz to "mbz" :-)
 972                  */
 973                 skb_push(skb, sizeof(struct iphdr));
 974                 skb_reset_network_header(skb);
 975                 skb_reset_transport_header(skb);
 976                 msg = (struct igmpmsg *)skb_network_header(skb);
 977                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
 978                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
 979                 msg->im_mbz = 0;
 980                 msg->im_vif = mrt->mroute_reg_vif_num;
 981                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
 982                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
 983                                              sizeof(struct iphdr));
 984         } else {
 985                 /* Copy the IP header */
 986                 skb_set_network_header(skb, skb->len);
 987                 skb_put(skb, ihl);
 988                 skb_copy_to_linear_data(skb, pkt->data, ihl);
 989                 /* Flag to the kernel this is a route add */
 990                 ip_hdr(skb)->protocol = 0;
 991                 msg = (struct igmpmsg *)skb_network_header(skb);
 992                 msg->im_vif = vifi;
 993                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
 994                 /* Add our header */
 995                 igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
 996                 igmp->type = assert;
 997                 msg->im_msgtype = assert;
 998                 igmp->code = 0;
 999                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1000                 skb->transport_header = skb->network_header;
1001         }
1002
1003         rcu_read_lock();
1004         mroute_sk = rcu_dereference(mrt->mroute_sk);
1005         if (!mroute_sk) {
1006                 rcu_read_unlock();
1007                 kfree_skb(skb);
1008                 return -EINVAL;
1009         }
1010
1011         /* Deliver to mrouted */
1012         ret = sock_queue_rcv_skb(mroute_sk, skb);
1013         rcu_read_unlock();
1014         if (ret < 0) {
1015                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1016                 kfree_skb(skb);
1017         }
1018
1019         return ret;
1020 }
1021
1022 /* Queue a packet for resolution. It gets locked cache entry! */
1023 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1024                                  struct sk_buff *skb)
1025 {
1026         bool found = false;
1027         int err;
1028         struct mfc_cache *c;
1029         const struct iphdr *iph = ip_hdr(skb);
1030
1031         spin_lock_bh(&mfc_unres_lock);
1032         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
1033                 if (c->mfc_mcastgrp == iph->daddr &&
1034                     c->mfc_origin == iph->saddr) {
1035                         found = true;
1036                         break;
1037                 }
1038         }
1039
1040         if (!found) {
1041                 /* Create a new entry if allowable */
1042                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1043                     (c = ipmr_cache_alloc_unres()) == NULL) {
1044                         spin_unlock_bh(&mfc_unres_lock);
1045
1046                         kfree_skb(skb);
1047                         return -ENOBUFS;
1048                 }
1049
1050                 /* Fill in the new cache entry */
1051                 c->mfc_parent   = -1;
1052                 c->mfc_origin   = iph->saddr;
1053                 c->mfc_mcastgrp = iph->daddr;
1054
1055                 /* Reflect first query at mrouted. */
1056                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1057                 if (err < 0) {
1058                         /* If the report failed throw the cache entry
1059                            out - Brad Parker
1060                          */
1061                         spin_unlock_bh(&mfc_unres_lock);
1062
1063                         ipmr_cache_free(c);
1064                         kfree_skb(skb);
1065                         return err;
1066                 }
1067
1068                 atomic_inc(&mrt->cache_resolve_queue_len);
1069                 list_add(&c->list, &mrt->mfc_unres_queue);
1070                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1071
1072                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1073                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1074         }
1075
1076         /* See if we can append the packet */
1077         if (c->mfc_un.unres.unresolved.qlen > 3) {
1078                 kfree_skb(skb);
1079                 err = -ENOBUFS;
1080         } else {
1081                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1082                 err = 0;
1083         }
1084
1085         spin_unlock_bh(&mfc_unres_lock);
1086         return err;
1087 }
1088
1089 /* MFC cache manipulation by user space mroute daemon */
1090
1091 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1092 {
1093         int line;
1094         struct mfc_cache *c, *next;
1095
1096         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1097
1098         list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1099                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1100                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr &&
1101                     (parent == -1 || parent == c->mfc_parent)) {
1102                         list_del_rcu(&c->list);
1103                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1104                         ipmr_cache_free(c);
1105                         return 0;
1106                 }
1107         }
1108         return -ENOENT;
1109 }
1110
1111 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1112                         struct mfcctl *mfc, int mrtsock, int parent)
1113 {
1114         bool found = false;
1115         int line;
1116         struct mfc_cache *uc, *c;
1117
1118         if (mfc->mfcc_parent >= MAXVIFS)
1119                 return -ENFILE;
1120
1121         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1122
1123         list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1124                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1125                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr &&
1126                     (parent == -1 || parent == c->mfc_parent)) {
1127                         found = true;
1128                         break;
1129                 }
1130         }
1131
1132         if (found) {
1133                 write_lock_bh(&mrt_lock);
1134                 c->mfc_parent = mfc->mfcc_parent;
1135                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1136                 if (!mrtsock)
1137                         c->mfc_flags |= MFC_STATIC;
1138                 write_unlock_bh(&mrt_lock);
1139                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1140                 return 0;
1141         }
1142
1143         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1144             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1145                 return -EINVAL;
1146
1147         c = ipmr_cache_alloc();
1148         if (!c)
1149                 return -ENOMEM;
1150
1151         c->mfc_origin = mfc->mfcc_origin.s_addr;
1152         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1153         c->mfc_parent = mfc->mfcc_parent;
1154         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1155         if (!mrtsock)
1156                 c->mfc_flags |= MFC_STATIC;
1157
1158         list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1159
1160         /* Check to see if we resolved a queued list. If so we
1161          * need to send on the frames and tidy up.
1162          */
1163         found = false;
1164         spin_lock_bh(&mfc_unres_lock);
1165         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1166                 if (uc->mfc_origin == c->mfc_origin &&
1167                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1168                         list_del(&uc->list);
1169                         atomic_dec(&mrt->cache_resolve_queue_len);
1170                         found = true;
1171                         break;
1172                 }
1173         }
1174         if (list_empty(&mrt->mfc_unres_queue))
1175                 del_timer(&mrt->ipmr_expire_timer);
1176         spin_unlock_bh(&mfc_unres_lock);
1177
1178         if (found) {
1179                 ipmr_cache_resolve(net, mrt, uc, c);
1180                 ipmr_cache_free(uc);
1181         }
1182         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1183         return 0;
1184 }
1185
1186 /* Close the multicast socket, and clear the vif tables etc */
1187 static void mroute_clean_tables(struct mr_table *mrt, bool all)
1188 {
1189         int i;
1190         LIST_HEAD(list);
1191         struct mfc_cache *c, *next;
1192
1193         /* Shut down all active vif entries */
1194         for (i = 0; i < mrt->maxvif; i++) {
1195                 if (!all && (mrt->vif_table[i].flags & VIFF_STATIC))
1196                         continue;
1197                 vif_delete(mrt, i, 0, &list);
1198         }
1199         unregister_netdevice_many(&list);
1200
1201         /* Wipe the cache */
1202         for (i = 0; i < MFC_LINES; i++) {
1203                 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1204                         if (!all && (c->mfc_flags & MFC_STATIC))
1205                                 continue;
1206                         list_del_rcu(&c->list);
1207                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1208                         ipmr_cache_free(c);
1209                 }
1210         }
1211
1212         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1213                 spin_lock_bh(&mfc_unres_lock);
1214                 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1215                         list_del(&c->list);
1216                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1217                         ipmr_destroy_unres(mrt, c);
1218                 }
1219                 spin_unlock_bh(&mfc_unres_lock);
1220         }
1221 }
1222
1223 /* called from ip_ra_control(), before an RCU grace period,
1224  * we dont need to call synchronize_rcu() here
1225  */
1226 static void mrtsock_destruct(struct sock *sk)
1227 {
1228         struct net *net = sock_net(sk);
1229         struct mr_table *mrt;
1230
1231         rtnl_lock();
1232         ipmr_for_each_table(mrt, net) {
1233                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1234                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1235                         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1236                                                     NETCONFA_IFINDEX_ALL,
1237                                                     net->ipv4.devconf_all);
1238                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1239                         mroute_clean_tables(mrt, false);
1240                 }
1241         }
1242         rtnl_unlock();
1243 }
1244
1245 /* Socket options and virtual interface manipulation. The whole
1246  * virtual interface system is a complete heap, but unfortunately
1247  * that's how BSD mrouted happens to think. Maybe one day with a proper
1248  * MOSPF/PIM router set up we can clean this up.
1249  */
1250
1251 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval,
1252                          unsigned int optlen)
1253 {
1254         struct net *net = sock_net(sk);
1255         int val, ret = 0, parent = 0;
1256         struct mr_table *mrt;
1257         struct vifctl vif;
1258         struct mfcctl mfc;
1259         u32 uval;
1260
1261         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1262         rtnl_lock();
1263         if (sk->sk_type != SOCK_RAW ||
1264             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1265                 ret = -EOPNOTSUPP;
1266                 goto out_unlock;
1267         }
1268
1269         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1270         if (!mrt) {
1271                 ret = -ENOENT;
1272                 goto out_unlock;
1273         }
1274         if (optname != MRT_INIT) {
1275                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1276                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1277                         ret = -EACCES;
1278                         goto out_unlock;
1279                 }
1280         }
1281
1282         switch (optname) {
1283         case MRT_INIT:
1284                 if (optlen != sizeof(int)) {
1285                         ret = -EINVAL;
1286                         break;
1287                 }
1288                 if (rtnl_dereference(mrt->mroute_sk)) {
1289                         ret = -EADDRINUSE;
1290                         break;
1291                 }
1292
1293                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1294                 if (ret == 0) {
1295                         rcu_assign_pointer(mrt->mroute_sk, sk);
1296                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1297                         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1298                                                     NETCONFA_IFINDEX_ALL,
1299                                                     net->ipv4.devconf_all);
1300                 }
1301                 break;
1302         case MRT_DONE:
1303                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1304                         ret = -EACCES;
1305                 } else {
1306                         /* We need to unlock here because mrtsock_destruct takes
1307                          * care of rtnl itself and we can't change that due to
1308                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1309                          */
1310                         rtnl_unlock();
1311                         ret = ip_ra_control(sk, 0, NULL);
1312                         goto out;
1313                 }
1314                 break;
1315         case MRT_ADD_VIF:
1316         case MRT_DEL_VIF:
1317                 if (optlen != sizeof(vif)) {
1318                         ret = -EINVAL;
1319                         break;
1320                 }
1321                 if (copy_from_user(&vif, optval, sizeof(vif))) {
1322                         ret = -EFAULT;
1323                         break;
1324                 }
1325                 if (vif.vifc_vifi >= MAXVIFS) {
1326                         ret = -ENFILE;
1327                         break;
1328                 }
1329                 if (optname == MRT_ADD_VIF) {
1330                         ret = vif_add(net, mrt, &vif,
1331                                       sk == rtnl_dereference(mrt->mroute_sk));
1332                 } else {
1333                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1334                 }
1335                 break;
1336         /* Manipulate the forwarding caches. These live
1337          * in a sort of kernel/user symbiosis.
1338          */
1339         case MRT_ADD_MFC:
1340         case MRT_DEL_MFC:
1341                 parent = -1;
1342         case MRT_ADD_MFC_PROXY:
1343         case MRT_DEL_MFC_PROXY:
1344                 if (optlen != sizeof(mfc)) {
1345                         ret = -EINVAL;
1346                         break;
1347                 }
1348                 if (copy_from_user(&mfc, optval, sizeof(mfc))) {
1349                         ret = -EFAULT;
1350                         break;
1351                 }
1352                 if (parent == 0)
1353                         parent = mfc.mfcc_parent;
1354                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1355                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1356                 else
1357                         ret = ipmr_mfc_add(net, mrt, &mfc,
1358                                            sk == rtnl_dereference(mrt->mroute_sk),
1359                                            parent);
1360                 break;
1361         /* Control PIM assert. */
1362         case MRT_ASSERT:
1363                 if (optlen != sizeof(val)) {
1364                         ret = -EINVAL;
1365                         break;
1366                 }
1367                 if (get_user(val, (int __user *)optval)) {
1368                         ret = -EFAULT;
1369                         break;
1370                 }
1371                 mrt->mroute_do_assert = val;
1372                 break;
1373         case MRT_PIM:
1374                 if (!ipmr_pimsm_enabled()) {
1375                         ret = -ENOPROTOOPT;
1376                         break;
1377                 }
1378                 if (optlen != sizeof(val)) {
1379                         ret = -EINVAL;
1380                         break;
1381                 }
1382                 if (get_user(val, (int __user *)optval)) {
1383                         ret = -EFAULT;
1384                         break;
1385                 }
1386
1387                 val = !!val;
1388                 if (val != mrt->mroute_do_pim) {
1389                         mrt->mroute_do_pim = val;
1390                         mrt->mroute_do_assert = val;
1391                 }
1392                 break;
1393         case MRT_TABLE:
1394                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1395                         ret = -ENOPROTOOPT;
1396                         break;
1397                 }
1398                 if (optlen != sizeof(uval)) {
1399                         ret = -EINVAL;
1400                         break;
1401                 }
1402                 if (get_user(uval, (u32 __user *)optval)) {
1403                         ret = -EFAULT;
1404                         break;
1405                 }
1406
1407                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1408                         ret = -EBUSY;
1409                 } else {
1410                         mrt = ipmr_new_table(net, uval);
1411                         if (IS_ERR(mrt))
1412                                 ret = PTR_ERR(mrt);
1413                         else
1414                                 raw_sk(sk)->ipmr_table = uval;
1415                 }
1416                 break;
1417         /* Spurious command, or MRT_VERSION which you cannot set. */
1418         default:
1419                 ret = -ENOPROTOOPT;
1420         }
1421 out_unlock:
1422         rtnl_unlock();
1423 out:
1424         return ret;
1425 }
1426
1427 /* Getsock opt support for the multicast routing system. */
1428 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1429 {
1430         int olr;
1431         int val;
1432         struct net *net = sock_net(sk);
1433         struct mr_table *mrt;
1434
1435         if (sk->sk_type != SOCK_RAW ||
1436             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1437                 return -EOPNOTSUPP;
1438
1439         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1440         if (!mrt)
1441                 return -ENOENT;
1442
1443         switch (optname) {
1444         case MRT_VERSION:
1445                 val = 0x0305;
1446                 break;
1447         case MRT_PIM:
1448                 if (!ipmr_pimsm_enabled())
1449                         return -ENOPROTOOPT;
1450                 val = mrt->mroute_do_pim;
1451                 break;
1452         case MRT_ASSERT:
1453                 val = mrt->mroute_do_assert;
1454                 break;
1455         default:
1456                 return -ENOPROTOOPT;
1457         }
1458
1459         if (get_user(olr, optlen))
1460                 return -EFAULT;
1461         olr = min_t(unsigned int, olr, sizeof(int));
1462         if (olr < 0)
1463                 return -EINVAL;
1464         if (put_user(olr, optlen))
1465                 return -EFAULT;
1466         if (copy_to_user(optval, &val, olr))
1467                 return -EFAULT;
1468         return 0;
1469 }
1470
1471 /* The IP multicast ioctl support routines. */
1472 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1473 {
1474         struct sioc_sg_req sr;
1475         struct sioc_vif_req vr;
1476         struct vif_device *vif;
1477         struct mfc_cache *c;
1478         struct net *net = sock_net(sk);
1479         struct mr_table *mrt;
1480
1481         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1482         if (!mrt)
1483                 return -ENOENT;
1484
1485         switch (cmd) {
1486         case SIOCGETVIFCNT:
1487                 if (copy_from_user(&vr, arg, sizeof(vr)))
1488                         return -EFAULT;
1489                 if (vr.vifi >= mrt->maxvif)
1490                         return -EINVAL;
1491                 read_lock(&mrt_lock);
1492                 vif = &mrt->vif_table[vr.vifi];
1493                 if (VIF_EXISTS(mrt, vr.vifi)) {
1494                         vr.icount = vif->pkt_in;
1495                         vr.ocount = vif->pkt_out;
1496                         vr.ibytes = vif->bytes_in;
1497                         vr.obytes = vif->bytes_out;
1498                         read_unlock(&mrt_lock);
1499
1500                         if (copy_to_user(arg, &vr, sizeof(vr)))
1501                                 return -EFAULT;
1502                         return 0;
1503                 }
1504                 read_unlock(&mrt_lock);
1505                 return -EADDRNOTAVAIL;
1506         case SIOCGETSGCNT:
1507                 if (copy_from_user(&sr, arg, sizeof(sr)))
1508                         return -EFAULT;
1509
1510                 rcu_read_lock();
1511                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1512                 if (c) {
1513                         sr.pktcnt = c->mfc_un.res.pkt;
1514                         sr.bytecnt = c->mfc_un.res.bytes;
1515                         sr.wrong_if = c->mfc_un.res.wrong_if;
1516                         rcu_read_unlock();
1517
1518                         if (copy_to_user(arg, &sr, sizeof(sr)))
1519                                 return -EFAULT;
1520                         return 0;
1521                 }
1522                 rcu_read_unlock();
1523                 return -EADDRNOTAVAIL;
1524         default:
1525                 return -ENOIOCTLCMD;
1526         }
1527 }
1528
1529 #ifdef CONFIG_COMPAT
1530 struct compat_sioc_sg_req {
1531         struct in_addr src;
1532         struct in_addr grp;
1533         compat_ulong_t pktcnt;
1534         compat_ulong_t bytecnt;
1535         compat_ulong_t wrong_if;
1536 };
1537
1538 struct compat_sioc_vif_req {
1539         vifi_t  vifi;           /* Which iface */
1540         compat_ulong_t icount;
1541         compat_ulong_t ocount;
1542         compat_ulong_t ibytes;
1543         compat_ulong_t obytes;
1544 };
1545
1546 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1547 {
1548         struct compat_sioc_sg_req sr;
1549         struct compat_sioc_vif_req vr;
1550         struct vif_device *vif;
1551         struct mfc_cache *c;
1552         struct net *net = sock_net(sk);
1553         struct mr_table *mrt;
1554
1555         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1556         if (!mrt)
1557                 return -ENOENT;
1558
1559         switch (cmd) {
1560         case SIOCGETVIFCNT:
1561                 if (copy_from_user(&vr, arg, sizeof(vr)))
1562                         return -EFAULT;
1563                 if (vr.vifi >= mrt->maxvif)
1564                         return -EINVAL;
1565                 read_lock(&mrt_lock);
1566                 vif = &mrt->vif_table[vr.vifi];
1567                 if (VIF_EXISTS(mrt, vr.vifi)) {
1568                         vr.icount = vif->pkt_in;
1569                         vr.ocount = vif->pkt_out;
1570                         vr.ibytes = vif->bytes_in;
1571                         vr.obytes = vif->bytes_out;
1572                         read_unlock(&mrt_lock);
1573
1574                         if (copy_to_user(arg, &vr, sizeof(vr)))
1575                                 return -EFAULT;
1576                         return 0;
1577                 }
1578                 read_unlock(&mrt_lock);
1579                 return -EADDRNOTAVAIL;
1580         case SIOCGETSGCNT:
1581                 if (copy_from_user(&sr, arg, sizeof(sr)))
1582                         return -EFAULT;
1583
1584                 rcu_read_lock();
1585                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1586                 if (c) {
1587                         sr.pktcnt = c->mfc_un.res.pkt;
1588                         sr.bytecnt = c->mfc_un.res.bytes;
1589                         sr.wrong_if = c->mfc_un.res.wrong_if;
1590                         rcu_read_unlock();
1591
1592                         if (copy_to_user(arg, &sr, sizeof(sr)))
1593                                 return -EFAULT;
1594                         return 0;
1595                 }
1596                 rcu_read_unlock();
1597                 return -EADDRNOTAVAIL;
1598         default:
1599                 return -ENOIOCTLCMD;
1600         }
1601 }
1602 #endif
1603
1604 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1605 {
1606         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1607         struct net *net = dev_net(dev);
1608         struct mr_table *mrt;
1609         struct vif_device *v;
1610         int ct;
1611
1612         if (event != NETDEV_UNREGISTER)
1613                 return NOTIFY_DONE;
1614
1615         ipmr_for_each_table(mrt, net) {
1616                 v = &mrt->vif_table[0];
1617                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1618                         if (v->dev == dev)
1619                                 vif_delete(mrt, ct, 1, NULL);
1620                 }
1621         }
1622         return NOTIFY_DONE;
1623 }
1624
1625 static struct notifier_block ip_mr_notifier = {
1626         .notifier_call = ipmr_device_event,
1627 };
1628
1629 /* Encapsulate a packet by attaching a valid IPIP header to it.
1630  * This avoids tunnel drivers and other mess and gives us the speed so
1631  * important for multicast video.
1632  */
1633 static void ip_encap(struct net *net, struct sk_buff *skb,
1634                      __be32 saddr, __be32 daddr)
1635 {
1636         struct iphdr *iph;
1637         const struct iphdr *old_iph = ip_hdr(skb);
1638
1639         skb_push(skb, sizeof(struct iphdr));
1640         skb->transport_header = skb->network_header;
1641         skb_reset_network_header(skb);
1642         iph = ip_hdr(skb);
1643
1644         iph->version    =       4;
1645         iph->tos        =       old_iph->tos;
1646         iph->ttl        =       old_iph->ttl;
1647         iph->frag_off   =       0;
1648         iph->daddr      =       daddr;
1649         iph->saddr      =       saddr;
1650         iph->protocol   =       IPPROTO_IPIP;
1651         iph->ihl        =       5;
1652         iph->tot_len    =       htons(skb->len);
1653         ip_select_ident(net, skb, NULL);
1654         ip_send_check(iph);
1655
1656         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1657         nf_reset(skb);
1658 }
1659
1660 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1661                                       struct sk_buff *skb)
1662 {
1663         struct ip_options *opt = &(IPCB(skb)->opt);
1664
1665         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1666         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1667
1668         if (unlikely(opt->optlen))
1669                 ip_forward_options(skb);
1670
1671         return dst_output(net, sk, skb);
1672 }
1673
1674 /* Processing handlers for ipmr_forward */
1675
1676 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1677                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1678 {
1679         const struct iphdr *iph = ip_hdr(skb);
1680         struct vif_device *vif = &mrt->vif_table[vifi];
1681         struct net_device *dev;
1682         struct rtable *rt;
1683         struct flowi4 fl4;
1684         int    encap = 0;
1685
1686         if (!vif->dev)
1687                 goto out_free;
1688
1689         if (vif->flags & VIFF_REGISTER) {
1690                 vif->pkt_out++;
1691                 vif->bytes_out += skb->len;
1692                 vif->dev->stats.tx_bytes += skb->len;
1693                 vif->dev->stats.tx_packets++;
1694                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1695                 goto out_free;
1696         }
1697
1698         if (vif->flags & VIFF_TUNNEL) {
1699                 rt = ip_route_output_ports(net, &fl4, NULL,
1700                                            vif->remote, vif->local,
1701                                            0, 0,
1702                                            IPPROTO_IPIP,
1703                                            RT_TOS(iph->tos), vif->link);
1704                 if (IS_ERR(rt))
1705                         goto out_free;
1706                 encap = sizeof(struct iphdr);
1707         } else {
1708                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1709                                            0, 0,
1710                                            IPPROTO_IPIP,
1711                                            RT_TOS(iph->tos), vif->link);
1712                 if (IS_ERR(rt))
1713                         goto out_free;
1714         }
1715
1716         dev = rt->dst.dev;
1717
1718         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1719                 /* Do not fragment multicasts. Alas, IPv4 does not
1720                  * allow to send ICMP, so that packets will disappear
1721                  * to blackhole.
1722                  */
1723                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1724                 ip_rt_put(rt);
1725                 goto out_free;
1726         }
1727
1728         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1729
1730         if (skb_cow(skb, encap)) {
1731                 ip_rt_put(rt);
1732                 goto out_free;
1733         }
1734
1735         vif->pkt_out++;
1736         vif->bytes_out += skb->len;
1737
1738         skb_dst_drop(skb);
1739         skb_dst_set(skb, &rt->dst);
1740         ip_decrease_ttl(ip_hdr(skb));
1741
1742         /* FIXME: forward and output firewalls used to be called here.
1743          * What do we do with netfilter? -- RR
1744          */
1745         if (vif->flags & VIFF_TUNNEL) {
1746                 ip_encap(net, skb, vif->local, vif->remote);
1747                 /* FIXME: extra output firewall step used to be here. --RR */
1748                 vif->dev->stats.tx_packets++;
1749                 vif->dev->stats.tx_bytes += skb->len;
1750         }
1751
1752         IPCB(skb)->flags |= IPSKB_FORWARDED | IPSKB_FRAG_SEGS;
1753
1754         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1755          * not only before forwarding, but after forwarding on all output
1756          * interfaces. It is clear, if mrouter runs a multicasting
1757          * program, it should receive packets not depending to what interface
1758          * program is joined.
1759          * If we will not make it, the program will have to join on all
1760          * interfaces. On the other hand, multihoming host (or router, but
1761          * not mrouter) cannot join to more than one interface - it will
1762          * result in receiving multiple packets.
1763          */
1764         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1765                 net, NULL, skb, skb->dev, dev,
1766                 ipmr_forward_finish);
1767         return;
1768
1769 out_free:
1770         kfree_skb(skb);
1771 }
1772
1773 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1774 {
1775         int ct;
1776
1777         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1778                 if (mrt->vif_table[ct].dev == dev)
1779                         break;
1780         }
1781         return ct;
1782 }
1783
1784 /* "local" means that we should preserve one skb (for local delivery) */
1785 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1786                           struct sk_buff *skb, struct mfc_cache *cache,
1787                           int local)
1788 {
1789         int psend = -1;
1790         int vif, ct;
1791         int true_vifi = ipmr_find_vif(mrt, skb->dev);
1792
1793         vif = cache->mfc_parent;
1794         cache->mfc_un.res.pkt++;
1795         cache->mfc_un.res.bytes += skb->len;
1796         cache->mfc_un.res.lastuse = jiffies;
1797
1798         if (cache->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1799                 struct mfc_cache *cache_proxy;
1800
1801                 /* For an (*,G) entry, we only check that the incomming
1802                  * interface is part of the static tree.
1803                  */
1804                 cache_proxy = ipmr_cache_find_any_parent(mrt, vif);
1805                 if (cache_proxy &&
1806                     cache_proxy->mfc_un.res.ttls[true_vifi] < 255)
1807                         goto forward;
1808         }
1809
1810         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1811         if (mrt->vif_table[vif].dev != skb->dev) {
1812                 if (rt_is_output_route(skb_rtable(skb))) {
1813                         /* It is our own packet, looped back.
1814                          * Very complicated situation...
1815                          *
1816                          * The best workaround until routing daemons will be
1817                          * fixed is not to redistribute packet, if it was
1818                          * send through wrong interface. It means, that
1819                          * multicast applications WILL NOT work for
1820                          * (S,G), which have default multicast route pointing
1821                          * to wrong oif. In any case, it is not a good
1822                          * idea to use multicasting applications on router.
1823                          */
1824                         goto dont_forward;
1825                 }
1826
1827                 cache->mfc_un.res.wrong_if++;
1828
1829                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1830                     /* pimsm uses asserts, when switching from RPT to SPT,
1831                      * so that we cannot check that packet arrived on an oif.
1832                      * It is bad, but otherwise we would need to move pretty
1833                      * large chunk of pimd to kernel. Ough... --ANK
1834                      */
1835                     (mrt->mroute_do_pim ||
1836                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1837                     time_after(jiffies,
1838                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1839                         cache->mfc_un.res.last_assert = jiffies;
1840                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1841                 }
1842                 goto dont_forward;
1843         }
1844
1845 forward:
1846         mrt->vif_table[vif].pkt_in++;
1847         mrt->vif_table[vif].bytes_in += skb->len;
1848
1849         /* Forward the frame */
1850         if (cache->mfc_origin == htonl(INADDR_ANY) &&
1851             cache->mfc_mcastgrp == htonl(INADDR_ANY)) {
1852                 if (true_vifi >= 0 &&
1853                     true_vifi != cache->mfc_parent &&
1854                     ip_hdr(skb)->ttl >
1855                                 cache->mfc_un.res.ttls[cache->mfc_parent]) {
1856                         /* It's an (*,*) entry and the packet is not coming from
1857                          * the upstream: forward the packet to the upstream
1858                          * only.
1859                          */
1860                         psend = cache->mfc_parent;
1861                         goto last_forward;
1862                 }
1863                 goto dont_forward;
1864         }
1865         for (ct = cache->mfc_un.res.maxvif - 1;
1866              ct >= cache->mfc_un.res.minvif; ct--) {
1867                 /* For (*,G) entry, don't forward to the incoming interface */
1868                 if ((cache->mfc_origin != htonl(INADDR_ANY) ||
1869                      ct != true_vifi) &&
1870                     ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1871                         if (psend != -1) {
1872                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1873
1874                                 if (skb2)
1875                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1876                                                         psend);
1877                         }
1878                         psend = ct;
1879                 }
1880         }
1881 last_forward:
1882         if (psend != -1) {
1883                 if (local) {
1884                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1885
1886                         if (skb2)
1887                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1888                 } else {
1889                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1890                         return;
1891                 }
1892         }
1893
1894 dont_forward:
1895         if (!local)
1896                 kfree_skb(skb);
1897 }
1898
1899 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1900 {
1901         struct rtable *rt = skb_rtable(skb);
1902         struct iphdr *iph = ip_hdr(skb);
1903         struct flowi4 fl4 = {
1904                 .daddr = iph->daddr,
1905                 .saddr = iph->saddr,
1906                 .flowi4_tos = RT_TOS(iph->tos),
1907                 .flowi4_oif = (rt_is_output_route(rt) ?
1908                                skb->dev->ifindex : 0),
1909                 .flowi4_iif = (rt_is_output_route(rt) ?
1910                                LOOPBACK_IFINDEX :
1911                                skb->dev->ifindex),
1912                 .flowi4_mark = skb->mark,
1913         };
1914         struct mr_table *mrt;
1915         int err;
1916
1917         err = ipmr_fib_lookup(net, &fl4, &mrt);
1918         if (err)
1919                 return ERR_PTR(err);
1920         return mrt;
1921 }
1922
1923 /* Multicast packets for forwarding arrive here
1924  * Called with rcu_read_lock();
1925  */
1926 int ip_mr_input(struct sk_buff *skb)
1927 {
1928         struct mfc_cache *cache;
1929         struct net *net = dev_net(skb->dev);
1930         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1931         struct mr_table *mrt;
1932
1933         /* Packet is looped back after forward, it should not be
1934          * forwarded second time, but still can be delivered locally.
1935          */
1936         if (IPCB(skb)->flags & IPSKB_FORWARDED)
1937                 goto dont_forward;
1938
1939         mrt = ipmr_rt_fib_lookup(net, skb);
1940         if (IS_ERR(mrt)) {
1941                 kfree_skb(skb);
1942                 return PTR_ERR(mrt);
1943         }
1944         if (!local) {
1945                 if (IPCB(skb)->opt.router_alert) {
1946                         if (ip_call_ra_chain(skb))
1947                                 return 0;
1948                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1949                         /* IGMPv1 (and broken IGMPv2 implementations sort of
1950                          * Cisco IOS <= 11.2(8)) do not put router alert
1951                          * option to IGMP packets destined to routable
1952                          * groups. It is very bad, because it means
1953                          * that we can forward NO IGMP messages.
1954                          */
1955                         struct sock *mroute_sk;
1956
1957                         mroute_sk = rcu_dereference(mrt->mroute_sk);
1958                         if (mroute_sk) {
1959                                 nf_reset(skb);
1960                                 raw_rcv(mroute_sk, skb);
1961                                 return 0;
1962                         }
1963                     }
1964         }
1965
1966         /* already under rcu_read_lock() */
1967         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1968         if (!cache) {
1969                 int vif = ipmr_find_vif(mrt, skb->dev);
1970
1971                 if (vif >= 0)
1972                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
1973                                                     vif);
1974         }
1975
1976         /* No usable cache entry */
1977         if (!cache) {
1978                 int vif;
1979
1980                 if (local) {
1981                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1982                         ip_local_deliver(skb);
1983                         if (!skb2)
1984                                 return -ENOBUFS;
1985                         skb = skb2;
1986                 }
1987
1988                 read_lock(&mrt_lock);
1989                 vif = ipmr_find_vif(mrt, skb->dev);
1990                 if (vif >= 0) {
1991                         int err2 = ipmr_cache_unresolved(mrt, vif, skb);
1992                         read_unlock(&mrt_lock);
1993
1994                         return err2;
1995                 }
1996                 read_unlock(&mrt_lock);
1997                 kfree_skb(skb);
1998                 return -ENODEV;
1999         }
2000
2001         read_lock(&mrt_lock);
2002         ip_mr_forward(net, mrt, skb, cache, local);
2003         read_unlock(&mrt_lock);
2004
2005         if (local)
2006                 return ip_local_deliver(skb);
2007
2008         return 0;
2009
2010 dont_forward:
2011         if (local)
2012                 return ip_local_deliver(skb);
2013         kfree_skb(skb);
2014         return 0;
2015 }
2016
2017 #ifdef CONFIG_IP_PIMSM_V1
2018 /* Handle IGMP messages of PIMv1 */
2019 int pim_rcv_v1(struct sk_buff *skb)
2020 {
2021         struct igmphdr *pim;
2022         struct net *net = dev_net(skb->dev);
2023         struct mr_table *mrt;
2024
2025         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2026                 goto drop;
2027
2028         pim = igmp_hdr(skb);
2029
2030         mrt = ipmr_rt_fib_lookup(net, skb);
2031         if (IS_ERR(mrt))
2032                 goto drop;
2033         if (!mrt->mroute_do_pim ||
2034             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2035                 goto drop;
2036
2037         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2038 drop:
2039                 kfree_skb(skb);
2040         }
2041         return 0;
2042 }
2043 #endif
2044
2045 #ifdef CONFIG_IP_PIMSM_V2
2046 static int pim_rcv(struct sk_buff *skb)
2047 {
2048         struct pimreghdr *pim;
2049         struct net *net = dev_net(skb->dev);
2050         struct mr_table *mrt;
2051
2052         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2053                 goto drop;
2054
2055         pim = (struct pimreghdr *)skb_transport_header(skb);
2056         if (pim->type != ((PIM_VERSION << 4) | (PIM_REGISTER)) ||
2057             (pim->flags & PIM_NULL_REGISTER) ||
2058             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2059              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2060                 goto drop;
2061
2062         mrt = ipmr_rt_fib_lookup(net, skb);
2063         if (IS_ERR(mrt))
2064                 goto drop;
2065         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2066 drop:
2067                 kfree_skb(skb);
2068         }
2069         return 0;
2070 }
2071 #endif
2072
2073 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2074                               struct mfc_cache *c, struct rtmsg *rtm)
2075 {
2076         struct rta_mfc_stats mfcs;
2077         struct nlattr *mp_attr;
2078         struct rtnexthop *nhp;
2079         unsigned long lastuse;
2080         int ct;
2081
2082         /* If cache is unresolved, don't try to parse IIF and OIF */
2083         if (c->mfc_parent >= MAXVIFS)
2084                 return -ENOENT;
2085
2086         if (VIF_EXISTS(mrt, c->mfc_parent) &&
2087             nla_put_u32(skb, RTA_IIF, mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
2088                 return -EMSGSIZE;
2089
2090         if (!(mp_attr = nla_nest_start(skb, RTA_MULTIPATH)))
2091                 return -EMSGSIZE;
2092
2093         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2094                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2095                         if (!(nhp = nla_reserve_nohdr(skb, sizeof(*nhp)))) {
2096                                 nla_nest_cancel(skb, mp_attr);
2097                                 return -EMSGSIZE;
2098                         }
2099
2100                         nhp->rtnh_flags = 0;
2101                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2102                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2103                         nhp->rtnh_len = sizeof(*nhp);
2104                 }
2105         }
2106
2107         nla_nest_end(skb, mp_attr);
2108
2109         lastuse = READ_ONCE(c->mfc_un.res.lastuse);
2110         lastuse = time_after_eq(jiffies, lastuse) ? jiffies - lastuse : 0;
2111
2112         mfcs.mfcs_packets = c->mfc_un.res.pkt;
2113         mfcs.mfcs_bytes = c->mfc_un.res.bytes;
2114         mfcs.mfcs_wrong_if = c->mfc_un.res.wrong_if;
2115         if (nla_put_64bit(skb, RTA_MFC_STATS, sizeof(mfcs), &mfcs, RTA_PAD) ||
2116             nla_put_u64_64bit(skb, RTA_EXPIRES, jiffies_to_clock_t(lastuse),
2117                               RTA_PAD))
2118                 return -EMSGSIZE;
2119
2120         rtm->rtm_type = RTN_MULTICAST;
2121         return 1;
2122 }
2123
2124 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2125                    __be32 saddr, __be32 daddr,
2126                    struct rtmsg *rtm, int nowait, u32 portid)
2127 {
2128         struct mfc_cache *cache;
2129         struct mr_table *mrt;
2130         int err;
2131
2132         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2133         if (!mrt)
2134                 return -ENOENT;
2135
2136         rcu_read_lock();
2137         cache = ipmr_cache_find(mrt, saddr, daddr);
2138         if (!cache && skb->dev) {
2139                 int vif = ipmr_find_vif(mrt, skb->dev);
2140
2141                 if (vif >= 0)
2142                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2143         }
2144         if (!cache) {
2145                 struct sk_buff *skb2;
2146                 struct iphdr *iph;
2147                 struct net_device *dev;
2148                 int vif = -1;
2149
2150                 if (nowait) {
2151                         rcu_read_unlock();
2152                         return -EAGAIN;
2153                 }
2154
2155                 dev = skb->dev;
2156                 read_lock(&mrt_lock);
2157                 if (dev)
2158                         vif = ipmr_find_vif(mrt, dev);
2159                 if (vif < 0) {
2160                         read_unlock(&mrt_lock);
2161                         rcu_read_unlock();
2162                         return -ENODEV;
2163                 }
2164                 skb2 = skb_clone(skb, GFP_ATOMIC);
2165                 if (!skb2) {
2166                         read_unlock(&mrt_lock);
2167                         rcu_read_unlock();
2168                         return -ENOMEM;
2169                 }
2170
2171                 NETLINK_CB(skb2).portid = portid;
2172                 skb_push(skb2, sizeof(struct iphdr));
2173                 skb_reset_network_header(skb2);
2174                 iph = ip_hdr(skb2);
2175                 iph->ihl = sizeof(struct iphdr) >> 2;
2176                 iph->saddr = saddr;
2177                 iph->daddr = daddr;
2178                 iph->version = 0;
2179                 err = ipmr_cache_unresolved(mrt, vif, skb2);
2180                 read_unlock(&mrt_lock);
2181                 rcu_read_unlock();
2182                 return err;
2183         }
2184
2185         read_lock(&mrt_lock);
2186         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2187         read_unlock(&mrt_lock);
2188         rcu_read_unlock();
2189         return err;
2190 }
2191
2192 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2193                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2194                             int flags)
2195 {
2196         struct nlmsghdr *nlh;
2197         struct rtmsg *rtm;
2198         int err;
2199
2200         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2201         if (!nlh)
2202                 return -EMSGSIZE;
2203
2204         rtm = nlmsg_data(nlh);
2205         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2206         rtm->rtm_dst_len  = 32;
2207         rtm->rtm_src_len  = 32;
2208         rtm->rtm_tos      = 0;
2209         rtm->rtm_table    = mrt->id;
2210         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2211                 goto nla_put_failure;
2212         rtm->rtm_type     = RTN_MULTICAST;
2213         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2214         if (c->mfc_flags & MFC_STATIC)
2215                 rtm->rtm_protocol = RTPROT_STATIC;
2216         else
2217                 rtm->rtm_protocol = RTPROT_MROUTED;
2218         rtm->rtm_flags    = 0;
2219
2220         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2221             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2222                 goto nla_put_failure;
2223         err = __ipmr_fill_mroute(mrt, skb, c, rtm);
2224         /* do not break the dump if cache is unresolved */
2225         if (err < 0 && err != -ENOENT)
2226                 goto nla_put_failure;
2227
2228         nlmsg_end(skb, nlh);
2229         return 0;
2230
2231 nla_put_failure:
2232         nlmsg_cancel(skb, nlh);
2233         return -EMSGSIZE;
2234 }
2235
2236 static size_t mroute_msgsize(bool unresolved, int maxvif)
2237 {
2238         size_t len =
2239                 NLMSG_ALIGN(sizeof(struct rtmsg))
2240                 + nla_total_size(4)     /* RTA_TABLE */
2241                 + nla_total_size(4)     /* RTA_SRC */
2242                 + nla_total_size(4)     /* RTA_DST */
2243                 ;
2244
2245         if (!unresolved)
2246                 len = len
2247                       + nla_total_size(4)       /* RTA_IIF */
2248                       + nla_total_size(0)       /* RTA_MULTIPATH */
2249                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2250                                                 /* RTA_MFC_STATS */
2251                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2252                 ;
2253
2254         return len;
2255 }
2256
2257 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2258                                  int cmd)
2259 {
2260         struct net *net = read_pnet(&mrt->net);
2261         struct sk_buff *skb;
2262         int err = -ENOBUFS;
2263
2264         skb = nlmsg_new(mroute_msgsize(mfc->mfc_parent >= MAXVIFS, mrt->maxvif),
2265                         GFP_ATOMIC);
2266         if (!skb)
2267                 goto errout;
2268
2269         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2270         if (err < 0)
2271                 goto errout;
2272
2273         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2274         return;
2275
2276 errout:
2277         kfree_skb(skb);
2278         if (err < 0)
2279                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2280 }
2281
2282 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2283 {
2284         struct net *net = sock_net(skb->sk);
2285         struct mr_table *mrt;
2286         struct mfc_cache *mfc;
2287         unsigned int t = 0, s_t;
2288         unsigned int h = 0, s_h;
2289         unsigned int e = 0, s_e;
2290
2291         s_t = cb->args[0];
2292         s_h = cb->args[1];
2293         s_e = cb->args[2];
2294
2295         rcu_read_lock();
2296         ipmr_for_each_table(mrt, net) {
2297                 if (t < s_t)
2298                         goto next_table;
2299                 if (t > s_t)
2300                         s_h = 0;
2301                 for (h = s_h; h < MFC_LINES; h++) {
2302                         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2303                                 if (e < s_e)
2304                                         goto next_entry;
2305                                 if (ipmr_fill_mroute(mrt, skb,
2306                                                      NETLINK_CB(cb->skb).portid,
2307                                                      cb->nlh->nlmsg_seq,
2308                                                      mfc, RTM_NEWROUTE,
2309                                                      NLM_F_MULTI) < 0)
2310                                         goto done;
2311 next_entry:
2312                                 e++;
2313                         }
2314                         e = s_e = 0;
2315                 }
2316                 spin_lock_bh(&mfc_unres_lock);
2317                 list_for_each_entry(mfc, &mrt->mfc_unres_queue, list) {
2318                         if (e < s_e)
2319                                 goto next_entry2;
2320                         if (ipmr_fill_mroute(mrt, skb,
2321                                              NETLINK_CB(cb->skb).portid,
2322                                              cb->nlh->nlmsg_seq,
2323                                              mfc, RTM_NEWROUTE,
2324                                              NLM_F_MULTI) < 0) {
2325                                 spin_unlock_bh(&mfc_unres_lock);
2326                                 goto done;
2327                         }
2328 next_entry2:
2329                         e++;
2330                 }
2331                 spin_unlock_bh(&mfc_unres_lock);
2332                 e = s_e = 0;
2333                 s_h = 0;
2334 next_table:
2335                 t++;
2336         }
2337 done:
2338         rcu_read_unlock();
2339
2340         cb->args[2] = e;
2341         cb->args[1] = h;
2342         cb->args[0] = t;
2343
2344         return skb->len;
2345 }
2346
2347 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2348         [RTA_SRC]       = { .type = NLA_U32 },
2349         [RTA_DST]       = { .type = NLA_U32 },
2350         [RTA_IIF]       = { .type = NLA_U32 },
2351         [RTA_TABLE]     = { .type = NLA_U32 },
2352         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2353 };
2354
2355 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2356 {
2357         switch (rtm_protocol) {
2358         case RTPROT_STATIC:
2359         case RTPROT_MROUTED:
2360                 return true;
2361         }
2362         return false;
2363 }
2364
2365 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2366 {
2367         struct rtnexthop *rtnh = nla_data(nla);
2368         int remaining = nla_len(nla), vifi = 0;
2369
2370         while (rtnh_ok(rtnh, remaining)) {
2371                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2372                 if (++vifi == MAXVIFS)
2373                         break;
2374                 rtnh = rtnh_next(rtnh, &remaining);
2375         }
2376
2377         return remaining > 0 ? -EINVAL : vifi;
2378 }
2379
2380 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2381 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2382                             struct mfcctl *mfcc, int *mrtsock,
2383                             struct mr_table **mrtret)
2384 {
2385         struct net_device *dev = NULL;
2386         u32 tblid = RT_TABLE_DEFAULT;
2387         struct mr_table *mrt;
2388         struct nlattr *attr;
2389         struct rtmsg *rtm;
2390         int ret, rem;
2391
2392         ret = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipmr_policy);
2393         if (ret < 0)
2394                 goto out;
2395         rtm = nlmsg_data(nlh);
2396
2397         ret = -EINVAL;
2398         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2399             rtm->rtm_type != RTN_MULTICAST ||
2400             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2401             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2402                 goto out;
2403
2404         memset(mfcc, 0, sizeof(*mfcc));
2405         mfcc->mfcc_parent = -1;
2406         ret = 0;
2407         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2408                 switch (nla_type(attr)) {
2409                 case RTA_SRC:
2410                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2411                         break;
2412                 case RTA_DST:
2413                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2414                         break;
2415                 case RTA_IIF:
2416                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2417                         if (!dev) {
2418                                 ret = -ENODEV;
2419                                 goto out;
2420                         }
2421                         break;
2422                 case RTA_MULTIPATH:
2423                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2424                                 ret = -EINVAL;
2425                                 goto out;
2426                         }
2427                         break;
2428                 case RTA_PREFSRC:
2429                         ret = 1;
2430                         break;
2431                 case RTA_TABLE:
2432                         tblid = nla_get_u32(attr);
2433                         break;
2434                 }
2435         }
2436         mrt = ipmr_get_table(net, tblid);
2437         if (!mrt) {
2438                 ret = -ENOENT;
2439                 goto out;
2440         }
2441         *mrtret = mrt;
2442         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2443         if (dev)
2444                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2445
2446 out:
2447         return ret;
2448 }
2449
2450 /* takes care of both newroute and delroute */
2451 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh)
2452 {
2453         struct net *net = sock_net(skb->sk);
2454         int ret, mrtsock, parent;
2455         struct mr_table *tbl;
2456         struct mfcctl mfcc;
2457
2458         mrtsock = 0;
2459         tbl = NULL;
2460         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl);
2461         if (ret < 0)
2462                 return ret;
2463
2464         parent = ret ? mfcc.mfcc_parent : -1;
2465         if (nlh->nlmsg_type == RTM_NEWROUTE)
2466                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2467         else
2468                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2469 }
2470
2471 #ifdef CONFIG_PROC_FS
2472 /* The /proc interfaces to multicast routing :
2473  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2474  */
2475 struct ipmr_vif_iter {
2476         struct seq_net_private p;
2477         struct mr_table *mrt;
2478         int ct;
2479 };
2480
2481 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2482                                            struct ipmr_vif_iter *iter,
2483                                            loff_t pos)
2484 {
2485         struct mr_table *mrt = iter->mrt;
2486
2487         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2488                 if (!VIF_EXISTS(mrt, iter->ct))
2489                         continue;
2490                 if (pos-- == 0)
2491                         return &mrt->vif_table[iter->ct];
2492         }
2493         return NULL;
2494 }
2495
2496 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2497         __acquires(mrt_lock)
2498 {
2499         struct ipmr_vif_iter *iter = seq->private;
2500         struct net *net = seq_file_net(seq);
2501         struct mr_table *mrt;
2502
2503         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2504         if (!mrt)
2505                 return ERR_PTR(-ENOENT);
2506
2507         iter->mrt = mrt;
2508
2509         read_lock(&mrt_lock);
2510         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2511                 : SEQ_START_TOKEN;
2512 }
2513
2514 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2515 {
2516         struct ipmr_vif_iter *iter = seq->private;
2517         struct net *net = seq_file_net(seq);
2518         struct mr_table *mrt = iter->mrt;
2519
2520         ++*pos;
2521         if (v == SEQ_START_TOKEN)
2522                 return ipmr_vif_seq_idx(net, iter, 0);
2523
2524         while (++iter->ct < mrt->maxvif) {
2525                 if (!VIF_EXISTS(mrt, iter->ct))
2526                         continue;
2527                 return &mrt->vif_table[iter->ct];
2528         }
2529         return NULL;
2530 }
2531
2532 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2533         __releases(mrt_lock)
2534 {
2535         read_unlock(&mrt_lock);
2536 }
2537
2538 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2539 {
2540         struct ipmr_vif_iter *iter = seq->private;
2541         struct mr_table *mrt = iter->mrt;
2542
2543         if (v == SEQ_START_TOKEN) {
2544                 seq_puts(seq,
2545                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2546         } else {
2547                 const struct vif_device *vif = v;
2548                 const char *name =  vif->dev ? vif->dev->name : "none";
2549
2550                 seq_printf(seq,
2551                            "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2552                            vif - mrt->vif_table,
2553                            name, vif->bytes_in, vif->pkt_in,
2554                            vif->bytes_out, vif->pkt_out,
2555                            vif->flags, vif->local, vif->remote);
2556         }
2557         return 0;
2558 }
2559
2560 static const struct seq_operations ipmr_vif_seq_ops = {
2561         .start = ipmr_vif_seq_start,
2562         .next  = ipmr_vif_seq_next,
2563         .stop  = ipmr_vif_seq_stop,
2564         .show  = ipmr_vif_seq_show,
2565 };
2566
2567 static int ipmr_vif_open(struct inode *inode, struct file *file)
2568 {
2569         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2570                             sizeof(struct ipmr_vif_iter));
2571 }
2572
2573 static const struct file_operations ipmr_vif_fops = {
2574         .owner   = THIS_MODULE,
2575         .open    = ipmr_vif_open,
2576         .read    = seq_read,
2577         .llseek  = seq_lseek,
2578         .release = seq_release_net,
2579 };
2580
2581 struct ipmr_mfc_iter {
2582         struct seq_net_private p;
2583         struct mr_table *mrt;
2584         struct list_head *cache;
2585         int ct;
2586 };
2587
2588
2589 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2590                                           struct ipmr_mfc_iter *it, loff_t pos)
2591 {
2592         struct mr_table *mrt = it->mrt;
2593         struct mfc_cache *mfc;
2594
2595         rcu_read_lock();
2596         for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2597                 it->cache = &mrt->mfc_cache_array[it->ct];
2598                 list_for_each_entry_rcu(mfc, it->cache, list)
2599                         if (pos-- == 0)
2600                                 return mfc;
2601         }
2602         rcu_read_unlock();
2603
2604         spin_lock_bh(&mfc_unres_lock);
2605         it->cache = &mrt->mfc_unres_queue;
2606         list_for_each_entry(mfc, it->cache, list)
2607                 if (pos-- == 0)
2608                         return mfc;
2609         spin_unlock_bh(&mfc_unres_lock);
2610
2611         it->cache = NULL;
2612         return NULL;
2613 }
2614
2615
2616 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2617 {
2618         struct ipmr_mfc_iter *it = seq->private;
2619         struct net *net = seq_file_net(seq);
2620         struct mr_table *mrt;
2621
2622         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2623         if (!mrt)
2624                 return ERR_PTR(-ENOENT);
2625
2626         it->mrt = mrt;
2627         it->cache = NULL;
2628         it->ct = 0;
2629         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2630                 : SEQ_START_TOKEN;
2631 }
2632
2633 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2634 {
2635         struct mfc_cache *mfc = v;
2636         struct ipmr_mfc_iter *it = seq->private;
2637         struct net *net = seq_file_net(seq);
2638         struct mr_table *mrt = it->mrt;
2639
2640         ++*pos;
2641
2642         if (v == SEQ_START_TOKEN)
2643                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2644
2645         if (mfc->list.next != it->cache)
2646                 return list_entry(mfc->list.next, struct mfc_cache, list);
2647
2648         if (it->cache == &mrt->mfc_unres_queue)
2649                 goto end_of_list;
2650
2651         BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2652
2653         while (++it->ct < MFC_LINES) {
2654                 it->cache = &mrt->mfc_cache_array[it->ct];
2655                 if (list_empty(it->cache))
2656                         continue;
2657                 return list_first_entry(it->cache, struct mfc_cache, list);
2658         }
2659
2660         /* exhausted cache_array, show unresolved */
2661         rcu_read_unlock();
2662         it->cache = &mrt->mfc_unres_queue;
2663         it->ct = 0;
2664
2665         spin_lock_bh(&mfc_unres_lock);
2666         if (!list_empty(it->cache))
2667                 return list_first_entry(it->cache, struct mfc_cache, list);
2668
2669 end_of_list:
2670         spin_unlock_bh(&mfc_unres_lock);
2671         it->cache = NULL;
2672
2673         return NULL;
2674 }
2675
2676 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2677 {
2678         struct ipmr_mfc_iter *it = seq->private;
2679         struct mr_table *mrt = it->mrt;
2680
2681         if (it->cache == &mrt->mfc_unres_queue)
2682                 spin_unlock_bh(&mfc_unres_lock);
2683         else if (it->cache == &mrt->mfc_cache_array[it->ct])
2684                 rcu_read_unlock();
2685 }
2686
2687 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2688 {
2689         int n;
2690
2691         if (v == SEQ_START_TOKEN) {
2692                 seq_puts(seq,
2693                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2694         } else {
2695                 const struct mfc_cache *mfc = v;
2696                 const struct ipmr_mfc_iter *it = seq->private;
2697                 const struct mr_table *mrt = it->mrt;
2698
2699                 seq_printf(seq, "%08X %08X %-3hd",
2700                            (__force u32) mfc->mfc_mcastgrp,
2701                            (__force u32) mfc->mfc_origin,
2702                            mfc->mfc_parent);
2703
2704                 if (it->cache != &mrt->mfc_unres_queue) {
2705                         seq_printf(seq, " %8lu %8lu %8lu",
2706                                    mfc->mfc_un.res.pkt,
2707                                    mfc->mfc_un.res.bytes,
2708                                    mfc->mfc_un.res.wrong_if);
2709                         for (n = mfc->mfc_un.res.minvif;
2710                              n < mfc->mfc_un.res.maxvif; n++) {
2711                                 if (VIF_EXISTS(mrt, n) &&
2712                                     mfc->mfc_un.res.ttls[n] < 255)
2713                                         seq_printf(seq,
2714                                            " %2d:%-3d",
2715                                            n, mfc->mfc_un.res.ttls[n]);
2716                         }
2717                 } else {
2718                         /* unresolved mfc_caches don't contain
2719                          * pkt, bytes and wrong_if values
2720                          */
2721                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2722                 }
2723                 seq_putc(seq, '\n');
2724         }
2725         return 0;
2726 }
2727
2728 static const struct seq_operations ipmr_mfc_seq_ops = {
2729         .start = ipmr_mfc_seq_start,
2730         .next  = ipmr_mfc_seq_next,
2731         .stop  = ipmr_mfc_seq_stop,
2732         .show  = ipmr_mfc_seq_show,
2733 };
2734
2735 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2736 {
2737         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2738                             sizeof(struct ipmr_mfc_iter));
2739 }
2740
2741 static const struct file_operations ipmr_mfc_fops = {
2742         .owner   = THIS_MODULE,
2743         .open    = ipmr_mfc_open,
2744         .read    = seq_read,
2745         .llseek  = seq_lseek,
2746         .release = seq_release_net,
2747 };
2748 #endif
2749
2750 #ifdef CONFIG_IP_PIMSM_V2
2751 static const struct net_protocol pim_protocol = {
2752         .handler        =       pim_rcv,
2753         .netns_ok       =       1,
2754 };
2755 #endif
2756
2757 /* Setup for IP multicast routing */
2758 static int __net_init ipmr_net_init(struct net *net)
2759 {
2760         int err;
2761
2762         err = ipmr_rules_init(net);
2763         if (err < 0)
2764                 goto fail;
2765
2766 #ifdef CONFIG_PROC_FS
2767         err = -ENOMEM;
2768         if (!proc_create("ip_mr_vif", 0, net->proc_net, &ipmr_vif_fops))
2769                 goto proc_vif_fail;
2770         if (!proc_create("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_fops))
2771                 goto proc_cache_fail;
2772 #endif
2773         return 0;
2774
2775 #ifdef CONFIG_PROC_FS
2776 proc_cache_fail:
2777         remove_proc_entry("ip_mr_vif", net->proc_net);
2778 proc_vif_fail:
2779         ipmr_rules_exit(net);
2780 #endif
2781 fail:
2782         return err;
2783 }
2784
2785 static void __net_exit ipmr_net_exit(struct net *net)
2786 {
2787 #ifdef CONFIG_PROC_FS
2788         remove_proc_entry("ip_mr_cache", net->proc_net);
2789         remove_proc_entry("ip_mr_vif", net->proc_net);
2790 #endif
2791         ipmr_rules_exit(net);
2792 }
2793
2794 static struct pernet_operations ipmr_net_ops = {
2795         .init = ipmr_net_init,
2796         .exit = ipmr_net_exit,
2797 };
2798
2799 int __init ip_mr_init(void)
2800 {
2801         int err;
2802
2803         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2804                                        sizeof(struct mfc_cache),
2805                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2806                                        NULL);
2807
2808         err = register_pernet_subsys(&ipmr_net_ops);
2809         if (err)
2810                 goto reg_pernet_fail;
2811
2812         err = register_netdevice_notifier(&ip_mr_notifier);
2813         if (err)
2814                 goto reg_notif_fail;
2815 #ifdef CONFIG_IP_PIMSM_V2
2816         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2817                 pr_err("%s: can't add PIM protocol\n", __func__);
2818                 err = -EAGAIN;
2819                 goto add_proto_fail;
2820         }
2821 #endif
2822         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
2823                       NULL, ipmr_rtm_dumproute, NULL);
2824         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
2825                       ipmr_rtm_route, NULL, NULL);
2826         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
2827                       ipmr_rtm_route, NULL, NULL);
2828         return 0;
2829
2830 #ifdef CONFIG_IP_PIMSM_V2
2831 add_proto_fail:
2832         unregister_netdevice_notifier(&ip_mr_notifier);
2833 #endif
2834 reg_notif_fail:
2835         unregister_pernet_subsys(&ipmr_net_ops);
2836 reg_pernet_fail:
2837         kmem_cache_destroy(mrt_cachep);
2838         return err;
2839 }