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 <linux/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 inline int ipmr_hash_cmp(struct rhashtable_compare_arg *arg,
 303                                 const void *ptr)
 304 {
 305         const struct mfc_cache_cmp_arg *cmparg = arg->key;
 306         struct mfc_cache *c = (struct mfc_cache *)ptr;
 307
 308         return cmparg->mfc_mcastgrp != c->mfc_mcastgrp ||
 309                cmparg->mfc_origin != c->mfc_origin;
 310 }
 311
 312 static const struct rhashtable_params ipmr_rht_params = {
 313         .head_offset = offsetof(struct mfc_cache, mnode),
 314         .key_offset = offsetof(struct mfc_cache, cmparg),
 315         .key_len = sizeof(struct mfc_cache_cmp_arg),
 316         .nelem_hint = 3,
 317         .locks_mul = 1,
 318         .obj_cmpfn = ipmr_hash_cmp,
 319         .automatic_shrinking = true,
 320 };
 321
 322 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 323 {
 324         struct mr_table *mrt;
 325
 326         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 327         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 328                 return ERR_PTR(-EINVAL);
 329
 330         mrt = ipmr_get_table(net, id);
 331         if (mrt)
 332                 return mrt;
 333
 334         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
 335         if (!mrt)
 336                 return ERR_PTR(-ENOMEM);
 337         write_pnet(&mrt->net, net);
 338         mrt->id = id;
 339
 340         rhltable_init(&mrt->mfc_hash, &ipmr_rht_params);
 341         INIT_LIST_HEAD(&mrt->mfc_cache_list);
 342         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
 343
 344         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
 345                     (unsigned long)mrt);
 346
 347         mrt->mroute_reg_vif_num = -1;
 348 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 349         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 350 #endif
 351         return mrt;
 352 }
 353
 354 static void ipmr_free_table(struct mr_table *mrt)
 355 {
 356         del_timer_sync(&mrt->ipmr_expire_timer);
 357         mroute_clean_tables(mrt, true);
 358         rhltable_destroy(&mrt->mfc_hash);
 359         kfree(mrt);
 360 }
 361
 362 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 363
 364 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 365 {
 366         struct net *net = dev_net(dev);
 367
 368         dev_close(dev);
 369
 370         dev = __dev_get_by_name(net, "tunl0");
 371         if (dev) {
 372                 const struct net_device_ops *ops = dev->netdev_ops;
 373                 struct ifreq ifr;
 374                 struct ip_tunnel_parm p;
 375
 376                 memset(&p, 0, sizeof(p));
 377                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 378                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 379                 p.iph.version = 4;
 380                 p.iph.ihl = 5;
 381                 p.iph.protocol = IPPROTO_IPIP;
 382                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 383                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 384
 385                 if (ops->ndo_do_ioctl) {
 386                         mm_segment_t oldfs = get_fs();
 387
 388                         set_fs(KERNEL_DS);
 389                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 390                         set_fs(oldfs);
 391                 }
 392         }
 393 }
 394
 395 /* Initialize ipmr pimreg/tunnel in_device */
 396 static bool ipmr_init_vif_indev(const struct net_device *dev)
 397 {
 398         struct in_device *in_dev;
 399
 400         ASSERT_RTNL();
 401
 402         in_dev = __in_dev_get_rtnl(dev);
 403         if (!in_dev)
 404                 return false;
 405         ipv4_devconf_setall(in_dev);
 406         neigh_parms_data_state_setall(in_dev->arp_parms);
 407         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 408
 409         return true;
 410 }
 411
 412 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 413 {
 414         struct net_device  *dev;
 415
 416         dev = __dev_get_by_name(net, "tunl0");
 417
 418         if (dev) {
 419                 const struct net_device_ops *ops = dev->netdev_ops;
 420                 int err;
 421                 struct ifreq ifr;
 422                 struct ip_tunnel_parm p;
 423
 424                 memset(&p, 0, sizeof(p));
 425                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 426                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 427                 p.iph.version = 4;
 428                 p.iph.ihl = 5;
 429                 p.iph.protocol = IPPROTO_IPIP;
 430                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 431                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 432
 433                 if (ops->ndo_do_ioctl) {
 434                         mm_segment_t oldfs = get_fs();
 435
 436                         set_fs(KERNEL_DS);
 437                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 438                         set_fs(oldfs);
 439                 } else {
 440                         err = -EOPNOTSUPP;
 441                 }
 442                 dev = NULL;
 443
 444                 if (err == 0 &&
 445                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 446                         dev->flags |= IFF_MULTICAST;
 447                         if (!ipmr_init_vif_indev(dev))
 448                                 goto failure;
 449                         if (dev_open(dev))
 450                                 goto failure;
 451                         dev_hold(dev);
 452                 }
 453         }
 454         return dev;
 455
 456 failure:
 457         unregister_netdevice(dev);
 458         return NULL;
 459 }
 460
 461 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 462 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 463 {
 464         struct net *net = dev_net(dev);
 465         struct mr_table *mrt;
 466         struct flowi4 fl4 = {
 467                 .flowi4_oif     = dev->ifindex,
 468                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 469                 .flowi4_mark    = skb->mark,
 470         };
 471         int err;
 472
 473         err = ipmr_fib_lookup(net, &fl4, &mrt);
 474         if (err < 0) {
 475                 kfree_skb(skb);
 476                 return err;
 477         }
 478
 479         read_lock(&mrt_lock);
 480         dev->stats.tx_bytes += skb->len;
 481         dev->stats.tx_packets++;
 482         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 483         read_unlock(&mrt_lock);
 484         kfree_skb(skb);
 485         return NETDEV_TX_OK;
 486 }
 487
 488 static int reg_vif_get_iflink(const struct net_device *dev)
 489 {
 490         return 0;
 491 }
 492
 493 static const struct net_device_ops reg_vif_netdev_ops = {
 494         .ndo_start_xmit = reg_vif_xmit,
 495         .ndo_get_iflink = reg_vif_get_iflink,
 496 };
 497
 498 static void reg_vif_setup(struct net_device *dev)
 499 {
 500         dev->type               = ARPHRD_PIMREG;
 501         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 502         dev->flags              = IFF_NOARP;
 503         dev->netdev_ops         = &reg_vif_netdev_ops;
 504         dev->destructor         = free_netdev;
 505         dev->features           |= NETIF_F_NETNS_LOCAL;
 506 }
 507
 508 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 509 {
 510         struct net_device *dev;
 511         char name[IFNAMSIZ];
 512
 513         if (mrt->id == RT_TABLE_DEFAULT)
 514                 sprintf(name, "pimreg");
 515         else
 516                 sprintf(name, "pimreg%u", mrt->id);
 517
 518         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 519
 520         if (!dev)
 521                 return NULL;
 522
 523         dev_net_set(dev, net);
 524
 525         if (register_netdevice(dev)) {
 526                 free_netdev(dev);
 527                 return NULL;
 528         }
 529
 530         if (!ipmr_init_vif_indev(dev))
 531                 goto failure;
 532         if (dev_open(dev))
 533                 goto failure;
 534
 535         dev_hold(dev);
 536
 537         return dev;
 538
 539 failure:
 540         unregister_netdevice(dev);
 541         return NULL;
 542 }
 543
 544 /* called with rcu_read_lock() */
 545 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 546                      unsigned int pimlen)
 547 {
 548         struct net_device *reg_dev = NULL;
 549         struct iphdr *encap;
 550
 551         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 552         /* Check that:
 553          * a. packet is really sent to a multicast group
 554          * b. packet is not a NULL-REGISTER
 555          * c. packet is not truncated
 556          */
 557         if (!ipv4_is_multicast(encap->daddr) ||
 558             encap->tot_len == 0 ||
 559             ntohs(encap->tot_len) + pimlen > skb->len)
 560                 return 1;
 561
 562         read_lock(&mrt_lock);
 563         if (mrt->mroute_reg_vif_num >= 0)
 564                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 565         read_unlock(&mrt_lock);
 566
 567         if (!reg_dev)
 568                 return 1;
 569
 570         skb->mac_header = skb->network_header;
 571         skb_pull(skb, (u8 *)encap - skb->data);
 572         skb_reset_network_header(skb);
 573         skb->protocol = htons(ETH_P_IP);
 574         skb->ip_summed = CHECKSUM_NONE;
 575
 576         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 577
 578         netif_rx(skb);
 579
 580         return NET_RX_SUCCESS;
 581 }
 582 #else
 583 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 584 {
 585         return NULL;
 586 }
 587 #endif
 588
 589 /**
 590  *      vif_delete - Delete a VIF entry
 591  *      @notify: Set to 1, if the caller is a notifier_call
 592  */
 593 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 594                       struct list_head *head)
 595 {
 596         struct vif_device *v;
 597         struct net_device *dev;
 598         struct in_device *in_dev;
 599
 600         if (vifi < 0 || vifi >= mrt->maxvif)
 601                 return -EADDRNOTAVAIL;
 602
 603         v = &mrt->vif_table[vifi];
 604
 605         write_lock_bh(&mrt_lock);
 606         dev = v->dev;
 607         v->dev = NULL;
 608
 609         if (!dev) {
 610                 write_unlock_bh(&mrt_lock);
 611                 return -EADDRNOTAVAIL;
 612         }
 613
 614         if (vifi == mrt->mroute_reg_vif_num)
 615                 mrt->mroute_reg_vif_num = -1;
 616
 617         if (vifi + 1 == mrt->maxvif) {
 618                 int tmp;
 619
 620                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 621                         if (VIF_EXISTS(mrt, tmp))
 622                                 break;
 623                 }
 624                 mrt->maxvif = tmp+1;
 625         }
 626
 627         write_unlock_bh(&mrt_lock);
 628
 629         dev_set_allmulti(dev, -1);
 630
 631         in_dev = __in_dev_get_rtnl(dev);
 632         if (in_dev) {
 633                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 634                 inet_netconf_notify_devconf(dev_net(dev), RTM_NEWNETCONF,
 635                                             NETCONFA_MC_FORWARDING,
 636                                             dev->ifindex, &in_dev->cnf);
 637                 ip_rt_multicast_event(in_dev);
 638         }
 639
 640         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 641                 unregister_netdevice_queue(dev, head);
 642
 643         dev_put(dev);
 644         return 0;
 645 }
 646
 647 static void ipmr_cache_free_rcu(struct rcu_head *head)
 648 {
 649         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
 650
 651         kmem_cache_free(mrt_cachep, c);
 652 }
 653
 654 static inline void ipmr_cache_free(struct mfc_cache *c)
 655 {
 656         call_rcu(&c->rcu, ipmr_cache_free_rcu);
 657 }
 658
 659 /* Destroy an unresolved cache entry, killing queued skbs
 660  * and reporting error to netlink readers.
 661  */
 662 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 663 {
 664         struct net *net = read_pnet(&mrt->net);
 665         struct sk_buff *skb;
 666         struct nlmsgerr *e;
 667
 668         atomic_dec(&mrt->cache_resolve_queue_len);
 669
 670         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
 671                 if (ip_hdr(skb)->version == 0) {
 672                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 673                         nlh->nlmsg_type = NLMSG_ERROR;
 674                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 675                         skb_trim(skb, nlh->nlmsg_len);
 676                         e = nlmsg_data(nlh);
 677                         e->error = -ETIMEDOUT;
 678                         memset(&e->msg, 0, sizeof(e->msg));
 679
 680                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 681                 } else {
 682                         kfree_skb(skb);
 683                 }
 684         }
 685
 686         ipmr_cache_free(c);
 687 }
 688
 689 /* Timer process for the unresolved queue. */
 690 static void ipmr_expire_process(unsigned long arg)
 691 {
 692         struct mr_table *mrt = (struct mr_table *)arg;
 693         unsigned long now;
 694         unsigned long expires;
 695         struct mfc_cache *c, *next;
 696
 697         if (!spin_trylock(&mfc_unres_lock)) {
 698                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 699                 return;
 700         }
 701
 702         if (list_empty(&mrt->mfc_unres_queue))
 703                 goto out;
 704
 705         now = jiffies;
 706         expires = 10*HZ;
 707
 708         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 709                 if (time_after(c->mfc_un.unres.expires, now)) {
 710                         unsigned long interval = c->mfc_un.unres.expires - now;
 711                         if (interval < expires)
 712                                 expires = interval;
 713                         continue;
 714                 }
 715
 716                 list_del(&c->list);
 717                 mroute_netlink_event(mrt, c, RTM_DELROUTE);
 718                 ipmr_destroy_unres(mrt, c);
 719         }
 720
 721         if (!list_empty(&mrt->mfc_unres_queue))
 722                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 723
 724 out:
 725         spin_unlock(&mfc_unres_lock);
 726 }
 727
 728 /* Fill oifs list. It is called under write locked mrt_lock. */
 729 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
 730                                    unsigned char *ttls)
 731 {
 732         int vifi;
 733
 734         cache->mfc_un.res.minvif = MAXVIFS;
 735         cache->mfc_un.res.maxvif = 0;
 736         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 737
 738         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 739                 if (VIF_EXISTS(mrt, vifi) &&
 740                     ttls[vifi] && ttls[vifi] < 255) {
 741                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 742                         if (cache->mfc_un.res.minvif > vifi)
 743                                 cache->mfc_un.res.minvif = vifi;
 744                         if (cache->mfc_un.res.maxvif <= vifi)
 745                                 cache->mfc_un.res.maxvif = vifi + 1;
 746                 }
 747         }
 748         cache->mfc_un.res.lastuse = jiffies;
 749 }
 750
 751 static int vif_add(struct net *net, struct mr_table *mrt,
 752                    struct vifctl *vifc, int mrtsock)
 753 {
 754         int vifi = vifc->vifc_vifi;
 755         struct vif_device *v = &mrt->vif_table[vifi];
 756         struct net_device *dev;
 757         struct in_device *in_dev;
 758         int err;
 759
 760         /* Is vif busy ? */
 761         if (VIF_EXISTS(mrt, vifi))
 762                 return -EADDRINUSE;
 763
 764         switch (vifc->vifc_flags) {
 765         case VIFF_REGISTER:
 766                 if (!ipmr_pimsm_enabled())
 767                         return -EINVAL;
 768                 /* Special Purpose VIF in PIM
 769                  * All the packets will be sent to the daemon
 770                  */
 771                 if (mrt->mroute_reg_vif_num >= 0)
 772                         return -EADDRINUSE;
 773                 dev = ipmr_reg_vif(net, mrt);
 774                 if (!dev)
 775                         return -ENOBUFS;
 776                 err = dev_set_allmulti(dev, 1);
 777                 if (err) {
 778                         unregister_netdevice(dev);
 779                         dev_put(dev);
 780                         return err;
 781                 }
 782                 break;
 783         case VIFF_TUNNEL:
 784                 dev = ipmr_new_tunnel(net, vifc);
 785                 if (!dev)
 786                         return -ENOBUFS;
 787                 err = dev_set_allmulti(dev, 1);
 788                 if (err) {
 789                         ipmr_del_tunnel(dev, vifc);
 790                         dev_put(dev);
 791                         return err;
 792                 }
 793                 break;
 794         case VIFF_USE_IFINDEX:
 795         case 0:
 796                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 797                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 798                         if (dev && !__in_dev_get_rtnl(dev)) {
 799                                 dev_put(dev);
 800                                 return -EADDRNOTAVAIL;
 801                         }
 802                 } else {
 803                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 804                 }
 805                 if (!dev)
 806                         return -EADDRNOTAVAIL;
 807                 err = dev_set_allmulti(dev, 1);
 808                 if (err) {
 809                         dev_put(dev);
 810                         return err;
 811                 }
 812                 break;
 813         default:
 814                 return -EINVAL;
 815         }
 816
 817         in_dev = __in_dev_get_rtnl(dev);
 818         if (!in_dev) {
 819                 dev_put(dev);
 820                 return -EADDRNOTAVAIL;
 821         }
 822         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 823         inet_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_MC_FORWARDING,
 824                                     dev->ifindex, &in_dev->cnf);
 825         ip_rt_multicast_event(in_dev);
 826
 827         /* Fill in the VIF structures */
 828
 829         v->rate_limit = vifc->vifc_rate_limit;
 830         v->local = vifc->vifc_lcl_addr.s_addr;
 831         v->remote = vifc->vifc_rmt_addr.s_addr;
 832         v->flags = vifc->vifc_flags;
 833         if (!mrtsock)
 834                 v->flags |= VIFF_STATIC;
 835         v->threshold = vifc->vifc_threshold;
 836         v->bytes_in = 0;
 837         v->bytes_out = 0;
 838         v->pkt_in = 0;
 839         v->pkt_out = 0;
 840         v->link = dev->ifindex;
 841         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
 842                 v->link = dev_get_iflink(dev);
 843
 844         /* And finish update writing critical data */
 845         write_lock_bh(&mrt_lock);
 846         v->dev = dev;
 847         if (v->flags & VIFF_REGISTER)
 848                 mrt->mroute_reg_vif_num = vifi;
 849         if (vifi+1 > mrt->maxvif)
 850                 mrt->maxvif = vifi+1;
 851         write_unlock_bh(&mrt_lock);
 852         return 0;
 853 }
 854
 855 /* called with rcu_read_lock() */
 856 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 857                                          __be32 origin,
 858                                          __be32 mcastgrp)
 859 {
 860         struct mfc_cache_cmp_arg arg = {
 861                         .mfc_mcastgrp = mcastgrp,
 862                         .mfc_origin = origin
 863         };
 864         struct rhlist_head *tmp, *list;
 865         struct mfc_cache *c;
 866
 867         list = rhltable_lookup(&mrt->mfc_hash, &arg, ipmr_rht_params);
 868         rhl_for_each_entry_rcu(c, tmp, list, mnode)
 869                 return c;
 870
 871         return NULL;
 872 }
 873
 874 /* Look for a (*,*,oif) entry */
 875 static struct mfc_cache *ipmr_cache_find_any_parent(struct mr_table *mrt,
 876                                                     int vifi)
 877 {
 878         struct mfc_cache_cmp_arg arg = {
 879                         .mfc_mcastgrp = htonl(INADDR_ANY),
 880                         .mfc_origin = htonl(INADDR_ANY)
 881         };
 882         struct rhlist_head *tmp, *list;
 883         struct mfc_cache *c;
 884
 885         list = rhltable_lookup(&mrt->mfc_hash, &arg, ipmr_rht_params);
 886         rhl_for_each_entry_rcu(c, tmp, list, mnode)
 887                 if (c->mfc_un.res.ttls[vifi] < 255)
 888                         return c;
 889
 890         return NULL;
 891 }
 892
 893 /* Look for a (*,G) entry */
 894 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 895                                              __be32 mcastgrp, int vifi)
 896 {
 897         struct mfc_cache_cmp_arg arg = {
 898                         .mfc_mcastgrp = mcastgrp,
 899                         .mfc_origin = htonl(INADDR_ANY)
 900         };
 901         struct rhlist_head *tmp, *list;
 902         struct mfc_cache *c, *proxy;
 903
 904         if (mcastgrp == htonl(INADDR_ANY))
 905                 goto skip;
 906
 907         list = rhltable_lookup(&mrt->mfc_hash, &arg, ipmr_rht_params);
 908         rhl_for_each_entry_rcu(c, tmp, list, mnode) {
 909                 if (c->mfc_un.res.ttls[vifi] < 255)
 910                         return c;
 911
 912                 /* It's ok if the vifi is part of the static tree */
 913                 proxy = ipmr_cache_find_any_parent(mrt, c->mfc_parent);
 914                 if (proxy && proxy->mfc_un.res.ttls[vifi] < 255)
 915                         return c;
 916         }
 917
 918 skip:
 919         return ipmr_cache_find_any_parent(mrt, vifi);
 920 }
 921
 922 /* Look for a (S,G,iif) entry if parent != -1 */
 923 static struct mfc_cache *ipmr_cache_find_parent(struct mr_table *mrt,
 924                                                 __be32 origin, __be32 mcastgrp,
 925                                                 int parent)
 926 {
 927         struct mfc_cache_cmp_arg arg = {
 928                         .mfc_mcastgrp = mcastgrp,
 929                         .mfc_origin = origin,
 930         };
 931         struct rhlist_head *tmp, *list;
 932         struct mfc_cache *c;
 933
 934         list = rhltable_lookup(&mrt->mfc_hash, &arg, ipmr_rht_params);
 935         rhl_for_each_entry_rcu(c, tmp, list, mnode)
 936                 if (parent == -1 || parent == c->mfc_parent)
 937                         return c;
 938
 939         return NULL;
 940 }
 941
 942 /* Allocate a multicast cache entry */
 943 static struct mfc_cache *ipmr_cache_alloc(void)
 944 {
 945         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 946
 947         if (c) {
 948                 c->mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 949                 c->mfc_un.res.minvif = MAXVIFS;
 950         }
 951         return c;
 952 }
 953
 954 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 955 {
 956         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 957
 958         if (c) {
 959                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
 960                 c->mfc_un.unres.expires = jiffies + 10*HZ;
 961         }
 962         return c;
 963 }
 964
 965 /* A cache entry has gone into a resolved state from queued */
 966 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 967                                struct mfc_cache *uc, struct mfc_cache *c)
 968 {
 969         struct sk_buff *skb;
 970         struct nlmsgerr *e;
 971
 972         /* Play the pending entries through our router */
 973         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
 974                 if (ip_hdr(skb)->version == 0) {
 975                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 976
 977                         if (__ipmr_fill_mroute(mrt, skb, c, nlmsg_data(nlh)) > 0) {
 978                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 979                                                  (u8 *)nlh;
 980                         } else {
 981                                 nlh->nlmsg_type = NLMSG_ERROR;
 982                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 983                                 skb_trim(skb, nlh->nlmsg_len);
 984                                 e = nlmsg_data(nlh);
 985                                 e->error = -EMSGSIZE;
 986                                 memset(&e->msg, 0, sizeof(e->msg));
 987                         }
 988
 989                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 990                 } else {
 991                         ip_mr_forward(net, mrt, skb, c, 0);
 992                 }
 993         }
 994 }
 995
 996 /* Bounce a cache query up to mrouted. We could use netlink for this but mrouted
 997  * expects the following bizarre scheme.
 998  *
 999  * Called under mrt_lock.
1000  */
1001 static int ipmr_cache_report(struct mr_table *mrt,
1002                              struct sk_buff *pkt, vifi_t vifi, int assert)
1003 {
1004         const int ihl = ip_hdrlen(pkt);
1005         struct sock *mroute_sk;
1006         struct igmphdr *igmp;
1007         struct igmpmsg *msg;
1008         struct sk_buff *skb;
1009         int ret;
1010
1011         if (assert == IGMPMSG_WHOLEPKT)
1012                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
1013         else
1014                 skb = alloc_skb(128, GFP_ATOMIC);
1015
1016         if (!skb)
1017                 return -ENOBUFS;
1018
1019         if (assert == IGMPMSG_WHOLEPKT) {
1020                 /* Ugly, but we have no choice with this interface.
1021                  * Duplicate old header, fix ihl, length etc.
1022                  * And all this only to mangle msg->im_msgtype and
1023                  * to set msg->im_mbz to "mbz" :-)
1024                  */
1025                 skb_push(skb, sizeof(struct iphdr));
1026                 skb_reset_network_header(skb);
1027                 skb_reset_transport_header(skb);
1028                 msg = (struct igmpmsg *)skb_network_header(skb);
1029                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
1030                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
1031                 msg->im_mbz = 0;
1032                 msg->im_vif = mrt->mroute_reg_vif_num;
1033                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
1034                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
1035                                              sizeof(struct iphdr));
1036         } else {
1037                 /* Copy the IP header */
1038                 skb_set_network_header(skb, skb->len);
1039                 skb_put(skb, ihl);
1040                 skb_copy_to_linear_data(skb, pkt->data, ihl);
1041                 /* Flag to the kernel this is a route add */
1042                 ip_hdr(skb)->protocol = 0;
1043                 msg = (struct igmpmsg *)skb_network_header(skb);
1044                 msg->im_vif = vifi;
1045                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1046                 /* Add our header */
1047                 igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
1048                 igmp->type = assert;
1049                 msg->im_msgtype = assert;
1050                 igmp->code = 0;
1051                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1052                 skb->transport_header = skb->network_header;
1053         }
1054
1055         rcu_read_lock();
1056         mroute_sk = rcu_dereference(mrt->mroute_sk);
1057         if (!mroute_sk) {
1058                 rcu_read_unlock();
1059                 kfree_skb(skb);
1060                 return -EINVAL;
1061         }
1062
1063         /* Deliver to mrouted */
1064         ret = sock_queue_rcv_skb(mroute_sk, skb);
1065         rcu_read_unlock();
1066         if (ret < 0) {
1067                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1068                 kfree_skb(skb);
1069         }
1070
1071         return ret;
1072 }
1073
1074 /* Queue a packet for resolution. It gets locked cache entry! */
1075 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1076                                  struct sk_buff *skb)
1077 {
1078         const struct iphdr *iph = ip_hdr(skb);
1079         struct mfc_cache *c;
1080         bool found = false;
1081         int err;
1082
1083         spin_lock_bh(&mfc_unres_lock);
1084         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
1085                 if (c->mfc_mcastgrp == iph->daddr &&
1086                     c->mfc_origin == iph->saddr) {
1087                         found = true;
1088                         break;
1089                 }
1090         }
1091
1092         if (!found) {
1093                 /* Create a new entry if allowable */
1094                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1095                     (c = ipmr_cache_alloc_unres()) == NULL) {
1096                         spin_unlock_bh(&mfc_unres_lock);
1097
1098                         kfree_skb(skb);
1099                         return -ENOBUFS;
1100                 }
1101
1102                 /* Fill in the new cache entry */
1103                 c->mfc_parent   = -1;
1104                 c->mfc_origin   = iph->saddr;
1105                 c->mfc_mcastgrp = iph->daddr;
1106
1107                 /* Reflect first query at mrouted. */
1108                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1109                 if (err < 0) {
1110                         /* If the report failed throw the cache entry
1111                            out - Brad Parker
1112                          */
1113                         spin_unlock_bh(&mfc_unres_lock);
1114
1115                         ipmr_cache_free(c);
1116                         kfree_skb(skb);
1117                         return err;
1118                 }
1119
1120                 atomic_inc(&mrt->cache_resolve_queue_len);
1121                 list_add(&c->list, &mrt->mfc_unres_queue);
1122                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1123
1124                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1125                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1126         }
1127
1128         /* See if we can append the packet */
1129         if (c->mfc_un.unres.unresolved.qlen > 3) {
1130                 kfree_skb(skb);
1131                 err = -ENOBUFS;
1132         } else {
1133                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1134                 err = 0;
1135         }
1136
1137         spin_unlock_bh(&mfc_unres_lock);
1138         return err;
1139 }
1140
1141 /* MFC cache manipulation by user space mroute daemon */
1142
1143 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1144 {
1145         struct mfc_cache *c;
1146
1147         /* The entries are added/deleted only under RTNL */
1148         rcu_read_lock();
1149         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1150                                    mfc->mfcc_mcastgrp.s_addr, parent);
1151         rcu_read_unlock();
1152         if (!c)
1153                 return -ENOENT;
1154         rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1155         list_del_rcu(&c->list);
1156         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1157         ipmr_cache_free(c);
1158
1159         return 0;
1160 }
1161
1162 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1163                         struct mfcctl *mfc, int mrtsock, int parent)
1164 {
1165         struct mfc_cache *uc, *c;
1166         bool found;
1167         int ret;
1168
1169         if (mfc->mfcc_parent >= MAXVIFS)
1170                 return -ENFILE;
1171
1172         /* The entries are added/deleted only under RTNL */
1173         rcu_read_lock();
1174         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1175                                    mfc->mfcc_mcastgrp.s_addr, parent);
1176         rcu_read_unlock();
1177         if (c) {
1178                 write_lock_bh(&mrt_lock);
1179                 c->mfc_parent = mfc->mfcc_parent;
1180                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1181                 if (!mrtsock)
1182                         c->mfc_flags |= MFC_STATIC;
1183                 write_unlock_bh(&mrt_lock);
1184                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1185                 return 0;
1186         }
1187
1188         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1189             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1190                 return -EINVAL;
1191
1192         c = ipmr_cache_alloc();
1193         if (!c)
1194                 return -ENOMEM;
1195
1196         c->mfc_origin = mfc->mfcc_origin.s_addr;
1197         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1198         c->mfc_parent = mfc->mfcc_parent;
1199         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1200         if (!mrtsock)
1201                 c->mfc_flags |= MFC_STATIC;
1202
1203         ret = rhltable_insert_key(&mrt->mfc_hash, &c->cmparg, &c->mnode,
1204                                   ipmr_rht_params);
1205         if (ret) {
1206                 pr_err("ipmr: rhtable insert error %d\n", ret);
1207                 ipmr_cache_free(c);
1208                 return ret;
1209         }
1210         list_add_tail_rcu(&c->list, &mrt->mfc_cache_list);
1211         /* Check to see if we resolved a queued list. If so we
1212          * need to send on the frames and tidy up.
1213          */
1214         found = false;
1215         spin_lock_bh(&mfc_unres_lock);
1216         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1217                 if (uc->mfc_origin == c->mfc_origin &&
1218                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1219                         list_del(&uc->list);
1220                         atomic_dec(&mrt->cache_resolve_queue_len);
1221                         found = true;
1222                         break;
1223                 }
1224         }
1225         if (list_empty(&mrt->mfc_unres_queue))
1226                 del_timer(&mrt->ipmr_expire_timer);
1227         spin_unlock_bh(&mfc_unres_lock);
1228
1229         if (found) {
1230                 ipmr_cache_resolve(net, mrt, uc, c);
1231                 ipmr_cache_free(uc);
1232         }
1233         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1234         return 0;
1235 }
1236
1237 /* Close the multicast socket, and clear the vif tables etc */
1238 static void mroute_clean_tables(struct mr_table *mrt, bool all)
1239 {
1240         struct mfc_cache *c, *tmp;
1241         LIST_HEAD(list);
1242         int i;
1243
1244         /* Shut down all active vif entries */
1245         for (i = 0; i < mrt->maxvif; i++) {
1246                 if (!all && (mrt->vif_table[i].flags & VIFF_STATIC))
1247                         continue;
1248                 vif_delete(mrt, i, 0, &list);
1249         }
1250         unregister_netdevice_many(&list);
1251
1252         /* Wipe the cache */
1253         list_for_each_entry_safe(c, tmp, &mrt->mfc_cache_list, list) {
1254                 if (!all && (c->mfc_flags & MFC_STATIC))
1255                         continue;
1256                 rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1257                 list_del_rcu(&c->list);
1258                 mroute_netlink_event(mrt, c, RTM_DELROUTE);
1259                 ipmr_cache_free(c);
1260         }
1261
1262         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1263                 spin_lock_bh(&mfc_unres_lock);
1264                 list_for_each_entry_safe(c, tmp, &mrt->mfc_unres_queue, list) {
1265                         list_del(&c->list);
1266                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1267                         ipmr_destroy_unres(mrt, c);
1268                 }
1269                 spin_unlock_bh(&mfc_unres_lock);
1270         }
1271 }
1272
1273 /* called from ip_ra_control(), before an RCU grace period,
1274  * we dont need to call synchronize_rcu() here
1275  */
1276 static void mrtsock_destruct(struct sock *sk)
1277 {
1278         struct net *net = sock_net(sk);
1279         struct mr_table *mrt;
1280
1281         rtnl_lock();
1282         ipmr_for_each_table(mrt, net) {
1283                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1284                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1285                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1286                                                     NETCONFA_MC_FORWARDING,
1287                                                     NETCONFA_IFINDEX_ALL,
1288                                                     net->ipv4.devconf_all);
1289                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1290                         mroute_clean_tables(mrt, false);
1291                 }
1292         }
1293         rtnl_unlock();
1294 }
1295
1296 /* Socket options and virtual interface manipulation. The whole
1297  * virtual interface system is a complete heap, but unfortunately
1298  * that's how BSD mrouted happens to think. Maybe one day with a proper
1299  * MOSPF/PIM router set up we can clean this up.
1300  */
1301
1302 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval,
1303                          unsigned int optlen)
1304 {
1305         struct net *net = sock_net(sk);
1306         int val, ret = 0, parent = 0;
1307         struct mr_table *mrt;
1308         struct vifctl vif;
1309         struct mfcctl mfc;
1310         u32 uval;
1311
1312         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1313         rtnl_lock();
1314         if (sk->sk_type != SOCK_RAW ||
1315             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1316                 ret = -EOPNOTSUPP;
1317                 goto out_unlock;
1318         }
1319
1320         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1321         if (!mrt) {
1322                 ret = -ENOENT;
1323                 goto out_unlock;
1324         }
1325         if (optname != MRT_INIT) {
1326                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1327                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1328                         ret = -EACCES;
1329                         goto out_unlock;
1330                 }
1331         }
1332
1333         switch (optname) {
1334         case MRT_INIT:
1335                 if (optlen != sizeof(int)) {
1336                         ret = -EINVAL;
1337                         break;
1338                 }
1339                 if (rtnl_dereference(mrt->mroute_sk)) {
1340                         ret = -EADDRINUSE;
1341                         break;
1342                 }
1343
1344                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1345                 if (ret == 0) {
1346                         rcu_assign_pointer(mrt->mroute_sk, sk);
1347                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1348                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1349                                                     NETCONFA_MC_FORWARDING,
1350                                                     NETCONFA_IFINDEX_ALL,
1351                                                     net->ipv4.devconf_all);
1352                 }
1353                 break;
1354         case MRT_DONE:
1355                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1356                         ret = -EACCES;
1357                 } else {
1358                         /* We need to unlock here because mrtsock_destruct takes
1359                          * care of rtnl itself and we can't change that due to
1360                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1361                          */
1362                         rtnl_unlock();
1363                         ret = ip_ra_control(sk, 0, NULL);
1364                         goto out;
1365                 }
1366                 break;
1367         case MRT_ADD_VIF:
1368         case MRT_DEL_VIF:
1369                 if (optlen != sizeof(vif)) {
1370                         ret = -EINVAL;
1371                         break;
1372                 }
1373                 if (copy_from_user(&vif, optval, sizeof(vif))) {
1374                         ret = -EFAULT;
1375                         break;
1376                 }
1377                 if (vif.vifc_vifi >= MAXVIFS) {
1378                         ret = -ENFILE;
1379                         break;
1380                 }
1381                 if (optname == MRT_ADD_VIF) {
1382                         ret = vif_add(net, mrt, &vif,
1383                                       sk == rtnl_dereference(mrt->mroute_sk));
1384                 } else {
1385                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1386                 }
1387                 break;
1388         /* Manipulate the forwarding caches. These live
1389          * in a sort of kernel/user symbiosis.
1390          */
1391         case MRT_ADD_MFC:
1392         case MRT_DEL_MFC:
1393                 parent = -1;
1394         case MRT_ADD_MFC_PROXY:
1395         case MRT_DEL_MFC_PROXY:
1396                 if (optlen != sizeof(mfc)) {
1397                         ret = -EINVAL;
1398                         break;
1399                 }
1400                 if (copy_from_user(&mfc, optval, sizeof(mfc))) {
1401                         ret = -EFAULT;
1402                         break;
1403                 }
1404                 if (parent == 0)
1405                         parent = mfc.mfcc_parent;
1406                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1407                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1408                 else
1409                         ret = ipmr_mfc_add(net, mrt, &mfc,
1410                                            sk == rtnl_dereference(mrt->mroute_sk),
1411                                            parent);
1412                 break;
1413         /* Control PIM assert. */
1414         case MRT_ASSERT:
1415                 if (optlen != sizeof(val)) {
1416                         ret = -EINVAL;
1417                         break;
1418                 }
1419                 if (get_user(val, (int __user *)optval)) {
1420                         ret = -EFAULT;
1421                         break;
1422                 }
1423                 mrt->mroute_do_assert = val;
1424                 break;
1425         case MRT_PIM:
1426                 if (!ipmr_pimsm_enabled()) {
1427                         ret = -ENOPROTOOPT;
1428                         break;
1429                 }
1430                 if (optlen != sizeof(val)) {
1431                         ret = -EINVAL;
1432                         break;
1433                 }
1434                 if (get_user(val, (int __user *)optval)) {
1435                         ret = -EFAULT;
1436                         break;
1437                 }
1438
1439                 val = !!val;
1440                 if (val != mrt->mroute_do_pim) {
1441                         mrt->mroute_do_pim = val;
1442                         mrt->mroute_do_assert = val;
1443                 }
1444                 break;
1445         case MRT_TABLE:
1446                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1447                         ret = -ENOPROTOOPT;
1448                         break;
1449                 }
1450                 if (optlen != sizeof(uval)) {
1451                         ret = -EINVAL;
1452                         break;
1453                 }
1454                 if (get_user(uval, (u32 __user *)optval)) {
1455                         ret = -EFAULT;
1456                         break;
1457                 }
1458
1459                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1460                         ret = -EBUSY;
1461                 } else {
1462                         mrt = ipmr_new_table(net, uval);
1463                         if (IS_ERR(mrt))
1464                                 ret = PTR_ERR(mrt);
1465                         else
1466                                 raw_sk(sk)->ipmr_table = uval;
1467                 }
1468                 break;
1469         /* Spurious command, or MRT_VERSION which you cannot set. */
1470         default:
1471                 ret = -ENOPROTOOPT;
1472         }
1473 out_unlock:
1474         rtnl_unlock();
1475 out:
1476         return ret;
1477 }
1478
1479 /* Getsock opt support for the multicast routing system. */
1480 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1481 {
1482         int olr;
1483         int val;
1484         struct net *net = sock_net(sk);
1485         struct mr_table *mrt;
1486
1487         if (sk->sk_type != SOCK_RAW ||
1488             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1489                 return -EOPNOTSUPP;
1490
1491         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1492         if (!mrt)
1493                 return -ENOENT;
1494
1495         switch (optname) {
1496         case MRT_VERSION:
1497                 val = 0x0305;
1498                 break;
1499         case MRT_PIM:
1500                 if (!ipmr_pimsm_enabled())
1501                         return -ENOPROTOOPT;
1502                 val = mrt->mroute_do_pim;
1503                 break;
1504         case MRT_ASSERT:
1505                 val = mrt->mroute_do_assert;
1506                 break;
1507         default:
1508                 return -ENOPROTOOPT;
1509         }
1510
1511         if (get_user(olr, optlen))
1512                 return -EFAULT;
1513         olr = min_t(unsigned int, olr, sizeof(int));
1514         if (olr < 0)
1515                 return -EINVAL;
1516         if (put_user(olr, optlen))
1517                 return -EFAULT;
1518         if (copy_to_user(optval, &val, olr))
1519                 return -EFAULT;
1520         return 0;
1521 }
1522
1523 /* The IP multicast ioctl support routines. */
1524 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1525 {
1526         struct sioc_sg_req sr;
1527         struct sioc_vif_req vr;
1528         struct vif_device *vif;
1529         struct mfc_cache *c;
1530         struct net *net = sock_net(sk);
1531         struct mr_table *mrt;
1532
1533         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1534         if (!mrt)
1535                 return -ENOENT;
1536
1537         switch (cmd) {
1538         case SIOCGETVIFCNT:
1539                 if (copy_from_user(&vr, arg, sizeof(vr)))
1540                         return -EFAULT;
1541                 if (vr.vifi >= mrt->maxvif)
1542                         return -EINVAL;
1543                 read_lock(&mrt_lock);
1544                 vif = &mrt->vif_table[vr.vifi];
1545                 if (VIF_EXISTS(mrt, vr.vifi)) {
1546                         vr.icount = vif->pkt_in;
1547                         vr.ocount = vif->pkt_out;
1548                         vr.ibytes = vif->bytes_in;
1549                         vr.obytes = vif->bytes_out;
1550                         read_unlock(&mrt_lock);
1551
1552                         if (copy_to_user(arg, &vr, sizeof(vr)))
1553                                 return -EFAULT;
1554                         return 0;
1555                 }
1556                 read_unlock(&mrt_lock);
1557                 return -EADDRNOTAVAIL;
1558         case SIOCGETSGCNT:
1559                 if (copy_from_user(&sr, arg, sizeof(sr)))
1560                         return -EFAULT;
1561
1562                 rcu_read_lock();
1563                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1564                 if (c) {
1565                         sr.pktcnt = c->mfc_un.res.pkt;
1566                         sr.bytecnt = c->mfc_un.res.bytes;
1567                         sr.wrong_if = c->mfc_un.res.wrong_if;
1568                         rcu_read_unlock();
1569
1570                         if (copy_to_user(arg, &sr, sizeof(sr)))
1571                                 return -EFAULT;
1572                         return 0;
1573                 }
1574                 rcu_read_unlock();
1575                 return -EADDRNOTAVAIL;
1576         default:
1577                 return -ENOIOCTLCMD;
1578         }
1579 }
1580
1581 #ifdef CONFIG_COMPAT
1582 struct compat_sioc_sg_req {
1583         struct in_addr src;
1584         struct in_addr grp;
1585         compat_ulong_t pktcnt;
1586         compat_ulong_t bytecnt;
1587         compat_ulong_t wrong_if;
1588 };
1589
1590 struct compat_sioc_vif_req {
1591         vifi_t  vifi;           /* Which iface */
1592         compat_ulong_t icount;
1593         compat_ulong_t ocount;
1594         compat_ulong_t ibytes;
1595         compat_ulong_t obytes;
1596 };
1597
1598 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1599 {
1600         struct compat_sioc_sg_req sr;
1601         struct compat_sioc_vif_req vr;
1602         struct vif_device *vif;
1603         struct mfc_cache *c;
1604         struct net *net = sock_net(sk);
1605         struct mr_table *mrt;
1606
1607         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1608         if (!mrt)
1609                 return -ENOENT;
1610
1611         switch (cmd) {
1612         case SIOCGETVIFCNT:
1613                 if (copy_from_user(&vr, arg, sizeof(vr)))
1614                         return -EFAULT;
1615                 if (vr.vifi >= mrt->maxvif)
1616                         return -EINVAL;
1617                 read_lock(&mrt_lock);
1618                 vif = &mrt->vif_table[vr.vifi];
1619                 if (VIF_EXISTS(mrt, vr.vifi)) {
1620                         vr.icount = vif->pkt_in;
1621                         vr.ocount = vif->pkt_out;
1622                         vr.ibytes = vif->bytes_in;
1623                         vr.obytes = vif->bytes_out;
1624                         read_unlock(&mrt_lock);
1625
1626                         if (copy_to_user(arg, &vr, sizeof(vr)))
1627                                 return -EFAULT;
1628                         return 0;
1629                 }
1630                 read_unlock(&mrt_lock);
1631                 return -EADDRNOTAVAIL;
1632         case SIOCGETSGCNT:
1633                 if (copy_from_user(&sr, arg, sizeof(sr)))
1634                         return -EFAULT;
1635
1636                 rcu_read_lock();
1637                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1638                 if (c) {
1639                         sr.pktcnt = c->mfc_un.res.pkt;
1640                         sr.bytecnt = c->mfc_un.res.bytes;
1641                         sr.wrong_if = c->mfc_un.res.wrong_if;
1642                         rcu_read_unlock();
1643
1644                         if (copy_to_user(arg, &sr, sizeof(sr)))
1645                                 return -EFAULT;
1646                         return 0;
1647                 }
1648                 rcu_read_unlock();
1649                 return -EADDRNOTAVAIL;
1650         default:
1651                 return -ENOIOCTLCMD;
1652         }
1653 }
1654 #endif
1655
1656 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1657 {
1658         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1659         struct net *net = dev_net(dev);
1660         struct mr_table *mrt;
1661         struct vif_device *v;
1662         int ct;
1663
1664         if (event != NETDEV_UNREGISTER)
1665                 return NOTIFY_DONE;
1666
1667         ipmr_for_each_table(mrt, net) {
1668                 v = &mrt->vif_table[0];
1669                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1670                         if (v->dev == dev)
1671                                 vif_delete(mrt, ct, 1, NULL);
1672                 }
1673         }
1674         return NOTIFY_DONE;
1675 }
1676
1677 static struct notifier_block ip_mr_notifier = {
1678         .notifier_call = ipmr_device_event,
1679 };
1680
1681 /* Encapsulate a packet by attaching a valid IPIP header to it.
1682  * This avoids tunnel drivers and other mess and gives us the speed so
1683  * important for multicast video.
1684  */
1685 static void ip_encap(struct net *net, struct sk_buff *skb,
1686                      __be32 saddr, __be32 daddr)
1687 {
1688         struct iphdr *iph;
1689         const struct iphdr *old_iph = ip_hdr(skb);
1690
1691         skb_push(skb, sizeof(struct iphdr));
1692         skb->transport_header = skb->network_header;
1693         skb_reset_network_header(skb);
1694         iph = ip_hdr(skb);
1695
1696         iph->version    =       4;
1697         iph->tos        =       old_iph->tos;
1698         iph->ttl        =       old_iph->ttl;
1699         iph->frag_off   =       0;
1700         iph->daddr      =       daddr;
1701         iph->saddr      =       saddr;
1702         iph->protocol   =       IPPROTO_IPIP;
1703         iph->ihl        =       5;
1704         iph->tot_len    =       htons(skb->len);
1705         ip_select_ident(net, skb, NULL);
1706         ip_send_check(iph);
1707
1708         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1709         nf_reset(skb);
1710 }
1711
1712 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1713                                       struct sk_buff *skb)
1714 {
1715         struct ip_options *opt = &(IPCB(skb)->opt);
1716
1717         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1718         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1719
1720         if (unlikely(opt->optlen))
1721                 ip_forward_options(skb);
1722
1723         return dst_output(net, sk, skb);
1724 }
1725
1726 /* Processing handlers for ipmr_forward */
1727
1728 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1729                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1730 {
1731         const struct iphdr *iph = ip_hdr(skb);
1732         struct vif_device *vif = &mrt->vif_table[vifi];
1733         struct net_device *dev;
1734         struct rtable *rt;
1735         struct flowi4 fl4;
1736         int    encap = 0;
1737
1738         if (!vif->dev)
1739                 goto out_free;
1740
1741         if (vif->flags & VIFF_REGISTER) {
1742                 vif->pkt_out++;
1743                 vif->bytes_out += skb->len;
1744                 vif->dev->stats.tx_bytes += skb->len;
1745                 vif->dev->stats.tx_packets++;
1746                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1747                 goto out_free;
1748         }
1749
1750         if (vif->flags & VIFF_TUNNEL) {
1751                 rt = ip_route_output_ports(net, &fl4, NULL,
1752                                            vif->remote, vif->local,
1753                                            0, 0,
1754                                            IPPROTO_IPIP,
1755                                            RT_TOS(iph->tos), vif->link);
1756                 if (IS_ERR(rt))
1757                         goto out_free;
1758                 encap = sizeof(struct iphdr);
1759         } else {
1760                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1761                                            0, 0,
1762                                            IPPROTO_IPIP,
1763                                            RT_TOS(iph->tos), vif->link);
1764                 if (IS_ERR(rt))
1765                         goto out_free;
1766         }
1767
1768         dev = rt->dst.dev;
1769
1770         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1771                 /* Do not fragment multicasts. Alas, IPv4 does not
1772                  * allow to send ICMP, so that packets will disappear
1773                  * to blackhole.
1774                  */
1775                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1776                 ip_rt_put(rt);
1777                 goto out_free;
1778         }
1779
1780         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1781
1782         if (skb_cow(skb, encap)) {
1783                 ip_rt_put(rt);
1784                 goto out_free;
1785         }
1786
1787         vif->pkt_out++;
1788         vif->bytes_out += skb->len;
1789
1790         skb_dst_drop(skb);
1791         skb_dst_set(skb, &rt->dst);
1792         ip_decrease_ttl(ip_hdr(skb));
1793
1794         /* FIXME: forward and output firewalls used to be called here.
1795          * What do we do with netfilter? -- RR
1796          */
1797         if (vif->flags & VIFF_TUNNEL) {
1798                 ip_encap(net, skb, vif->local, vif->remote);
1799                 /* FIXME: extra output firewall step used to be here. --RR */
1800                 vif->dev->stats.tx_packets++;
1801                 vif->dev->stats.tx_bytes += skb->len;
1802         }
1803
1804         IPCB(skb)->flags |= IPSKB_FORWARDED;
1805
1806         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1807          * not only before forwarding, but after forwarding on all output
1808          * interfaces. It is clear, if mrouter runs a multicasting
1809          * program, it should receive packets not depending to what interface
1810          * program is joined.
1811          * If we will not make it, the program will have to join on all
1812          * interfaces. On the other hand, multihoming host (or router, but
1813          * not mrouter) cannot join to more than one interface - it will
1814          * result in receiving multiple packets.
1815          */
1816         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1817                 net, NULL, skb, skb->dev, dev,
1818                 ipmr_forward_finish);
1819         return;
1820
1821 out_free:
1822         kfree_skb(skb);
1823 }
1824
1825 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1826 {
1827         int ct;
1828
1829         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1830                 if (mrt->vif_table[ct].dev == dev)
1831                         break;
1832         }
1833         return ct;
1834 }
1835
1836 /* "local" means that we should preserve one skb (for local delivery) */
1837 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1838                           struct sk_buff *skb, struct mfc_cache *cache,
1839                           int local)
1840 {
1841         int true_vifi = ipmr_find_vif(mrt, skb->dev);
1842         int psend = -1;
1843         int vif, ct;
1844
1845         vif = cache->mfc_parent;
1846         cache->mfc_un.res.pkt++;
1847         cache->mfc_un.res.bytes += skb->len;
1848         cache->mfc_un.res.lastuse = jiffies;
1849
1850         if (cache->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1851                 struct mfc_cache *cache_proxy;
1852
1853                 /* For an (*,G) entry, we only check that the incomming
1854                  * interface is part of the static tree.
1855                  */
1856                 cache_proxy = ipmr_cache_find_any_parent(mrt, vif);
1857                 if (cache_proxy &&
1858                     cache_proxy->mfc_un.res.ttls[true_vifi] < 255)
1859                         goto forward;
1860         }
1861
1862         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1863         if (mrt->vif_table[vif].dev != skb->dev) {
1864                 struct net_device *mdev;
1865
1866                 mdev = l3mdev_master_dev_rcu(mrt->vif_table[vif].dev);
1867                 if (mdev == skb->dev)
1868                         goto forward;
1869
1870                 if (rt_is_output_route(skb_rtable(skb))) {
1871                         /* It is our own packet, looped back.
1872                          * Very complicated situation...
1873                          *
1874                          * The best workaround until routing daemons will be
1875                          * fixed is not to redistribute packet, if it was
1876                          * send through wrong interface. It means, that
1877                          * multicast applications WILL NOT work for
1878                          * (S,G), which have default multicast route pointing
1879                          * to wrong oif. In any case, it is not a good
1880                          * idea to use multicasting applications on router.
1881                          */
1882                         goto dont_forward;
1883                 }
1884
1885                 cache->mfc_un.res.wrong_if++;
1886
1887                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1888                     /* pimsm uses asserts, when switching from RPT to SPT,
1889                      * so that we cannot check that packet arrived on an oif.
1890                      * It is bad, but otherwise we would need to move pretty
1891                      * large chunk of pimd to kernel. Ough... --ANK
1892                      */
1893                     (mrt->mroute_do_pim ||
1894                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1895                     time_after(jiffies,
1896                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1897                         cache->mfc_un.res.last_assert = jiffies;
1898                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1899                 }
1900                 goto dont_forward;
1901         }
1902
1903 forward:
1904         mrt->vif_table[vif].pkt_in++;
1905         mrt->vif_table[vif].bytes_in += skb->len;
1906
1907         /* Forward the frame */
1908         if (cache->mfc_origin == htonl(INADDR_ANY) &&
1909             cache->mfc_mcastgrp == htonl(INADDR_ANY)) {
1910                 if (true_vifi >= 0 &&
1911                     true_vifi != cache->mfc_parent &&
1912                     ip_hdr(skb)->ttl >
1913                                 cache->mfc_un.res.ttls[cache->mfc_parent]) {
1914                         /* It's an (*,*) entry and the packet is not coming from
1915                          * the upstream: forward the packet to the upstream
1916                          * only.
1917                          */
1918                         psend = cache->mfc_parent;
1919                         goto last_forward;
1920                 }
1921                 goto dont_forward;
1922         }
1923         for (ct = cache->mfc_un.res.maxvif - 1;
1924              ct >= cache->mfc_un.res.minvif; ct--) {
1925                 /* For (*,G) entry, don't forward to the incoming interface */
1926                 if ((cache->mfc_origin != htonl(INADDR_ANY) ||
1927                      ct != true_vifi) &&
1928                     ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1929                         if (psend != -1) {
1930                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1931
1932                                 if (skb2)
1933                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1934                                                         psend);
1935                         }
1936                         psend = ct;
1937                 }
1938         }
1939 last_forward:
1940         if (psend != -1) {
1941                 if (local) {
1942                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1943
1944                         if (skb2)
1945                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1946                 } else {
1947                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1948                         return;
1949                 }
1950         }
1951
1952 dont_forward:
1953         if (!local)
1954                 kfree_skb(skb);
1955 }
1956
1957 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1958 {
1959         struct rtable *rt = skb_rtable(skb);
1960         struct iphdr *iph = ip_hdr(skb);
1961         struct flowi4 fl4 = {
1962                 .daddr = iph->daddr,
1963                 .saddr = iph->saddr,
1964                 .flowi4_tos = RT_TOS(iph->tos),
1965                 .flowi4_oif = (rt_is_output_route(rt) ?
1966                                skb->dev->ifindex : 0),
1967                 .flowi4_iif = (rt_is_output_route(rt) ?
1968                                LOOPBACK_IFINDEX :
1969                                skb->dev->ifindex),
1970                 .flowi4_mark = skb->mark,
1971         };
1972         struct mr_table *mrt;
1973         int err;
1974
1975         err = ipmr_fib_lookup(net, &fl4, &mrt);
1976         if (err)
1977                 return ERR_PTR(err);
1978         return mrt;
1979 }
1980
1981 /* Multicast packets for forwarding arrive here
1982  * Called with rcu_read_lock();
1983  */
1984 int ip_mr_input(struct sk_buff *skb)
1985 {
1986         struct mfc_cache *cache;
1987         struct net *net = dev_net(skb->dev);
1988         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1989         struct mr_table *mrt;
1990
1991         /* Packet is looped back after forward, it should not be
1992          * forwarded second time, but still can be delivered locally.
1993          */
1994         if (IPCB(skb)->flags & IPSKB_FORWARDED)
1995                 goto dont_forward;
1996
1997         mrt = ipmr_rt_fib_lookup(net, skb);
1998         if (IS_ERR(mrt)) {
1999                 kfree_skb(skb);
2000                 return PTR_ERR(mrt);
2001         }
2002         if (!local) {
2003                 if (IPCB(skb)->opt.router_alert) {
2004                         if (ip_call_ra_chain(skb))
2005                                 return 0;
2006                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
2007                         /* IGMPv1 (and broken IGMPv2 implementations sort of
2008                          * Cisco IOS <= 11.2(8)) do not put router alert
2009                          * option to IGMP packets destined to routable
2010                          * groups. It is very bad, because it means
2011                          * that we can forward NO IGMP messages.
2012                          */
2013                         struct sock *mroute_sk;
2014
2015                         mroute_sk = rcu_dereference(mrt->mroute_sk);
2016                         if (mroute_sk) {
2017                                 nf_reset(skb);
2018                                 raw_rcv(mroute_sk, skb);
2019                                 return 0;
2020                         }
2021                     }
2022         }
2023
2024         /* already under rcu_read_lock() */
2025         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
2026         if (!cache) {
2027                 int vif = ipmr_find_vif(mrt, skb->dev);
2028
2029                 if (vif >= 0)
2030                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2031                                                     vif);
2032         }
2033
2034         /* No usable cache entry */
2035         if (!cache) {
2036                 int vif;
2037
2038                 if (local) {
2039                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2040                         ip_local_deliver(skb);
2041                         if (!skb2)
2042                                 return -ENOBUFS;
2043                         skb = skb2;
2044                 }
2045
2046                 read_lock(&mrt_lock);
2047                 vif = ipmr_find_vif(mrt, skb->dev);
2048                 if (vif >= 0) {
2049                         int err2 = ipmr_cache_unresolved(mrt, vif, skb);
2050                         read_unlock(&mrt_lock);
2051
2052                         return err2;
2053                 }
2054                 read_unlock(&mrt_lock);
2055                 kfree_skb(skb);
2056                 return -ENODEV;
2057         }
2058
2059         read_lock(&mrt_lock);
2060         ip_mr_forward(net, mrt, skb, cache, local);
2061         read_unlock(&mrt_lock);
2062
2063         if (local)
2064                 return ip_local_deliver(skb);
2065
2066         return 0;
2067
2068 dont_forward:
2069         if (local)
2070                 return ip_local_deliver(skb);
2071         kfree_skb(skb);
2072         return 0;
2073 }
2074
2075 #ifdef CONFIG_IP_PIMSM_V1
2076 /* Handle IGMP messages of PIMv1 */
2077 int pim_rcv_v1(struct sk_buff *skb)
2078 {
2079         struct igmphdr *pim;
2080         struct net *net = dev_net(skb->dev);
2081         struct mr_table *mrt;
2082
2083         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2084                 goto drop;
2085
2086         pim = igmp_hdr(skb);
2087
2088         mrt = ipmr_rt_fib_lookup(net, skb);
2089         if (IS_ERR(mrt))
2090                 goto drop;
2091         if (!mrt->mroute_do_pim ||
2092             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2093                 goto drop;
2094
2095         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2096 drop:
2097                 kfree_skb(skb);
2098         }
2099         return 0;
2100 }
2101 #endif
2102
2103 #ifdef CONFIG_IP_PIMSM_V2
2104 static int pim_rcv(struct sk_buff *skb)
2105 {
2106         struct pimreghdr *pim;
2107         struct net *net = dev_net(skb->dev);
2108         struct mr_table *mrt;
2109
2110         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2111                 goto drop;
2112
2113         pim = (struct pimreghdr *)skb_transport_header(skb);
2114         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2115             (pim->flags & PIM_NULL_REGISTER) ||
2116             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2117              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2118                 goto drop;
2119
2120         mrt = ipmr_rt_fib_lookup(net, skb);
2121         if (IS_ERR(mrt))
2122                 goto drop;
2123         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2124 drop:
2125                 kfree_skb(skb);
2126         }
2127         return 0;
2128 }
2129 #endif
2130
2131 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2132                               struct mfc_cache *c, struct rtmsg *rtm)
2133 {
2134         struct rta_mfc_stats mfcs;
2135         struct nlattr *mp_attr;
2136         struct rtnexthop *nhp;
2137         unsigned long lastuse;
2138         int ct;
2139
2140         /* If cache is unresolved, don't try to parse IIF and OIF */
2141         if (c->mfc_parent >= MAXVIFS) {
2142                 rtm->rtm_flags |= RTNH_F_UNRESOLVED;
2143                 return -ENOENT;
2144         }
2145
2146         if (VIF_EXISTS(mrt, c->mfc_parent) &&
2147             nla_put_u32(skb, RTA_IIF, mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
2148                 return -EMSGSIZE;
2149
2150         if (!(mp_attr = nla_nest_start(skb, RTA_MULTIPATH)))
2151                 return -EMSGSIZE;
2152
2153         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2154                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2155                         if (!(nhp = nla_reserve_nohdr(skb, sizeof(*nhp)))) {
2156                                 nla_nest_cancel(skb, mp_attr);
2157                                 return -EMSGSIZE;
2158                         }
2159
2160                         nhp->rtnh_flags = 0;
2161                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2162                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2163                         nhp->rtnh_len = sizeof(*nhp);
2164                 }
2165         }
2166
2167         nla_nest_end(skb, mp_attr);
2168
2169         lastuse = READ_ONCE(c->mfc_un.res.lastuse);
2170         lastuse = time_after_eq(jiffies, lastuse) ? jiffies - lastuse : 0;
2171
2172         mfcs.mfcs_packets = c->mfc_un.res.pkt;
2173         mfcs.mfcs_bytes = c->mfc_un.res.bytes;
2174         mfcs.mfcs_wrong_if = c->mfc_un.res.wrong_if;
2175         if (nla_put_64bit(skb, RTA_MFC_STATS, sizeof(mfcs), &mfcs, RTA_PAD) ||
2176             nla_put_u64_64bit(skb, RTA_EXPIRES, jiffies_to_clock_t(lastuse),
2177                               RTA_PAD))
2178                 return -EMSGSIZE;
2179
2180         rtm->rtm_type = RTN_MULTICAST;
2181         return 1;
2182 }
2183
2184 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2185                    __be32 saddr, __be32 daddr,
2186                    struct rtmsg *rtm, u32 portid)
2187 {
2188         struct mfc_cache *cache;
2189         struct mr_table *mrt;
2190         int err;
2191
2192         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2193         if (!mrt)
2194                 return -ENOENT;
2195
2196         rcu_read_lock();
2197         cache = ipmr_cache_find(mrt, saddr, daddr);
2198         if (!cache && skb->dev) {
2199                 int vif = ipmr_find_vif(mrt, skb->dev);
2200
2201                 if (vif >= 0)
2202                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2203         }
2204         if (!cache) {
2205                 struct sk_buff *skb2;
2206                 struct iphdr *iph;
2207                 struct net_device *dev;
2208                 int vif = -1;
2209
2210                 dev = skb->dev;
2211                 read_lock(&mrt_lock);
2212                 if (dev)
2213                         vif = ipmr_find_vif(mrt, dev);
2214                 if (vif < 0) {
2215                         read_unlock(&mrt_lock);
2216                         rcu_read_unlock();
2217                         return -ENODEV;
2218                 }
2219                 skb2 = skb_clone(skb, GFP_ATOMIC);
2220                 if (!skb2) {
2221                         read_unlock(&mrt_lock);
2222                         rcu_read_unlock();
2223                         return -ENOMEM;
2224                 }
2225
2226                 NETLINK_CB(skb2).portid = portid;
2227                 skb_push(skb2, sizeof(struct iphdr));
2228                 skb_reset_network_header(skb2);
2229                 iph = ip_hdr(skb2);
2230                 iph->ihl = sizeof(struct iphdr) >> 2;
2231                 iph->saddr = saddr;
2232                 iph->daddr = daddr;
2233                 iph->version = 0;
2234                 err = ipmr_cache_unresolved(mrt, vif, skb2);
2235                 read_unlock(&mrt_lock);
2236                 rcu_read_unlock();
2237                 return err;
2238         }
2239
2240         read_lock(&mrt_lock);
2241         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2242         read_unlock(&mrt_lock);
2243         rcu_read_unlock();
2244         return err;
2245 }
2246
2247 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2248                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2249                             int flags)
2250 {
2251         struct nlmsghdr *nlh;
2252         struct rtmsg *rtm;
2253         int err;
2254
2255         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2256         if (!nlh)
2257                 return -EMSGSIZE;
2258
2259         rtm = nlmsg_data(nlh);
2260         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2261         rtm->rtm_dst_len  = 32;
2262         rtm->rtm_src_len  = 32;
2263         rtm->rtm_tos      = 0;
2264         rtm->rtm_table    = mrt->id;
2265         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2266                 goto nla_put_failure;
2267         rtm->rtm_type     = RTN_MULTICAST;
2268         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2269         if (c->mfc_flags & MFC_STATIC)
2270                 rtm->rtm_protocol = RTPROT_STATIC;
2271         else
2272                 rtm->rtm_protocol = RTPROT_MROUTED;
2273         rtm->rtm_flags    = 0;
2274
2275         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2276             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2277                 goto nla_put_failure;
2278         err = __ipmr_fill_mroute(mrt, skb, c, rtm);
2279         /* do not break the dump if cache is unresolved */
2280         if (err < 0 && err != -ENOENT)
2281                 goto nla_put_failure;
2282
2283         nlmsg_end(skb, nlh);
2284         return 0;
2285
2286 nla_put_failure:
2287         nlmsg_cancel(skb, nlh);
2288         return -EMSGSIZE;
2289 }
2290
2291 static size_t mroute_msgsize(bool unresolved, int maxvif)
2292 {
2293         size_t len =
2294                 NLMSG_ALIGN(sizeof(struct rtmsg))
2295                 + nla_total_size(4)     /* RTA_TABLE */
2296                 + nla_total_size(4)     /* RTA_SRC */
2297                 + nla_total_size(4)     /* RTA_DST */
2298                 ;
2299
2300         if (!unresolved)
2301                 len = len
2302                       + nla_total_size(4)       /* RTA_IIF */
2303                       + nla_total_size(0)       /* RTA_MULTIPATH */
2304                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2305                                                 /* RTA_MFC_STATS */
2306                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2307                 ;
2308
2309         return len;
2310 }
2311
2312 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2313                                  int cmd)
2314 {
2315         struct net *net = read_pnet(&mrt->net);
2316         struct sk_buff *skb;
2317         int err = -ENOBUFS;
2318
2319         skb = nlmsg_new(mroute_msgsize(mfc->mfc_parent >= MAXVIFS, mrt->maxvif),
2320                         GFP_ATOMIC);
2321         if (!skb)
2322                 goto errout;
2323
2324         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2325         if (err < 0)
2326                 goto errout;
2327
2328         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2329         return;
2330
2331 errout:
2332         kfree_skb(skb);
2333         if (err < 0)
2334                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2335 }
2336
2337 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2338 {
2339         struct net *net = sock_net(skb->sk);
2340         struct mr_table *mrt;
2341         struct mfc_cache *mfc;
2342         unsigned int t = 0, s_t;
2343         unsigned int e = 0, s_e;
2344
2345         s_t = cb->args[0];
2346         s_e = cb->args[1];
2347
2348         rcu_read_lock();
2349         ipmr_for_each_table(mrt, net) {
2350                 if (t < s_t)
2351                         goto next_table;
2352                 list_for_each_entry_rcu(mfc, &mrt->mfc_cache_list, list) {
2353                         if (e < s_e)
2354                                 goto next_entry;
2355                         if (ipmr_fill_mroute(mrt, skb,
2356                                              NETLINK_CB(cb->skb).portid,
2357                                              cb->nlh->nlmsg_seq,
2358                                              mfc, RTM_NEWROUTE,
2359                                              NLM_F_MULTI) < 0)
2360                                 goto done;
2361 next_entry:
2362                         e++;
2363                 }
2364                 e = 0;
2365                 s_e = 0;
2366
2367                 spin_lock_bh(&mfc_unres_lock);
2368                 list_for_each_entry(mfc, &mrt->mfc_unres_queue, list) {
2369                         if (e < s_e)
2370                                 goto next_entry2;
2371                         if (ipmr_fill_mroute(mrt, skb,
2372                                              NETLINK_CB(cb->skb).portid,
2373                                              cb->nlh->nlmsg_seq,
2374                                              mfc, RTM_NEWROUTE,
2375                                              NLM_F_MULTI) < 0) {
2376                                 spin_unlock_bh(&mfc_unres_lock);
2377                                 goto done;
2378                         }
2379 next_entry2:
2380                         e++;
2381                 }
2382                 spin_unlock_bh(&mfc_unres_lock);
2383                 e = 0;
2384                 s_e = 0;
2385 next_table:
2386                 t++;
2387         }
2388 done:
2389         rcu_read_unlock();
2390
2391         cb->args[1] = e;
2392         cb->args[0] = t;
2393
2394         return skb->len;
2395 }
2396
2397 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2398         [RTA_SRC]       = { .type = NLA_U32 },
2399         [RTA_DST]       = { .type = NLA_U32 },
2400         [RTA_IIF]       = { .type = NLA_U32 },
2401         [RTA_TABLE]     = { .type = NLA_U32 },
2402         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2403 };
2404
2405 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2406 {
2407         switch (rtm_protocol) {
2408         case RTPROT_STATIC:
2409         case RTPROT_MROUTED:
2410                 return true;
2411         }
2412         return false;
2413 }
2414
2415 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2416 {
2417         struct rtnexthop *rtnh = nla_data(nla);
2418         int remaining = nla_len(nla), vifi = 0;
2419
2420         while (rtnh_ok(rtnh, remaining)) {
2421                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2422                 if (++vifi == MAXVIFS)
2423                         break;
2424                 rtnh = rtnh_next(rtnh, &remaining);
2425         }
2426
2427         return remaining > 0 ? -EINVAL : vifi;
2428 }
2429
2430 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2431 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2432                             struct mfcctl *mfcc, int *mrtsock,
2433                             struct mr_table **mrtret)
2434 {
2435         struct net_device *dev = NULL;
2436         u32 tblid = RT_TABLE_DEFAULT;
2437         struct mr_table *mrt;
2438         struct nlattr *attr;
2439         struct rtmsg *rtm;
2440         int ret, rem;
2441
2442         ret = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipmr_policy,
2443                              NULL);
2444         if (ret < 0)
2445                 goto out;
2446         rtm = nlmsg_data(nlh);
2447
2448         ret = -EINVAL;
2449         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2450             rtm->rtm_type != RTN_MULTICAST ||
2451             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2452             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2453                 goto out;
2454
2455         memset(mfcc, 0, sizeof(*mfcc));
2456         mfcc->mfcc_parent = -1;
2457         ret = 0;
2458         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2459                 switch (nla_type(attr)) {
2460                 case RTA_SRC:
2461                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2462                         break;
2463                 case RTA_DST:
2464                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2465                         break;
2466                 case RTA_IIF:
2467                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2468                         if (!dev) {
2469                                 ret = -ENODEV;
2470                                 goto out;
2471                         }
2472                         break;
2473                 case RTA_MULTIPATH:
2474                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2475                                 ret = -EINVAL;
2476                                 goto out;
2477                         }
2478                         break;
2479                 case RTA_PREFSRC:
2480                         ret = 1;
2481                         break;
2482                 case RTA_TABLE:
2483                         tblid = nla_get_u32(attr);
2484                         break;
2485                 }
2486         }
2487         mrt = ipmr_get_table(net, tblid);
2488         if (!mrt) {
2489                 ret = -ENOENT;
2490                 goto out;
2491         }
2492         *mrtret = mrt;
2493         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2494         if (dev)
2495                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2496
2497 out:
2498         return ret;
2499 }
2500
2501 /* takes care of both newroute and delroute */
2502 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh)
2503 {
2504         struct net *net = sock_net(skb->sk);
2505         int ret, mrtsock, parent;
2506         struct mr_table *tbl;
2507         struct mfcctl mfcc;
2508
2509         mrtsock = 0;
2510         tbl = NULL;
2511         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl);
2512         if (ret < 0)
2513                 return ret;
2514
2515         parent = ret ? mfcc.mfcc_parent : -1;
2516         if (nlh->nlmsg_type == RTM_NEWROUTE)
2517                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2518         else
2519                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2520 }
2521
2522 #ifdef CONFIG_PROC_FS
2523 /* The /proc interfaces to multicast routing :
2524  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2525  */
2526 struct ipmr_vif_iter {
2527         struct seq_net_private p;
2528         struct mr_table *mrt;
2529         int ct;
2530 };
2531
2532 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2533                                            struct ipmr_vif_iter *iter,
2534                                            loff_t pos)
2535 {
2536         struct mr_table *mrt = iter->mrt;
2537
2538         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2539                 if (!VIF_EXISTS(mrt, iter->ct))
2540                         continue;
2541                 if (pos-- == 0)
2542                         return &mrt->vif_table[iter->ct];
2543         }
2544         return NULL;
2545 }
2546
2547 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2548         __acquires(mrt_lock)
2549 {
2550         struct ipmr_vif_iter *iter = seq->private;
2551         struct net *net = seq_file_net(seq);
2552         struct mr_table *mrt;
2553
2554         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2555         if (!mrt)
2556                 return ERR_PTR(-ENOENT);
2557
2558         iter->mrt = mrt;
2559
2560         read_lock(&mrt_lock);
2561         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2562                 : SEQ_START_TOKEN;
2563 }
2564
2565 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2566 {
2567         struct ipmr_vif_iter *iter = seq->private;
2568         struct net *net = seq_file_net(seq);
2569         struct mr_table *mrt = iter->mrt;
2570
2571         ++*pos;
2572         if (v == SEQ_START_TOKEN)
2573                 return ipmr_vif_seq_idx(net, iter, 0);
2574
2575         while (++iter->ct < mrt->maxvif) {
2576                 if (!VIF_EXISTS(mrt, iter->ct))
2577                         continue;
2578                 return &mrt->vif_table[iter->ct];
2579         }
2580         return NULL;
2581 }
2582
2583 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2584         __releases(mrt_lock)
2585 {
2586         read_unlock(&mrt_lock);
2587 }
2588
2589 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2590 {
2591         struct ipmr_vif_iter *iter = seq->private;
2592         struct mr_table *mrt = iter->mrt;
2593
2594         if (v == SEQ_START_TOKEN) {
2595                 seq_puts(seq,
2596                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2597         } else {
2598                 const struct vif_device *vif = v;
2599                 const char *name =  vif->dev ? vif->dev->name : "none";
2600
2601                 seq_printf(seq,
2602                            "%2zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2603                            vif - mrt->vif_table,
2604                            name, vif->bytes_in, vif->pkt_in,
2605                            vif->bytes_out, vif->pkt_out,
2606                            vif->flags, vif->local, vif->remote);
2607         }
2608         return 0;
2609 }
2610
2611 static const struct seq_operations ipmr_vif_seq_ops = {
2612         .start = ipmr_vif_seq_start,
2613         .next  = ipmr_vif_seq_next,
2614         .stop  = ipmr_vif_seq_stop,
2615         .show  = ipmr_vif_seq_show,
2616 };
2617
2618 static int ipmr_vif_open(struct inode *inode, struct file *file)
2619 {
2620         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2621                             sizeof(struct ipmr_vif_iter));
2622 }
2623
2624 static const struct file_operations ipmr_vif_fops = {
2625         .owner   = THIS_MODULE,
2626         .open    = ipmr_vif_open,
2627         .read    = seq_read,
2628         .llseek  = seq_lseek,
2629         .release = seq_release_net,
2630 };
2631
2632 struct ipmr_mfc_iter {
2633         struct seq_net_private p;
2634         struct mr_table *mrt;
2635         struct list_head *cache;
2636 };
2637
2638 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2639                                           struct ipmr_mfc_iter *it, loff_t pos)
2640 {
2641         struct mr_table *mrt = it->mrt;
2642         struct mfc_cache *mfc;
2643
2644         rcu_read_lock();
2645         it->cache = &mrt->mfc_cache_list;
2646         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_list, list)
2647                 if (pos-- == 0)
2648                         return mfc;
2649         rcu_read_unlock();
2650
2651         spin_lock_bh(&mfc_unres_lock);
2652         it->cache = &mrt->mfc_unres_queue;
2653         list_for_each_entry(mfc, it->cache, list)
2654                 if (pos-- == 0)
2655                         return mfc;
2656         spin_unlock_bh(&mfc_unres_lock);
2657
2658         it->cache = NULL;
2659         return NULL;
2660 }
2661
2662
2663 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2664 {
2665         struct ipmr_mfc_iter *it = seq->private;
2666         struct net *net = seq_file_net(seq);
2667         struct mr_table *mrt;
2668
2669         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2670         if (!mrt)
2671                 return ERR_PTR(-ENOENT);
2672
2673         it->mrt = mrt;
2674         it->cache = NULL;
2675         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2676                 : SEQ_START_TOKEN;
2677 }
2678
2679 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2680 {
2681         struct ipmr_mfc_iter *it = seq->private;
2682         struct net *net = seq_file_net(seq);
2683         struct mr_table *mrt = it->mrt;
2684         struct mfc_cache *mfc = v;
2685
2686         ++*pos;
2687
2688         if (v == SEQ_START_TOKEN)
2689                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2690
2691         if (mfc->list.next != it->cache)
2692                 return list_entry(mfc->list.next, struct mfc_cache, list);
2693
2694         if (it->cache == &mrt->mfc_unres_queue)
2695                 goto end_of_list;
2696
2697         /* exhausted cache_array, show unresolved */
2698         rcu_read_unlock();
2699         it->cache = &mrt->mfc_unres_queue;
2700
2701         spin_lock_bh(&mfc_unres_lock);
2702         if (!list_empty(it->cache))
2703                 return list_first_entry(it->cache, struct mfc_cache, list);
2704
2705 end_of_list:
2706         spin_unlock_bh(&mfc_unres_lock);
2707         it->cache = NULL;
2708
2709         return NULL;
2710 }
2711
2712 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2713 {
2714         struct ipmr_mfc_iter *it = seq->private;
2715         struct mr_table *mrt = it->mrt;
2716
2717         if (it->cache == &mrt->mfc_unres_queue)
2718                 spin_unlock_bh(&mfc_unres_lock);
2719         else if (it->cache == &mrt->mfc_cache_list)
2720                 rcu_read_unlock();
2721 }
2722
2723 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2724 {
2725         int n;
2726
2727         if (v == SEQ_START_TOKEN) {
2728                 seq_puts(seq,
2729                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2730         } else {
2731                 const struct mfc_cache *mfc = v;
2732                 const struct ipmr_mfc_iter *it = seq->private;
2733                 const struct mr_table *mrt = it->mrt;
2734
2735                 seq_printf(seq, "%08X %08X %-3hd",
2736                            (__force u32) mfc->mfc_mcastgrp,
2737                            (__force u32) mfc->mfc_origin,
2738                            mfc->mfc_parent);
2739
2740                 if (it->cache != &mrt->mfc_unres_queue) {
2741                         seq_printf(seq, " %8lu %8lu %8lu",
2742                                    mfc->mfc_un.res.pkt,
2743                                    mfc->mfc_un.res.bytes,
2744                                    mfc->mfc_un.res.wrong_if);
2745                         for (n = mfc->mfc_un.res.minvif;
2746                              n < mfc->mfc_un.res.maxvif; n++) {
2747                                 if (VIF_EXISTS(mrt, n) &&
2748                                     mfc->mfc_un.res.ttls[n] < 255)
2749                                         seq_printf(seq,
2750                                            " %2d:%-3d",
2751                                            n, mfc->mfc_un.res.ttls[n]);
2752                         }
2753                 } else {
2754                         /* unresolved mfc_caches don't contain
2755                          * pkt, bytes and wrong_if values
2756                          */
2757                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2758                 }
2759                 seq_putc(seq, '\n');
2760         }
2761         return 0;
2762 }
2763
2764 static const struct seq_operations ipmr_mfc_seq_ops = {
2765         .start = ipmr_mfc_seq_start,
2766         .next  = ipmr_mfc_seq_next,
2767         .stop  = ipmr_mfc_seq_stop,
2768         .show  = ipmr_mfc_seq_show,
2769 };
2770
2771 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2772 {
2773         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2774                             sizeof(struct ipmr_mfc_iter));
2775 }
2776
2777 static const struct file_operations ipmr_mfc_fops = {
2778         .owner   = THIS_MODULE,
2779         .open    = ipmr_mfc_open,
2780         .read    = seq_read,
2781         .llseek  = seq_lseek,
2782         .release = seq_release_net,
2783 };
2784 #endif
2785
2786 #ifdef CONFIG_IP_PIMSM_V2
2787 static const struct net_protocol pim_protocol = {
2788         .handler        =       pim_rcv,
2789         .netns_ok       =       1,
2790 };
2791 #endif
2792
2793 /* Setup for IP multicast routing */
2794 static int __net_init ipmr_net_init(struct net *net)
2795 {
2796         int err;
2797
2798         err = ipmr_rules_init(net);
2799         if (err < 0)
2800                 goto fail;
2801
2802 #ifdef CONFIG_PROC_FS
2803         err = -ENOMEM;
2804         if (!proc_create("ip_mr_vif", 0, net->proc_net, &ipmr_vif_fops))
2805                 goto proc_vif_fail;
2806         if (!proc_create("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_fops))
2807                 goto proc_cache_fail;
2808 #endif
2809         return 0;
2810
2811 #ifdef CONFIG_PROC_FS
2812 proc_cache_fail:
2813         remove_proc_entry("ip_mr_vif", net->proc_net);
2814 proc_vif_fail:
2815         ipmr_rules_exit(net);
2816 #endif
2817 fail:
2818         return err;
2819 }
2820
2821 static void __net_exit ipmr_net_exit(struct net *net)
2822 {
2823 #ifdef CONFIG_PROC_FS
2824         remove_proc_entry("ip_mr_cache", net->proc_net);
2825         remove_proc_entry("ip_mr_vif", net->proc_net);
2826 #endif
2827         ipmr_rules_exit(net);
2828 }
2829
2830 static struct pernet_operations ipmr_net_ops = {
2831         .init = ipmr_net_init,
2832         .exit = ipmr_net_exit,
2833 };
2834
2835 int __init ip_mr_init(void)
2836 {
2837         int err;
2838
2839         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2840                                        sizeof(struct mfc_cache),
2841                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2842                                        NULL);
2843
2844         err = register_pernet_subsys(&ipmr_net_ops);
2845         if (err)
2846                 goto reg_pernet_fail;
2847
2848         err = register_netdevice_notifier(&ip_mr_notifier);
2849         if (err)
2850                 goto reg_notif_fail;
2851 #ifdef CONFIG_IP_PIMSM_V2
2852         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2853                 pr_err("%s: can't add PIM protocol\n", __func__);
2854                 err = -EAGAIN;
2855                 goto add_proto_fail;
2856         }
2857 #endif
2858         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
2859                       NULL, ipmr_rtm_dumproute, NULL);
2860         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
2861                       ipmr_rtm_route, NULL, NULL);
2862         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
2863                       ipmr_rtm_route, NULL, NULL);
2864         return 0;
2865
2866 #ifdef CONFIG_IP_PIMSM_V2
2867 add_proto_fail:
2868         unregister_netdevice_notifier(&ip_mr_notifier);
2869 #endif
2870 reg_notif_fail:
2871         unregister_pernet_subsys(&ipmr_net_ops);
2872 reg_pernet_fail:
2873         kmem_cache_destroy(mrt_cachep);
2874         return err;
2875 }