net/ipv4/ipmr.c

   1 /*
   2  *      IP multicast routing support for mrouted 3.6/3.8
   3  *
   4  *              (c) 1995 Alan Cox, <alan@lxorguk.ukuu.org.uk>
   5  *        Linux Consultancy and Custom Driver Development
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  *
  12  *      Fixes:
  13  *      Michael Chastain        :       Incorrect size of copying.
  14  *      Alan Cox                :       Added the cache manager code
  15  *      Alan Cox                :       Fixed the clone/copy bug and device race.
  16  *      Mike McLagan            :       Routing by source
  17  *      Malcolm Beattie         :       Buffer handling fixes.
  18  *      Alexey Kuznetsov        :       Double buffer free and other fixes.
  19  *      SVR Anand               :       Fixed several multicast bugs and problems.
  20  *      Alexey Kuznetsov        :       Status, optimisations and more.
  21  *      Brad Parker             :       Better behaviour on mrouted upcall
  22  *                                      overflow.
  23  *      Carlos Picoto           :       PIMv1 Support
  24  *      Pavlin Ivanov Radoslavov:       PIMv2 Registers must checksum only PIM header
  25  *                                      Relax this requirement to work with older peers.
  26  *
  27  */
  28
  29 #include <asm/uaccess.h>
  30 #include <linux/types.h>
  31 #include <linux/capability.h>
  32 #include <linux/errno.h>
  33 #include <linux/timer.h>
  34 #include <linux/mm.h>
  35 #include <linux/kernel.h>
  36 #include <linux/fcntl.h>
  37 #include <linux/stat.h>
  38 #include <linux/socket.h>
  39 #include <linux/in.h>
  40 #include <linux/inet.h>
  41 #include <linux/netdevice.h>
  42 #include <linux/inetdevice.h>
  43 #include <linux/igmp.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/seq_file.h>
  46 #include <linux/mroute.h>
  47 #include <linux/init.h>
  48 #include <linux/if_ether.h>
  49 #include <linux/slab.h>
  50 #include <net/net_namespace.h>
  51 #include <net/ip.h>
  52 #include <net/protocol.h>
  53 #include <linux/skbuff.h>
  54 #include <net/route.h>
  55 #include <net/sock.h>
  56 #include <net/icmp.h>
  57 #include <net/udp.h>
  58 #include <net/raw.h>
  59 #include <linux/notifier.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/netfilter_ipv4.h>
  62 #include <linux/compat.h>
  63 #include <linux/export.h>
  64 #include <net/ip_tunnels.h>
  65 #include <net/checksum.h>
  66 #include <net/netlink.h>
  67 #include <net/fib_rules.h>
  68 #include <linux/netconf.h>
  69 #include <net/nexthop.h>
  70
  71 struct ipmr_rule {
  72         struct fib_rule         common;
  73 };
  74
  75 struct ipmr_result {
  76         struct mr_table         *mrt;
  77 };
  78
  79 /* Big lock, protecting vif table, mrt cache and mroute socket state.
  80  * Note that the changes are semaphored via rtnl_lock.
  81  */
  82
  83 static DEFINE_RWLOCK(mrt_lock);
  84
  85 /* Multicast router control variables */
  86
  87 /* Special spinlock for queue of unresolved entries */
  88 static DEFINE_SPINLOCK(mfc_unres_lock);
  89
  90 /* We return to original Alan's scheme. Hash table of resolved
  91  * entries is changed only in process context and protected
  92  * with weak lock mrt_lock. Queue of unresolved entries is protected
  93  * with strong spinlock mfc_unres_lock.
  94  *
  95  * In this case data path is free of exclusive locks at all.
  96  */
  97
  98 static struct kmem_cache *mrt_cachep __read_mostly;
  99
 100 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 101 static void ipmr_free_table(struct mr_table *mrt);
 102
 103 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
 104                           struct sk_buff *skb, struct mfc_cache *cache,
 105                           int local);
 106 static int ipmr_cache_report(struct mr_table *mrt,
 107                              struct sk_buff *pkt, vifi_t vifi, int assert);
 108 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
 109                               struct mfc_cache *c, struct rtmsg *rtm);
 110 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
 111                                  int cmd);
 112 static void mroute_clean_tables(struct mr_table *mrt, bool all);
 113 static void ipmr_expire_process(unsigned long arg);
 114
 115 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 116 #define ipmr_for_each_table(mrt, net) \
 117         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
 118
 119 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 120 {
 121         struct mr_table *mrt;
 122
 123         ipmr_for_each_table(mrt, net) {
 124                 if (mrt->id == id)
 125                         return mrt;
 126         }
 127         return NULL;
 128 }
 129
 130 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 131                            struct mr_table **mrt)
 132 {
 133         int err;
 134         struct ipmr_result res;
 135         struct fib_lookup_arg arg = {
 136                 .result = &res,
 137                 .flags = FIB_LOOKUP_NOREF,
 138         };
 139
 140         /* update flow if oif or iif point to device enslaved to l3mdev */
 141         l3mdev_update_flow(net, flowi4_to_flowi(flp4));
 142
 143         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 144                                flowi4_to_flowi(flp4), 0, &arg);
 145         if (err < 0)
 146                 return err;
 147         *mrt = res.mrt;
 148         return 0;
 149 }
 150
 151 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 152                             int flags, struct fib_lookup_arg *arg)
 153 {
 154         struct ipmr_result *res = arg->result;
 155         struct mr_table *mrt;
 156
 157         switch (rule->action) {
 158         case FR_ACT_TO_TBL:
 159                 break;
 160         case FR_ACT_UNREACHABLE:
 161                 return -ENETUNREACH;
 162         case FR_ACT_PROHIBIT:
 163                 return -EACCES;
 164         case FR_ACT_BLACKHOLE:
 165         default:
 166                 return -EINVAL;
 167         }
 168
 169         arg->table = fib_rule_get_table(rule, arg);
 170
 171         mrt = ipmr_get_table(rule->fr_net, arg->table);
 172         if (!mrt)
 173                 return -EAGAIN;
 174         res->mrt = mrt;
 175         return 0;
 176 }
 177
 178 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 179 {
 180         return 1;
 181 }
 182
 183 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 184         FRA_GENERIC_POLICY,
 185 };
 186
 187 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 188                                struct fib_rule_hdr *frh, struct nlattr **tb)
 189 {
 190         return 0;
 191 }
 192
 193 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 194                              struct nlattr **tb)
 195 {
 196         return 1;
 197 }
 198
 199 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 200                           struct fib_rule_hdr *frh)
 201 {
 202         frh->dst_len = 0;
 203         frh->src_len = 0;
 204         frh->tos     = 0;
 205         return 0;
 206 }
 207
 208 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
 209         .family         = RTNL_FAMILY_IPMR,
 210         .rule_size      = sizeof(struct ipmr_rule),
 211         .addr_size      = sizeof(u32),
 212         .action         = ipmr_rule_action,
 213         .match          = ipmr_rule_match,
 214         .configure      = ipmr_rule_configure,
 215         .compare        = ipmr_rule_compare,
 216         .fill           = ipmr_rule_fill,
 217         .nlgroup        = RTNLGRP_IPV4_RULE,
 218         .policy         = ipmr_rule_policy,
 219         .owner          = THIS_MODULE,
 220 };
 221
 222 static int __net_init ipmr_rules_init(struct net *net)
 223 {
 224         struct fib_rules_ops *ops;
 225         struct mr_table *mrt;
 226         int err;
 227
 228         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 229         if (IS_ERR(ops))
 230                 return PTR_ERR(ops);
 231
 232         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 233
 234         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 235         if (IS_ERR(mrt)) {
 236                 err = PTR_ERR(mrt);
 237                 goto err1;
 238         }
 239
 240         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 241         if (err < 0)
 242                 goto err2;
 243
 244         net->ipv4.mr_rules_ops = ops;
 245         return 0;
 246
 247 err2:
 248         ipmr_free_table(mrt);
 249 err1:
 250         fib_rules_unregister(ops);
 251         return err;
 252 }
 253
 254 static void __net_exit ipmr_rules_exit(struct net *net)
 255 {
 256         struct mr_table *mrt, *next;
 257
 258         rtnl_lock();
 259         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 260                 list_del(&mrt->list);
 261                 ipmr_free_table(mrt);
 262         }
 263         fib_rules_unregister(net->ipv4.mr_rules_ops);
 264         rtnl_unlock();
 265 }
 266 #else
 267 #define ipmr_for_each_table(mrt, net) \
 268         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 269
 270 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 271 {
 272         return net->ipv4.mrt;
 273 }
 274
 275 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 276                            struct mr_table **mrt)
 277 {
 278         *mrt = net->ipv4.mrt;
 279         return 0;
 280 }
 281
 282 static int __net_init ipmr_rules_init(struct net *net)
 283 {
 284         struct mr_table *mrt;
 285
 286         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 287         if (IS_ERR(mrt))
 288                 return PTR_ERR(mrt);
 289         net->ipv4.mrt = mrt;
 290         return 0;
 291 }
 292
 293 static void __net_exit ipmr_rules_exit(struct net *net)
 294 {
 295         rtnl_lock();
 296         ipmr_free_table(net->ipv4.mrt);
 297         net->ipv4.mrt = NULL;
 298         rtnl_unlock();
 299 }
 300 #endif
 301
 302 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 303 {
 304         struct mr_table *mrt;
 305         unsigned int i;
 306
 307         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 308         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 309                 return ERR_PTR(-EINVAL);
 310
 311         mrt = ipmr_get_table(net, id);
 312         if (mrt)
 313                 return mrt;
 314
 315         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
 316         if (!mrt)
 317                 return ERR_PTR(-ENOMEM);
 318         write_pnet(&mrt->net, net);
 319         mrt->id = id;
 320
 321         /* Forwarding cache */
 322         for (i = 0; i < MFC_LINES; i++)
 323                 INIT_LIST_HEAD(&mrt->mfc_cache_array[i]);
 324
 325         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
 326
 327         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
 328                     (unsigned long)mrt);
 329
 330         mrt->mroute_reg_vif_num = -1;
 331 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 332         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 333 #endif
 334         return mrt;
 335 }
 336
 337 static void ipmr_free_table(struct mr_table *mrt)
 338 {
 339         del_timer_sync(&mrt->ipmr_expire_timer);
 340         mroute_clean_tables(mrt, true);
 341         kfree(mrt);
 342 }
 343
 344 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 345
 346 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 347 {
 348         struct net *net = dev_net(dev);
 349
 350         dev_close(dev);
 351
 352         dev = __dev_get_by_name(net, "tunl0");
 353         if (dev) {
 354                 const struct net_device_ops *ops = dev->netdev_ops;
 355                 struct ifreq ifr;
 356                 struct ip_tunnel_parm p;
 357
 358                 memset(&p, 0, sizeof(p));
 359                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 360                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 361                 p.iph.version = 4;
 362                 p.iph.ihl = 5;
 363                 p.iph.protocol = IPPROTO_IPIP;
 364                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 365                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 366
 367                 if (ops->ndo_do_ioctl) {
 368                         mm_segment_t oldfs = get_fs();
 369
 370                         set_fs(KERNEL_DS);
 371                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 372                         set_fs(oldfs);
 373                 }
 374         }
 375 }
 376
 377 /* Initialize ipmr pimreg/tunnel in_device */
 378 static bool ipmr_init_vif_indev(const struct net_device *dev)
 379 {
 380         struct in_device *in_dev;
 381
 382         ASSERT_RTNL();
 383
 384         in_dev = __in_dev_get_rtnl(dev);
 385         if (!in_dev)
 386                 return false;
 387         ipv4_devconf_setall(in_dev);
 388         neigh_parms_data_state_setall(in_dev->arp_parms);
 389         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 390
 391         return true;
 392 }
 393
 394 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 395 {
 396         struct net_device  *dev;
 397
 398         dev = __dev_get_by_name(net, "tunl0");
 399
 400         if (dev) {
 401                 const struct net_device_ops *ops = dev->netdev_ops;
 402                 int err;
 403                 struct ifreq ifr;
 404                 struct ip_tunnel_parm p;
 405
 406                 memset(&p, 0, sizeof(p));
 407                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 408                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 409                 p.iph.version = 4;
 410                 p.iph.ihl = 5;
 411                 p.iph.protocol = IPPROTO_IPIP;
 412                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 413                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 414
 415                 if (ops->ndo_do_ioctl) {
 416                         mm_segment_t oldfs = get_fs();
 417
 418                         set_fs(KERNEL_DS);
 419                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 420                         set_fs(oldfs);
 421                 } else {
 422                         err = -EOPNOTSUPP;
 423                 }
 424                 dev = NULL;
 425
 426                 if (err == 0 &&
 427                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 428                         dev->flags |= IFF_MULTICAST;
 429                         if (!ipmr_init_vif_indev(dev))
 430                                 goto failure;
 431                         if (dev_open(dev))
 432                                 goto failure;
 433                         dev_hold(dev);
 434                 }
 435         }
 436         return dev;
 437
 438 failure:
 439         unregister_netdevice(dev);
 440         return NULL;
 441 }
 442
 443 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 444 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 445 {
 446         struct net *net = dev_net(dev);
 447         struct mr_table *mrt;
 448         struct flowi4 fl4 = {
 449                 .flowi4_oif     = dev->ifindex,
 450                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 451                 .flowi4_mark    = skb->mark,
 452         };
 453         int err;
 454
 455         err = ipmr_fib_lookup(net, &fl4, &mrt);
 456         if (err < 0) {
 457                 kfree_skb(skb);
 458                 return err;
 459         }
 460
 461         read_lock(&mrt_lock);
 462         dev->stats.tx_bytes += skb->len;
 463         dev->stats.tx_packets++;
 464         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 465         read_unlock(&mrt_lock);
 466         kfree_skb(skb);
 467         return NETDEV_TX_OK;
 468 }
 469
 470 static int reg_vif_get_iflink(const struct net_device *dev)
 471 {
 472         return 0;
 473 }
 474
 475 static const struct net_device_ops reg_vif_netdev_ops = {
 476         .ndo_start_xmit = reg_vif_xmit,
 477         .ndo_get_iflink = reg_vif_get_iflink,
 478 };
 479
 480 static void reg_vif_setup(struct net_device *dev)
 481 {
 482         dev->type               = ARPHRD_PIMREG;
 483         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 484         dev->flags              = IFF_NOARP;
 485         dev->netdev_ops         = &reg_vif_netdev_ops;
 486         dev->destructor         = free_netdev;
 487         dev->features           |= NETIF_F_NETNS_LOCAL;
 488 }
 489
 490 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 491 {
 492         struct net_device *dev;
 493         char name[IFNAMSIZ];
 494
 495         if (mrt->id == RT_TABLE_DEFAULT)
 496                 sprintf(name, "pimreg");
 497         else
 498                 sprintf(name, "pimreg%u", mrt->id);
 499
 500         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 501
 502         if (!dev)
 503                 return NULL;
 504
 505         dev_net_set(dev, net);
 506
 507         if (register_netdevice(dev)) {
 508                 free_netdev(dev);
 509                 return NULL;
 510         }
 511
 512         if (!ipmr_init_vif_indev(dev))
 513                 goto failure;
 514         if (dev_open(dev))
 515                 goto failure;
 516
 517         dev_hold(dev);
 518
 519         return dev;
 520
 521 failure:
 522         unregister_netdevice(dev);
 523         return NULL;
 524 }
 525
 526 /* called with rcu_read_lock() */
 527 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 528                      unsigned int pimlen)
 529 {
 530         struct net_device *reg_dev = NULL;
 531         struct iphdr *encap;
 532
 533         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 534         /* Check that:
 535          * a. packet is really sent to a multicast group
 536          * b. packet is not a NULL-REGISTER
 537          * c. packet is not truncated
 538          */
 539         if (!ipv4_is_multicast(encap->daddr) ||
 540             encap->tot_len == 0 ||
 541             ntohs(encap->tot_len) + pimlen > skb->len)
 542                 return 1;
 543
 544         read_lock(&mrt_lock);
 545         if (mrt->mroute_reg_vif_num >= 0)
 546                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 547         read_unlock(&mrt_lock);
 548
 549         if (!reg_dev)
 550                 return 1;
 551
 552         skb->mac_header = skb->network_header;
 553         skb_pull(skb, (u8 *)encap - skb->data);
 554         skb_reset_network_header(skb);
 555         skb->protocol = htons(ETH_P_IP);
 556         skb->ip_summed = CHECKSUM_NONE;
 557
 558         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 559
 560         netif_rx(skb);
 561
 562         return NET_RX_SUCCESS;
 563 }
 564 #else
 565 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 566 {
 567         return NULL;
 568 }
 569 #endif
 570
 571 /**
 572  *      vif_delete - Delete a VIF entry
 573  *      @notify: Set to 1, if the caller is a notifier_call
 574  */
 575 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 576                       struct list_head *head)
 577 {
 578         struct vif_device *v;
 579         struct net_device *dev;
 580         struct in_device *in_dev;
 581
 582         if (vifi < 0 || vifi >= mrt->maxvif)
 583                 return -EADDRNOTAVAIL;
 584
 585         v = &mrt->vif_table[vifi];
 586
 587         write_lock_bh(&mrt_lock);
 588         dev = v->dev;
 589         v->dev = NULL;
 590
 591         if (!dev) {
 592                 write_unlock_bh(&mrt_lock);
 593                 return -EADDRNOTAVAIL;
 594         }
 595
 596         if (vifi == mrt->mroute_reg_vif_num)
 597                 mrt->mroute_reg_vif_num = -1;
 598
 599         if (vifi + 1 == mrt->maxvif) {
 600                 int tmp;
 601
 602                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 603                         if (VIF_EXISTS(mrt, tmp))
 604                                 break;
 605                 }
 606                 mrt->maxvif = tmp+1;
 607         }
 608
 609         write_unlock_bh(&mrt_lock);
 610
 611         dev_set_allmulti(dev, -1);
 612
 613         in_dev = __in_dev_get_rtnl(dev);
 614         if (in_dev) {
 615                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 616                 inet_netconf_notify_devconf(dev_net(dev),
 617                                             NETCONFA_MC_FORWARDING,
 618                                             dev->ifindex, &in_dev->cnf);
 619                 ip_rt_multicast_event(in_dev);
 620         }
 621
 622         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 623                 unregister_netdevice_queue(dev, head);
 624
 625         dev_put(dev);
 626         return 0;
 627 }
 628
 629 static void ipmr_cache_free_rcu(struct rcu_head *head)
 630 {
 631         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
 632
 633         kmem_cache_free(mrt_cachep, c);
 634 }
 635
 636 static inline void ipmr_cache_free(struct mfc_cache *c)
 637 {
 638         call_rcu(&c->rcu, ipmr_cache_free_rcu);
 639 }
 640
 641 /* Destroy an unresolved cache entry, killing queued skbs
 642  * and reporting error to netlink readers.
 643  */
 644 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 645 {
 646         struct net *net = read_pnet(&mrt->net);
 647         struct sk_buff *skb;
 648         struct nlmsgerr *e;
 649
 650         atomic_dec(&mrt->cache_resolve_queue_len);
 651
 652         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
 653                 if (ip_hdr(skb)->version == 0) {
 654                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 655                         nlh->nlmsg_type = NLMSG_ERROR;
 656                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 657                         skb_trim(skb, nlh->nlmsg_len);
 658                         e = nlmsg_data(nlh);
 659                         e->error = -ETIMEDOUT;
 660                         memset(&e->msg, 0, sizeof(e->msg));
 661
 662                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 663                 } else {
 664                         kfree_skb(skb);
 665                 }
 666         }
 667
 668         ipmr_cache_free(c);
 669 }
 670
 671 /* Timer process for the unresolved queue. */
 672 static void ipmr_expire_process(unsigned long arg)
 673 {
 674         struct mr_table *mrt = (struct mr_table *)arg;
 675         unsigned long now;
 676         unsigned long expires;
 677         struct mfc_cache *c, *next;
 678
 679         if (!spin_trylock(&mfc_unres_lock)) {
 680                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 681                 return;
 682         }
 683
 684         if (list_empty(&mrt->mfc_unres_queue))
 685                 goto out;
 686
 687         now = jiffies;
 688         expires = 10*HZ;
 689
 690         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 691                 if (time_after(c->mfc_un.unres.expires, now)) {
 692                         unsigned long interval = c->mfc_un.unres.expires - now;
 693                         if (interval < expires)
 694                                 expires = interval;
 695                         continue;
 696                 }
 697
 698                 list_del(&c->list);
 699                 mroute_netlink_event(mrt, c, RTM_DELROUTE);
 700                 ipmr_destroy_unres(mrt, c);
 701         }
 702
 703         if (!list_empty(&mrt->mfc_unres_queue))
 704                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 705
 706 out:
 707         spin_unlock(&mfc_unres_lock);
 708 }
 709
 710 /* Fill oifs list. It is called under write locked mrt_lock. */
 711 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
 712                                    unsigned char *ttls)
 713 {
 714         int vifi;
 715
 716         cache->mfc_un.res.minvif = MAXVIFS;
 717         cache->mfc_un.res.maxvif = 0;
 718         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 719
 720         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 721                 if (VIF_EXISTS(mrt, vifi) &&
 722                     ttls[vifi] && ttls[vifi] < 255) {
 723                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 724                         if (cache->mfc_un.res.minvif > vifi)
 725                                 cache->mfc_un.res.minvif = vifi;
 726                         if (cache->mfc_un.res.maxvif <= vifi)
 727                                 cache->mfc_un.res.maxvif = vifi + 1;
 728                 }
 729         }
 730         cache->mfc_un.res.lastuse = jiffies;
 731 }
 732
 733 static int vif_add(struct net *net, struct mr_table *mrt,
 734                    struct vifctl *vifc, int mrtsock)
 735 {
 736         int vifi = vifc->vifc_vifi;
 737         struct vif_device *v = &mrt->vif_table[vifi];
 738         struct net_device *dev;
 739         struct in_device *in_dev;
 740         int err;
 741
 742         /* Is vif busy ? */
 743         if (VIF_EXISTS(mrt, vifi))
 744                 return -EADDRINUSE;
 745
 746         switch (vifc->vifc_flags) {
 747         case VIFF_REGISTER:
 748                 if (!ipmr_pimsm_enabled())
 749                         return -EINVAL;
 750                 /* Special Purpose VIF in PIM
 751                  * All the packets will be sent to the daemon
 752                  */
 753                 if (mrt->mroute_reg_vif_num >= 0)
 754                         return -EADDRINUSE;
 755                 dev = ipmr_reg_vif(net, mrt);
 756                 if (!dev)
 757                         return -ENOBUFS;
 758                 err = dev_set_allmulti(dev, 1);
 759                 if (err) {
 760                         unregister_netdevice(dev);
 761                         dev_put(dev);
 762                         return err;
 763                 }
 764                 break;
 765         case VIFF_TUNNEL:
 766                 dev = ipmr_new_tunnel(net, vifc);
 767                 if (!dev)
 768                         return -ENOBUFS;
 769                 err = dev_set_allmulti(dev, 1);
 770                 if (err) {
 771                         ipmr_del_tunnel(dev, vifc);
 772                         dev_put(dev);
 773                         return err;
 774                 }
 775                 break;
 776         case VIFF_USE_IFINDEX:
 777         case 0:
 778                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 779                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 780                         if (dev && !__in_dev_get_rtnl(dev)) {
 781                                 dev_put(dev);
 782                                 return -EADDRNOTAVAIL;
 783                         }
 784                 } else {
 785                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 786                 }
 787                 if (!dev)
 788                         return -EADDRNOTAVAIL;
 789                 err = dev_set_allmulti(dev, 1);
 790                 if (err) {
 791                         dev_put(dev);
 792                         return err;
 793                 }
 794                 break;
 795         default:
 796                 return -EINVAL;
 797         }
 798
 799         in_dev = __in_dev_get_rtnl(dev);
 800         if (!in_dev) {
 801                 dev_put(dev);
 802                 return -EADDRNOTAVAIL;
 803         }
 804         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 805         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING, dev->ifindex,
 806                                     &in_dev->cnf);
 807         ip_rt_multicast_event(in_dev);
 808
 809         /* Fill in the VIF structures */
 810
 811         v->rate_limit = vifc->vifc_rate_limit;
 812         v->local = vifc->vifc_lcl_addr.s_addr;
 813         v->remote = vifc->vifc_rmt_addr.s_addr;
 814         v->flags = vifc->vifc_flags;
 815         if (!mrtsock)
 816                 v->flags |= VIFF_STATIC;
 817         v->threshold = vifc->vifc_threshold;
 818         v->bytes_in = 0;
 819         v->bytes_out = 0;
 820         v->pkt_in = 0;
 821         v->pkt_out = 0;
 822         v->link = dev->ifindex;
 823         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
 824                 v->link = dev_get_iflink(dev);
 825
 826         /* And finish update writing critical data */
 827         write_lock_bh(&mrt_lock);
 828         v->dev = dev;
 829         if (v->flags & VIFF_REGISTER)
 830                 mrt->mroute_reg_vif_num = vifi;
 831         if (vifi+1 > mrt->maxvif)
 832                 mrt->maxvif = vifi+1;
 833         write_unlock_bh(&mrt_lock);
 834         return 0;
 835 }
 836
 837 /* called with rcu_read_lock() */
 838 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 839                                          __be32 origin,
 840                                          __be32 mcastgrp)
 841 {
 842         int line = MFC_HASH(mcastgrp, origin);
 843         struct mfc_cache *c;
 844
 845         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
 846                 if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
 847                         return c;
 848         }
 849         return NULL;
 850 }
 851
 852 /* Look for a (*,*,oif) entry */
 853 static struct mfc_cache *ipmr_cache_find_any_parent(struct mr_table *mrt,
 854                                                     int vifi)
 855 {
 856         int line = MFC_HASH(htonl(INADDR_ANY), htonl(INADDR_ANY));
 857         struct mfc_cache *c;
 858
 859         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list)
 860                 if (c->mfc_origin == htonl(INADDR_ANY) &&
 861                     c->mfc_mcastgrp == htonl(INADDR_ANY) &&
 862                     c->mfc_un.res.ttls[vifi] < 255)
 863                         return c;
 864
 865         return NULL;
 866 }
 867
 868 /* Look for a (*,G) entry */
 869 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 870                                              __be32 mcastgrp, int vifi)
 871 {
 872         int line = MFC_HASH(mcastgrp, htonl(INADDR_ANY));
 873         struct mfc_cache *c, *proxy;
 874
 875         if (mcastgrp == htonl(INADDR_ANY))
 876                 goto skip;
 877
 878         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list)
 879                 if (c->mfc_origin == htonl(INADDR_ANY) &&
 880                     c->mfc_mcastgrp == mcastgrp) {
 881                         if (c->mfc_un.res.ttls[vifi] < 255)
 882                                 return c;
 883
 884                         /* It's ok if the vifi is part of the static tree */
 885                         proxy = ipmr_cache_find_any_parent(mrt,
 886                                                            c->mfc_parent);
 887                         if (proxy && proxy->mfc_un.res.ttls[vifi] < 255)
 888                                 return c;
 889                 }
 890
 891 skip:
 892         return ipmr_cache_find_any_parent(mrt, vifi);
 893 }
 894
 895 /* Allocate a multicast cache entry */
 896 static struct mfc_cache *ipmr_cache_alloc(void)
 897 {
 898         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 899
 900         if (c) {
 901                 c->mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 902                 c->mfc_un.res.minvif = MAXVIFS;
 903         }
 904         return c;
 905 }
 906
 907 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 908 {
 909         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 910
 911         if (c) {
 912                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
 913                 c->mfc_un.unres.expires = jiffies + 10*HZ;
 914         }
 915         return c;
 916 }
 917
 918 /* A cache entry has gone into a resolved state from queued */
 919 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 920                                struct mfc_cache *uc, struct mfc_cache *c)
 921 {
 922         struct sk_buff *skb;
 923         struct nlmsgerr *e;
 924
 925         /* Play the pending entries through our router */
 926         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
 927                 if (ip_hdr(skb)->version == 0) {
 928                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 929
 930                         if (__ipmr_fill_mroute(mrt, skb, c, nlmsg_data(nlh)) > 0) {
 931                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 932                                                  (u8 *)nlh;
 933                         } else {
 934                                 nlh->nlmsg_type = NLMSG_ERROR;
 935                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 936                                 skb_trim(skb, nlh->nlmsg_len);
 937                                 e = nlmsg_data(nlh);
 938                                 e->error = -EMSGSIZE;
 939                                 memset(&e->msg, 0, sizeof(e->msg));
 940                         }
 941
 942                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 943                 } else {
 944                         ip_mr_forward(net, mrt, skb, c, 0);
 945                 }
 946         }
 947 }
 948
 949 /* Bounce a cache query up to mrouted. We could use netlink for this but mrouted
 950  * expects the following bizarre scheme.
 951  *
 952  * Called under mrt_lock.
 953  */
 954 static int ipmr_cache_report(struct mr_table *mrt,
 955                              struct sk_buff *pkt, vifi_t vifi, int assert)
 956 {
 957         const int ihl = ip_hdrlen(pkt);
 958         struct sock *mroute_sk;
 959         struct igmphdr *igmp;
 960         struct igmpmsg *msg;
 961         struct sk_buff *skb;
 962         int ret;
 963
 964         if (assert == IGMPMSG_WHOLEPKT)
 965                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
 966         else
 967                 skb = alloc_skb(128, GFP_ATOMIC);
 968
 969         if (!skb)
 970                 return -ENOBUFS;
 971
 972         if (assert == IGMPMSG_WHOLEPKT) {
 973                 /* Ugly, but we have no choice with this interface.
 974                  * Duplicate old header, fix ihl, length etc.
 975                  * And all this only to mangle msg->im_msgtype and
 976                  * to set msg->im_mbz to "mbz" :-)
 977                  */
 978                 skb_push(skb, sizeof(struct iphdr));
 979                 skb_reset_network_header(skb);
 980                 skb_reset_transport_header(skb);
 981                 msg = (struct igmpmsg *)skb_network_header(skb);
 982                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
 983                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
 984                 msg->im_mbz = 0;
 985                 msg->im_vif = mrt->mroute_reg_vif_num;
 986                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
 987                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
 988                                              sizeof(struct iphdr));
 989         } else {
 990                 /* Copy the IP header */
 991                 skb_set_network_header(skb, skb->len);
 992                 skb_put(skb, ihl);
 993                 skb_copy_to_linear_data(skb, pkt->data, ihl);
 994                 /* Flag to the kernel this is a route add */
 995                 ip_hdr(skb)->protocol = 0;
 996                 msg = (struct igmpmsg *)skb_network_header(skb);
 997                 msg->im_vif = vifi;
 998                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
 999                 /* Add our header */
1000                 igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
1001                 igmp->type = assert;
1002                 msg->im_msgtype = assert;
1003                 igmp->code = 0;
1004                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1005                 skb->transport_header = skb->network_header;
1006         }
1007
1008         rcu_read_lock();
1009         mroute_sk = rcu_dereference(mrt->mroute_sk);
1010         if (!mroute_sk) {
1011                 rcu_read_unlock();
1012                 kfree_skb(skb);
1013                 return -EINVAL;
1014         }
1015
1016         /* Deliver to mrouted */
1017         ret = sock_queue_rcv_skb(mroute_sk, skb);
1018         rcu_read_unlock();
1019         if (ret < 0) {
1020                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1021                 kfree_skb(skb);
1022         }
1023
1024         return ret;
1025 }
1026
1027 /* Queue a packet for resolution. It gets locked cache entry! */
1028 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1029                                  struct sk_buff *skb)
1030 {
1031         bool found = false;
1032         int err;
1033         struct mfc_cache *c;
1034         const struct iphdr *iph = ip_hdr(skb);
1035
1036         spin_lock_bh(&mfc_unres_lock);
1037         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
1038                 if (c->mfc_mcastgrp == iph->daddr &&
1039                     c->mfc_origin == iph->saddr) {
1040                         found = true;
1041                         break;
1042                 }
1043         }
1044
1045         if (!found) {
1046                 /* Create a new entry if allowable */
1047                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1048                     (c = ipmr_cache_alloc_unres()) == NULL) {
1049                         spin_unlock_bh(&mfc_unres_lock);
1050
1051                         kfree_skb(skb);
1052                         return -ENOBUFS;
1053                 }
1054
1055                 /* Fill in the new cache entry */
1056                 c->mfc_parent   = -1;
1057                 c->mfc_origin   = iph->saddr;
1058                 c->mfc_mcastgrp = iph->daddr;
1059
1060                 /* Reflect first query at mrouted. */
1061                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1062                 if (err < 0) {
1063                         /* If the report failed throw the cache entry
1064                            out - Brad Parker
1065                          */
1066                         spin_unlock_bh(&mfc_unres_lock);
1067
1068                         ipmr_cache_free(c);
1069                         kfree_skb(skb);
1070                         return err;
1071                 }
1072
1073                 atomic_inc(&mrt->cache_resolve_queue_len);
1074                 list_add(&c->list, &mrt->mfc_unres_queue);
1075                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1076
1077                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1078                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1079         }
1080
1081         /* See if we can append the packet */
1082         if (c->mfc_un.unres.unresolved.qlen > 3) {
1083                 kfree_skb(skb);
1084                 err = -ENOBUFS;
1085         } else {
1086                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1087                 err = 0;
1088         }
1089
1090         spin_unlock_bh(&mfc_unres_lock);
1091         return err;
1092 }
1093
1094 /* MFC cache manipulation by user space mroute daemon */
1095
1096 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1097 {
1098         int line;
1099         struct mfc_cache *c, *next;
1100
1101         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1102
1103         list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1104                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1105                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr &&
1106                     (parent == -1 || parent == c->mfc_parent)) {
1107                         list_del_rcu(&c->list);
1108                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1109                         ipmr_cache_free(c);
1110                         return 0;
1111                 }
1112         }
1113         return -ENOENT;
1114 }
1115
1116 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1117                         struct mfcctl *mfc, int mrtsock, int parent)
1118 {
1119         bool found = false;
1120         int line;
1121         struct mfc_cache *uc, *c;
1122
1123         if (mfc->mfcc_parent >= MAXVIFS)
1124                 return -ENFILE;
1125
1126         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1127
1128         list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1129                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1130                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr &&
1131                     (parent == -1 || parent == c->mfc_parent)) {
1132                         found = true;
1133                         break;
1134                 }
1135         }
1136
1137         if (found) {
1138                 write_lock_bh(&mrt_lock);
1139                 c->mfc_parent = mfc->mfcc_parent;
1140                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1141                 if (!mrtsock)
1142                         c->mfc_flags |= MFC_STATIC;
1143                 write_unlock_bh(&mrt_lock);
1144                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1145                 return 0;
1146         }
1147
1148         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1149             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1150                 return -EINVAL;
1151
1152         c = ipmr_cache_alloc();
1153         if (!c)
1154                 return -ENOMEM;
1155
1156         c->mfc_origin = mfc->mfcc_origin.s_addr;
1157         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1158         c->mfc_parent = mfc->mfcc_parent;
1159         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1160         if (!mrtsock)
1161                 c->mfc_flags |= MFC_STATIC;
1162
1163         list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1164
1165         /* Check to see if we resolved a queued list. If so we
1166          * need to send on the frames and tidy up.
1167          */
1168         found = false;
1169         spin_lock_bh(&mfc_unres_lock);
1170         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1171                 if (uc->mfc_origin == c->mfc_origin &&
1172                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1173                         list_del(&uc->list);
1174                         atomic_dec(&mrt->cache_resolve_queue_len);
1175                         found = true;
1176                         break;
1177                 }
1178         }
1179         if (list_empty(&mrt->mfc_unres_queue))
1180                 del_timer(&mrt->ipmr_expire_timer);
1181         spin_unlock_bh(&mfc_unres_lock);
1182
1183         if (found) {
1184                 ipmr_cache_resolve(net, mrt, uc, c);
1185                 ipmr_cache_free(uc);
1186         }
1187         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1188         return 0;
1189 }
1190
1191 /* Close the multicast socket, and clear the vif tables etc */
1192 static void mroute_clean_tables(struct mr_table *mrt, bool all)
1193 {
1194         int i;
1195         LIST_HEAD(list);
1196         struct mfc_cache *c, *next;
1197
1198         /* Shut down all active vif entries */
1199         for (i = 0; i < mrt->maxvif; i++) {
1200                 if (!all && (mrt->vif_table[i].flags & VIFF_STATIC))
1201                         continue;
1202                 vif_delete(mrt, i, 0, &list);
1203         }
1204         unregister_netdevice_many(&list);
1205
1206         /* Wipe the cache */
1207         for (i = 0; i < MFC_LINES; i++) {
1208                 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1209                         if (!all && (c->mfc_flags & MFC_STATIC))
1210                                 continue;
1211                         list_del_rcu(&c->list);
1212                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1213                         ipmr_cache_free(c);
1214                 }
1215         }
1216
1217         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1218                 spin_lock_bh(&mfc_unres_lock);
1219                 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1220                         list_del(&c->list);
1221                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1222                         ipmr_destroy_unres(mrt, c);
1223                 }
1224                 spin_unlock_bh(&mfc_unres_lock);
1225         }
1226 }
1227
1228 /* called from ip_ra_control(), before an RCU grace period,
1229  * we dont need to call synchronize_rcu() here
1230  */
1231 static void mrtsock_destruct(struct sock *sk)
1232 {
1233         struct net *net = sock_net(sk);
1234         struct mr_table *mrt;
1235
1236         rtnl_lock();
1237         ipmr_for_each_table(mrt, net) {
1238                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1239                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1240                         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1241                                                     NETCONFA_IFINDEX_ALL,
1242                                                     net->ipv4.devconf_all);
1243                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1244                         mroute_clean_tables(mrt, false);
1245                 }
1246         }
1247         rtnl_unlock();
1248 }
1249
1250 /* Socket options and virtual interface manipulation. The whole
1251  * virtual interface system is a complete heap, but unfortunately
1252  * that's how BSD mrouted happens to think. Maybe one day with a proper
1253  * MOSPF/PIM router set up we can clean this up.
1254  */
1255
1256 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval,
1257                          unsigned int optlen)
1258 {
1259         struct net *net = sock_net(sk);
1260         int val, ret = 0, parent = 0;
1261         struct mr_table *mrt;
1262         struct vifctl vif;
1263         struct mfcctl mfc;
1264         u32 uval;
1265
1266         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1267         rtnl_lock();
1268         if (sk->sk_type != SOCK_RAW ||
1269             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1270                 ret = -EOPNOTSUPP;
1271                 goto out_unlock;
1272         }
1273
1274         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1275         if (!mrt) {
1276                 ret = -ENOENT;
1277                 goto out_unlock;
1278         }
1279         if (optname != MRT_INIT) {
1280                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1281                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1282                         ret = -EACCES;
1283                         goto out_unlock;
1284                 }
1285         }
1286
1287         switch (optname) {
1288         case MRT_INIT:
1289                 if (optlen != sizeof(int)) {
1290                         ret = -EINVAL;
1291                         break;
1292                 }
1293                 if (rtnl_dereference(mrt->mroute_sk)) {
1294                         ret = -EADDRINUSE;
1295                         break;
1296                 }
1297
1298                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1299                 if (ret == 0) {
1300                         rcu_assign_pointer(mrt->mroute_sk, sk);
1301                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1302                         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1303                                                     NETCONFA_IFINDEX_ALL,
1304                                                     net->ipv4.devconf_all);
1305                 }
1306                 break;
1307         case MRT_DONE:
1308                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1309                         ret = -EACCES;
1310                 } else {
1311                         /* We need to unlock here because mrtsock_destruct takes
1312                          * care of rtnl itself and we can't change that due to
1313                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1314                          */
1315                         rtnl_unlock();
1316                         ret = ip_ra_control(sk, 0, NULL);
1317                         goto out;
1318                 }
1319                 break;
1320         case MRT_ADD_VIF:
1321         case MRT_DEL_VIF:
1322                 if (optlen != sizeof(vif)) {
1323                         ret = -EINVAL;
1324                         break;
1325                 }
1326                 if (copy_from_user(&vif, optval, sizeof(vif))) {
1327                         ret = -EFAULT;
1328                         break;
1329                 }
1330                 if (vif.vifc_vifi >= MAXVIFS) {
1331                         ret = -ENFILE;
1332                         break;
1333                 }
1334                 if (optname == MRT_ADD_VIF) {
1335                         ret = vif_add(net, mrt, &vif,
1336                                       sk == rtnl_dereference(mrt->mroute_sk));
1337                 } else {
1338                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1339                 }
1340                 break;
1341         /* Manipulate the forwarding caches. These live
1342          * in a sort of kernel/user symbiosis.
1343          */
1344         case MRT_ADD_MFC:
1345         case MRT_DEL_MFC:
1346                 parent = -1;
1347         case MRT_ADD_MFC_PROXY:
1348         case MRT_DEL_MFC_PROXY:
1349                 if (optlen != sizeof(mfc)) {
1350                         ret = -EINVAL;
1351                         break;
1352                 }
1353                 if (copy_from_user(&mfc, optval, sizeof(mfc))) {
1354                         ret = -EFAULT;
1355                         break;
1356                 }
1357                 if (parent == 0)
1358                         parent = mfc.mfcc_parent;
1359                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1360                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1361                 else
1362                         ret = ipmr_mfc_add(net, mrt, &mfc,
1363                                            sk == rtnl_dereference(mrt->mroute_sk),
1364                                            parent);
1365                 break;
1366         /* Control PIM assert. */
1367         case MRT_ASSERT:
1368                 if (optlen != sizeof(val)) {
1369                         ret = -EINVAL;
1370                         break;
1371                 }
1372                 if (get_user(val, (int __user *)optval)) {
1373                         ret = -EFAULT;
1374                         break;
1375                 }
1376                 mrt->mroute_do_assert = val;
1377                 break;
1378         case MRT_PIM:
1379                 if (!ipmr_pimsm_enabled()) {
1380                         ret = -ENOPROTOOPT;
1381                         break;
1382                 }
1383                 if (optlen != sizeof(val)) {
1384                         ret = -EINVAL;
1385                         break;
1386                 }
1387                 if (get_user(val, (int __user *)optval)) {
1388                         ret = -EFAULT;
1389                         break;
1390                 }
1391
1392                 val = !!val;
1393                 if (val != mrt->mroute_do_pim) {
1394                         mrt->mroute_do_pim = val;
1395                         mrt->mroute_do_assert = val;
1396                 }
1397                 break;
1398         case MRT_TABLE:
1399                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1400                         ret = -ENOPROTOOPT;
1401                         break;
1402                 }
1403                 if (optlen != sizeof(uval)) {
1404                         ret = -EINVAL;
1405                         break;
1406                 }
1407                 if (get_user(uval, (u32 __user *)optval)) {
1408                         ret = -EFAULT;
1409                         break;
1410                 }
1411
1412                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1413                         ret = -EBUSY;
1414                 } else {
1415                         mrt = ipmr_new_table(net, uval);
1416                         if (IS_ERR(mrt))
1417                                 ret = PTR_ERR(mrt);
1418                         else
1419                                 raw_sk(sk)->ipmr_table = uval;
1420                 }
1421                 break;
1422         /* Spurious command, or MRT_VERSION which you cannot set. */
1423         default:
1424                 ret = -ENOPROTOOPT;
1425         }
1426 out_unlock:
1427         rtnl_unlock();
1428 out:
1429         return ret;
1430 }
1431
1432 /* Getsock opt support for the multicast routing system. */
1433 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1434 {
1435         int olr;
1436         int val;
1437         struct net *net = sock_net(sk);
1438         struct mr_table *mrt;
1439
1440         if (sk->sk_type != SOCK_RAW ||
1441             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1442                 return -EOPNOTSUPP;
1443
1444         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1445         if (!mrt)
1446                 return -ENOENT;
1447
1448         switch (optname) {
1449         case MRT_VERSION:
1450                 val = 0x0305;
1451                 break;
1452         case MRT_PIM:
1453                 if (!ipmr_pimsm_enabled())
1454                         return -ENOPROTOOPT;
1455                 val = mrt->mroute_do_pim;
1456                 break;
1457         case MRT_ASSERT:
1458                 val = mrt->mroute_do_assert;
1459                 break;
1460         default:
1461                 return -ENOPROTOOPT;
1462         }
1463
1464         if (get_user(olr, optlen))
1465                 return -EFAULT;
1466         olr = min_t(unsigned int, olr, sizeof(int));
1467         if (olr < 0)
1468                 return -EINVAL;
1469         if (put_user(olr, optlen))
1470                 return -EFAULT;
1471         if (copy_to_user(optval, &val, olr))
1472                 return -EFAULT;
1473         return 0;
1474 }
1475
1476 /* The IP multicast ioctl support routines. */
1477 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1478 {
1479         struct sioc_sg_req sr;
1480         struct sioc_vif_req vr;
1481         struct vif_device *vif;
1482         struct mfc_cache *c;
1483         struct net *net = sock_net(sk);
1484         struct mr_table *mrt;
1485
1486         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1487         if (!mrt)
1488                 return -ENOENT;
1489
1490         switch (cmd) {
1491         case SIOCGETVIFCNT:
1492                 if (copy_from_user(&vr, arg, sizeof(vr)))
1493                         return -EFAULT;
1494                 if (vr.vifi >= mrt->maxvif)
1495                         return -EINVAL;
1496                 read_lock(&mrt_lock);
1497                 vif = &mrt->vif_table[vr.vifi];
1498                 if (VIF_EXISTS(mrt, vr.vifi)) {
1499                         vr.icount = vif->pkt_in;
1500                         vr.ocount = vif->pkt_out;
1501                         vr.ibytes = vif->bytes_in;
1502                         vr.obytes = vif->bytes_out;
1503                         read_unlock(&mrt_lock);
1504
1505                         if (copy_to_user(arg, &vr, sizeof(vr)))
1506                                 return -EFAULT;
1507                         return 0;
1508                 }
1509                 read_unlock(&mrt_lock);
1510                 return -EADDRNOTAVAIL;
1511         case SIOCGETSGCNT:
1512                 if (copy_from_user(&sr, arg, sizeof(sr)))
1513                         return -EFAULT;
1514
1515                 rcu_read_lock();
1516                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1517                 if (c) {
1518                         sr.pktcnt = c->mfc_un.res.pkt;
1519                         sr.bytecnt = c->mfc_un.res.bytes;
1520                         sr.wrong_if = c->mfc_un.res.wrong_if;
1521                         rcu_read_unlock();
1522
1523                         if (copy_to_user(arg, &sr, sizeof(sr)))
1524                                 return -EFAULT;
1525                         return 0;
1526                 }
1527                 rcu_read_unlock();
1528                 return -EADDRNOTAVAIL;
1529         default:
1530                 return -ENOIOCTLCMD;
1531         }
1532 }
1533
1534 #ifdef CONFIG_COMPAT
1535 struct compat_sioc_sg_req {
1536         struct in_addr src;
1537         struct in_addr grp;
1538         compat_ulong_t pktcnt;
1539         compat_ulong_t bytecnt;
1540         compat_ulong_t wrong_if;
1541 };
1542
1543 struct compat_sioc_vif_req {
1544         vifi_t  vifi;           /* Which iface */
1545         compat_ulong_t icount;
1546         compat_ulong_t ocount;
1547         compat_ulong_t ibytes;
1548         compat_ulong_t obytes;
1549 };
1550
1551 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1552 {
1553         struct compat_sioc_sg_req sr;
1554         struct compat_sioc_vif_req vr;
1555         struct vif_device *vif;
1556         struct mfc_cache *c;
1557         struct net *net = sock_net(sk);
1558         struct mr_table *mrt;
1559
1560         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1561         if (!mrt)
1562                 return -ENOENT;
1563
1564         switch (cmd) {
1565         case SIOCGETVIFCNT:
1566                 if (copy_from_user(&vr, arg, sizeof(vr)))
1567                         return -EFAULT;
1568                 if (vr.vifi >= mrt->maxvif)
1569                         return -EINVAL;
1570                 read_lock(&mrt_lock);
1571                 vif = &mrt->vif_table[vr.vifi];
1572                 if (VIF_EXISTS(mrt, vr.vifi)) {
1573                         vr.icount = vif->pkt_in;
1574                         vr.ocount = vif->pkt_out;
1575                         vr.ibytes = vif->bytes_in;
1576                         vr.obytes = vif->bytes_out;
1577                         read_unlock(&mrt_lock);
1578
1579                         if (copy_to_user(arg, &vr, sizeof(vr)))
1580                                 return -EFAULT;
1581                         return 0;
1582                 }
1583                 read_unlock(&mrt_lock);
1584                 return -EADDRNOTAVAIL;
1585         case SIOCGETSGCNT:
1586                 if (copy_from_user(&sr, arg, sizeof(sr)))
1587                         return -EFAULT;
1588
1589                 rcu_read_lock();
1590                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1591                 if (c) {
1592                         sr.pktcnt = c->mfc_un.res.pkt;
1593                         sr.bytecnt = c->mfc_un.res.bytes;
1594                         sr.wrong_if = c->mfc_un.res.wrong_if;
1595                         rcu_read_unlock();
1596
1597                         if (copy_to_user(arg, &sr, sizeof(sr)))
1598                                 return -EFAULT;
1599                         return 0;
1600                 }
1601                 rcu_read_unlock();
1602                 return -EADDRNOTAVAIL;
1603         default:
1604                 return -ENOIOCTLCMD;
1605         }
1606 }
1607 #endif
1608
1609 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1610 {
1611         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1612         struct net *net = dev_net(dev);
1613         struct mr_table *mrt;
1614         struct vif_device *v;
1615         int ct;
1616
1617         if (event != NETDEV_UNREGISTER)
1618                 return NOTIFY_DONE;
1619
1620         ipmr_for_each_table(mrt, net) {
1621                 v = &mrt->vif_table[0];
1622                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1623                         if (v->dev == dev)
1624                                 vif_delete(mrt, ct, 1, NULL);
1625                 }
1626         }
1627         return NOTIFY_DONE;
1628 }
1629
1630 static struct notifier_block ip_mr_notifier = {
1631         .notifier_call = ipmr_device_event,
1632 };
1633
1634 /* Encapsulate a packet by attaching a valid IPIP header to it.
1635  * This avoids tunnel drivers and other mess and gives us the speed so
1636  * important for multicast video.
1637  */
1638 static void ip_encap(struct net *net, struct sk_buff *skb,
1639                      __be32 saddr, __be32 daddr)
1640 {
1641         struct iphdr *iph;
1642         const struct iphdr *old_iph = ip_hdr(skb);
1643
1644         skb_push(skb, sizeof(struct iphdr));
1645         skb->transport_header = skb->network_header;
1646         skb_reset_network_header(skb);
1647         iph = ip_hdr(skb);
1648
1649         iph->version    =       4;
1650         iph->tos        =       old_iph->tos;
1651         iph->ttl        =       old_iph->ttl;
1652         iph->frag_off   =       0;
1653         iph->daddr      =       daddr;
1654         iph->saddr      =       saddr;
1655         iph->protocol   =       IPPROTO_IPIP;
1656         iph->ihl        =       5;
1657         iph->tot_len    =       htons(skb->len);
1658         ip_select_ident(net, skb, NULL);
1659         ip_send_check(iph);
1660
1661         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1662         nf_reset(skb);
1663 }
1664
1665 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1666                                       struct sk_buff *skb)
1667 {
1668         struct ip_options *opt = &(IPCB(skb)->opt);
1669
1670         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1671         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1672
1673         if (unlikely(opt->optlen))
1674                 ip_forward_options(skb);
1675
1676         return dst_output(net, sk, skb);
1677 }
1678
1679 /* Processing handlers for ipmr_forward */
1680
1681 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1682                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1683 {
1684         const struct iphdr *iph = ip_hdr(skb);
1685         struct vif_device *vif = &mrt->vif_table[vifi];
1686         struct net_device *dev;
1687         struct rtable *rt;
1688         struct flowi4 fl4;
1689         int    encap = 0;
1690
1691         if (!vif->dev)
1692                 goto out_free;
1693
1694         if (vif->flags & VIFF_REGISTER) {
1695                 vif->pkt_out++;
1696                 vif->bytes_out += skb->len;
1697                 vif->dev->stats.tx_bytes += skb->len;
1698                 vif->dev->stats.tx_packets++;
1699                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1700                 goto out_free;
1701         }
1702
1703         if (vif->flags & VIFF_TUNNEL) {
1704                 rt = ip_route_output_ports(net, &fl4, NULL,
1705                                            vif->remote, vif->local,
1706                                            0, 0,
1707                                            IPPROTO_IPIP,
1708                                            RT_TOS(iph->tos), vif->link);
1709                 if (IS_ERR(rt))
1710                         goto out_free;
1711                 encap = sizeof(struct iphdr);
1712         } else {
1713                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1714                                            0, 0,
1715                                            IPPROTO_IPIP,
1716                                            RT_TOS(iph->tos), vif->link);
1717                 if (IS_ERR(rt))
1718                         goto out_free;
1719         }
1720
1721         dev = rt->dst.dev;
1722
1723         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1724                 /* Do not fragment multicasts. Alas, IPv4 does not
1725                  * allow to send ICMP, so that packets will disappear
1726                  * to blackhole.
1727                  */
1728                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1729                 ip_rt_put(rt);
1730                 goto out_free;
1731         }
1732
1733         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1734
1735         if (skb_cow(skb, encap)) {
1736                 ip_rt_put(rt);
1737                 goto out_free;
1738         }
1739
1740         vif->pkt_out++;
1741         vif->bytes_out += skb->len;
1742
1743         skb_dst_drop(skb);
1744         skb_dst_set(skb, &rt->dst);
1745         ip_decrease_ttl(ip_hdr(skb));
1746
1747         /* FIXME: forward and output firewalls used to be called here.
1748          * What do we do with netfilter? -- RR
1749          */
1750         if (vif->flags & VIFF_TUNNEL) {
1751                 ip_encap(net, skb, vif->local, vif->remote);
1752                 /* FIXME: extra output firewall step used to be here. --RR */
1753                 vif->dev->stats.tx_packets++;
1754                 vif->dev->stats.tx_bytes += skb->len;
1755         }
1756
1757         IPCB(skb)->flags |= IPSKB_FORWARDED;
1758
1759         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1760          * not only before forwarding, but after forwarding on all output
1761          * interfaces. It is clear, if mrouter runs a multicasting
1762          * program, it should receive packets not depending to what interface
1763          * program is joined.
1764          * If we will not make it, the program will have to join on all
1765          * interfaces. On the other hand, multihoming host (or router, but
1766          * not mrouter) cannot join to more than one interface - it will
1767          * result in receiving multiple packets.
1768          */
1769         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1770                 net, NULL, skb, skb->dev, dev,
1771                 ipmr_forward_finish);
1772         return;
1773
1774 out_free:
1775         kfree_skb(skb);
1776 }
1777
1778 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1779 {
1780         int ct;
1781
1782         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1783                 if (mrt->vif_table[ct].dev == dev)
1784                         break;
1785         }
1786         return ct;
1787 }
1788
1789 /* "local" means that we should preserve one skb (for local delivery) */
1790 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1791                           struct sk_buff *skb, struct mfc_cache *cache,
1792                           int local)
1793 {
1794         int psend = -1;
1795         int vif, ct;
1796         int true_vifi = ipmr_find_vif(mrt, skb->dev);
1797
1798         vif = cache->mfc_parent;
1799         cache->mfc_un.res.pkt++;
1800         cache->mfc_un.res.bytes += skb->len;
1801         cache->mfc_un.res.lastuse = jiffies;
1802
1803         if (cache->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1804                 struct mfc_cache *cache_proxy;
1805
1806                 /* For an (*,G) entry, we only check that the incomming
1807                  * interface is part of the static tree.
1808                  */
1809                 cache_proxy = ipmr_cache_find_any_parent(mrt, vif);
1810                 if (cache_proxy &&
1811                     cache_proxy->mfc_un.res.ttls[true_vifi] < 255)
1812                         goto forward;
1813         }
1814
1815         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1816         if (mrt->vif_table[vif].dev != skb->dev) {
1817                 struct net_device *mdev;
1818
1819                 mdev = l3mdev_master_dev_rcu(mrt->vif_table[vif].dev);
1820                 if (mdev == skb->dev)
1821                         goto forward;
1822
1823                 if (rt_is_output_route(skb_rtable(skb))) {
1824                         /* It is our own packet, looped back.
1825                          * Very complicated situation...
1826                          *
1827                          * The best workaround until routing daemons will be
1828                          * fixed is not to redistribute packet, if it was
1829                          * send through wrong interface. It means, that
1830                          * multicast applications WILL NOT work for
1831                          * (S,G), which have default multicast route pointing
1832                          * to wrong oif. In any case, it is not a good
1833                          * idea to use multicasting applications on router.
1834                          */
1835                         goto dont_forward;
1836                 }
1837
1838                 cache->mfc_un.res.wrong_if++;
1839
1840                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1841                     /* pimsm uses asserts, when switching from RPT to SPT,
1842                      * so that we cannot check that packet arrived on an oif.
1843                      * It is bad, but otherwise we would need to move pretty
1844                      * large chunk of pimd to kernel. Ough... --ANK
1845                      */
1846                     (mrt->mroute_do_pim ||
1847                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1848                     time_after(jiffies,
1849                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1850                         cache->mfc_un.res.last_assert = jiffies;
1851                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1852                 }
1853                 goto dont_forward;
1854         }
1855
1856 forward:
1857         mrt->vif_table[vif].pkt_in++;
1858         mrt->vif_table[vif].bytes_in += skb->len;
1859
1860         /* Forward the frame */
1861         if (cache->mfc_origin == htonl(INADDR_ANY) &&
1862             cache->mfc_mcastgrp == htonl(INADDR_ANY)) {
1863                 if (true_vifi >= 0 &&
1864                     true_vifi != cache->mfc_parent &&
1865                     ip_hdr(skb)->ttl >
1866                                 cache->mfc_un.res.ttls[cache->mfc_parent]) {
1867                         /* It's an (*,*) entry and the packet is not coming from
1868                          * the upstream: forward the packet to the upstream
1869                          * only.
1870                          */
1871                         psend = cache->mfc_parent;
1872                         goto last_forward;
1873                 }
1874                 goto dont_forward;
1875         }
1876         for (ct = cache->mfc_un.res.maxvif - 1;
1877              ct >= cache->mfc_un.res.minvif; ct--) {
1878                 /* For (*,G) entry, don't forward to the incoming interface */
1879                 if ((cache->mfc_origin != htonl(INADDR_ANY) ||
1880                      ct != true_vifi) &&
1881                     ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1882                         if (psend != -1) {
1883                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1884
1885                                 if (skb2)
1886                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1887                                                         psend);
1888                         }
1889                         psend = ct;
1890                 }
1891         }
1892 last_forward:
1893         if (psend != -1) {
1894                 if (local) {
1895                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1896
1897                         if (skb2)
1898                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1899                 } else {
1900                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1901                         return;
1902                 }
1903         }
1904
1905 dont_forward:
1906         if (!local)
1907                 kfree_skb(skb);
1908 }
1909
1910 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1911 {
1912         struct rtable *rt = skb_rtable(skb);
1913         struct iphdr *iph = ip_hdr(skb);
1914         struct flowi4 fl4 = {
1915                 .daddr = iph->daddr,
1916                 .saddr = iph->saddr,
1917                 .flowi4_tos = RT_TOS(iph->tos),
1918                 .flowi4_oif = (rt_is_output_route(rt) ?
1919                                skb->dev->ifindex : 0),
1920                 .flowi4_iif = (rt_is_output_route(rt) ?
1921                                LOOPBACK_IFINDEX :
1922                                skb->dev->ifindex),
1923                 .flowi4_mark = skb->mark,
1924         };
1925         struct mr_table *mrt;
1926         int err;
1927
1928         err = ipmr_fib_lookup(net, &fl4, &mrt);
1929         if (err)
1930                 return ERR_PTR(err);
1931         return mrt;
1932 }
1933
1934 /* Multicast packets for forwarding arrive here
1935  * Called with rcu_read_lock();
1936  */
1937 int ip_mr_input(struct sk_buff *skb)
1938 {
1939         struct mfc_cache *cache;
1940         struct net *net = dev_net(skb->dev);
1941         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1942         struct mr_table *mrt;
1943
1944         /* Packet is looped back after forward, it should not be
1945          * forwarded second time, but still can be delivered locally.
1946          */
1947         if (IPCB(skb)->flags & IPSKB_FORWARDED)
1948                 goto dont_forward;
1949
1950         mrt = ipmr_rt_fib_lookup(net, skb);
1951         if (IS_ERR(mrt)) {
1952                 kfree_skb(skb);
1953                 return PTR_ERR(mrt);
1954         }
1955         if (!local) {
1956                 if (IPCB(skb)->opt.router_alert) {
1957                         if (ip_call_ra_chain(skb))
1958                                 return 0;
1959                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1960                         /* IGMPv1 (and broken IGMPv2 implementations sort of
1961                          * Cisco IOS <= 11.2(8)) do not put router alert
1962                          * option to IGMP packets destined to routable
1963                          * groups. It is very bad, because it means
1964                          * that we can forward NO IGMP messages.
1965                          */
1966                         struct sock *mroute_sk;
1967
1968                         mroute_sk = rcu_dereference(mrt->mroute_sk);
1969                         if (mroute_sk) {
1970                                 nf_reset(skb);
1971                                 raw_rcv(mroute_sk, skb);
1972                                 return 0;
1973                         }
1974                     }
1975         }
1976
1977         /* already under rcu_read_lock() */
1978         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1979         if (!cache) {
1980                 int vif = ipmr_find_vif(mrt, skb->dev);
1981
1982                 if (vif >= 0)
1983                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
1984                                                     vif);
1985         }
1986
1987         /* No usable cache entry */
1988         if (!cache) {
1989                 int vif;
1990
1991                 if (local) {
1992                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1993                         ip_local_deliver(skb);
1994                         if (!skb2)
1995                                 return -ENOBUFS;
1996                         skb = skb2;
1997                 }
1998
1999                 read_lock(&mrt_lock);
2000                 vif = ipmr_find_vif(mrt, skb->dev);
2001                 if (vif >= 0) {
2002                         int err2 = ipmr_cache_unresolved(mrt, vif, skb);
2003                         read_unlock(&mrt_lock);
2004
2005                         return err2;
2006                 }
2007                 read_unlock(&mrt_lock);
2008                 kfree_skb(skb);
2009                 return -ENODEV;
2010         }
2011
2012         read_lock(&mrt_lock);
2013         ip_mr_forward(net, mrt, skb, cache, local);
2014         read_unlock(&mrt_lock);
2015
2016         if (local)
2017                 return ip_local_deliver(skb);
2018
2019         return 0;
2020
2021 dont_forward:
2022         if (local)
2023                 return ip_local_deliver(skb);
2024         kfree_skb(skb);
2025         return 0;
2026 }
2027
2028 #ifdef CONFIG_IP_PIMSM_V1
2029 /* Handle IGMP messages of PIMv1 */
2030 int pim_rcv_v1(struct sk_buff *skb)
2031 {
2032         struct igmphdr *pim;
2033         struct net *net = dev_net(skb->dev);
2034         struct mr_table *mrt;
2035
2036         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2037                 goto drop;
2038
2039         pim = igmp_hdr(skb);
2040
2041         mrt = ipmr_rt_fib_lookup(net, skb);
2042         if (IS_ERR(mrt))
2043                 goto drop;
2044         if (!mrt->mroute_do_pim ||
2045             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2046                 goto drop;
2047
2048         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2049 drop:
2050                 kfree_skb(skb);
2051         }
2052         return 0;
2053 }
2054 #endif
2055
2056 #ifdef CONFIG_IP_PIMSM_V2
2057 static int pim_rcv(struct sk_buff *skb)
2058 {
2059         struct pimreghdr *pim;
2060         struct net *net = dev_net(skb->dev);
2061         struct mr_table *mrt;
2062
2063         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2064                 goto drop;
2065
2066         pim = (struct pimreghdr *)skb_transport_header(skb);
2067         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2068             (pim->flags & PIM_NULL_REGISTER) ||
2069             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2070              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2071                 goto drop;
2072
2073         mrt = ipmr_rt_fib_lookup(net, skb);
2074         if (IS_ERR(mrt))
2075                 goto drop;
2076         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2077 drop:
2078                 kfree_skb(skb);
2079         }
2080         return 0;
2081 }
2082 #endif
2083
2084 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2085                               struct mfc_cache *c, struct rtmsg *rtm)
2086 {
2087         struct rta_mfc_stats mfcs;
2088         struct nlattr *mp_attr;
2089         struct rtnexthop *nhp;
2090         unsigned long lastuse;
2091         int ct;
2092
2093         /* If cache is unresolved, don't try to parse IIF and OIF */
2094         if (c->mfc_parent >= MAXVIFS)
2095                 return -ENOENT;
2096
2097         if (VIF_EXISTS(mrt, c->mfc_parent) &&
2098             nla_put_u32(skb, RTA_IIF, mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
2099                 return -EMSGSIZE;
2100
2101         if (!(mp_attr = nla_nest_start(skb, RTA_MULTIPATH)))
2102                 return -EMSGSIZE;
2103
2104         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2105                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2106                         if (!(nhp = nla_reserve_nohdr(skb, sizeof(*nhp)))) {
2107                                 nla_nest_cancel(skb, mp_attr);
2108                                 return -EMSGSIZE;
2109                         }
2110
2111                         nhp->rtnh_flags = 0;
2112                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2113                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2114                         nhp->rtnh_len = sizeof(*nhp);
2115                 }
2116         }
2117
2118         nla_nest_end(skb, mp_attr);
2119
2120         lastuse = READ_ONCE(c->mfc_un.res.lastuse);
2121         lastuse = time_after_eq(jiffies, lastuse) ? jiffies - lastuse : 0;
2122
2123         mfcs.mfcs_packets = c->mfc_un.res.pkt;
2124         mfcs.mfcs_bytes = c->mfc_un.res.bytes;
2125         mfcs.mfcs_wrong_if = c->mfc_un.res.wrong_if;
2126         if (nla_put_64bit(skb, RTA_MFC_STATS, sizeof(mfcs), &mfcs, RTA_PAD) ||
2127             nla_put_u64_64bit(skb, RTA_EXPIRES, jiffies_to_clock_t(lastuse),
2128                               RTA_PAD))
2129                 return -EMSGSIZE;
2130
2131         rtm->rtm_type = RTN_MULTICAST;
2132         return 1;
2133 }
2134
2135 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2136                    __be32 saddr, __be32 daddr,
2137                    struct rtmsg *rtm, int nowait, u32 portid)
2138 {
2139         struct mfc_cache *cache;
2140         struct mr_table *mrt;
2141         int err;
2142
2143         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2144         if (!mrt)
2145                 return -ENOENT;
2146
2147         rcu_read_lock();
2148         cache = ipmr_cache_find(mrt, saddr, daddr);
2149         if (!cache && skb->dev) {
2150                 int vif = ipmr_find_vif(mrt, skb->dev);
2151
2152                 if (vif >= 0)
2153                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2154         }
2155         if (!cache) {
2156                 struct sk_buff *skb2;
2157                 struct iphdr *iph;
2158                 struct net_device *dev;
2159                 int vif = -1;
2160
2161                 if (nowait) {
2162                         rcu_read_unlock();
2163                         return -EAGAIN;
2164                 }
2165
2166                 dev = skb->dev;
2167                 read_lock(&mrt_lock);
2168                 if (dev)
2169                         vif = ipmr_find_vif(mrt, dev);
2170                 if (vif < 0) {
2171                         read_unlock(&mrt_lock);
2172                         rcu_read_unlock();
2173                         return -ENODEV;
2174                 }
2175                 skb2 = skb_clone(skb, GFP_ATOMIC);
2176                 if (!skb2) {
2177                         read_unlock(&mrt_lock);
2178                         rcu_read_unlock();
2179                         return -ENOMEM;
2180                 }
2181
2182                 NETLINK_CB(skb2).portid = portid;
2183                 skb_push(skb2, sizeof(struct iphdr));
2184                 skb_reset_network_header(skb2);
2185                 iph = ip_hdr(skb2);
2186                 iph->ihl = sizeof(struct iphdr) >> 2;
2187                 iph->saddr = saddr;
2188                 iph->daddr = daddr;
2189                 iph->version = 0;
2190                 err = ipmr_cache_unresolved(mrt, vif, skb2);
2191                 read_unlock(&mrt_lock);
2192                 rcu_read_unlock();
2193                 return err;
2194         }
2195
2196         read_lock(&mrt_lock);
2197         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2198         read_unlock(&mrt_lock);
2199         rcu_read_unlock();
2200         return err;
2201 }
2202
2203 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2204                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2205                             int flags)
2206 {
2207         struct nlmsghdr *nlh;
2208         struct rtmsg *rtm;
2209         int err;
2210
2211         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2212         if (!nlh)
2213                 return -EMSGSIZE;
2214
2215         rtm = nlmsg_data(nlh);
2216         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2217         rtm->rtm_dst_len  = 32;
2218         rtm->rtm_src_len  = 32;
2219         rtm->rtm_tos      = 0;
2220         rtm->rtm_table    = mrt->id;
2221         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2222                 goto nla_put_failure;
2223         rtm->rtm_type     = RTN_MULTICAST;
2224         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2225         if (c->mfc_flags & MFC_STATIC)
2226                 rtm->rtm_protocol = RTPROT_STATIC;
2227         else
2228                 rtm->rtm_protocol = RTPROT_MROUTED;
2229         rtm->rtm_flags    = 0;
2230
2231         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2232             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2233                 goto nla_put_failure;
2234         err = __ipmr_fill_mroute(mrt, skb, c, rtm);
2235         /* do not break the dump if cache is unresolved */
2236         if (err < 0 && err != -ENOENT)
2237                 goto nla_put_failure;
2238
2239         nlmsg_end(skb, nlh);
2240         return 0;
2241
2242 nla_put_failure:
2243         nlmsg_cancel(skb, nlh);
2244         return -EMSGSIZE;
2245 }
2246
2247 static size_t mroute_msgsize(bool unresolved, int maxvif)
2248 {
2249         size_t len =
2250                 NLMSG_ALIGN(sizeof(struct rtmsg))
2251                 + nla_total_size(4)     /* RTA_TABLE */
2252                 + nla_total_size(4)     /* RTA_SRC */
2253                 + nla_total_size(4)     /* RTA_DST */
2254                 ;
2255
2256         if (!unresolved)
2257                 len = len
2258                       + nla_total_size(4)       /* RTA_IIF */
2259                       + nla_total_size(0)       /* RTA_MULTIPATH */
2260                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2261                                                 /* RTA_MFC_STATS */
2262                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2263                 ;
2264
2265         return len;
2266 }
2267
2268 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2269                                  int cmd)
2270 {
2271         struct net *net = read_pnet(&mrt->net);
2272         struct sk_buff *skb;
2273         int err = -ENOBUFS;
2274
2275         skb = nlmsg_new(mroute_msgsize(mfc->mfc_parent >= MAXVIFS, mrt->maxvif),
2276                         GFP_ATOMIC);
2277         if (!skb)
2278                 goto errout;
2279
2280         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2281         if (err < 0)
2282                 goto errout;
2283
2284         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2285         return;
2286
2287 errout:
2288         kfree_skb(skb);
2289         if (err < 0)
2290                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2291 }
2292
2293 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2294 {
2295         struct net *net = sock_net(skb->sk);
2296         struct mr_table *mrt;
2297         struct mfc_cache *mfc;
2298         unsigned int t = 0, s_t;
2299         unsigned int h = 0, s_h;
2300         unsigned int e = 0, s_e;
2301
2302         s_t = cb->args[0];
2303         s_h = cb->args[1];
2304         s_e = cb->args[2];
2305
2306         rcu_read_lock();
2307         ipmr_for_each_table(mrt, net) {
2308                 if (t < s_t)
2309                         goto next_table;
2310                 if (t > s_t)
2311                         s_h = 0;
2312                 for (h = s_h; h < MFC_LINES; h++) {
2313                         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2314                                 if (e < s_e)
2315                                         goto next_entry;
2316                                 if (ipmr_fill_mroute(mrt, skb,
2317                                                      NETLINK_CB(cb->skb).portid,
2318                                                      cb->nlh->nlmsg_seq,
2319                                                      mfc, RTM_NEWROUTE,
2320                                                      NLM_F_MULTI) < 0)
2321                                         goto done;
2322 next_entry:
2323                                 e++;
2324                         }
2325                         e = s_e = 0;
2326                 }
2327                 spin_lock_bh(&mfc_unres_lock);
2328                 list_for_each_entry(mfc, &mrt->mfc_unres_queue, list) {
2329                         if (e < s_e)
2330                                 goto next_entry2;
2331                         if (ipmr_fill_mroute(mrt, skb,
2332                                              NETLINK_CB(cb->skb).portid,
2333                                              cb->nlh->nlmsg_seq,
2334                                              mfc, RTM_NEWROUTE,
2335                                              NLM_F_MULTI) < 0) {
2336                                 spin_unlock_bh(&mfc_unres_lock);
2337                                 goto done;
2338                         }
2339 next_entry2:
2340                         e++;
2341                 }
2342                 spin_unlock_bh(&mfc_unres_lock);
2343                 e = s_e = 0;
2344                 s_h = 0;
2345 next_table:
2346                 t++;
2347         }
2348 done:
2349         rcu_read_unlock();
2350
2351         cb->args[2] = e;
2352         cb->args[1] = h;
2353         cb->args[0] = t;
2354
2355         return skb->len;
2356 }
2357
2358 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2359         [RTA_SRC]       = { .type = NLA_U32 },
2360         [RTA_DST]       = { .type = NLA_U32 },
2361         [RTA_IIF]       = { .type = NLA_U32 },
2362         [RTA_TABLE]     = { .type = NLA_U32 },
2363         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2364 };
2365
2366 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2367 {
2368         switch (rtm_protocol) {
2369         case RTPROT_STATIC:
2370         case RTPROT_MROUTED:
2371                 return true;
2372         }
2373         return false;
2374 }
2375
2376 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2377 {
2378         struct rtnexthop *rtnh = nla_data(nla);
2379         int remaining = nla_len(nla), vifi = 0;
2380
2381         while (rtnh_ok(rtnh, remaining)) {
2382                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2383                 if (++vifi == MAXVIFS)
2384                         break;
2385                 rtnh = rtnh_next(rtnh, &remaining);
2386         }
2387
2388         return remaining > 0 ? -EINVAL : vifi;
2389 }
2390
2391 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2392 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2393                             struct mfcctl *mfcc, int *mrtsock,
2394                             struct mr_table **mrtret)
2395 {
2396         struct net_device *dev = NULL;
2397         u32 tblid = RT_TABLE_DEFAULT;
2398         struct mr_table *mrt;
2399         struct nlattr *attr;
2400         struct rtmsg *rtm;
2401         int ret, rem;
2402
2403         ret = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipmr_policy);
2404         if (ret < 0)
2405                 goto out;
2406         rtm = nlmsg_data(nlh);
2407
2408         ret = -EINVAL;
2409         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2410             rtm->rtm_type != RTN_MULTICAST ||
2411             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2412             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2413                 goto out;
2414
2415         memset(mfcc, 0, sizeof(*mfcc));
2416         mfcc->mfcc_parent = -1;
2417         ret = 0;
2418         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2419                 switch (nla_type(attr)) {
2420                 case RTA_SRC:
2421                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2422                         break;
2423                 case RTA_DST:
2424                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2425                         break;
2426                 case RTA_IIF:
2427                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2428                         if (!dev) {
2429                                 ret = -ENODEV;
2430                                 goto out;
2431                         }
2432                         break;
2433                 case RTA_MULTIPATH:
2434                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2435                                 ret = -EINVAL;
2436                                 goto out;
2437                         }
2438                         break;
2439                 case RTA_PREFSRC:
2440                         ret = 1;
2441                         break;
2442                 case RTA_TABLE:
2443                         tblid = nla_get_u32(attr);
2444                         break;
2445                 }
2446         }
2447         mrt = ipmr_get_table(net, tblid);
2448         if (!mrt) {
2449                 ret = -ENOENT;
2450                 goto out;
2451         }
2452         *mrtret = mrt;
2453         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2454         if (dev)
2455                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2456
2457 out:
2458         return ret;
2459 }
2460
2461 /* takes care of both newroute and delroute */
2462 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh)
2463 {
2464         struct net *net = sock_net(skb->sk);
2465         int ret, mrtsock, parent;
2466         struct mr_table *tbl;
2467         struct mfcctl mfcc;
2468
2469         mrtsock = 0;
2470         tbl = NULL;
2471         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl);
2472         if (ret < 0)
2473                 return ret;
2474
2475         parent = ret ? mfcc.mfcc_parent : -1;
2476         if (nlh->nlmsg_type == RTM_NEWROUTE)
2477                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2478         else
2479                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2480 }
2481
2482 #ifdef CONFIG_PROC_FS
2483 /* The /proc interfaces to multicast routing :
2484  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2485  */
2486 struct ipmr_vif_iter {
2487         struct seq_net_private p;
2488         struct mr_table *mrt;
2489         int ct;
2490 };
2491
2492 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2493                                            struct ipmr_vif_iter *iter,
2494                                            loff_t pos)
2495 {
2496         struct mr_table *mrt = iter->mrt;
2497
2498         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2499                 if (!VIF_EXISTS(mrt, iter->ct))
2500                         continue;
2501                 if (pos-- == 0)
2502                         return &mrt->vif_table[iter->ct];
2503         }
2504         return NULL;
2505 }
2506
2507 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2508         __acquires(mrt_lock)
2509 {
2510         struct ipmr_vif_iter *iter = seq->private;
2511         struct net *net = seq_file_net(seq);
2512         struct mr_table *mrt;
2513
2514         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2515         if (!mrt)
2516                 return ERR_PTR(-ENOENT);
2517
2518         iter->mrt = mrt;
2519
2520         read_lock(&mrt_lock);
2521         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2522                 : SEQ_START_TOKEN;
2523 }
2524
2525 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2526 {
2527         struct ipmr_vif_iter *iter = seq->private;
2528         struct net *net = seq_file_net(seq);
2529         struct mr_table *mrt = iter->mrt;
2530
2531         ++*pos;
2532         if (v == SEQ_START_TOKEN)
2533                 return ipmr_vif_seq_idx(net, iter, 0);
2534
2535         while (++iter->ct < mrt->maxvif) {
2536                 if (!VIF_EXISTS(mrt, iter->ct))
2537                         continue;
2538                 return &mrt->vif_table[iter->ct];
2539         }
2540         return NULL;
2541 }
2542
2543 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2544         __releases(mrt_lock)
2545 {
2546         read_unlock(&mrt_lock);
2547 }
2548
2549 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2550 {
2551         struct ipmr_vif_iter *iter = seq->private;
2552         struct mr_table *mrt = iter->mrt;
2553
2554         if (v == SEQ_START_TOKEN) {
2555                 seq_puts(seq,
2556                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2557         } else {
2558                 const struct vif_device *vif = v;
2559                 const char *name =  vif->dev ? vif->dev->name : "none";
2560
2561                 seq_printf(seq,
2562                            "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2563                            vif - mrt->vif_table,
2564                            name, vif->bytes_in, vif->pkt_in,
2565                            vif->bytes_out, vif->pkt_out,
2566                            vif->flags, vif->local, vif->remote);
2567         }
2568         return 0;
2569 }
2570
2571 static const struct seq_operations ipmr_vif_seq_ops = {
2572         .start = ipmr_vif_seq_start,
2573         .next  = ipmr_vif_seq_next,
2574         .stop  = ipmr_vif_seq_stop,
2575         .show  = ipmr_vif_seq_show,
2576 };
2577
2578 static int ipmr_vif_open(struct inode *inode, struct file *file)
2579 {
2580         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2581                             sizeof(struct ipmr_vif_iter));
2582 }
2583
2584 static const struct file_operations ipmr_vif_fops = {
2585         .owner   = THIS_MODULE,
2586         .open    = ipmr_vif_open,
2587         .read    = seq_read,
2588         .llseek  = seq_lseek,
2589         .release = seq_release_net,
2590 };
2591
2592 struct ipmr_mfc_iter {
2593         struct seq_net_private p;
2594         struct mr_table *mrt;
2595         struct list_head *cache;
2596         int ct;
2597 };
2598
2599
2600 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2601                                           struct ipmr_mfc_iter *it, loff_t pos)
2602 {
2603         struct mr_table *mrt = it->mrt;
2604         struct mfc_cache *mfc;
2605
2606         rcu_read_lock();
2607         for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2608                 it->cache = &mrt->mfc_cache_array[it->ct];
2609                 list_for_each_entry_rcu(mfc, it->cache, list)
2610                         if (pos-- == 0)
2611                                 return mfc;
2612         }
2613         rcu_read_unlock();
2614
2615         spin_lock_bh(&mfc_unres_lock);
2616         it->cache = &mrt->mfc_unres_queue;
2617         list_for_each_entry(mfc, it->cache, list)
2618                 if (pos-- == 0)
2619                         return mfc;
2620         spin_unlock_bh(&mfc_unres_lock);
2621
2622         it->cache = NULL;
2623         return NULL;
2624 }
2625
2626
2627 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2628 {
2629         struct ipmr_mfc_iter *it = seq->private;
2630         struct net *net = seq_file_net(seq);
2631         struct mr_table *mrt;
2632
2633         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2634         if (!mrt)
2635                 return ERR_PTR(-ENOENT);
2636
2637         it->mrt = mrt;
2638         it->cache = NULL;
2639         it->ct = 0;
2640         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2641                 : SEQ_START_TOKEN;
2642 }
2643
2644 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2645 {
2646         struct mfc_cache *mfc = v;
2647         struct ipmr_mfc_iter *it = seq->private;
2648         struct net *net = seq_file_net(seq);
2649         struct mr_table *mrt = it->mrt;
2650
2651         ++*pos;
2652
2653         if (v == SEQ_START_TOKEN)
2654                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2655
2656         if (mfc->list.next != it->cache)
2657                 return list_entry(mfc->list.next, struct mfc_cache, list);
2658
2659         if (it->cache == &mrt->mfc_unres_queue)
2660                 goto end_of_list;
2661
2662         BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2663
2664         while (++it->ct < MFC_LINES) {
2665                 it->cache = &mrt->mfc_cache_array[it->ct];
2666                 if (list_empty(it->cache))
2667                         continue;
2668                 return list_first_entry(it->cache, struct mfc_cache, list);
2669         }
2670
2671         /* exhausted cache_array, show unresolved */
2672         rcu_read_unlock();
2673         it->cache = &mrt->mfc_unres_queue;
2674         it->ct = 0;
2675
2676         spin_lock_bh(&mfc_unres_lock);
2677         if (!list_empty(it->cache))
2678                 return list_first_entry(it->cache, struct mfc_cache, list);
2679
2680 end_of_list:
2681         spin_unlock_bh(&mfc_unres_lock);
2682         it->cache = NULL;
2683
2684         return NULL;
2685 }
2686
2687 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2688 {
2689         struct ipmr_mfc_iter *it = seq->private;
2690         struct mr_table *mrt = it->mrt;
2691
2692         if (it->cache == &mrt->mfc_unres_queue)
2693                 spin_unlock_bh(&mfc_unres_lock);
2694         else if (it->cache == &mrt->mfc_cache_array[it->ct])
2695                 rcu_read_unlock();
2696 }
2697
2698 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2699 {
2700         int n;
2701
2702         if (v == SEQ_START_TOKEN) {
2703                 seq_puts(seq,
2704                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2705         } else {
2706                 const struct mfc_cache *mfc = v;
2707                 const struct ipmr_mfc_iter *it = seq->private;
2708                 const struct mr_table *mrt = it->mrt;
2709
2710                 seq_printf(seq, "%08X %08X %-3hd",
2711                            (__force u32) mfc->mfc_mcastgrp,
2712                            (__force u32) mfc->mfc_origin,
2713                            mfc->mfc_parent);
2714
2715                 if (it->cache != &mrt->mfc_unres_queue) {
2716                         seq_printf(seq, " %8lu %8lu %8lu",
2717                                    mfc->mfc_un.res.pkt,
2718                                    mfc->mfc_un.res.bytes,
2719                                    mfc->mfc_un.res.wrong_if);
2720                         for (n = mfc->mfc_un.res.minvif;
2721                              n < mfc->mfc_un.res.maxvif; n++) {
2722                                 if (VIF_EXISTS(mrt, n) &&
2723                                     mfc->mfc_un.res.ttls[n] < 255)
2724                                         seq_printf(seq,
2725                                            " %2d:%-3d",
2726                                            n, mfc->mfc_un.res.ttls[n]);
2727                         }
2728                 } else {
2729                         /* unresolved mfc_caches don't contain
2730                          * pkt, bytes and wrong_if values
2731                          */
2732                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2733                 }
2734                 seq_putc(seq, '\n');
2735         }
2736         return 0;
2737 }
2738
2739 static const struct seq_operations ipmr_mfc_seq_ops = {
2740         .start = ipmr_mfc_seq_start,
2741         .next  = ipmr_mfc_seq_next,
2742         .stop  = ipmr_mfc_seq_stop,
2743         .show  = ipmr_mfc_seq_show,
2744 };
2745
2746 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2747 {
2748         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2749                             sizeof(struct ipmr_mfc_iter));
2750 }
2751
2752 static const struct file_operations ipmr_mfc_fops = {
2753         .owner   = THIS_MODULE,
2754         .open    = ipmr_mfc_open,
2755         .read    = seq_read,
2756         .llseek  = seq_lseek,
2757         .release = seq_release_net,
2758 };
2759 #endif
2760
2761 #ifdef CONFIG_IP_PIMSM_V2
2762 static const struct net_protocol pim_protocol = {
2763         .handler        =       pim_rcv,
2764         .netns_ok       =       1,
2765 };
2766 #endif
2767
2768 /* Setup for IP multicast routing */
2769 static int __net_init ipmr_net_init(struct net *net)
2770 {
2771         int err;
2772
2773         err = ipmr_rules_init(net);
2774         if (err < 0)
2775                 goto fail;
2776
2777 #ifdef CONFIG_PROC_FS
2778         err = -ENOMEM;
2779         if (!proc_create("ip_mr_vif", 0, net->proc_net, &ipmr_vif_fops))
2780                 goto proc_vif_fail;
2781         if (!proc_create("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_fops))
2782                 goto proc_cache_fail;
2783 #endif
2784         return 0;
2785
2786 #ifdef CONFIG_PROC_FS
2787 proc_cache_fail:
2788         remove_proc_entry("ip_mr_vif", net->proc_net);
2789 proc_vif_fail:
2790         ipmr_rules_exit(net);
2791 #endif
2792 fail:
2793         return err;
2794 }
2795
2796 static void __net_exit ipmr_net_exit(struct net *net)
2797 {
2798 #ifdef CONFIG_PROC_FS
2799         remove_proc_entry("ip_mr_cache", net->proc_net);
2800         remove_proc_entry("ip_mr_vif", net->proc_net);
2801 #endif
2802         ipmr_rules_exit(net);
2803 }
2804
2805 static struct pernet_operations ipmr_net_ops = {
2806         .init = ipmr_net_init,
2807         .exit = ipmr_net_exit,
2808 };
2809
2810 int __init ip_mr_init(void)
2811 {
2812         int err;
2813
2814         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2815                                        sizeof(struct mfc_cache),
2816                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2817                                        NULL);
2818
2819         err = register_pernet_subsys(&ipmr_net_ops);
2820         if (err)
2821                 goto reg_pernet_fail;
2822
2823         err = register_netdevice_notifier(&ip_mr_notifier);
2824         if (err)
2825                 goto reg_notif_fail;
2826 #ifdef CONFIG_IP_PIMSM_V2
2827         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2828                 pr_err("%s: can't add PIM protocol\n", __func__);
2829                 err = -EAGAIN;
2830                 goto add_proto_fail;
2831         }
2832 #endif
2833         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
2834                       NULL, ipmr_rtm_dumproute, NULL);
2835         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
2836                       ipmr_rtm_route, NULL, NULL);
2837         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
2838                       ipmr_rtm_route, NULL, NULL);
2839         return 0;
2840
2841 #ifdef CONFIG_IP_PIMSM_V2
2842 add_proto_fail:
2843         unregister_netdevice_notifier(&ip_mr_notifier);
2844 #endif
2845 reg_notif_fail:
2846         unregister_pernet_subsys(&ipmr_net_ops);
2847 reg_pernet_fail:
2848         kmem_cache_destroy(mrt_cachep);
2849         return err;
2850 }