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 #include <linux/nospec.h>
  72
  73 struct ipmr_rule {
  74         struct fib_rule         common;
  75 };
  76
  77 struct ipmr_result {
  78         struct mr_table         *mrt;
  79 };
  80
  81 /* Big lock, protecting vif table, mrt cache and mroute socket state.
  82  * Note that the changes are semaphored via rtnl_lock.
  83  */
  84
  85 static DEFINE_RWLOCK(mrt_lock);
  86
  87 /* Multicast router control variables */
  88
  89 /* Special spinlock for queue of unresolved entries */
  90 static DEFINE_SPINLOCK(mfc_unres_lock);
  91
  92 /* We return to original Alan's scheme. Hash table of resolved
  93  * entries is changed only in process context and protected
  94  * with weak lock mrt_lock. Queue of unresolved entries is protected
  95  * with strong spinlock mfc_unres_lock.
  96  *
  97  * In this case data path is free of exclusive locks at all.
  98  */
  99
 100 static struct kmem_cache *mrt_cachep __read_mostly;
 101
 102 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 103 static void ipmr_free_table(struct mr_table *mrt);
 104
 105 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
 106                           struct sk_buff *skb, struct mfc_cache *cache,
 107                           int local);
 108 static int ipmr_cache_report(struct mr_table *mrt,
 109                              struct sk_buff *pkt, vifi_t vifi, int assert);
 110 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
 111                               struct mfc_cache *c, struct rtmsg *rtm);
 112 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
 113                                  int cmd);
 114 static void mroute_clean_tables(struct mr_table *mrt, bool all);
 115 static void ipmr_expire_process(unsigned long arg);
 116
 117 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 118 #define ipmr_for_each_table(mrt, net) \
 119         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
 120
 121 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 122 {
 123         struct mr_table *mrt;
 124
 125         ipmr_for_each_table(mrt, net) {
 126                 if (mrt->id == id)
 127                         return mrt;
 128         }
 129         return NULL;
 130 }
 131
 132 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 133                            struct mr_table **mrt)
 134 {
 135         int err;
 136         struct ipmr_result res;
 137         struct fib_lookup_arg arg = {
 138                 .result = &res,
 139                 .flags = FIB_LOOKUP_NOREF,
 140         };
 141
 142         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 143                                flowi4_to_flowi(flp4), 0, &arg);
 144         if (err < 0)
 145                 return err;
 146         *mrt = res.mrt;
 147         return 0;
 148 }
 149
 150 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 151                             int flags, struct fib_lookup_arg *arg)
 152 {
 153         struct ipmr_result *res = arg->result;
 154         struct mr_table *mrt;
 155
 156         switch (rule->action) {
 157         case FR_ACT_TO_TBL:
 158                 break;
 159         case FR_ACT_UNREACHABLE:
 160                 return -ENETUNREACH;
 161         case FR_ACT_PROHIBIT:
 162                 return -EACCES;
 163         case FR_ACT_BLACKHOLE:
 164         default:
 165                 return -EINVAL;
 166         }
 167
 168         mrt = ipmr_get_table(rule->fr_net, rule->table);
 169         if (!mrt)
 170                 return -EAGAIN;
 171         res->mrt = mrt;
 172         return 0;
 173 }
 174
 175 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 176 {
 177         return 1;
 178 }
 179
 180 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 181         FRA_GENERIC_POLICY,
 182 };
 183
 184 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 185                                struct fib_rule_hdr *frh, struct nlattr **tb)
 186 {
 187         return 0;
 188 }
 189
 190 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 191                              struct nlattr **tb)
 192 {
 193         return 1;
 194 }
 195
 196 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 197                           struct fib_rule_hdr *frh)
 198 {
 199         frh->dst_len = 0;
 200         frh->src_len = 0;
 201         frh->tos     = 0;
 202         return 0;
 203 }
 204
 205 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
 206         .family         = RTNL_FAMILY_IPMR,
 207         .rule_size      = sizeof(struct ipmr_rule),
 208         .addr_size      = sizeof(u32),
 209         .action         = ipmr_rule_action,
 210         .match          = ipmr_rule_match,
 211         .configure      = ipmr_rule_configure,
 212         .compare        = ipmr_rule_compare,
 213         .fill           = ipmr_rule_fill,
 214         .nlgroup        = RTNLGRP_IPV4_RULE,
 215         .policy         = ipmr_rule_policy,
 216         .owner          = THIS_MODULE,
 217 };
 218
 219 static int __net_init ipmr_rules_init(struct net *net)
 220 {
 221         struct fib_rules_ops *ops;
 222         struct mr_table *mrt;
 223         int err;
 224
 225         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 226         if (IS_ERR(ops))
 227                 return PTR_ERR(ops);
 228
 229         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 230
 231         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 232         if (IS_ERR(mrt)) {
 233                 err = PTR_ERR(mrt);
 234                 goto err1;
 235         }
 236
 237         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 238         if (err < 0)
 239                 goto err2;
 240
 241         net->ipv4.mr_rules_ops = ops;
 242         return 0;
 243
 244 err2:
 245         ipmr_free_table(mrt);
 246 err1:
 247         fib_rules_unregister(ops);
 248         return err;
 249 }
 250
 251 static void __net_exit ipmr_rules_exit(struct net *net)
 252 {
 253         struct mr_table *mrt, *next;
 254
 255         rtnl_lock();
 256         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 257                 list_del(&mrt->list);
 258                 ipmr_free_table(mrt);
 259         }
 260         fib_rules_unregister(net->ipv4.mr_rules_ops);
 261         rtnl_unlock();
 262 }
 263 #else
 264 #define ipmr_for_each_table(mrt, net) \
 265         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 266
 267 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 268 {
 269         return net->ipv4.mrt;
 270 }
 271
 272 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 273                            struct mr_table **mrt)
 274 {
 275         *mrt = net->ipv4.mrt;
 276         return 0;
 277 }
 278
 279 static int __net_init ipmr_rules_init(struct net *net)
 280 {
 281         struct mr_table *mrt;
 282
 283         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 284         if (IS_ERR(mrt))
 285                 return PTR_ERR(mrt);
 286         net->ipv4.mrt = mrt;
 287         return 0;
 288 }
 289
 290 static void __net_exit ipmr_rules_exit(struct net *net)
 291 {
 292         rtnl_lock();
 293         ipmr_free_table(net->ipv4.mrt);
 294         net->ipv4.mrt = NULL;
 295         rtnl_unlock();
 296 }
 297 #endif
 298
 299 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 300 {
 301         struct mr_table *mrt;
 302         unsigned int i;
 303
 304         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 305         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 306                 return ERR_PTR(-EINVAL);
 307
 308         mrt = ipmr_get_table(net, id);
 309         if (mrt)
 310                 return mrt;
 311
 312         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
 313         if (!mrt)
 314                 return ERR_PTR(-ENOMEM);
 315         write_pnet(&mrt->net, net);
 316         mrt->id = id;
 317
 318         /* Forwarding cache */
 319         for (i = 0; i < MFC_LINES; i++)
 320                 INIT_LIST_HEAD(&mrt->mfc_cache_array[i]);
 321
 322         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
 323
 324         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
 325                     (unsigned long)mrt);
 326
 327         mrt->mroute_reg_vif_num = -1;
 328 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 329         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 330 #endif
 331         return mrt;
 332 }
 333
 334 static void ipmr_free_table(struct mr_table *mrt)
 335 {
 336         del_timer_sync(&mrt->ipmr_expire_timer);
 337         mroute_clean_tables(mrt, true);
 338         kfree(mrt);
 339 }
 340
 341 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 342
 343 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 344 {
 345         struct net *net = dev_net(dev);
 346
 347         dev_close(dev);
 348
 349         dev = __dev_get_by_name(net, "tunl0");
 350         if (dev) {
 351                 const struct net_device_ops *ops = dev->netdev_ops;
 352                 struct ifreq ifr;
 353                 struct ip_tunnel_parm p;
 354
 355                 memset(&p, 0, sizeof(p));
 356                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 357                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 358                 p.iph.version = 4;
 359                 p.iph.ihl = 5;
 360                 p.iph.protocol = IPPROTO_IPIP;
 361                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 362                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 363
 364                 if (ops->ndo_do_ioctl) {
 365                         mm_segment_t oldfs = get_fs();
 366
 367                         set_fs(KERNEL_DS);
 368                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 369                         set_fs(oldfs);
 370                 }
 371         }
 372 }
 373
 374 /* Initialize ipmr pimreg/tunnel in_device */
 375 static bool ipmr_init_vif_indev(const struct net_device *dev)
 376 {
 377         struct in_device *in_dev;
 378
 379         ASSERT_RTNL();
 380
 381         in_dev = __in_dev_get_rtnl(dev);
 382         if (!in_dev)
 383                 return false;
 384         ipv4_devconf_setall(in_dev);
 385         neigh_parms_data_state_setall(in_dev->arp_parms);
 386         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 387
 388         return true;
 389 }
 390
 391 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 392 {
 393         struct net_device  *dev;
 394
 395         dev = __dev_get_by_name(net, "tunl0");
 396
 397         if (dev) {
 398                 const struct net_device_ops *ops = dev->netdev_ops;
 399                 int err;
 400                 struct ifreq ifr;
 401                 struct ip_tunnel_parm p;
 402
 403                 memset(&p, 0, sizeof(p));
 404                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 405                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 406                 p.iph.version = 4;
 407                 p.iph.ihl = 5;
 408                 p.iph.protocol = IPPROTO_IPIP;
 409                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 410                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 411
 412                 if (ops->ndo_do_ioctl) {
 413                         mm_segment_t oldfs = get_fs();
 414
 415                         set_fs(KERNEL_DS);
 416                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 417                         set_fs(oldfs);
 418                 } else {
 419                         err = -EOPNOTSUPP;
 420                 }
 421                 dev = NULL;
 422
 423                 if (err == 0 &&
 424                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 425                         dev->flags |= IFF_MULTICAST;
 426                         if (!ipmr_init_vif_indev(dev))
 427                                 goto failure;
 428                         if (dev_open(dev))
 429                                 goto failure;
 430                         dev_hold(dev);
 431                 }
 432         }
 433         return dev;
 434
 435 failure:
 436         unregister_netdevice(dev);
 437         return NULL;
 438 }
 439
 440 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 441 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 442 {
 443         struct net *net = dev_net(dev);
 444         struct mr_table *mrt;
 445         struct flowi4 fl4 = {
 446                 .flowi4_oif     = dev->ifindex,
 447                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 448                 .flowi4_mark    = skb->mark,
 449         };
 450         int err;
 451
 452         err = ipmr_fib_lookup(net, &fl4, &mrt);
 453         if (err < 0) {
 454                 kfree_skb(skb);
 455                 return err;
 456         }
 457
 458         read_lock(&mrt_lock);
 459         dev->stats.tx_bytes += skb->len;
 460         dev->stats.tx_packets++;
 461         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 462         read_unlock(&mrt_lock);
 463         kfree_skb(skb);
 464         return NETDEV_TX_OK;
 465 }
 466
 467 static int reg_vif_get_iflink(const struct net_device *dev)
 468 {
 469         return 0;
 470 }
 471
 472 static const struct net_device_ops reg_vif_netdev_ops = {
 473         .ndo_start_xmit = reg_vif_xmit,
 474         .ndo_get_iflink = reg_vif_get_iflink,
 475 };
 476
 477 static void reg_vif_setup(struct net_device *dev)
 478 {
 479         dev->type               = ARPHRD_PIMREG;
 480         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 481         dev->flags              = IFF_NOARP;
 482         dev->netdev_ops         = &reg_vif_netdev_ops;
 483         dev->destructor         = free_netdev;
 484         dev->features           |= NETIF_F_NETNS_LOCAL;
 485 }
 486
 487 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 488 {
 489         struct net_device *dev;
 490         char name[IFNAMSIZ];
 491
 492         if (mrt->id == RT_TABLE_DEFAULT)
 493                 sprintf(name, "pimreg");
 494         else
 495                 sprintf(name, "pimreg%u", mrt->id);
 496
 497         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 498
 499         if (!dev)
 500                 return NULL;
 501
 502         dev_net_set(dev, net);
 503
 504         if (register_netdevice(dev)) {
 505                 free_netdev(dev);
 506                 return NULL;
 507         }
 508
 509         if (!ipmr_init_vif_indev(dev))
 510                 goto failure;
 511         if (dev_open(dev))
 512                 goto failure;
 513
 514         dev_hold(dev);
 515
 516         return dev;
 517
 518 failure:
 519         unregister_netdevice(dev);
 520         return NULL;
 521 }
 522
 523 /* called with rcu_read_lock() */
 524 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 525                      unsigned int pimlen)
 526 {
 527         struct net_device *reg_dev = NULL;
 528         struct iphdr *encap;
 529
 530         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 531         /* Check that:
 532          * a. packet is really sent to a multicast group
 533          * b. packet is not a NULL-REGISTER
 534          * c. packet is not truncated
 535          */
 536         if (!ipv4_is_multicast(encap->daddr) ||
 537             encap->tot_len == 0 ||
 538             ntohs(encap->tot_len) + pimlen > skb->len)
 539                 return 1;
 540
 541         read_lock(&mrt_lock);
 542         if (mrt->mroute_reg_vif_num >= 0)
 543                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 544         read_unlock(&mrt_lock);
 545
 546         if (!reg_dev)
 547                 return 1;
 548
 549         skb->mac_header = skb->network_header;
 550         skb_pull(skb, (u8 *)encap - skb->data);
 551         skb_reset_network_header(skb);
 552         skb->protocol = htons(ETH_P_IP);
 553         skb->ip_summed = CHECKSUM_NONE;
 554
 555         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 556
 557         netif_rx(skb);
 558
 559         return NET_RX_SUCCESS;
 560 }
 561 #else
 562 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 563 {
 564         return NULL;
 565 }
 566 #endif
 567
 568 /**
 569  *      vif_delete - Delete a VIF entry
 570  *      @notify: Set to 1, if the caller is a notifier_call
 571  */
 572 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 573                       struct list_head *head)
 574 {
 575         struct vif_device *v;
 576         struct net_device *dev;
 577         struct in_device *in_dev;
 578
 579         if (vifi < 0 || vifi >= mrt->maxvif)
 580                 return -EADDRNOTAVAIL;
 581
 582         v = &mrt->vif_table[vifi];
 583
 584         write_lock_bh(&mrt_lock);
 585         dev = v->dev;
 586         v->dev = NULL;
 587
 588         if (!dev) {
 589                 write_unlock_bh(&mrt_lock);
 590                 return -EADDRNOTAVAIL;
 591         }
 592
 593         if (vifi == mrt->mroute_reg_vif_num)
 594                 mrt->mroute_reg_vif_num = -1;
 595
 596         if (vifi + 1 == mrt->maxvif) {
 597                 int tmp;
 598
 599                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 600                         if (VIF_EXISTS(mrt, tmp))
 601                                 break;
 602                 }
 603                 mrt->maxvif = tmp+1;
 604         }
 605
 606         write_unlock_bh(&mrt_lock);
 607
 608         dev_set_allmulti(dev, -1);
 609
 610         in_dev = __in_dev_get_rtnl(dev);
 611         if (in_dev) {
 612                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 613                 inet_netconf_notify_devconf(dev_net(dev),
 614                                             NETCONFA_MC_FORWARDING,
 615                                             dev->ifindex, &in_dev->cnf);
 616                 ip_rt_multicast_event(in_dev);
 617         }
 618
 619         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 620                 unregister_netdevice_queue(dev, head);
 621
 622         dev_put(dev);
 623         return 0;
 624 }
 625
 626 static void ipmr_cache_free_rcu(struct rcu_head *head)
 627 {
 628         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
 629
 630         kmem_cache_free(mrt_cachep, c);
 631 }
 632
 633 static inline void ipmr_cache_free(struct mfc_cache *c)
 634 {
 635         call_rcu(&c->rcu, ipmr_cache_free_rcu);
 636 }
 637
 638 /* Destroy an unresolved cache entry, killing queued skbs
 639  * and reporting error to netlink readers.
 640  */
 641 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 642 {
 643         struct net *net = read_pnet(&mrt->net);
 644         struct sk_buff *skb;
 645         struct nlmsgerr *e;
 646
 647         atomic_dec(&mrt->cache_resolve_queue_len);
 648
 649         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
 650                 if (ip_hdr(skb)->version == 0) {
 651                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 652                         nlh->nlmsg_type = NLMSG_ERROR;
 653                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 654                         skb_trim(skb, nlh->nlmsg_len);
 655                         e = nlmsg_data(nlh);
 656                         e->error = -ETIMEDOUT;
 657                         memset(&e->msg, 0, sizeof(e->msg));
 658
 659                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 660                 } else {
 661                         kfree_skb(skb);
 662                 }
 663         }
 664
 665         ipmr_cache_free(c);
 666 }
 667
 668 /* Timer process for the unresolved queue. */
 669 static void ipmr_expire_process(unsigned long arg)
 670 {
 671         struct mr_table *mrt = (struct mr_table *)arg;
 672         unsigned long now;
 673         unsigned long expires;
 674         struct mfc_cache *c, *next;
 675
 676         if (!spin_trylock(&mfc_unres_lock)) {
 677                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 678                 return;
 679         }
 680
 681         if (list_empty(&mrt->mfc_unres_queue))
 682                 goto out;
 683
 684         now = jiffies;
 685         expires = 10*HZ;
 686
 687         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 688                 if (time_after(c->mfc_un.unres.expires, now)) {
 689                         unsigned long interval = c->mfc_un.unres.expires - now;
 690                         if (interval < expires)
 691                                 expires = interval;
 692                         continue;
 693                 }
 694
 695                 list_del(&c->list);
 696                 mroute_netlink_event(mrt, c, RTM_DELROUTE);
 697                 ipmr_destroy_unres(mrt, c);
 698         }
 699
 700         if (!list_empty(&mrt->mfc_unres_queue))
 701                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 702
 703 out:
 704         spin_unlock(&mfc_unres_lock);
 705 }
 706
 707 /* Fill oifs list. It is called under write locked mrt_lock. */
 708 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
 709                                    unsigned char *ttls)
 710 {
 711         int vifi;
 712
 713         cache->mfc_un.res.minvif = MAXVIFS;
 714         cache->mfc_un.res.maxvif = 0;
 715         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 716
 717         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 718                 if (VIF_EXISTS(mrt, vifi) &&
 719                     ttls[vifi] && ttls[vifi] < 255) {
 720                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 721                         if (cache->mfc_un.res.minvif > vifi)
 722                                 cache->mfc_un.res.minvif = vifi;
 723                         if (cache->mfc_un.res.maxvif <= vifi)
 724                                 cache->mfc_un.res.maxvif = vifi + 1;
 725                 }
 726         }
 727         cache->mfc_un.res.lastuse = jiffies;
 728 }
 729
 730 static int vif_add(struct net *net, struct mr_table *mrt,
 731                    struct vifctl *vifc, int mrtsock)
 732 {
 733         int vifi = vifc->vifc_vifi;
 734         struct vif_device *v = &mrt->vif_table[vifi];
 735         struct net_device *dev;
 736         struct in_device *in_dev;
 737         int err;
 738
 739         /* Is vif busy ? */
 740         if (VIF_EXISTS(mrt, vifi))
 741                 return -EADDRINUSE;
 742
 743         switch (vifc->vifc_flags) {
 744         case VIFF_REGISTER:
 745                 if (!ipmr_pimsm_enabled())
 746                         return -EINVAL;
 747                 /* Special Purpose VIF in PIM
 748                  * All the packets will be sent to the daemon
 749                  */
 750                 if (mrt->mroute_reg_vif_num >= 0)
 751                         return -EADDRINUSE;
 752                 dev = ipmr_reg_vif(net, mrt);
 753                 if (!dev)
 754                         return -ENOBUFS;
 755                 err = dev_set_allmulti(dev, 1);
 756                 if (err) {
 757                         unregister_netdevice(dev);
 758                         dev_put(dev);
 759                         return err;
 760                 }
 761                 break;
 762         case VIFF_TUNNEL:
 763                 dev = ipmr_new_tunnel(net, vifc);
 764                 if (!dev)
 765                         return -ENOBUFS;
 766                 err = dev_set_allmulti(dev, 1);
 767                 if (err) {
 768                         ipmr_del_tunnel(dev, vifc);
 769                         dev_put(dev);
 770                         return err;
 771                 }
 772                 break;
 773         case VIFF_USE_IFINDEX:
 774         case 0:
 775                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 776                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 777                         if (dev && !__in_dev_get_rtnl(dev)) {
 778                                 dev_put(dev);
 779                                 return -EADDRNOTAVAIL;
 780                         }
 781                 } else {
 782                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 783                 }
 784                 if (!dev)
 785                         return -EADDRNOTAVAIL;
 786                 err = dev_set_allmulti(dev, 1);
 787                 if (err) {
 788                         dev_put(dev);
 789                         return err;
 790                 }
 791                 break;
 792         default:
 793                 return -EINVAL;
 794         }
 795
 796         in_dev = __in_dev_get_rtnl(dev);
 797         if (!in_dev) {
 798                 dev_put(dev);
 799                 return -EADDRNOTAVAIL;
 800         }
 801         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 802         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING, dev->ifindex,
 803                                     &in_dev->cnf);
 804         ip_rt_multicast_event(in_dev);
 805
 806         /* Fill in the VIF structures */
 807
 808         v->rate_limit = vifc->vifc_rate_limit;
 809         v->local = vifc->vifc_lcl_addr.s_addr;
 810         v->remote = vifc->vifc_rmt_addr.s_addr;
 811         v->flags = vifc->vifc_flags;
 812         if (!mrtsock)
 813                 v->flags |= VIFF_STATIC;
 814         v->threshold = vifc->vifc_threshold;
 815         v->bytes_in = 0;
 816         v->bytes_out = 0;
 817         v->pkt_in = 0;
 818         v->pkt_out = 0;
 819         v->link = dev->ifindex;
 820         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
 821                 v->link = dev_get_iflink(dev);
 822
 823         /* And finish update writing critical data */
 824         write_lock_bh(&mrt_lock);
 825         v->dev = dev;
 826         if (v->flags & VIFF_REGISTER)
 827                 mrt->mroute_reg_vif_num = vifi;
 828         if (vifi+1 > mrt->maxvif)
 829                 mrt->maxvif = vifi+1;
 830         write_unlock_bh(&mrt_lock);
 831         return 0;
 832 }
 833
 834 /* called with rcu_read_lock() */
 835 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 836                                          __be32 origin,
 837                                          __be32 mcastgrp)
 838 {
 839         int line = MFC_HASH(mcastgrp, origin);
 840         struct mfc_cache *c;
 841
 842         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
 843                 if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
 844                         return c;
 845         }
 846         return NULL;
 847 }
 848
 849 /* Look for a (*,*,oif) entry */
 850 static struct mfc_cache *ipmr_cache_find_any_parent(struct mr_table *mrt,
 851                                                     int vifi)
 852 {
 853         int line = MFC_HASH(htonl(INADDR_ANY), htonl(INADDR_ANY));
 854         struct mfc_cache *c;
 855
 856         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list)
 857                 if (c->mfc_origin == htonl(INADDR_ANY) &&
 858                     c->mfc_mcastgrp == htonl(INADDR_ANY) &&
 859                     c->mfc_un.res.ttls[vifi] < 255)
 860                         return c;
 861
 862         return NULL;
 863 }
 864
 865 /* Look for a (*,G) entry */
 866 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 867                                              __be32 mcastgrp, int vifi)
 868 {
 869         int line = MFC_HASH(mcastgrp, htonl(INADDR_ANY));
 870         struct mfc_cache *c, *proxy;
 871
 872         if (mcastgrp == htonl(INADDR_ANY))
 873                 goto skip;
 874
 875         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list)
 876                 if (c->mfc_origin == htonl(INADDR_ANY) &&
 877                     c->mfc_mcastgrp == mcastgrp) {
 878                         if (c->mfc_un.res.ttls[vifi] < 255)
 879                                 return c;
 880
 881                         /* It's ok if the vifi is part of the static tree */
 882                         proxy = ipmr_cache_find_any_parent(mrt,
 883                                                            c->mfc_parent);
 884                         if (proxy && proxy->mfc_un.res.ttls[vifi] < 255)
 885                                 return c;
 886                 }
 887
 888 skip:
 889         return ipmr_cache_find_any_parent(mrt, vifi);
 890 }
 891
 892 /* Allocate a multicast cache entry */
 893 static struct mfc_cache *ipmr_cache_alloc(void)
 894 {
 895         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 896
 897         if (c) {
 898                 c->mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 899                 c->mfc_un.res.minvif = MAXVIFS;
 900         }
 901         return c;
 902 }
 903
 904 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 905 {
 906         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 907
 908         if (c) {
 909                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
 910                 c->mfc_un.unres.expires = jiffies + 10*HZ;
 911         }
 912         return c;
 913 }
 914
 915 /* A cache entry has gone into a resolved state from queued */
 916 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 917                                struct mfc_cache *uc, struct mfc_cache *c)
 918 {
 919         struct sk_buff *skb;
 920         struct nlmsgerr *e;
 921
 922         /* Play the pending entries through our router */
 923         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
 924                 if (ip_hdr(skb)->version == 0) {
 925                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 926
 927                         if (__ipmr_fill_mroute(mrt, skb, c, nlmsg_data(nlh)) > 0) {
 928                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 929                                                  (u8 *)nlh;
 930                         } else {
 931                                 nlh->nlmsg_type = NLMSG_ERROR;
 932                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 933                                 skb_trim(skb, nlh->nlmsg_len);
 934                                 e = nlmsg_data(nlh);
 935                                 e->error = -EMSGSIZE;
 936                                 memset(&e->msg, 0, sizeof(e->msg));
 937                         }
 938
 939                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 940                 } else {
 941                         ip_mr_forward(net, mrt, skb, c, 0);
 942                 }
 943         }
 944 }
 945
 946 /* Bounce a cache query up to mrouted. We could use netlink for this but mrouted
 947  * expects the following bizarre scheme.
 948  *
 949  * Called under mrt_lock.
 950  */
 951 static int ipmr_cache_report(struct mr_table *mrt,
 952                              struct sk_buff *pkt, vifi_t vifi, int assert)
 953 {
 954         const int ihl = ip_hdrlen(pkt);
 955         struct sock *mroute_sk;
 956         struct igmphdr *igmp;
 957         struct igmpmsg *msg;
 958         struct sk_buff *skb;
 959         int ret;
 960
 961         if (assert == IGMPMSG_WHOLEPKT)
 962                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
 963         else
 964                 skb = alloc_skb(128, GFP_ATOMIC);
 965
 966         if (!skb)
 967                 return -ENOBUFS;
 968
 969         if (assert == IGMPMSG_WHOLEPKT) {
 970                 /* Ugly, but we have no choice with this interface.
 971                  * Duplicate old header, fix ihl, length etc.
 972                  * And all this only to mangle msg->im_msgtype and
 973                  * to set msg->im_mbz to "mbz" :-)
 974                  */
 975                 skb_push(skb, sizeof(struct iphdr));
 976                 skb_reset_network_header(skb);
 977                 skb_reset_transport_header(skb);
 978                 msg = (struct igmpmsg *)skb_network_header(skb);
 979                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
 980                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
 981                 msg->im_mbz = 0;
 982                 msg->im_vif = mrt->mroute_reg_vif_num;
 983                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
 984                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
 985                                              sizeof(struct iphdr));
 986         } else {
 987                 /* Copy the IP header */
 988                 skb_set_network_header(skb, skb->len);
 989                 skb_put(skb, ihl);
 990                 skb_copy_to_linear_data(skb, pkt->data, ihl);
 991                 /* Flag to the kernel this is a route add */
 992                 ip_hdr(skb)->protocol = 0;
 993                 msg = (struct igmpmsg *)skb_network_header(skb);
 994                 msg->im_vif = vifi;
 995                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
 996                 /* Add our header */
 997                 igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
 998                 igmp->type = assert;
 999                 msg->im_msgtype = assert;
1000                 igmp->code = 0;
1001                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1002                 skb->transport_header = skb->network_header;
1003         }
1004
1005         rcu_read_lock();
1006         mroute_sk = rcu_dereference(mrt->mroute_sk);
1007         if (!mroute_sk) {
1008                 rcu_read_unlock();
1009                 kfree_skb(skb);
1010                 return -EINVAL;
1011         }
1012
1013         /* Deliver to mrouted */
1014         ret = sock_queue_rcv_skb(mroute_sk, skb);
1015         rcu_read_unlock();
1016         if (ret < 0) {
1017                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1018                 kfree_skb(skb);
1019         }
1020
1021         return ret;
1022 }
1023
1024 /* Queue a packet for resolution. It gets locked cache entry! */
1025 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1026                                  struct sk_buff *skb)
1027 {
1028         bool found = false;
1029         int err;
1030         struct mfc_cache *c;
1031         const struct iphdr *iph = ip_hdr(skb);
1032
1033         spin_lock_bh(&mfc_unres_lock);
1034         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
1035                 if (c->mfc_mcastgrp == iph->daddr &&
1036                     c->mfc_origin == iph->saddr) {
1037                         found = true;
1038                         break;
1039                 }
1040         }
1041
1042         if (!found) {
1043                 /* Create a new entry if allowable */
1044                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1045                     (c = ipmr_cache_alloc_unres()) == NULL) {
1046                         spin_unlock_bh(&mfc_unres_lock);
1047
1048                         kfree_skb(skb);
1049                         return -ENOBUFS;
1050                 }
1051
1052                 /* Fill in the new cache entry */
1053                 c->mfc_parent   = -1;
1054                 c->mfc_origin   = iph->saddr;
1055                 c->mfc_mcastgrp = iph->daddr;
1056
1057                 /* Reflect first query at mrouted. */
1058                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1059                 if (err < 0) {
1060                         /* If the report failed throw the cache entry
1061                            out - Brad Parker
1062                          */
1063                         spin_unlock_bh(&mfc_unres_lock);
1064
1065                         ipmr_cache_free(c);
1066                         kfree_skb(skb);
1067                         return err;
1068                 }
1069
1070                 atomic_inc(&mrt->cache_resolve_queue_len);
1071                 list_add(&c->list, &mrt->mfc_unres_queue);
1072                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1073
1074                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1075                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1076         }
1077
1078         /* See if we can append the packet */
1079         if (c->mfc_un.unres.unresolved.qlen > 3) {
1080                 kfree_skb(skb);
1081                 err = -ENOBUFS;
1082         } else {
1083                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1084                 err = 0;
1085         }
1086
1087         spin_unlock_bh(&mfc_unres_lock);
1088         return err;
1089 }
1090
1091 /* MFC cache manipulation by user space mroute daemon */
1092
1093 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1094 {
1095         int line;
1096         struct mfc_cache *c, *next;
1097
1098         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1099
1100         list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1101                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1102                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr &&
1103                     (parent == -1 || parent == c->mfc_parent)) {
1104                         list_del_rcu(&c->list);
1105                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1106                         ipmr_cache_free(c);
1107                         return 0;
1108                 }
1109         }
1110         return -ENOENT;
1111 }
1112
1113 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1114                         struct mfcctl *mfc, int mrtsock, int parent)
1115 {
1116         bool found = false;
1117         int line;
1118         struct mfc_cache *uc, *c;
1119
1120         if (mfc->mfcc_parent >= MAXVIFS)
1121                 return -ENFILE;
1122
1123         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1124
1125         list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1126                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1127                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr &&
1128                     (parent == -1 || parent == c->mfc_parent)) {
1129                         found = true;
1130                         break;
1131                 }
1132         }
1133
1134         if (found) {
1135                 write_lock_bh(&mrt_lock);
1136                 c->mfc_parent = mfc->mfcc_parent;
1137                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1138                 if (!mrtsock)
1139                         c->mfc_flags |= MFC_STATIC;
1140                 write_unlock_bh(&mrt_lock);
1141                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1142                 return 0;
1143         }
1144
1145         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1146             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1147                 return -EINVAL;
1148
1149         c = ipmr_cache_alloc();
1150         if (!c)
1151                 return -ENOMEM;
1152
1153         c->mfc_origin = mfc->mfcc_origin.s_addr;
1154         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1155         c->mfc_parent = mfc->mfcc_parent;
1156         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1157         if (!mrtsock)
1158                 c->mfc_flags |= MFC_STATIC;
1159
1160         list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1161
1162         /* Check to see if we resolved a queued list. If so we
1163          * need to send on the frames and tidy up.
1164          */
1165         found = false;
1166         spin_lock_bh(&mfc_unres_lock);
1167         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1168                 if (uc->mfc_origin == c->mfc_origin &&
1169                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1170                         list_del(&uc->list);
1171                         atomic_dec(&mrt->cache_resolve_queue_len);
1172                         found = true;
1173                         break;
1174                 }
1175         }
1176         if (list_empty(&mrt->mfc_unres_queue))
1177                 del_timer(&mrt->ipmr_expire_timer);
1178         spin_unlock_bh(&mfc_unres_lock);
1179
1180         if (found) {
1181                 ipmr_cache_resolve(net, mrt, uc, c);
1182                 ipmr_cache_free(uc);
1183         }
1184         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1185         return 0;
1186 }
1187
1188 /* Close the multicast socket, and clear the vif tables etc */
1189 static void mroute_clean_tables(struct mr_table *mrt, bool all)
1190 {
1191         int i;
1192         LIST_HEAD(list);
1193         struct mfc_cache *c, *next;
1194
1195         /* Shut down all active vif entries */
1196         for (i = 0; i < mrt->maxvif; i++) {
1197                 if (!all && (mrt->vif_table[i].flags & VIFF_STATIC))
1198                         continue;
1199                 vif_delete(mrt, i, 0, &list);
1200         }
1201         unregister_netdevice_many(&list);
1202
1203         /* Wipe the cache */
1204         for (i = 0; i < MFC_LINES; i++) {
1205                 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1206                         if (!all && (c->mfc_flags & MFC_STATIC))
1207                                 continue;
1208                         list_del_rcu(&c->list);
1209                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1210                         ipmr_cache_free(c);
1211                 }
1212         }
1213
1214         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1215                 spin_lock_bh(&mfc_unres_lock);
1216                 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1217                         list_del(&c->list);
1218                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1219                         ipmr_destroy_unres(mrt, c);
1220                 }
1221                 spin_unlock_bh(&mfc_unres_lock);
1222         }
1223 }
1224
1225 /* called from ip_ra_control(), before an RCU grace period,
1226  * we dont need to call synchronize_rcu() here
1227  */
1228 static void mrtsock_destruct(struct sock *sk)
1229 {
1230         struct net *net = sock_net(sk);
1231         struct mr_table *mrt;
1232
1233         rtnl_lock();
1234         ipmr_for_each_table(mrt, net) {
1235                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1236                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1237                         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1238                                                     NETCONFA_IFINDEX_ALL,
1239                                                     net->ipv4.devconf_all);
1240                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1241                         mroute_clean_tables(mrt, false);
1242                 }
1243         }
1244         rtnl_unlock();
1245 }
1246
1247 /* Socket options and virtual interface manipulation. The whole
1248  * virtual interface system is a complete heap, but unfortunately
1249  * that's how BSD mrouted happens to think. Maybe one day with a proper
1250  * MOSPF/PIM router set up we can clean this up.
1251  */
1252
1253 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval,
1254                          unsigned int optlen)
1255 {
1256         struct net *net = sock_net(sk);
1257         int val, ret = 0, parent = 0;
1258         struct mr_table *mrt;
1259         struct vifctl vif;
1260         struct mfcctl mfc;
1261         u32 uval;
1262
1263         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1264         rtnl_lock();
1265         if (sk->sk_type != SOCK_RAW ||
1266             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1267                 ret = -EOPNOTSUPP;
1268                 goto out_unlock;
1269         }
1270
1271         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1272         if (!mrt) {
1273                 ret = -ENOENT;
1274                 goto out_unlock;
1275         }
1276         if (optname != MRT_INIT) {
1277                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1278                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1279                         ret = -EACCES;
1280                         goto out_unlock;
1281                 }
1282         }
1283
1284         switch (optname) {
1285         case MRT_INIT:
1286                 if (optlen != sizeof(int)) {
1287                         ret = -EINVAL;
1288                         break;
1289                 }
1290                 if (rtnl_dereference(mrt->mroute_sk)) {
1291                         ret = -EADDRINUSE;
1292                         break;
1293                 }
1294
1295                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1296                 if (ret == 0) {
1297                         rcu_assign_pointer(mrt->mroute_sk, sk);
1298                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1299                         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1300                                                     NETCONFA_IFINDEX_ALL,
1301                                                     net->ipv4.devconf_all);
1302                 }
1303                 break;
1304         case MRT_DONE:
1305                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1306                         ret = -EACCES;
1307                 } else {
1308                         /* We need to unlock here because mrtsock_destruct takes
1309                          * care of rtnl itself and we can't change that due to
1310                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1311                          */
1312                         rtnl_unlock();
1313                         ret = ip_ra_control(sk, 0, NULL);
1314                         goto out;
1315                 }
1316                 break;
1317         case MRT_ADD_VIF:
1318         case MRT_DEL_VIF:
1319                 if (optlen != sizeof(vif)) {
1320                         ret = -EINVAL;
1321                         break;
1322                 }
1323                 if (copy_from_user(&vif, optval, sizeof(vif))) {
1324                         ret = -EFAULT;
1325                         break;
1326                 }
1327                 if (vif.vifc_vifi >= MAXVIFS) {
1328                         ret = -ENFILE;
1329                         break;
1330                 }
1331                 if (optname == MRT_ADD_VIF) {
1332                         ret = vif_add(net, mrt, &vif,
1333                                       sk == rtnl_dereference(mrt->mroute_sk));
1334                 } else {
1335                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1336                 }
1337                 break;
1338         /* Manipulate the forwarding caches. These live
1339          * in a sort of kernel/user symbiosis.
1340          */
1341         case MRT_ADD_MFC:
1342         case MRT_DEL_MFC:
1343                 parent = -1;
1344         case MRT_ADD_MFC_PROXY:
1345         case MRT_DEL_MFC_PROXY:
1346                 if (optlen != sizeof(mfc)) {
1347                         ret = -EINVAL;
1348                         break;
1349                 }
1350                 if (copy_from_user(&mfc, optval, sizeof(mfc))) {
1351                         ret = -EFAULT;
1352                         break;
1353                 }
1354                 if (parent == 0)
1355                         parent = mfc.mfcc_parent;
1356                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1357                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1358                 else
1359                         ret = ipmr_mfc_add(net, mrt, &mfc,
1360                                            sk == rtnl_dereference(mrt->mroute_sk),
1361                                            parent);
1362                 break;
1363         /* Control PIM assert. */
1364         case MRT_ASSERT:
1365                 if (optlen != sizeof(val)) {
1366                         ret = -EINVAL;
1367                         break;
1368                 }
1369                 if (get_user(val, (int __user *)optval)) {
1370                         ret = -EFAULT;
1371                         break;
1372                 }
1373                 mrt->mroute_do_assert = val;
1374                 break;
1375         case MRT_PIM:
1376                 if (!ipmr_pimsm_enabled()) {
1377                         ret = -ENOPROTOOPT;
1378                         break;
1379                 }
1380                 if (optlen != sizeof(val)) {
1381                         ret = -EINVAL;
1382                         break;
1383                 }
1384                 if (get_user(val, (int __user *)optval)) {
1385                         ret = -EFAULT;
1386                         break;
1387                 }
1388
1389                 val = !!val;
1390                 if (val != mrt->mroute_do_pim) {
1391                         mrt->mroute_do_pim = val;
1392                         mrt->mroute_do_assert = val;
1393                 }
1394                 break;
1395         case MRT_TABLE:
1396                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1397                         ret = -ENOPROTOOPT;
1398                         break;
1399                 }
1400                 if (optlen != sizeof(uval)) {
1401                         ret = -EINVAL;
1402                         break;
1403                 }
1404                 if (get_user(uval, (u32 __user *)optval)) {
1405                         ret = -EFAULT;
1406                         break;
1407                 }
1408
1409                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1410                         ret = -EBUSY;
1411                 } else {
1412                         mrt = ipmr_new_table(net, uval);
1413                         if (IS_ERR(mrt))
1414                                 ret = PTR_ERR(mrt);
1415                         else
1416                                 raw_sk(sk)->ipmr_table = uval;
1417                 }
1418                 break;
1419         /* Spurious command, or MRT_VERSION which you cannot set. */
1420         default:
1421                 ret = -ENOPROTOOPT;
1422         }
1423 out_unlock:
1424         rtnl_unlock();
1425 out:
1426         return ret;
1427 }
1428
1429 /* Getsock opt support for the multicast routing system. */
1430 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1431 {
1432         int olr;
1433         int val;
1434         struct net *net = sock_net(sk);
1435         struct mr_table *mrt;
1436
1437         if (sk->sk_type != SOCK_RAW ||
1438             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1439                 return -EOPNOTSUPP;
1440
1441         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1442         if (!mrt)
1443                 return -ENOENT;
1444
1445         switch (optname) {
1446         case MRT_VERSION:
1447                 val = 0x0305;
1448                 break;
1449         case MRT_PIM:
1450                 if (!ipmr_pimsm_enabled())
1451                         return -ENOPROTOOPT;
1452                 val = mrt->mroute_do_pim;
1453                 break;
1454         case MRT_ASSERT:
1455                 val = mrt->mroute_do_assert;
1456                 break;
1457         default:
1458                 return -ENOPROTOOPT;
1459         }
1460
1461         if (get_user(olr, optlen))
1462                 return -EFAULT;
1463         olr = min_t(unsigned int, olr, sizeof(int));
1464         if (olr < 0)
1465                 return -EINVAL;
1466         if (put_user(olr, optlen))
1467                 return -EFAULT;
1468         if (copy_to_user(optval, &val, olr))
1469                 return -EFAULT;
1470         return 0;
1471 }
1472
1473 /* The IP multicast ioctl support routines. */
1474 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1475 {
1476         struct sioc_sg_req sr;
1477         struct sioc_vif_req vr;
1478         struct vif_device *vif;
1479         struct mfc_cache *c;
1480         struct net *net = sock_net(sk);
1481         struct mr_table *mrt;
1482
1483         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1484         if (!mrt)
1485                 return -ENOENT;
1486
1487         switch (cmd) {
1488         case SIOCGETVIFCNT:
1489                 if (copy_from_user(&vr, arg, sizeof(vr)))
1490                         return -EFAULT;
1491                 if (vr.vifi >= mrt->maxvif)
1492                         return -EINVAL;
1493                 read_lock(&mrt_lock);
1494                 vif = &mrt->vif_table[vr.vifi];
1495                 if (VIF_EXISTS(mrt, vr.vifi)) {
1496                         vr.icount = vif->pkt_in;
1497                         vr.ocount = vif->pkt_out;
1498                         vr.ibytes = vif->bytes_in;
1499                         vr.obytes = vif->bytes_out;
1500                         read_unlock(&mrt_lock);
1501
1502                         if (copy_to_user(arg, &vr, sizeof(vr)))
1503                                 return -EFAULT;
1504                         return 0;
1505                 }
1506                 read_unlock(&mrt_lock);
1507                 return -EADDRNOTAVAIL;
1508         case SIOCGETSGCNT:
1509                 if (copy_from_user(&sr, arg, sizeof(sr)))
1510                         return -EFAULT;
1511
1512                 rcu_read_lock();
1513                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1514                 if (c) {
1515                         sr.pktcnt = c->mfc_un.res.pkt;
1516                         sr.bytecnt = c->mfc_un.res.bytes;
1517                         sr.wrong_if = c->mfc_un.res.wrong_if;
1518                         rcu_read_unlock();
1519
1520                         if (copy_to_user(arg, &sr, sizeof(sr)))
1521                                 return -EFAULT;
1522                         return 0;
1523                 }
1524                 rcu_read_unlock();
1525                 return -EADDRNOTAVAIL;
1526         default:
1527                 return -ENOIOCTLCMD;
1528         }
1529 }
1530
1531 #ifdef CONFIG_COMPAT
1532 struct compat_sioc_sg_req {
1533         struct in_addr src;
1534         struct in_addr grp;
1535         compat_ulong_t pktcnt;
1536         compat_ulong_t bytecnt;
1537         compat_ulong_t wrong_if;
1538 };
1539
1540 struct compat_sioc_vif_req {
1541         vifi_t  vifi;           /* Which iface */
1542         compat_ulong_t icount;
1543         compat_ulong_t ocount;
1544         compat_ulong_t ibytes;
1545         compat_ulong_t obytes;
1546 };
1547
1548 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1549 {
1550         struct compat_sioc_sg_req sr;
1551         struct compat_sioc_vif_req vr;
1552         struct vif_device *vif;
1553         struct mfc_cache *c;
1554         struct net *net = sock_net(sk);
1555         struct mr_table *mrt;
1556
1557         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1558         if (!mrt)
1559                 return -ENOENT;
1560
1561         switch (cmd) {
1562         case SIOCGETVIFCNT:
1563                 if (copy_from_user(&vr, arg, sizeof(vr)))
1564                         return -EFAULT;
1565                 if (vr.vifi >= mrt->maxvif)
1566                         return -EINVAL;
1567                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1568                 read_lock(&mrt_lock);
1569                 vif = &mrt->vif_table[vr.vifi];
1570                 if (VIF_EXISTS(mrt, vr.vifi)) {
1571                         vr.icount = vif->pkt_in;
1572                         vr.ocount = vif->pkt_out;
1573                         vr.ibytes = vif->bytes_in;
1574                         vr.obytes = vif->bytes_out;
1575                         read_unlock(&mrt_lock);
1576
1577                         if (copy_to_user(arg, &vr, sizeof(vr)))
1578                                 return -EFAULT;
1579                         return 0;
1580                 }
1581                 read_unlock(&mrt_lock);
1582                 return -EADDRNOTAVAIL;
1583         case SIOCGETSGCNT:
1584                 if (copy_from_user(&sr, arg, sizeof(sr)))
1585                         return -EFAULT;
1586
1587                 rcu_read_lock();
1588                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1589                 if (c) {
1590                         sr.pktcnt = c->mfc_un.res.pkt;
1591                         sr.bytecnt = c->mfc_un.res.bytes;
1592                         sr.wrong_if = c->mfc_un.res.wrong_if;
1593                         rcu_read_unlock();
1594
1595                         if (copy_to_user(arg, &sr, sizeof(sr)))
1596                                 return -EFAULT;
1597                         return 0;
1598                 }
1599                 rcu_read_unlock();
1600                 return -EADDRNOTAVAIL;
1601         default:
1602                 return -ENOIOCTLCMD;
1603         }
1604 }
1605 #endif
1606
1607 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1608 {
1609         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1610         struct net *net = dev_net(dev);
1611         struct mr_table *mrt;
1612         struct vif_device *v;
1613         int ct;
1614
1615         if (event != NETDEV_UNREGISTER)
1616                 return NOTIFY_DONE;
1617
1618         ipmr_for_each_table(mrt, net) {
1619                 v = &mrt->vif_table[0];
1620                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1621                         if (v->dev == dev)
1622                                 vif_delete(mrt, ct, 1, NULL);
1623                 }
1624         }
1625         return NOTIFY_DONE;
1626 }
1627
1628 static struct notifier_block ip_mr_notifier = {
1629         .notifier_call = ipmr_device_event,
1630 };
1631
1632 /* Encapsulate a packet by attaching a valid IPIP header to it.
1633  * This avoids tunnel drivers and other mess and gives us the speed so
1634  * important for multicast video.
1635  */
1636 static void ip_encap(struct net *net, struct sk_buff *skb,
1637                      __be32 saddr, __be32 daddr)
1638 {
1639         struct iphdr *iph;
1640         const struct iphdr *old_iph = ip_hdr(skb);
1641
1642         skb_push(skb, sizeof(struct iphdr));
1643         skb->transport_header = skb->network_header;
1644         skb_reset_network_header(skb);
1645         iph = ip_hdr(skb);
1646
1647         iph->version    =       4;
1648         iph->tos        =       old_iph->tos;
1649         iph->ttl        =       old_iph->ttl;
1650         iph->frag_off   =       0;
1651         iph->daddr      =       daddr;
1652         iph->saddr      =       saddr;
1653         iph->protocol   =       IPPROTO_IPIP;
1654         iph->ihl        =       5;
1655         iph->tot_len    =       htons(skb->len);
1656         ip_select_ident(net, skb, NULL);
1657         ip_send_check(iph);
1658
1659         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1660         nf_reset(skb);
1661 }
1662
1663 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1664                                       struct sk_buff *skb)
1665 {
1666         struct ip_options *opt = &(IPCB(skb)->opt);
1667
1668         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1669         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1670
1671         if (unlikely(opt->optlen))
1672                 ip_forward_options(skb);
1673
1674         return dst_output(net, sk, skb);
1675 }
1676
1677 /* Processing handlers for ipmr_forward */
1678
1679 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1680                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1681 {
1682         const struct iphdr *iph = ip_hdr(skb);
1683         struct vif_device *vif = &mrt->vif_table[vifi];
1684         struct net_device *dev;
1685         struct rtable *rt;
1686         struct flowi4 fl4;
1687         int    encap = 0;
1688
1689         if (!vif->dev)
1690                 goto out_free;
1691
1692         if (vif->flags & VIFF_REGISTER) {
1693                 vif->pkt_out++;
1694                 vif->bytes_out += skb->len;
1695                 vif->dev->stats.tx_bytes += skb->len;
1696                 vif->dev->stats.tx_packets++;
1697                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1698                 goto out_free;
1699         }
1700
1701         if (vif->flags & VIFF_TUNNEL) {
1702                 rt = ip_route_output_ports(net, &fl4, NULL,
1703                                            vif->remote, vif->local,
1704                                            0, 0,
1705                                            IPPROTO_IPIP,
1706                                            RT_TOS(iph->tos), vif->link);
1707                 if (IS_ERR(rt))
1708                         goto out_free;
1709                 encap = sizeof(struct iphdr);
1710         } else {
1711                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1712                                            0, 0,
1713                                            IPPROTO_IPIP,
1714                                            RT_TOS(iph->tos), vif->link);
1715                 if (IS_ERR(rt))
1716                         goto out_free;
1717         }
1718
1719         dev = rt->dst.dev;
1720
1721         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1722                 /* Do not fragment multicasts. Alas, IPv4 does not
1723                  * allow to send ICMP, so that packets will disappear
1724                  * to blackhole.
1725                  */
1726                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1727                 ip_rt_put(rt);
1728                 goto out_free;
1729         }
1730
1731         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1732
1733         if (skb_cow(skb, encap)) {
1734                 ip_rt_put(rt);
1735                 goto out_free;
1736         }
1737
1738         vif->pkt_out++;
1739         vif->bytes_out += skb->len;
1740
1741         skb_dst_drop(skb);
1742         skb_dst_set(skb, &rt->dst);
1743         ip_decrease_ttl(ip_hdr(skb));
1744
1745         /* FIXME: forward and output firewalls used to be called here.
1746          * What do we do with netfilter? -- RR
1747          */
1748         if (vif->flags & VIFF_TUNNEL) {
1749                 ip_encap(net, skb, vif->local, vif->remote);
1750                 /* FIXME: extra output firewall step used to be here. --RR */
1751                 vif->dev->stats.tx_packets++;
1752                 vif->dev->stats.tx_bytes += skb->len;
1753         }
1754
1755         IPCB(skb)->flags |= IPSKB_FORWARDED;
1756
1757         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1758          * not only before forwarding, but after forwarding on all output
1759          * interfaces. It is clear, if mrouter runs a multicasting
1760          * program, it should receive packets not depending to what interface
1761          * program is joined.
1762          * If we will not make it, the program will have to join on all
1763          * interfaces. On the other hand, multihoming host (or router, but
1764          * not mrouter) cannot join to more than one interface - it will
1765          * result in receiving multiple packets.
1766          */
1767         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1768                 net, NULL, skb, skb->dev, dev,
1769                 ipmr_forward_finish);
1770         return;
1771
1772 out_free:
1773         kfree_skb(skb);
1774 }
1775
1776 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1777 {
1778         int ct;
1779
1780         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1781                 if (mrt->vif_table[ct].dev == dev)
1782                         break;
1783         }
1784         return ct;
1785 }
1786
1787 /* "local" means that we should preserve one skb (for local delivery) */
1788 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1789                           struct sk_buff *skb, struct mfc_cache *cache,
1790                           int local)
1791 {
1792         int psend = -1;
1793         int vif, ct;
1794         int true_vifi = ipmr_find_vif(mrt, skb->dev);
1795
1796         vif = cache->mfc_parent;
1797         cache->mfc_un.res.pkt++;
1798         cache->mfc_un.res.bytes += skb->len;
1799         cache->mfc_un.res.lastuse = jiffies;
1800
1801         if (cache->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1802                 struct mfc_cache *cache_proxy;
1803
1804                 /* For an (*,G) entry, we only check that the incomming
1805                  * interface is part of the static tree.
1806                  */
1807                 cache_proxy = ipmr_cache_find_any_parent(mrt, vif);
1808                 if (cache_proxy &&
1809                     cache_proxy->mfc_un.res.ttls[true_vifi] < 255)
1810                         goto forward;
1811         }
1812
1813         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1814         if (mrt->vif_table[vif].dev != skb->dev) {
1815                 if (rt_is_output_route(skb_rtable(skb))) {
1816                         /* It is our own packet, looped back.
1817                          * Very complicated situation...
1818                          *
1819                          * The best workaround until routing daemons will be
1820                          * fixed is not to redistribute packet, if it was
1821                          * send through wrong interface. It means, that
1822                          * multicast applications WILL NOT work for
1823                          * (S,G), which have default multicast route pointing
1824                          * to wrong oif. In any case, it is not a good
1825                          * idea to use multicasting applications on router.
1826                          */
1827                         goto dont_forward;
1828                 }
1829
1830                 cache->mfc_un.res.wrong_if++;
1831
1832                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1833                     /* pimsm uses asserts, when switching from RPT to SPT,
1834                      * so that we cannot check that packet arrived on an oif.
1835                      * It is bad, but otherwise we would need to move pretty
1836                      * large chunk of pimd to kernel. Ough... --ANK
1837                      */
1838                     (mrt->mroute_do_pim ||
1839                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1840                     time_after(jiffies,
1841                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1842                         cache->mfc_un.res.last_assert = jiffies;
1843                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1844                 }
1845                 goto dont_forward;
1846         }
1847
1848 forward:
1849         mrt->vif_table[vif].pkt_in++;
1850         mrt->vif_table[vif].bytes_in += skb->len;
1851
1852         /* Forward the frame */
1853         if (cache->mfc_origin == htonl(INADDR_ANY) &&
1854             cache->mfc_mcastgrp == htonl(INADDR_ANY)) {
1855                 if (true_vifi >= 0 &&
1856                     true_vifi != cache->mfc_parent &&
1857                     ip_hdr(skb)->ttl >
1858                                 cache->mfc_un.res.ttls[cache->mfc_parent]) {
1859                         /* It's an (*,*) entry and the packet is not coming from
1860                          * the upstream: forward the packet to the upstream
1861                          * only.
1862                          */
1863                         psend = cache->mfc_parent;
1864                         goto last_forward;
1865                 }
1866                 goto dont_forward;
1867         }
1868         for (ct = cache->mfc_un.res.maxvif - 1;
1869              ct >= cache->mfc_un.res.minvif; ct--) {
1870                 /* For (*,G) entry, don't forward to the incoming interface */
1871                 if ((cache->mfc_origin != htonl(INADDR_ANY) ||
1872                      ct != true_vifi) &&
1873                     ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1874                         if (psend != -1) {
1875                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1876
1877                                 if (skb2)
1878                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1879                                                         psend);
1880                         }
1881                         psend = ct;
1882                 }
1883         }
1884 last_forward:
1885         if (psend != -1) {
1886                 if (local) {
1887                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1888
1889                         if (skb2)
1890                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1891                 } else {
1892                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1893                         return;
1894                 }
1895         }
1896
1897 dont_forward:
1898         if (!local)
1899                 kfree_skb(skb);
1900 }
1901
1902 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1903 {
1904         struct rtable *rt = skb_rtable(skb);
1905         struct iphdr *iph = ip_hdr(skb);
1906         struct flowi4 fl4 = {
1907                 .daddr = iph->daddr,
1908                 .saddr = iph->saddr,
1909                 .flowi4_tos = RT_TOS(iph->tos),
1910                 .flowi4_oif = (rt_is_output_route(rt) ?
1911                                skb->dev->ifindex : 0),
1912                 .flowi4_iif = (rt_is_output_route(rt) ?
1913                                LOOPBACK_IFINDEX :
1914                                skb->dev->ifindex),
1915                 .flowi4_mark = skb->mark,
1916         };
1917         struct mr_table *mrt;
1918         int err;
1919
1920         err = ipmr_fib_lookup(net, &fl4, &mrt);
1921         if (err)
1922                 return ERR_PTR(err);
1923         return mrt;
1924 }
1925
1926 /* Multicast packets for forwarding arrive here
1927  * Called with rcu_read_lock();
1928  */
1929 int ip_mr_input(struct sk_buff *skb)
1930 {
1931         struct mfc_cache *cache;
1932         struct net *net = dev_net(skb->dev);
1933         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1934         struct mr_table *mrt;
1935         struct net_device *dev;
1936
1937         /* skb->dev passed in is the loX master dev for vrfs.
1938          * As there are no vifs associated with loopback devices,
1939          * get the proper interface that does have a vif associated with it.
1940          */
1941         dev = skb->dev;
1942         if (netif_is_l3_master(skb->dev)) {
1943                 dev = dev_get_by_index_rcu(net, IPCB(skb)->iif);
1944                 if (!dev) {
1945                         kfree_skb(skb);
1946                         return -ENODEV;
1947                 }
1948         }
1949
1950         /* Packet is looped back after forward, it should not be
1951          * forwarded second time, but still can be delivered locally.
1952          */
1953         if (IPCB(skb)->flags & IPSKB_FORWARDED)
1954                 goto dont_forward;
1955
1956         mrt = ipmr_rt_fib_lookup(net, skb);
1957         if (IS_ERR(mrt)) {
1958                 kfree_skb(skb);
1959                 return PTR_ERR(mrt);
1960         }
1961         if (!local) {
1962                 if (IPCB(skb)->opt.router_alert) {
1963                         if (ip_call_ra_chain(skb))
1964                                 return 0;
1965                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1966                         /* IGMPv1 (and broken IGMPv2 implementations sort of
1967                          * Cisco IOS <= 11.2(8)) do not put router alert
1968                          * option to IGMP packets destined to routable
1969                          * groups. It is very bad, because it means
1970                          * that we can forward NO IGMP messages.
1971                          */
1972                         struct sock *mroute_sk;
1973
1974                         mroute_sk = rcu_dereference(mrt->mroute_sk);
1975                         if (mroute_sk) {
1976                                 nf_reset(skb);
1977                                 raw_rcv(mroute_sk, skb);
1978                                 return 0;
1979                         }
1980                     }
1981         }
1982
1983         /* already under rcu_read_lock() */
1984         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1985         if (!cache) {
1986                 int vif = ipmr_find_vif(mrt, dev);
1987
1988                 if (vif >= 0)
1989                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
1990                                                     vif);
1991         }
1992
1993         /* No usable cache entry */
1994         if (!cache) {
1995                 int vif;
1996
1997                 if (local) {
1998                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1999                         ip_local_deliver(skb);
2000                         if (!skb2)
2001                                 return -ENOBUFS;
2002                         skb = skb2;
2003                 }
2004
2005                 read_lock(&mrt_lock);
2006                 vif = ipmr_find_vif(mrt, dev);
2007                 if (vif >= 0) {
2008                         int err2 = ipmr_cache_unresolved(mrt, vif, skb);
2009                         read_unlock(&mrt_lock);
2010
2011                         return err2;
2012                 }
2013                 read_unlock(&mrt_lock);
2014                 kfree_skb(skb);
2015                 return -ENODEV;
2016         }
2017
2018         read_lock(&mrt_lock);
2019         ip_mr_forward(net, mrt, skb, cache, local);
2020         read_unlock(&mrt_lock);
2021
2022         if (local)
2023                 return ip_local_deliver(skb);
2024
2025         return 0;
2026
2027 dont_forward:
2028         if (local)
2029                 return ip_local_deliver(skb);
2030         kfree_skb(skb);
2031         return 0;
2032 }
2033
2034 #ifdef CONFIG_IP_PIMSM_V1
2035 /* Handle IGMP messages of PIMv1 */
2036 int pim_rcv_v1(struct sk_buff *skb)
2037 {
2038         struct igmphdr *pim;
2039         struct net *net = dev_net(skb->dev);
2040         struct mr_table *mrt;
2041
2042         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2043                 goto drop;
2044
2045         pim = igmp_hdr(skb);
2046
2047         mrt = ipmr_rt_fib_lookup(net, skb);
2048         if (IS_ERR(mrt))
2049                 goto drop;
2050         if (!mrt->mroute_do_pim ||
2051             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2052                 goto drop;
2053
2054         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2055 drop:
2056                 kfree_skb(skb);
2057         }
2058         return 0;
2059 }
2060 #endif
2061
2062 #ifdef CONFIG_IP_PIMSM_V2
2063 static int pim_rcv(struct sk_buff *skb)
2064 {
2065         struct pimreghdr *pim;
2066         struct net *net = dev_net(skb->dev);
2067         struct mr_table *mrt;
2068
2069         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2070                 goto drop;
2071
2072         pim = (struct pimreghdr *)skb_transport_header(skb);
2073         if (pim->type != ((PIM_VERSION << 4) | (PIM_REGISTER)) ||
2074             (pim->flags & PIM_NULL_REGISTER) ||
2075             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2076              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2077                 goto drop;
2078
2079         mrt = ipmr_rt_fib_lookup(net, skb);
2080         if (IS_ERR(mrt))
2081                 goto drop;
2082         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2083 drop:
2084                 kfree_skb(skb);
2085         }
2086         return 0;
2087 }
2088 #endif
2089
2090 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2091                               struct mfc_cache *c, struct rtmsg *rtm)
2092 {
2093         struct rta_mfc_stats mfcs;
2094         struct nlattr *mp_attr;
2095         struct rtnexthop *nhp;
2096         unsigned long lastuse;
2097         int ct;
2098
2099         /* If cache is unresolved, don't try to parse IIF and OIF */
2100         if (c->mfc_parent >= MAXVIFS)
2101                 return -ENOENT;
2102
2103         if (VIF_EXISTS(mrt, c->mfc_parent) &&
2104             nla_put_u32(skb, RTA_IIF, mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
2105                 return -EMSGSIZE;
2106
2107         if (!(mp_attr = nla_nest_start(skb, RTA_MULTIPATH)))
2108                 return -EMSGSIZE;
2109
2110         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2111                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2112                         if (!(nhp = nla_reserve_nohdr(skb, sizeof(*nhp)))) {
2113                                 nla_nest_cancel(skb, mp_attr);
2114                                 return -EMSGSIZE;
2115                         }
2116
2117                         nhp->rtnh_flags = 0;
2118                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2119                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2120                         nhp->rtnh_len = sizeof(*nhp);
2121                 }
2122         }
2123
2124         nla_nest_end(skb, mp_attr);
2125
2126         lastuse = READ_ONCE(c->mfc_un.res.lastuse);
2127         lastuse = time_after_eq(jiffies, lastuse) ? jiffies - lastuse : 0;
2128
2129         mfcs.mfcs_packets = c->mfc_un.res.pkt;
2130         mfcs.mfcs_bytes = c->mfc_un.res.bytes;
2131         mfcs.mfcs_wrong_if = c->mfc_un.res.wrong_if;
2132         if (nla_put_64bit(skb, RTA_MFC_STATS, sizeof(mfcs), &mfcs, RTA_PAD) ||
2133             nla_put_u64_64bit(skb, RTA_EXPIRES, jiffies_to_clock_t(lastuse),
2134                               RTA_PAD))
2135                 return -EMSGSIZE;
2136
2137         rtm->rtm_type = RTN_MULTICAST;
2138         return 1;
2139 }
2140
2141 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2142                    __be32 saddr, __be32 daddr,
2143                    struct rtmsg *rtm, int nowait, u32 portid)
2144 {
2145         struct mfc_cache *cache;
2146         struct mr_table *mrt;
2147         int err;
2148
2149         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2150         if (!mrt)
2151                 return -ENOENT;
2152
2153         rcu_read_lock();
2154         cache = ipmr_cache_find(mrt, saddr, daddr);
2155         if (!cache && skb->dev) {
2156                 int vif = ipmr_find_vif(mrt, skb->dev);
2157
2158                 if (vif >= 0)
2159                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2160         }
2161         if (!cache) {
2162                 struct sk_buff *skb2;
2163                 struct iphdr *iph;
2164                 struct net_device *dev;
2165                 int vif = -1;
2166
2167                 if (nowait) {
2168                         rcu_read_unlock();
2169                         return -EAGAIN;
2170                 }
2171
2172                 dev = skb->dev;
2173                 read_lock(&mrt_lock);
2174                 if (dev)
2175                         vif = ipmr_find_vif(mrt, dev);
2176                 if (vif < 0) {
2177                         read_unlock(&mrt_lock);
2178                         rcu_read_unlock();
2179                         return -ENODEV;
2180                 }
2181                 skb2 = skb_clone(skb, GFP_ATOMIC);
2182                 if (!skb2) {
2183                         read_unlock(&mrt_lock);
2184                         rcu_read_unlock();
2185                         return -ENOMEM;
2186                 }
2187
2188                 NETLINK_CB(skb2).portid = portid;
2189                 skb_push(skb2, sizeof(struct iphdr));
2190                 skb_reset_network_header(skb2);
2191                 iph = ip_hdr(skb2);
2192                 iph->ihl = sizeof(struct iphdr) >> 2;
2193                 iph->saddr = saddr;
2194                 iph->daddr = daddr;
2195                 iph->version = 0;
2196                 err = ipmr_cache_unresolved(mrt, vif, skb2);
2197                 read_unlock(&mrt_lock);
2198                 rcu_read_unlock();
2199                 return err;
2200         }
2201
2202         read_lock(&mrt_lock);
2203         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2204         read_unlock(&mrt_lock);
2205         rcu_read_unlock();
2206         return err;
2207 }
2208
2209 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2210                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2211                             int flags)
2212 {
2213         struct nlmsghdr *nlh;
2214         struct rtmsg *rtm;
2215         int err;
2216
2217         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2218         if (!nlh)
2219                 return -EMSGSIZE;
2220
2221         rtm = nlmsg_data(nlh);
2222         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2223         rtm->rtm_dst_len  = 32;
2224         rtm->rtm_src_len  = 32;
2225         rtm->rtm_tos      = 0;
2226         rtm->rtm_table    = mrt->id;
2227         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2228                 goto nla_put_failure;
2229         rtm->rtm_type     = RTN_MULTICAST;
2230         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2231         if (c->mfc_flags & MFC_STATIC)
2232                 rtm->rtm_protocol = RTPROT_STATIC;
2233         else
2234                 rtm->rtm_protocol = RTPROT_MROUTED;
2235         rtm->rtm_flags    = 0;
2236
2237         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2238             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2239                 goto nla_put_failure;
2240         err = __ipmr_fill_mroute(mrt, skb, c, rtm);
2241         /* do not break the dump if cache is unresolved */
2242         if (err < 0 && err != -ENOENT)
2243                 goto nla_put_failure;
2244
2245         nlmsg_end(skb, nlh);
2246         return 0;
2247
2248 nla_put_failure:
2249         nlmsg_cancel(skb, nlh);
2250         return -EMSGSIZE;
2251 }
2252
2253 static size_t mroute_msgsize(bool unresolved, int maxvif)
2254 {
2255         size_t len =
2256                 NLMSG_ALIGN(sizeof(struct rtmsg))
2257                 + nla_total_size(4)     /* RTA_TABLE */
2258                 + nla_total_size(4)     /* RTA_SRC */
2259                 + nla_total_size(4)     /* RTA_DST */
2260                 ;
2261
2262         if (!unresolved)
2263                 len = len
2264                       + nla_total_size(4)       /* RTA_IIF */
2265                       + nla_total_size(0)       /* RTA_MULTIPATH */
2266                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2267                                                 /* RTA_MFC_STATS */
2268                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2269                 ;
2270
2271         return len;
2272 }
2273
2274 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2275                                  int cmd)
2276 {
2277         struct net *net = read_pnet(&mrt->net);
2278         struct sk_buff *skb;
2279         int err = -ENOBUFS;
2280
2281         skb = nlmsg_new(mroute_msgsize(mfc->mfc_parent >= MAXVIFS, mrt->maxvif),
2282                         GFP_ATOMIC);
2283         if (!skb)
2284                 goto errout;
2285
2286         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2287         if (err < 0)
2288                 goto errout;
2289
2290         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2291         return;
2292
2293 errout:
2294         kfree_skb(skb);
2295         if (err < 0)
2296                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2297 }
2298
2299 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2300 {
2301         struct net *net = sock_net(skb->sk);
2302         struct mr_table *mrt;
2303         struct mfc_cache *mfc;
2304         unsigned int t = 0, s_t;
2305         unsigned int h = 0, s_h;
2306         unsigned int e = 0, s_e;
2307
2308         s_t = cb->args[0];
2309         s_h = cb->args[1];
2310         s_e = cb->args[2];
2311
2312         rcu_read_lock();
2313         ipmr_for_each_table(mrt, net) {
2314                 if (t < s_t)
2315                         goto next_table;
2316                 if (t > s_t)
2317                         s_h = 0;
2318                 for (h = s_h; h < MFC_LINES; h++) {
2319                         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2320                                 if (e < s_e)
2321                                         goto next_entry;
2322                                 if (ipmr_fill_mroute(mrt, skb,
2323                                                      NETLINK_CB(cb->skb).portid,
2324                                                      cb->nlh->nlmsg_seq,
2325                                                      mfc, RTM_NEWROUTE,
2326                                                      NLM_F_MULTI) < 0)
2327                                         goto done;
2328 next_entry:
2329                                 e++;
2330                         }
2331                         e = s_e = 0;
2332                 }
2333                 spin_lock_bh(&mfc_unres_lock);
2334                 list_for_each_entry(mfc, &mrt->mfc_unres_queue, list) {
2335                         if (e < s_e)
2336                                 goto next_entry2;
2337                         if (ipmr_fill_mroute(mrt, skb,
2338                                              NETLINK_CB(cb->skb).portid,
2339                                              cb->nlh->nlmsg_seq,
2340                                              mfc, RTM_NEWROUTE,
2341                                              NLM_F_MULTI) < 0) {
2342                                 spin_unlock_bh(&mfc_unres_lock);
2343                                 goto done;
2344                         }
2345 next_entry2:
2346                         e++;
2347                 }
2348                 spin_unlock_bh(&mfc_unres_lock);
2349                 e = s_e = 0;
2350                 s_h = 0;
2351 next_table:
2352                 t++;
2353         }
2354 done:
2355         rcu_read_unlock();
2356
2357         cb->args[2] = e;
2358         cb->args[1] = h;
2359         cb->args[0] = t;
2360
2361         return skb->len;
2362 }
2363
2364 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2365         [RTA_SRC]       = { .type = NLA_U32 },
2366         [RTA_DST]       = { .type = NLA_U32 },
2367         [RTA_IIF]       = { .type = NLA_U32 },
2368         [RTA_TABLE]     = { .type = NLA_U32 },
2369         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2370 };
2371
2372 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2373 {
2374         switch (rtm_protocol) {
2375         case RTPROT_STATIC:
2376         case RTPROT_MROUTED:
2377                 return true;
2378         }
2379         return false;
2380 }
2381
2382 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2383 {
2384         struct rtnexthop *rtnh = nla_data(nla);
2385         int remaining = nla_len(nla), vifi = 0;
2386
2387         while (rtnh_ok(rtnh, remaining)) {
2388                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2389                 if (++vifi == MAXVIFS)
2390                         break;
2391                 rtnh = rtnh_next(rtnh, &remaining);
2392         }
2393
2394         return remaining > 0 ? -EINVAL : vifi;
2395 }
2396
2397 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2398 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2399                             struct mfcctl *mfcc, int *mrtsock,
2400                             struct mr_table **mrtret)
2401 {
2402         struct net_device *dev = NULL;
2403         u32 tblid = RT_TABLE_DEFAULT;
2404         struct mr_table *mrt;
2405         struct nlattr *attr;
2406         struct rtmsg *rtm;
2407         int ret, rem;
2408
2409         ret = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipmr_policy);
2410         if (ret < 0)
2411                 goto out;
2412         rtm = nlmsg_data(nlh);
2413
2414         ret = -EINVAL;
2415         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2416             rtm->rtm_type != RTN_MULTICAST ||
2417             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2418             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2419                 goto out;
2420
2421         memset(mfcc, 0, sizeof(*mfcc));
2422         mfcc->mfcc_parent = -1;
2423         ret = 0;
2424         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2425                 switch (nla_type(attr)) {
2426                 case RTA_SRC:
2427                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2428                         break;
2429                 case RTA_DST:
2430                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2431                         break;
2432                 case RTA_IIF:
2433                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2434                         if (!dev) {
2435                                 ret = -ENODEV;
2436                                 goto out;
2437                         }
2438                         break;
2439                 case RTA_MULTIPATH:
2440                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2441                                 ret = -EINVAL;
2442                                 goto out;
2443                         }
2444                         break;
2445                 case RTA_PREFSRC:
2446                         ret = 1;
2447                         break;
2448                 case RTA_TABLE:
2449                         tblid = nla_get_u32(attr);
2450                         break;
2451                 }
2452         }
2453         mrt = ipmr_get_table(net, tblid);
2454         if (!mrt) {
2455                 ret = -ENOENT;
2456                 goto out;
2457         }
2458         *mrtret = mrt;
2459         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2460         if (dev)
2461                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2462
2463 out:
2464         return ret;
2465 }
2466
2467 /* takes care of both newroute and delroute */
2468 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh)
2469 {
2470         struct net *net = sock_net(skb->sk);
2471         int ret, mrtsock, parent;
2472         struct mr_table *tbl;
2473         struct mfcctl mfcc;
2474
2475         mrtsock = 0;
2476         tbl = NULL;
2477         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl);
2478         if (ret < 0)
2479                 return ret;
2480
2481         parent = ret ? mfcc.mfcc_parent : -1;
2482         if (nlh->nlmsg_type == RTM_NEWROUTE)
2483                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2484         else
2485                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2486 }
2487
2488 #ifdef CONFIG_PROC_FS
2489 /* The /proc interfaces to multicast routing :
2490  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2491  */
2492 struct ipmr_vif_iter {
2493         struct seq_net_private p;
2494         struct mr_table *mrt;
2495         int ct;
2496 };
2497
2498 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2499                                            struct ipmr_vif_iter *iter,
2500                                            loff_t pos)
2501 {
2502         struct mr_table *mrt = iter->mrt;
2503
2504         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2505                 if (!VIF_EXISTS(mrt, iter->ct))
2506                         continue;
2507                 if (pos-- == 0)
2508                         return &mrt->vif_table[iter->ct];
2509         }
2510         return NULL;
2511 }
2512
2513 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2514         __acquires(mrt_lock)
2515 {
2516         struct ipmr_vif_iter *iter = seq->private;
2517         struct net *net = seq_file_net(seq);
2518         struct mr_table *mrt;
2519
2520         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2521         if (!mrt)
2522                 return ERR_PTR(-ENOENT);
2523
2524         iter->mrt = mrt;
2525
2526         read_lock(&mrt_lock);
2527         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2528                 : SEQ_START_TOKEN;
2529 }
2530
2531 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2532 {
2533         struct ipmr_vif_iter *iter = seq->private;
2534         struct net *net = seq_file_net(seq);
2535         struct mr_table *mrt = iter->mrt;
2536
2537         ++*pos;
2538         if (v == SEQ_START_TOKEN)
2539                 return ipmr_vif_seq_idx(net, iter, 0);
2540
2541         while (++iter->ct < mrt->maxvif) {
2542                 if (!VIF_EXISTS(mrt, iter->ct))
2543                         continue;
2544                 return &mrt->vif_table[iter->ct];
2545         }
2546         return NULL;
2547 }
2548
2549 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2550         __releases(mrt_lock)
2551 {
2552         read_unlock(&mrt_lock);
2553 }
2554
2555 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2556 {
2557         struct ipmr_vif_iter *iter = seq->private;
2558         struct mr_table *mrt = iter->mrt;
2559
2560         if (v == SEQ_START_TOKEN) {
2561                 seq_puts(seq,
2562                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2563         } else {
2564                 const struct vif_device *vif = v;
2565                 const char *name =  vif->dev ? vif->dev->name : "none";
2566
2567                 seq_printf(seq,
2568                            "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2569                            vif - mrt->vif_table,
2570                            name, vif->bytes_in, vif->pkt_in,
2571                            vif->bytes_out, vif->pkt_out,
2572                            vif->flags, vif->local, vif->remote);
2573         }
2574         return 0;
2575 }
2576
2577 static const struct seq_operations ipmr_vif_seq_ops = {
2578         .start = ipmr_vif_seq_start,
2579         .next  = ipmr_vif_seq_next,
2580         .stop  = ipmr_vif_seq_stop,
2581         .show  = ipmr_vif_seq_show,
2582 };
2583
2584 static int ipmr_vif_open(struct inode *inode, struct file *file)
2585 {
2586         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2587                             sizeof(struct ipmr_vif_iter));
2588 }
2589
2590 static const struct file_operations ipmr_vif_fops = {
2591         .owner   = THIS_MODULE,
2592         .open    = ipmr_vif_open,
2593         .read    = seq_read,
2594         .llseek  = seq_lseek,
2595         .release = seq_release_net,
2596 };
2597
2598 struct ipmr_mfc_iter {
2599         struct seq_net_private p;
2600         struct mr_table *mrt;
2601         struct list_head *cache;
2602         int ct;
2603 };
2604
2605
2606 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2607                                           struct ipmr_mfc_iter *it, loff_t pos)
2608 {
2609         struct mr_table *mrt = it->mrt;
2610         struct mfc_cache *mfc;
2611
2612         rcu_read_lock();
2613         for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2614                 it->cache = &mrt->mfc_cache_array[it->ct];
2615                 list_for_each_entry_rcu(mfc, it->cache, list)
2616                         if (pos-- == 0)
2617                                 return mfc;
2618         }
2619         rcu_read_unlock();
2620
2621         spin_lock_bh(&mfc_unres_lock);
2622         it->cache = &mrt->mfc_unres_queue;
2623         list_for_each_entry(mfc, it->cache, list)
2624                 if (pos-- == 0)
2625                         return mfc;
2626         spin_unlock_bh(&mfc_unres_lock);
2627
2628         it->cache = NULL;
2629         return NULL;
2630 }
2631
2632
2633 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2634 {
2635         struct ipmr_mfc_iter *it = seq->private;
2636         struct net *net = seq_file_net(seq);
2637         struct mr_table *mrt;
2638
2639         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2640         if (!mrt)
2641                 return ERR_PTR(-ENOENT);
2642
2643         it->mrt = mrt;
2644         it->cache = NULL;
2645         it->ct = 0;
2646         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2647                 : SEQ_START_TOKEN;
2648 }
2649
2650 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2651 {
2652         struct mfc_cache *mfc = v;
2653         struct ipmr_mfc_iter *it = seq->private;
2654         struct net *net = seq_file_net(seq);
2655         struct mr_table *mrt = it->mrt;
2656
2657         ++*pos;
2658
2659         if (v == SEQ_START_TOKEN)
2660                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2661
2662         if (mfc->list.next != it->cache)
2663                 return list_entry(mfc->list.next, struct mfc_cache, list);
2664
2665         if (it->cache == &mrt->mfc_unres_queue)
2666                 goto end_of_list;
2667
2668         BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2669
2670         while (++it->ct < MFC_LINES) {
2671                 it->cache = &mrt->mfc_cache_array[it->ct];
2672                 if (list_empty(it->cache))
2673                         continue;
2674                 return list_first_entry(it->cache, struct mfc_cache, list);
2675         }
2676
2677         /* exhausted cache_array, show unresolved */
2678         rcu_read_unlock();
2679         it->cache = &mrt->mfc_unres_queue;
2680         it->ct = 0;
2681
2682         spin_lock_bh(&mfc_unres_lock);
2683         if (!list_empty(it->cache))
2684                 return list_first_entry(it->cache, struct mfc_cache, list);
2685
2686 end_of_list:
2687         spin_unlock_bh(&mfc_unres_lock);
2688         it->cache = NULL;
2689
2690         return NULL;
2691 }
2692
2693 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2694 {
2695         struct ipmr_mfc_iter *it = seq->private;
2696         struct mr_table *mrt = it->mrt;
2697
2698         if (it->cache == &mrt->mfc_unres_queue)
2699                 spin_unlock_bh(&mfc_unres_lock);
2700         else if (it->cache == &mrt->mfc_cache_array[it->ct])
2701                 rcu_read_unlock();
2702 }
2703
2704 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2705 {
2706         int n;
2707
2708         if (v == SEQ_START_TOKEN) {
2709                 seq_puts(seq,
2710                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2711         } else {
2712                 const struct mfc_cache *mfc = v;
2713                 const struct ipmr_mfc_iter *it = seq->private;
2714                 const struct mr_table *mrt = it->mrt;
2715
2716                 seq_printf(seq, "%08X %08X %-3hd",
2717                            (__force u32) mfc->mfc_mcastgrp,
2718                            (__force u32) mfc->mfc_origin,
2719                            mfc->mfc_parent);
2720
2721                 if (it->cache != &mrt->mfc_unres_queue) {
2722                         seq_printf(seq, " %8lu %8lu %8lu",
2723                                    mfc->mfc_un.res.pkt,
2724                                    mfc->mfc_un.res.bytes,
2725                                    mfc->mfc_un.res.wrong_if);
2726                         for (n = mfc->mfc_un.res.minvif;
2727                              n < mfc->mfc_un.res.maxvif; n++) {
2728                                 if (VIF_EXISTS(mrt, n) &&
2729                                     mfc->mfc_un.res.ttls[n] < 255)
2730                                         seq_printf(seq,
2731                                            " %2d:%-3d",
2732                                            n, mfc->mfc_un.res.ttls[n]);
2733                         }
2734                 } else {
2735                         /* unresolved mfc_caches don't contain
2736                          * pkt, bytes and wrong_if values
2737                          */
2738                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2739                 }
2740                 seq_putc(seq, '\n');
2741         }
2742         return 0;
2743 }
2744
2745 static const struct seq_operations ipmr_mfc_seq_ops = {
2746         .start = ipmr_mfc_seq_start,
2747         .next  = ipmr_mfc_seq_next,
2748         .stop  = ipmr_mfc_seq_stop,
2749         .show  = ipmr_mfc_seq_show,
2750 };
2751
2752 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2753 {
2754         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2755                             sizeof(struct ipmr_mfc_iter));
2756 }
2757
2758 static const struct file_operations ipmr_mfc_fops = {
2759         .owner   = THIS_MODULE,
2760         .open    = ipmr_mfc_open,
2761         .read    = seq_read,
2762         .llseek  = seq_lseek,
2763         .release = seq_release_net,
2764 };
2765 #endif
2766
2767 #ifdef CONFIG_IP_PIMSM_V2
2768 static const struct net_protocol pim_protocol = {
2769         .handler        =       pim_rcv,
2770         .netns_ok       =       1,
2771 };
2772 #endif
2773
2774 /* Setup for IP multicast routing */
2775 static int __net_init ipmr_net_init(struct net *net)
2776 {
2777         int err;
2778
2779         err = ipmr_rules_init(net);
2780         if (err < 0)
2781                 goto fail;
2782
2783 #ifdef CONFIG_PROC_FS
2784         err = -ENOMEM;
2785         if (!proc_create("ip_mr_vif", 0, net->proc_net, &ipmr_vif_fops))
2786                 goto proc_vif_fail;
2787         if (!proc_create("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_fops))
2788                 goto proc_cache_fail;
2789 #endif
2790         return 0;
2791
2792 #ifdef CONFIG_PROC_FS
2793 proc_cache_fail:
2794         remove_proc_entry("ip_mr_vif", net->proc_net);
2795 proc_vif_fail:
2796         ipmr_rules_exit(net);
2797 #endif
2798 fail:
2799         return err;
2800 }
2801
2802 static void __net_exit ipmr_net_exit(struct net *net)
2803 {
2804 #ifdef CONFIG_PROC_FS
2805         remove_proc_entry("ip_mr_cache", net->proc_net);
2806         remove_proc_entry("ip_mr_vif", net->proc_net);
2807 #endif
2808         ipmr_rules_exit(net);
2809 }
2810
2811 static struct pernet_operations ipmr_net_ops = {
2812         .init = ipmr_net_init,
2813         .exit = ipmr_net_exit,
2814 };
2815
2816 int __init ip_mr_init(void)
2817 {
2818         int err;
2819
2820         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2821                                        sizeof(struct mfc_cache),
2822                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2823                                        NULL);
2824
2825         err = register_pernet_subsys(&ipmr_net_ops);
2826         if (err)
2827                 goto reg_pernet_fail;
2828
2829         err = register_netdevice_notifier(&ip_mr_notifier);
2830         if (err)
2831                 goto reg_notif_fail;
2832 #ifdef CONFIG_IP_PIMSM_V2
2833         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2834                 pr_err("%s: can't add PIM protocol\n", __func__);
2835                 err = -EAGAIN;
2836                 goto add_proto_fail;
2837         }
2838 #endif
2839         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
2840                       NULL, ipmr_rtm_dumproute, NULL);
2841         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
2842                       ipmr_rtm_route, NULL, NULL);
2843         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
2844                       ipmr_rtm_route, NULL, NULL);
2845         return 0;
2846
2847 #ifdef CONFIG_IP_PIMSM_V2
2848 add_proto_fail:
2849         unregister_netdevice_notifier(&ip_mr_notifier);
2850 #endif
2851 reg_notif_fail:
2852         unregister_pernet_subsys(&ipmr_net_ops);
2853 reg_pernet_fail:
2854         kmem_cache_destroy(mrt_cachep);
2855         return err;
2856 }