net/ipv4/ipmr.c

   1 /*
   2  *      IP multicast routing support for mrouted 3.6/3.8
   3  *
   4  *              (c) 1995 Alan Cox, <alan@lxorguk.ukuu.org.uk>
   5  *        Linux Consultancy and Custom Driver Development
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  *
  12  *      Fixes:
  13  *      Michael Chastain        :       Incorrect size of copying.
  14  *      Alan Cox                :       Added the cache manager code
  15  *      Alan Cox                :       Fixed the clone/copy bug and device race.
  16  *      Mike McLagan            :       Routing by source
  17  *      Malcolm Beattie         :       Buffer handling fixes.
  18  *      Alexey Kuznetsov        :       Double buffer free and other fixes.
  19  *      SVR Anand               :       Fixed several multicast bugs and problems.
  20  *      Alexey Kuznetsov        :       Status, optimisations and more.
  21  *      Brad Parker             :       Better behaviour on mrouted upcall
  22  *                                      overflow.
  23  *      Carlos Picoto           :       PIMv1 Support
  24  *      Pavlin Ivanov Radoslavov:       PIMv2 Registers must checksum only PIM header
  25  *                                      Relax this requirement to work with older peers.
  26  *
  27  */
  28
  29 #include <asm/system.h>
  30 #include <asm/uaccess.h>
  31 #include <linux/types.h>
  32 #include <linux/capability.h>
  33 #include <linux/errno.h>
  34 #include <linux/timer.h>
  35 #include <linux/mm.h>
  36 #include <linux/kernel.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/stat.h>
  39 #include <linux/socket.h>
  40 #include <linux/in.h>
  41 #include <linux/inet.h>
  42 #include <linux/netdevice.h>
  43 #include <linux/inetdevice.h>
  44 #include <linux/igmp.h>
  45 #include <linux/proc_fs.h>
  46 #include <linux/seq_file.h>
  47 #include <linux/mroute.h>
  48 #include <linux/init.h>
  49 #include <linux/if_ether.h>
  50 #include <linux/slab.h>
  51 #include <net/net_namespace.h>
  52 #include <net/ip.h>
  53 #include <net/protocol.h>
  54 #include <linux/skbuff.h>
  55 #include <net/route.h>
  56 #include <net/sock.h>
  57 #include <net/icmp.h>
  58 #include <net/udp.h>
  59 #include <net/raw.h>
  60 #include <linux/notifier.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/netfilter_ipv4.h>
  63 #include <linux/compat.h>
  64 #include <linux/export.h>
  65 #include <net/ipip.h>
  66 #include <net/checksum.h>
  67 #include <net/netlink.h>
  68 #include <net/fib_rules.h>
  69
  70 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
  71 #define CONFIG_IP_PIMSM 1
  72 #endif
  73
  74 struct mr_table {
  75         struct list_head        list;
  76 #ifdef CONFIG_NET_NS
  77         struct net              *net;
  78 #endif
  79         u32                     id;
  80         struct sock __rcu       *mroute_sk;
  81         struct timer_list       ipmr_expire_timer;
  82         struct list_head        mfc_unres_queue;
  83         struct list_head        mfc_cache_array[MFC_LINES];
  84         struct vif_device       vif_table[MAXVIFS];
  85         int                     maxvif;
  86         atomic_t                cache_resolve_queue_len;
  87         int                     mroute_do_assert;
  88         int                     mroute_do_pim;
  89 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
  90         int                     mroute_reg_vif_num;
  91 #endif
  92 };
  93
  94 struct ipmr_rule {
  95         struct fib_rule         common;
  96 };
  97
  98 struct ipmr_result {
  99         struct mr_table         *mrt;
 100 };
 101
 102 /* Big lock, protecting vif table, mrt cache and mroute socket state.
 103  * Note that the changes are semaphored via rtnl_lock.
 104  */
 105
 106 static DEFINE_RWLOCK(mrt_lock);
 107
 108 /*
 109  *      Multicast router control variables
 110  */
 111
 112 #define VIF_EXISTS(_mrt, _idx) ((_mrt)->vif_table[_idx].dev != NULL)
 113
 114 /* Special spinlock for queue of unresolved entries */
 115 static DEFINE_SPINLOCK(mfc_unres_lock);
 116
 117 /* We return to original Alan's scheme. Hash table of resolved
 118  * entries is changed only in process context and protected
 119  * with weak lock mrt_lock. Queue of unresolved entries is protected
 120  * with strong spinlock mfc_unres_lock.
 121  *
 122  * In this case data path is free of exclusive locks at all.
 123  */
 124
 125 static struct kmem_cache *mrt_cachep __read_mostly;
 126
 127 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 128 static void ipmr_free_table(struct mr_table *mrt);
 129
 130 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
 131                          struct sk_buff *skb, struct mfc_cache *cache,
 132                          int local);
 133 static int ipmr_cache_report(struct mr_table *mrt,
 134                              struct sk_buff *pkt, vifi_t vifi, int assert);
 135 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
 136                               struct mfc_cache *c, struct rtmsg *rtm);
 137 static void mroute_clean_tables(struct mr_table *mrt, bool all);
 138 static void ipmr_expire_process(unsigned long arg);
 139
 140 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 141 #define ipmr_for_each_table(mrt, net) \
 142         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
 143
 144 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 145 {
 146         struct mr_table *mrt;
 147
 148         ipmr_for_each_table(mrt, net) {
 149                 if (mrt->id == id)
 150                         return mrt;
 151         }
 152         return NULL;
 153 }
 154
 155 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 156                            struct mr_table **mrt)
 157 {
 158         int err;
 159         struct ipmr_result res;
 160         struct fib_lookup_arg arg = {
 161                 .result = &res,
 162                 .flags = FIB_LOOKUP_NOREF,
 163         };
 164
 165         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 166                                flowi4_to_flowi(flp4), 0, &arg);
 167         if (err < 0)
 168                 return err;
 169         *mrt = res.mrt;
 170         return 0;
 171 }
 172
 173 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 174                             int flags, struct fib_lookup_arg *arg)
 175 {
 176         struct ipmr_result *res = arg->result;
 177         struct mr_table *mrt;
 178
 179         switch (rule->action) {
 180         case FR_ACT_TO_TBL:
 181                 break;
 182         case FR_ACT_UNREACHABLE:
 183                 return -ENETUNREACH;
 184         case FR_ACT_PROHIBIT:
 185                 return -EACCES;
 186         case FR_ACT_BLACKHOLE:
 187         default:
 188                 return -EINVAL;
 189         }
 190
 191         mrt = ipmr_get_table(rule->fr_net, rule->table);
 192         if (mrt == NULL)
 193                 return -EAGAIN;
 194         res->mrt = mrt;
 195         return 0;
 196 }
 197
 198 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 199 {
 200         return 1;
 201 }
 202
 203 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 204         FRA_GENERIC_POLICY,
 205 };
 206
 207 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 208                                struct fib_rule_hdr *frh, struct nlattr **tb)
 209 {
 210         return 0;
 211 }
 212
 213 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 214                              struct nlattr **tb)
 215 {
 216         return 1;
 217 }
 218
 219 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 220                           struct fib_rule_hdr *frh)
 221 {
 222         frh->dst_len = 0;
 223         frh->src_len = 0;
 224         frh->tos     = 0;
 225         return 0;
 226 }
 227
 228 static const struct fib_rules_ops __net_initdata ipmr_rules_ops_template = {
 229         .family         = RTNL_FAMILY_IPMR,
 230         .rule_size      = sizeof(struct ipmr_rule),
 231         .addr_size      = sizeof(u32),
 232         .action         = ipmr_rule_action,
 233         .match          = ipmr_rule_match,
 234         .configure      = ipmr_rule_configure,
 235         .compare        = ipmr_rule_compare,
 236         .default_pref   = fib_default_rule_pref,
 237         .fill           = ipmr_rule_fill,
 238         .nlgroup        = RTNLGRP_IPV4_RULE,
 239         .policy         = ipmr_rule_policy,
 240         .owner          = THIS_MODULE,
 241 };
 242
 243 static int __net_init ipmr_rules_init(struct net *net)
 244 {
 245         struct fib_rules_ops *ops;
 246         struct mr_table *mrt;
 247         int err;
 248
 249         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 250         if (IS_ERR(ops))
 251                 return PTR_ERR(ops);
 252
 253         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 254
 255         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 256         if (mrt == NULL) {
 257                 err = -ENOMEM;
 258                 goto err1;
 259         }
 260
 261         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 262         if (err < 0)
 263                 goto err2;
 264
 265         net->ipv4.mr_rules_ops = ops;
 266         return 0;
 267
 268 err2:
 269         kfree(mrt);
 270 err1:
 271         fib_rules_unregister(ops);
 272         return err;
 273 }
 274
 275 static void __net_exit ipmr_rules_exit(struct net *net)
 276 {
 277         struct mr_table *mrt, *next;
 278
 279         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 280                 list_del(&mrt->list);
 281                 ipmr_free_table(mrt);
 282         }
 283         fib_rules_unregister(net->ipv4.mr_rules_ops);
 284 }
 285 #else
 286 #define ipmr_for_each_table(mrt, net) \
 287         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 288
 289 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 290 {
 291         return net->ipv4.mrt;
 292 }
 293
 294 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 295                            struct mr_table **mrt)
 296 {
 297         *mrt = net->ipv4.mrt;
 298         return 0;
 299 }
 300
 301 static int __net_init ipmr_rules_init(struct net *net)
 302 {
 303         net->ipv4.mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 304         return net->ipv4.mrt ? 0 : -ENOMEM;
 305 }
 306
 307 static void __net_exit ipmr_rules_exit(struct net *net)
 308 {
 309         ipmr_free_table(net->ipv4.mrt);
 310 }
 311 #endif
 312
 313 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 314 {
 315         struct mr_table *mrt;
 316         unsigned int i;
 317
 318         mrt = ipmr_get_table(net, id);
 319         if (mrt != NULL)
 320                 return mrt;
 321
 322         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
 323         if (mrt == NULL)
 324                 return NULL;
 325         write_pnet(&mrt->net, net);
 326         mrt->id = id;
 327
 328         /* Forwarding cache */
 329         for (i = 0; i < MFC_LINES; i++)
 330                 INIT_LIST_HEAD(&mrt->mfc_cache_array[i]);
 331
 332         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
 333
 334         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
 335                     (unsigned long)mrt);
 336
 337 #ifdef CONFIG_IP_PIMSM
 338         mrt->mroute_reg_vif_num = -1;
 339 #endif
 340 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 341         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 342 #endif
 343         return mrt;
 344 }
 345
 346 static void ipmr_free_table(struct mr_table *mrt)
 347 {
 348         del_timer_sync(&mrt->ipmr_expire_timer);
 349         mroute_clean_tables(mrt, true);
 350         kfree(mrt);
 351 }
 352
 353 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 354
 355 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 356 {
 357         struct net *net = dev_net(dev);
 358
 359         dev_close(dev);
 360
 361         dev = __dev_get_by_name(net, "tunl0");
 362         if (dev) {
 363                 const struct net_device_ops *ops = dev->netdev_ops;
 364                 struct ifreq ifr;
 365                 struct ip_tunnel_parm p;
 366
 367                 memset(&p, 0, sizeof(p));
 368                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 369                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 370                 p.iph.version = 4;
 371                 p.iph.ihl = 5;
 372                 p.iph.protocol = IPPROTO_IPIP;
 373                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 374                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 375
 376                 if (ops->ndo_do_ioctl) {
 377                         mm_segment_t oldfs = get_fs();
 378
 379                         set_fs(KERNEL_DS);
 380                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 381                         set_fs(oldfs);
 382                 }
 383         }
 384 }
 385
 386 static
 387 struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 388 {
 389         struct net_device  *dev;
 390
 391         dev = __dev_get_by_name(net, "tunl0");
 392
 393         if (dev) {
 394                 const struct net_device_ops *ops = dev->netdev_ops;
 395                 int err;
 396                 struct ifreq ifr;
 397                 struct ip_tunnel_parm p;
 398                 struct in_device  *in_dev;
 399
 400                 memset(&p, 0, sizeof(p));
 401                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 402                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 403                 p.iph.version = 4;
 404                 p.iph.ihl = 5;
 405                 p.iph.protocol = IPPROTO_IPIP;
 406                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 407                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 408
 409                 if (ops->ndo_do_ioctl) {
 410                         mm_segment_t oldfs = get_fs();
 411
 412                         set_fs(KERNEL_DS);
 413                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 414                         set_fs(oldfs);
 415                 } else {
 416                         err = -EOPNOTSUPP;
 417                 }
 418                 dev = NULL;
 419
 420                 if (err == 0 &&
 421                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 422                         dev->flags |= IFF_MULTICAST;
 423
 424                         in_dev = __in_dev_get_rtnl(dev);
 425                         if (in_dev == NULL)
 426                                 goto failure;
 427
 428                         ipv4_devconf_setall(in_dev);
 429                         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 430
 431                         if (dev_open(dev))
 432                                 goto failure;
 433                         dev_hold(dev);
 434                 }
 435         }
 436         return dev;
 437
 438 failure:
 439         /* allow the register to be completed before unregistering. */
 440         rtnl_unlock();
 441         rtnl_lock();
 442
 443         unregister_netdevice(dev);
 444         return NULL;
 445 }
 446
 447 #ifdef CONFIG_IP_PIMSM
 448
 449 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 450 {
 451         struct net *net = dev_net(dev);
 452         struct mr_table *mrt;
 453         struct flowi4 fl4 = {
 454                 .flowi4_oif     = dev->ifindex,
 455                 .flowi4_iif     = skb->skb_iif,
 456                 .flowi4_mark    = skb->mark,
 457         };
 458         int err;
 459
 460         err = ipmr_fib_lookup(net, &fl4, &mrt);
 461         if (err < 0) {
 462                 kfree_skb(skb);
 463                 return err;
 464         }
 465
 466         read_lock(&mrt_lock);
 467         dev->stats.tx_bytes += skb->len;
 468         dev->stats.tx_packets++;
 469         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 470         read_unlock(&mrt_lock);
 471         kfree_skb(skb);
 472         return NETDEV_TX_OK;
 473 }
 474
 475 static const struct net_device_ops reg_vif_netdev_ops = {
 476         .ndo_start_xmit = reg_vif_xmit,
 477 };
 478
 479 static void reg_vif_setup(struct net_device *dev)
 480 {
 481         dev->type               = ARPHRD_PIMREG;
 482         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 483         dev->flags              = IFF_NOARP;
 484         dev->netdev_ops         = &reg_vif_netdev_ops,
 485         dev->destructor         = free_netdev;
 486         dev->features           |= NETIF_F_NETNS_LOCAL;
 487 }
 488
 489 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 490 {
 491         struct net_device *dev;
 492         struct in_device *in_dev;
 493         char name[IFNAMSIZ];
 494
 495         if (mrt->id == RT_TABLE_DEFAULT)
 496                 sprintf(name, "pimreg");
 497         else
 498                 sprintf(name, "pimreg%u", mrt->id);
 499
 500         dev = alloc_netdev(0, name, reg_vif_setup);
 501
 502         if (dev == NULL)
 503                 return NULL;
 504
 505         dev_net_set(dev, net);
 506
 507         if (register_netdevice(dev)) {
 508                 free_netdev(dev);
 509                 return NULL;
 510         }
 511         dev->iflink = 0;
 512
 513         rcu_read_lock();
 514         in_dev = __in_dev_get_rcu(dev);
 515         if (!in_dev) {
 516                 rcu_read_unlock();
 517                 goto failure;
 518         }
 519
 520         ipv4_devconf_setall(in_dev);
 521         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 522         rcu_read_unlock();
 523
 524         if (dev_open(dev))
 525                 goto failure;
 526
 527         dev_hold(dev);
 528
 529         return dev;
 530
 531 failure:
 532         /* allow the register to be completed before unregistering. */
 533         rtnl_unlock();
 534         rtnl_lock();
 535
 536         unregister_netdevice(dev);
 537         return NULL;
 538 }
 539 #endif
 540
 541 /*
 542  *      Delete a VIF entry
 543  *      @notify: Set to 1, if the caller is a notifier_call
 544  */
 545
 546 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 547                       struct list_head *head)
 548 {
 549         struct vif_device *v;
 550         struct net_device *dev;
 551         struct in_device *in_dev;
 552
 553         if (vifi < 0 || vifi >= mrt->maxvif)
 554                 return -EADDRNOTAVAIL;
 555
 556         v = &mrt->vif_table[vifi];
 557
 558         write_lock_bh(&mrt_lock);
 559         dev = v->dev;
 560         v->dev = NULL;
 561
 562         if (!dev) {
 563                 write_unlock_bh(&mrt_lock);
 564                 return -EADDRNOTAVAIL;
 565         }
 566
 567 #ifdef CONFIG_IP_PIMSM
 568         if (vifi == mrt->mroute_reg_vif_num)
 569                 mrt->mroute_reg_vif_num = -1;
 570 #endif
 571
 572         if (vifi + 1 == mrt->maxvif) {
 573                 int tmp;
 574
 575                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 576                         if (VIF_EXISTS(mrt, tmp))
 577                                 break;
 578                 }
 579                 mrt->maxvif = tmp+1;
 580         }
 581
 582         write_unlock_bh(&mrt_lock);
 583
 584         dev_set_allmulti(dev, -1);
 585
 586         in_dev = __in_dev_get_rtnl(dev);
 587         if (in_dev) {
 588                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 589                 ip_rt_multicast_event(in_dev);
 590         }
 591
 592         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 593                 unregister_netdevice_queue(dev, head);
 594
 595         dev_put(dev);
 596         return 0;
 597 }
 598
 599 static void ipmr_cache_free_rcu(struct rcu_head *head)
 600 {
 601         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
 602
 603         kmem_cache_free(mrt_cachep, c);
 604 }
 605
 606 static inline void ipmr_cache_free(struct mfc_cache *c)
 607 {
 608         call_rcu(&c->rcu, ipmr_cache_free_rcu);
 609 }
 610
 611 /* Destroy an unresolved cache entry, killing queued skbs
 612  * and reporting error to netlink readers.
 613  */
 614
 615 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 616 {
 617         struct net *net = read_pnet(&mrt->net);
 618         struct sk_buff *skb;
 619         struct nlmsgerr *e;
 620
 621         atomic_dec(&mrt->cache_resolve_queue_len);
 622
 623         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
 624                 if (ip_hdr(skb)->version == 0) {
 625                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 626                         nlh->nlmsg_type = NLMSG_ERROR;
 627                         nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
 628                         skb_trim(skb, nlh->nlmsg_len);
 629                         e = NLMSG_DATA(nlh);
 630                         e->error = -ETIMEDOUT;
 631                         memset(&e->msg, 0, sizeof(e->msg));
 632
 633                         rtnl_unicast(skb, net, NETLINK_CB(skb).pid);
 634                 } else {
 635                         kfree_skb(skb);
 636                 }
 637         }
 638
 639         ipmr_cache_free(c);
 640 }
 641
 642
 643 /* Timer process for the unresolved queue. */
 644
 645 static void ipmr_expire_process(unsigned long arg)
 646 {
 647         struct mr_table *mrt = (struct mr_table *)arg;
 648         unsigned long now;
 649         unsigned long expires;
 650         struct mfc_cache *c, *next;
 651
 652         if (!spin_trylock(&mfc_unres_lock)) {
 653                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 654                 return;
 655         }
 656
 657         if (list_empty(&mrt->mfc_unres_queue))
 658                 goto out;
 659
 660         now = jiffies;
 661         expires = 10*HZ;
 662
 663         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 664                 if (time_after(c->mfc_un.unres.expires, now)) {
 665                         unsigned long interval = c->mfc_un.unres.expires - now;
 666                         if (interval < expires)
 667                                 expires = interval;
 668                         continue;
 669                 }
 670
 671                 list_del(&c->list);
 672                 ipmr_destroy_unres(mrt, c);
 673         }
 674
 675         if (!list_empty(&mrt->mfc_unres_queue))
 676                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 677
 678 out:
 679         spin_unlock(&mfc_unres_lock);
 680 }
 681
 682 /* Fill oifs list. It is called under write locked mrt_lock. */
 683
 684 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
 685                                    unsigned char *ttls)
 686 {
 687         int vifi;
 688
 689         cache->mfc_un.res.minvif = MAXVIFS;
 690         cache->mfc_un.res.maxvif = 0;
 691         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 692
 693         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 694                 if (VIF_EXISTS(mrt, vifi) &&
 695                     ttls[vifi] && ttls[vifi] < 255) {
 696                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 697                         if (cache->mfc_un.res.minvif > vifi)
 698                                 cache->mfc_un.res.minvif = vifi;
 699                         if (cache->mfc_un.res.maxvif <= vifi)
 700                                 cache->mfc_un.res.maxvif = vifi + 1;
 701                 }
 702         }
 703 }
 704
 705 static int vif_add(struct net *net, struct mr_table *mrt,
 706                    struct vifctl *vifc, int mrtsock)
 707 {
 708         int vifi = vifc->vifc_vifi;
 709         struct vif_device *v = &mrt->vif_table[vifi];
 710         struct net_device *dev;
 711         struct in_device *in_dev;
 712         int err;
 713
 714         /* Is vif busy ? */
 715         if (VIF_EXISTS(mrt, vifi))
 716                 return -EADDRINUSE;
 717
 718         switch (vifc->vifc_flags) {
 719 #ifdef CONFIG_IP_PIMSM
 720         case VIFF_REGISTER:
 721                 /*
 722                  * Special Purpose VIF in PIM
 723                  * All the packets will be sent to the daemon
 724                  */
 725                 if (mrt->mroute_reg_vif_num >= 0)
 726                         return -EADDRINUSE;
 727                 dev = ipmr_reg_vif(net, mrt);
 728                 if (!dev)
 729                         return -ENOBUFS;
 730                 err = dev_set_allmulti(dev, 1);
 731                 if (err) {
 732                         unregister_netdevice(dev);
 733                         dev_put(dev);
 734                         return err;
 735                 }
 736                 break;
 737 #endif
 738         case VIFF_TUNNEL:
 739                 dev = ipmr_new_tunnel(net, vifc);
 740                 if (!dev)
 741                         return -ENOBUFS;
 742                 err = dev_set_allmulti(dev, 1);
 743                 if (err) {
 744                         ipmr_del_tunnel(dev, vifc);
 745                         dev_put(dev);
 746                         return err;
 747                 }
 748                 break;
 749
 750         case VIFF_USE_IFINDEX:
 751         case 0:
 752                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 753                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 754                         if (dev && __in_dev_get_rtnl(dev) == NULL) {
 755                                 dev_put(dev);
 756                                 return -EADDRNOTAVAIL;
 757                         }
 758                 } else {
 759                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 760                 }
 761                 if (!dev)
 762                         return -EADDRNOTAVAIL;
 763                 err = dev_set_allmulti(dev, 1);
 764                 if (err) {
 765                         dev_put(dev);
 766                         return err;
 767                 }
 768                 break;
 769         default:
 770                 return -EINVAL;
 771         }
 772
 773         in_dev = __in_dev_get_rtnl(dev);
 774         if (!in_dev) {
 775                 dev_put(dev);
 776                 return -EADDRNOTAVAIL;
 777         }
 778         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 779         ip_rt_multicast_event(in_dev);
 780
 781         /* Fill in the VIF structures */
 782
 783         v->rate_limit = vifc->vifc_rate_limit;
 784         v->local = vifc->vifc_lcl_addr.s_addr;
 785         v->remote = vifc->vifc_rmt_addr.s_addr;
 786         v->flags = vifc->vifc_flags;
 787         if (!mrtsock)
 788                 v->flags |= VIFF_STATIC;
 789         v->threshold = vifc->vifc_threshold;
 790         v->bytes_in = 0;
 791         v->bytes_out = 0;
 792         v->pkt_in = 0;
 793         v->pkt_out = 0;
 794         v->link = dev->ifindex;
 795         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
 796                 v->link = dev->iflink;
 797
 798         /* And finish update writing critical data */
 799         write_lock_bh(&mrt_lock);
 800         v->dev = dev;
 801 #ifdef CONFIG_IP_PIMSM
 802         if (v->flags & VIFF_REGISTER)
 803                 mrt->mroute_reg_vif_num = vifi;
 804 #endif
 805         if (vifi+1 > mrt->maxvif)
 806                 mrt->maxvif = vifi+1;
 807         write_unlock_bh(&mrt_lock);
 808         return 0;
 809 }
 810
 811 /* called with rcu_read_lock() */
 812 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 813                                          __be32 origin,
 814                                          __be32 mcastgrp)
 815 {
 816         int line = MFC_HASH(mcastgrp, origin);
 817         struct mfc_cache *c;
 818
 819         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
 820                 if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
 821                         return c;
 822         }
 823         return NULL;
 824 }
 825
 826 /*
 827  *      Allocate a multicast cache entry
 828  */
 829 static struct mfc_cache *ipmr_cache_alloc(void)
 830 {
 831         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 832
 833         if (c)
 834                 c->mfc_un.res.minvif = MAXVIFS;
 835         return c;
 836 }
 837
 838 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 839 {
 840         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 841
 842         if (c) {
 843                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
 844                 c->mfc_un.unres.expires = jiffies + 10*HZ;
 845         }
 846         return c;
 847 }
 848
 849 /*
 850  *      A cache entry has gone into a resolved state from queued
 851  */
 852
 853 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 854                                struct mfc_cache *uc, struct mfc_cache *c)
 855 {
 856         struct sk_buff *skb;
 857         struct nlmsgerr *e;
 858
 859         /* Play the pending entries through our router */
 860
 861         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
 862                 if (ip_hdr(skb)->version == 0) {
 863                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 864
 865                         if (__ipmr_fill_mroute(mrt, skb, c, NLMSG_DATA(nlh)) > 0) {
 866                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 867                                                  (u8 *)nlh;
 868                         } else {
 869                                 nlh->nlmsg_type = NLMSG_ERROR;
 870                                 nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
 871                                 skb_trim(skb, nlh->nlmsg_len);
 872                                 e = NLMSG_DATA(nlh);
 873                                 e->error = -EMSGSIZE;
 874                                 memset(&e->msg, 0, sizeof(e->msg));
 875                         }
 876
 877                         rtnl_unicast(skb, net, NETLINK_CB(skb).pid);
 878                 } else {
 879                         ip_mr_forward(net, mrt, skb, c, 0);
 880                 }
 881         }
 882 }
 883
 884 /*
 885  *      Bounce a cache query up to mrouted. We could use netlink for this but mrouted
 886  *      expects the following bizarre scheme.
 887  *
 888  *      Called under mrt_lock.
 889  */
 890
 891 static int ipmr_cache_report(struct mr_table *mrt,
 892                              struct sk_buff *pkt, vifi_t vifi, int assert)
 893 {
 894         struct sk_buff *skb;
 895         const int ihl = ip_hdrlen(pkt);
 896         struct igmphdr *igmp;
 897         struct igmpmsg *msg;
 898         struct sock *mroute_sk;
 899         int ret;
 900
 901 #ifdef CONFIG_IP_PIMSM
 902         if (assert == IGMPMSG_WHOLEPKT)
 903                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
 904         else
 905 #endif
 906                 skb = alloc_skb(128, GFP_ATOMIC);
 907
 908         if (!skb)
 909                 return -ENOBUFS;
 910
 911 #ifdef CONFIG_IP_PIMSM
 912         if (assert == IGMPMSG_WHOLEPKT) {
 913                 /* Ugly, but we have no choice with this interface.
 914                  * Duplicate old header, fix ihl, length etc.
 915                  * And all this only to mangle msg->im_msgtype and
 916                  * to set msg->im_mbz to "mbz" :-)
 917                  */
 918                 skb_push(skb, sizeof(struct iphdr));
 919                 skb_reset_network_header(skb);
 920                 skb_reset_transport_header(skb);
 921                 msg = (struct igmpmsg *)skb_network_header(skb);
 922                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
 923                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
 924                 msg->im_mbz = 0;
 925                 msg->im_vif = mrt->mroute_reg_vif_num;
 926                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
 927                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
 928                                              sizeof(struct iphdr));
 929         } else
 930 #endif
 931         {
 932
 933         /* Copy the IP header */
 934
 935         skb->network_header = skb->tail;
 936         skb_put(skb, ihl);
 937         skb_copy_to_linear_data(skb, pkt->data, ihl);
 938         ip_hdr(skb)->protocol = 0;      /* Flag to the kernel this is a route add */
 939         msg = (struct igmpmsg *)skb_network_header(skb);
 940         msg->im_vif = vifi;
 941         skb_dst_set(skb, dst_clone(skb_dst(pkt)));
 942
 943         /* Add our header */
 944
 945         igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
 946         igmp->type      =
 947         msg->im_msgtype = assert;
 948         igmp->code      = 0;
 949         ip_hdr(skb)->tot_len = htons(skb->len);         /* Fix the length */
 950         skb->transport_header = skb->network_header;
 951         }
 952
 953         rcu_read_lock();
 954         mroute_sk = rcu_dereference(mrt->mroute_sk);
 955         if (mroute_sk == NULL) {
 956                 rcu_read_unlock();
 957                 kfree_skb(skb);
 958                 return -EINVAL;
 959         }
 960
 961         /* Deliver to mrouted */
 962
 963         ret = sock_queue_rcv_skb(mroute_sk, skb);
 964         rcu_read_unlock();
 965         if (ret < 0) {
 966                 if (net_ratelimit())
 967                         printk(KERN_WARNING "mroute: pending queue full, dropping entries.\n");
 968                 kfree_skb(skb);
 969         }
 970
 971         return ret;
 972 }
 973
 974 /*
 975  *      Queue a packet for resolution. It gets locked cache entry!
 976  */
 977
 978 static int
 979 ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi, struct sk_buff *skb)
 980 {
 981         bool found = false;
 982         int err;
 983         struct mfc_cache *c;
 984         const struct iphdr *iph = ip_hdr(skb);
 985
 986         spin_lock_bh(&mfc_unres_lock);
 987         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
 988                 if (c->mfc_mcastgrp == iph->daddr &&
 989                     c->mfc_origin == iph->saddr) {
 990                         found = true;
 991                         break;
 992                 }
 993         }
 994
 995         if (!found) {
 996                 /* Create a new entry if allowable */
 997
 998                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
 999                     (c = ipmr_cache_alloc_unres()) == NULL) {
1000                         spin_unlock_bh(&mfc_unres_lock);
1001
1002                         kfree_skb(skb);
1003                         return -ENOBUFS;
1004                 }
1005
1006                 /* Fill in the new cache entry */
1007
1008                 c->mfc_parent   = -1;
1009                 c->mfc_origin   = iph->saddr;
1010                 c->mfc_mcastgrp = iph->daddr;
1011
1012                 /* Reflect first query at mrouted. */
1013
1014                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1015                 if (err < 0) {
1016                         /* If the report failed throw the cache entry
1017                            out - Brad Parker
1018                          */
1019                         spin_unlock_bh(&mfc_unres_lock);
1020
1021                         ipmr_cache_free(c);
1022                         kfree_skb(skb);
1023                         return err;
1024                 }
1025
1026                 atomic_inc(&mrt->cache_resolve_queue_len);
1027                 list_add(&c->list, &mrt->mfc_unres_queue);
1028
1029                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1030                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1031         }
1032
1033         /* See if we can append the packet */
1034
1035         if (c->mfc_un.unres.unresolved.qlen > 3) {
1036                 kfree_skb(skb);
1037                 err = -ENOBUFS;
1038         } else {
1039                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1040                 err = 0;
1041         }
1042
1043         spin_unlock_bh(&mfc_unres_lock);
1044         return err;
1045 }
1046
1047 /*
1048  *      MFC cache manipulation by user space mroute daemon
1049  */
1050
1051 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc)
1052 {
1053         int line;
1054         struct mfc_cache *c, *next;
1055
1056         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1057
1058         list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1059                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1060                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1061                         list_del_rcu(&c->list);
1062
1063                         ipmr_cache_free(c);
1064                         return 0;
1065                 }
1066         }
1067         return -ENOENT;
1068 }
1069
1070 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1071                         struct mfcctl *mfc, int mrtsock)
1072 {
1073         bool found = false;
1074         int line;
1075         struct mfc_cache *uc, *c;
1076
1077         if (mfc->mfcc_parent >= MAXVIFS)
1078                 return -ENFILE;
1079
1080         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1081
1082         list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1083                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1084                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1085                         found = true;
1086                         break;
1087                 }
1088         }
1089
1090         if (found) {
1091                 write_lock_bh(&mrt_lock);
1092                 c->mfc_parent = mfc->mfcc_parent;
1093                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1094                 if (!mrtsock)
1095                         c->mfc_flags |= MFC_STATIC;
1096                 write_unlock_bh(&mrt_lock);
1097                 return 0;
1098         }
1099
1100         if (!ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1101                 return -EINVAL;
1102
1103         c = ipmr_cache_alloc();
1104         if (c == NULL)
1105                 return -ENOMEM;
1106
1107         c->mfc_origin = mfc->mfcc_origin.s_addr;
1108         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1109         c->mfc_parent = mfc->mfcc_parent;
1110         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1111         if (!mrtsock)
1112                 c->mfc_flags |= MFC_STATIC;
1113
1114         list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1115
1116         /*
1117          *      Check to see if we resolved a queued list. If so we
1118          *      need to send on the frames and tidy up.
1119          */
1120         found = false;
1121         spin_lock_bh(&mfc_unres_lock);
1122         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1123                 if (uc->mfc_origin == c->mfc_origin &&
1124                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1125                         list_del(&uc->list);
1126                         atomic_dec(&mrt->cache_resolve_queue_len);
1127                         found = true;
1128                         break;
1129                 }
1130         }
1131         if (list_empty(&mrt->mfc_unres_queue))
1132                 del_timer(&mrt->ipmr_expire_timer);
1133         spin_unlock_bh(&mfc_unres_lock);
1134
1135         if (found) {
1136                 ipmr_cache_resolve(net, mrt, uc, c);
1137                 ipmr_cache_free(uc);
1138         }
1139         return 0;
1140 }
1141
1142 /*
1143  *      Close the multicast socket, and clear the vif tables etc
1144  */
1145
1146 static void mroute_clean_tables(struct mr_table *mrt, bool all)
1147 {
1148         int i;
1149         LIST_HEAD(list);
1150         struct mfc_cache *c, *next;
1151
1152         /* Shut down all active vif entries */
1153
1154         for (i = 0; i < mrt->maxvif; i++) {
1155                 if (!all && (mrt->vif_table[i].flags & VIFF_STATIC))
1156                         continue;
1157                 vif_delete(mrt, i, 0, &list);
1158         }
1159         unregister_netdevice_many(&list);
1160
1161         /* Wipe the cache */
1162
1163         for (i = 0; i < MFC_LINES; i++) {
1164                 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1165                         if (!all && (c->mfc_flags & MFC_STATIC))
1166                                 continue;
1167                         list_del_rcu(&c->list);
1168                         ipmr_cache_free(c);
1169                 }
1170         }
1171
1172         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1173                 spin_lock_bh(&mfc_unres_lock);
1174                 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1175                         list_del(&c->list);
1176                         ipmr_destroy_unres(mrt, c);
1177                 }
1178                 spin_unlock_bh(&mfc_unres_lock);
1179         }
1180 }
1181
1182 /* called from ip_ra_control(), before an RCU grace period,
1183  * we dont need to call synchronize_rcu() here
1184  */
1185 static void mrtsock_destruct(struct sock *sk)
1186 {
1187         struct net *net = sock_net(sk);
1188         struct mr_table *mrt;
1189
1190         rtnl_lock();
1191         ipmr_for_each_table(mrt, net) {
1192                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1193                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1194                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1195                         mroute_clean_tables(mrt, false);
1196                 }
1197         }
1198         rtnl_unlock();
1199 }
1200
1201 /*
1202  *      Socket options and virtual interface manipulation. The whole
1203  *      virtual interface system is a complete heap, but unfortunately
1204  *      that's how BSD mrouted happens to think. Maybe one day with a proper
1205  *      MOSPF/PIM router set up we can clean this up.
1206  */
1207
1208 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval, unsigned int optlen)
1209 {
1210         int ret;
1211         struct vifctl vif;
1212         struct mfcctl mfc;
1213         struct net *net = sock_net(sk);
1214         struct mr_table *mrt;
1215
1216         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1217         if (mrt == NULL)
1218                 return -ENOENT;
1219
1220         if (optname != MRT_INIT) {
1221                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1222                     !capable(CAP_NET_ADMIN))
1223                         return -EACCES;
1224         }
1225
1226         switch (optname) {
1227         case MRT_INIT:
1228                 if (sk->sk_type != SOCK_RAW ||
1229                     inet_sk(sk)->inet_num != IPPROTO_IGMP)
1230                         return -EOPNOTSUPP;
1231                 if (optlen != sizeof(int))
1232                         return -ENOPROTOOPT;
1233
1234                 rtnl_lock();
1235                 if (rtnl_dereference(mrt->mroute_sk)) {
1236                         rtnl_unlock();
1237                         return -EADDRINUSE;
1238                 }
1239
1240                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1241                 if (ret == 0) {
1242                         rcu_assign_pointer(mrt->mroute_sk, sk);
1243                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1244                 }
1245                 rtnl_unlock();
1246                 return ret;
1247         case MRT_DONE:
1248                 if (sk != rcu_access_pointer(mrt->mroute_sk))
1249                         return -EACCES;
1250                 return ip_ra_control(sk, 0, NULL);
1251         case MRT_ADD_VIF:
1252         case MRT_DEL_VIF:
1253                 if (optlen != sizeof(vif))
1254                         return -EINVAL;
1255                 if (copy_from_user(&vif, optval, sizeof(vif)))
1256                         return -EFAULT;
1257                 if (vif.vifc_vifi >= MAXVIFS)
1258                         return -ENFILE;
1259                 rtnl_lock();
1260                 if (optname == MRT_ADD_VIF) {
1261                         ret = vif_add(net, mrt, &vif,
1262                                       sk == rtnl_dereference(mrt->mroute_sk));
1263                 } else {
1264                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1265                 }
1266                 rtnl_unlock();
1267                 return ret;
1268
1269                 /*
1270                  *      Manipulate the forwarding caches. These live
1271                  *      in a sort of kernel/user symbiosis.
1272                  */
1273         case MRT_ADD_MFC:
1274         case MRT_DEL_MFC:
1275                 if (optlen != sizeof(mfc))
1276                         return -EINVAL;
1277                 if (copy_from_user(&mfc, optval, sizeof(mfc)))
1278                         return -EFAULT;
1279                 rtnl_lock();
1280                 if (optname == MRT_DEL_MFC)
1281                         ret = ipmr_mfc_delete(mrt, &mfc);
1282                 else
1283                         ret = ipmr_mfc_add(net, mrt, &mfc,
1284                                            sk == rtnl_dereference(mrt->mroute_sk));
1285                 rtnl_unlock();
1286                 return ret;
1287                 /*
1288                  *      Control PIM assert.
1289                  */
1290         case MRT_ASSERT:
1291         {
1292                 int v;
1293                 if (get_user(v, (int __user *)optval))
1294                         return -EFAULT;
1295                 mrt->mroute_do_assert = (v) ? 1 : 0;
1296                 return 0;
1297         }
1298 #ifdef CONFIG_IP_PIMSM
1299         case MRT_PIM:
1300         {
1301                 int v;
1302
1303                 if (get_user(v, (int __user *)optval))
1304                         return -EFAULT;
1305                 v = (v) ? 1 : 0;
1306
1307                 rtnl_lock();
1308                 ret = 0;
1309                 if (v != mrt->mroute_do_pim) {
1310                         mrt->mroute_do_pim = v;
1311                         mrt->mroute_do_assert = v;
1312                 }
1313                 rtnl_unlock();
1314                 return ret;
1315         }
1316 #endif
1317 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
1318         case MRT_TABLE:
1319         {
1320                 u32 v;
1321
1322                 if (optlen != sizeof(u32))
1323                         return -EINVAL;
1324                 if (get_user(v, (u32 __user *)optval))
1325                         return -EFAULT;
1326
1327                 rtnl_lock();
1328                 ret = 0;
1329                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1330                         ret = -EBUSY;
1331                 } else {
1332                         if (!ipmr_new_table(net, v))
1333                                 ret = -ENOMEM;
1334                         raw_sk(sk)->ipmr_table = v;
1335                 }
1336                 rtnl_unlock();
1337                 return ret;
1338         }
1339 #endif
1340         /*
1341          *      Spurious command, or MRT_VERSION which you cannot
1342          *      set.
1343          */
1344         default:
1345                 return -ENOPROTOOPT;
1346         }
1347 }
1348
1349 /*
1350  *      Getsock opt support for the multicast routing system.
1351  */
1352
1353 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1354 {
1355         int olr;
1356         int val;
1357         struct net *net = sock_net(sk);
1358         struct mr_table *mrt;
1359
1360         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1361         if (mrt == NULL)
1362                 return -ENOENT;
1363
1364         if (optname != MRT_VERSION &&
1365 #ifdef CONFIG_IP_PIMSM
1366            optname != MRT_PIM &&
1367 #endif
1368            optname != MRT_ASSERT)
1369                 return -ENOPROTOOPT;
1370
1371         if (get_user(olr, optlen))
1372                 return -EFAULT;
1373
1374         olr = min_t(unsigned int, olr, sizeof(int));
1375         if (olr < 0)
1376                 return -EINVAL;
1377
1378         if (put_user(olr, optlen))
1379                 return -EFAULT;
1380         if (optname == MRT_VERSION)
1381                 val = 0x0305;
1382 #ifdef CONFIG_IP_PIMSM
1383         else if (optname == MRT_PIM)
1384                 val = mrt->mroute_do_pim;
1385 #endif
1386         else
1387                 val = mrt->mroute_do_assert;
1388         if (copy_to_user(optval, &val, olr))
1389                 return -EFAULT;
1390         return 0;
1391 }
1392
1393 /*
1394  *      The IP multicast ioctl support routines.
1395  */
1396
1397 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1398 {
1399         struct sioc_sg_req sr;
1400         struct sioc_vif_req vr;
1401         struct vif_device *vif;
1402         struct mfc_cache *c;
1403         struct net *net = sock_net(sk);
1404         struct mr_table *mrt;
1405
1406         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1407         if (mrt == NULL)
1408                 return -ENOENT;
1409
1410         switch (cmd) {
1411         case SIOCGETVIFCNT:
1412                 if (copy_from_user(&vr, arg, sizeof(vr)))
1413                         return -EFAULT;
1414                 if (vr.vifi >= mrt->maxvif)
1415                         return -EINVAL;
1416                 read_lock(&mrt_lock);
1417                 vif = &mrt->vif_table[vr.vifi];
1418                 if (VIF_EXISTS(mrt, vr.vifi)) {
1419                         vr.icount = vif->pkt_in;
1420                         vr.ocount = vif->pkt_out;
1421                         vr.ibytes = vif->bytes_in;
1422                         vr.obytes = vif->bytes_out;
1423                         read_unlock(&mrt_lock);
1424
1425                         if (copy_to_user(arg, &vr, sizeof(vr)))
1426                                 return -EFAULT;
1427                         return 0;
1428                 }
1429                 read_unlock(&mrt_lock);
1430                 return -EADDRNOTAVAIL;
1431         case SIOCGETSGCNT:
1432                 if (copy_from_user(&sr, arg, sizeof(sr)))
1433                         return -EFAULT;
1434
1435                 rcu_read_lock();
1436                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1437                 if (c) {
1438                         sr.pktcnt = c->mfc_un.res.pkt;
1439                         sr.bytecnt = c->mfc_un.res.bytes;
1440                         sr.wrong_if = c->mfc_un.res.wrong_if;
1441                         rcu_read_unlock();
1442
1443                         if (copy_to_user(arg, &sr, sizeof(sr)))
1444                                 return -EFAULT;
1445                         return 0;
1446                 }
1447                 rcu_read_unlock();
1448                 return -EADDRNOTAVAIL;
1449         default:
1450                 return -ENOIOCTLCMD;
1451         }
1452 }
1453
1454 #ifdef CONFIG_COMPAT
1455 struct compat_sioc_sg_req {
1456         struct in_addr src;
1457         struct in_addr grp;
1458         compat_ulong_t pktcnt;
1459         compat_ulong_t bytecnt;
1460         compat_ulong_t wrong_if;
1461 };
1462
1463 struct compat_sioc_vif_req {
1464         vifi_t  vifi;           /* Which iface */
1465         compat_ulong_t icount;
1466         compat_ulong_t ocount;
1467         compat_ulong_t ibytes;
1468         compat_ulong_t obytes;
1469 };
1470
1471 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1472 {
1473         struct compat_sioc_sg_req sr;
1474         struct compat_sioc_vif_req vr;
1475         struct vif_device *vif;
1476         struct mfc_cache *c;
1477         struct net *net = sock_net(sk);
1478         struct mr_table *mrt;
1479
1480         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1481         if (mrt == NULL)
1482                 return -ENOENT;
1483
1484         switch (cmd) {
1485         case SIOCGETVIFCNT:
1486                 if (copy_from_user(&vr, arg, sizeof(vr)))
1487                         return -EFAULT;
1488                 if (vr.vifi >= mrt->maxvif)
1489                         return -EINVAL;
1490                 read_lock(&mrt_lock);
1491                 vif = &mrt->vif_table[vr.vifi];
1492                 if (VIF_EXISTS(mrt, vr.vifi)) {
1493                         vr.icount = vif->pkt_in;
1494                         vr.ocount = vif->pkt_out;
1495                         vr.ibytes = vif->bytes_in;
1496                         vr.obytes = vif->bytes_out;
1497                         read_unlock(&mrt_lock);
1498
1499                         if (copy_to_user(arg, &vr, sizeof(vr)))
1500                                 return -EFAULT;
1501                         return 0;
1502                 }
1503                 read_unlock(&mrt_lock);
1504                 return -EADDRNOTAVAIL;
1505         case SIOCGETSGCNT:
1506                 if (copy_from_user(&sr, arg, sizeof(sr)))
1507                         return -EFAULT;
1508
1509                 rcu_read_lock();
1510                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1511                 if (c) {
1512                         sr.pktcnt = c->mfc_un.res.pkt;
1513                         sr.bytecnt = c->mfc_un.res.bytes;
1514                         sr.wrong_if = c->mfc_un.res.wrong_if;
1515                         rcu_read_unlock();
1516
1517                         if (copy_to_user(arg, &sr, sizeof(sr)))
1518                                 return -EFAULT;
1519                         return 0;
1520                 }
1521                 rcu_read_unlock();
1522                 return -EADDRNOTAVAIL;
1523         default:
1524                 return -ENOIOCTLCMD;
1525         }
1526 }
1527 #endif
1528
1529
1530 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1531 {
1532         struct net_device *dev = ptr;
1533         struct net *net = dev_net(dev);
1534         struct mr_table *mrt;
1535         struct vif_device *v;
1536         int ct;
1537         LIST_HEAD(list);
1538
1539         if (event != NETDEV_UNREGISTER)
1540                 return NOTIFY_DONE;
1541
1542         ipmr_for_each_table(mrt, net) {
1543                 v = &mrt->vif_table[0];
1544                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1545                         if (v->dev == dev)
1546                                 vif_delete(mrt, ct, 1, &list);
1547                 }
1548         }
1549         unregister_netdevice_many(&list);
1550         return NOTIFY_DONE;
1551 }
1552
1553
1554 static struct notifier_block ip_mr_notifier = {
1555         .notifier_call = ipmr_device_event,
1556 };
1557
1558 /*
1559  *      Encapsulate a packet by attaching a valid IPIP header to it.
1560  *      This avoids tunnel drivers and other mess and gives us the speed so
1561  *      important for multicast video.
1562  */
1563
1564 static void ip_encap(struct sk_buff *skb, __be32 saddr, __be32 daddr)
1565 {
1566         struct iphdr *iph;
1567         const struct iphdr *old_iph = ip_hdr(skb);
1568
1569         skb_push(skb, sizeof(struct iphdr));
1570         skb->transport_header = skb->network_header;
1571         skb_reset_network_header(skb);
1572         iph = ip_hdr(skb);
1573
1574         iph->version    =       4;
1575         iph->tos        =       old_iph->tos;
1576         iph->ttl        =       old_iph->ttl;
1577         iph->frag_off   =       0;
1578         iph->daddr      =       daddr;
1579         iph->saddr      =       saddr;
1580         iph->protocol   =       IPPROTO_IPIP;
1581         iph->ihl        =       5;
1582         iph->tot_len    =       htons(skb->len);
1583         ip_select_ident(skb, NULL);
1584         ip_send_check(iph);
1585
1586         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1587         nf_reset(skb);
1588 }
1589
1590 static inline int ipmr_forward_finish(struct sk_buff *skb)
1591 {
1592         struct ip_options *opt = &(IPCB(skb)->opt);
1593
1594         IP_INC_STATS(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTFORWDATAGRAMS);
1595
1596         if (unlikely(opt->optlen))
1597                 ip_forward_options(skb);
1598
1599         return dst_output(skb);
1600 }
1601
1602 /*
1603  *      Processing handlers for ipmr_forward
1604  */
1605
1606 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1607                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1608 {
1609         const struct iphdr *iph = ip_hdr(skb);
1610         struct vif_device *vif = &mrt->vif_table[vifi];
1611         struct net_device *dev;
1612         struct rtable *rt;
1613         struct flowi4 fl4;
1614         int    encap = 0;
1615
1616         if (vif->dev == NULL)
1617                 goto out_free;
1618
1619 #ifdef CONFIG_IP_PIMSM
1620         if (vif->flags & VIFF_REGISTER) {
1621                 vif->pkt_out++;
1622                 vif->bytes_out += skb->len;
1623                 vif->dev->stats.tx_bytes += skb->len;
1624                 vif->dev->stats.tx_packets++;
1625                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1626                 goto out_free;
1627         }
1628 #endif
1629
1630         if (vif->flags & VIFF_TUNNEL) {
1631                 rt = ip_route_output_ports(net, &fl4, NULL,
1632                                            vif->remote, vif->local,
1633                                            0, 0,
1634                                            IPPROTO_IPIP,
1635                                            RT_TOS(iph->tos), vif->link);
1636                 if (IS_ERR(rt))
1637                         goto out_free;
1638                 encap = sizeof(struct iphdr);
1639         } else {
1640                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1641                                            0, 0,
1642                                            IPPROTO_IPIP,
1643                                            RT_TOS(iph->tos), vif->link);
1644                 if (IS_ERR(rt))
1645                         goto out_free;
1646         }
1647
1648         dev = rt->dst.dev;
1649
1650         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1651                 /* Do not fragment multicasts. Alas, IPv4 does not
1652                  * allow to send ICMP, so that packets will disappear
1653                  * to blackhole.
1654                  */
1655
1656                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
1657                 ip_rt_put(rt);
1658                 goto out_free;
1659         }
1660
1661         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1662
1663         if (skb_cow(skb, encap)) {
1664                 ip_rt_put(rt);
1665                 goto out_free;
1666         }
1667
1668         vif->pkt_out++;
1669         vif->bytes_out += skb->len;
1670
1671         skb_dst_drop(skb);
1672         skb_dst_set(skb, &rt->dst);
1673         ip_decrease_ttl(ip_hdr(skb));
1674
1675         /* FIXME: forward and output firewalls used to be called here.
1676          * What do we do with netfilter? -- RR
1677          */
1678         if (vif->flags & VIFF_TUNNEL) {
1679                 ip_encap(skb, vif->local, vif->remote);
1680                 /* FIXME: extra output firewall step used to be here. --RR */
1681                 vif->dev->stats.tx_packets++;
1682                 vif->dev->stats.tx_bytes += skb->len;
1683         }
1684
1685         IPCB(skb)->flags |= IPSKB_FORWARDED;
1686
1687         /*
1688          * RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1689          * not only before forwarding, but after forwarding on all output
1690          * interfaces. It is clear, if mrouter runs a multicasting
1691          * program, it should receive packets not depending to what interface
1692          * program is joined.
1693          * If we will not make it, the program will have to join on all
1694          * interfaces. On the other hand, multihoming host (or router, but
1695          * not mrouter) cannot join to more than one interface - it will
1696          * result in receiving multiple packets.
1697          */
1698         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD, skb, skb->dev, dev,
1699                 ipmr_forward_finish);
1700         return;
1701
1702 out_free:
1703         kfree_skb(skb);
1704 }
1705
1706 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1707 {
1708         int ct;
1709
1710         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1711                 if (mrt->vif_table[ct].dev == dev)
1712                         break;
1713         }
1714         return ct;
1715 }
1716
1717 /* "local" means that we should preserve one skb (for local delivery) */
1718
1719 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
1720                          struct sk_buff *skb, struct mfc_cache *cache,
1721                          int local)
1722 {
1723         int psend = -1;
1724         int vif, ct;
1725
1726         vif = cache->mfc_parent;
1727         cache->mfc_un.res.pkt++;
1728         cache->mfc_un.res.bytes += skb->len;
1729
1730         /*
1731          * Wrong interface: drop packet and (maybe) send PIM assert.
1732          */
1733         if (mrt->vif_table[vif].dev != skb->dev) {
1734                 int true_vifi;
1735
1736                 if (rt_is_output_route(skb_rtable(skb))) {
1737                         /* It is our own packet, looped back.
1738                          * Very complicated situation...
1739                          *
1740                          * The best workaround until routing daemons will be
1741                          * fixed is not to redistribute packet, if it was
1742                          * send through wrong interface. It means, that
1743                          * multicast applications WILL NOT work for
1744                          * (S,G), which have default multicast route pointing
1745                          * to wrong oif. In any case, it is not a good
1746                          * idea to use multicasting applications on router.
1747                          */
1748                         goto dont_forward;
1749                 }
1750
1751                 cache->mfc_un.res.wrong_if++;
1752                 true_vifi = ipmr_find_vif(mrt, skb->dev);
1753
1754                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1755                     /* pimsm uses asserts, when switching from RPT to SPT,
1756                      * so that we cannot check that packet arrived on an oif.
1757                      * It is bad, but otherwise we would need to move pretty
1758                      * large chunk of pimd to kernel. Ough... --ANK
1759                      */
1760                     (mrt->mroute_do_pim ||
1761                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1762                     time_after(jiffies,
1763                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1764                         cache->mfc_un.res.last_assert = jiffies;
1765                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1766                 }
1767                 goto dont_forward;
1768         }
1769
1770         mrt->vif_table[vif].pkt_in++;
1771         mrt->vif_table[vif].bytes_in += skb->len;
1772
1773         /*
1774          *      Forward the frame
1775          */
1776         for (ct = cache->mfc_un.res.maxvif - 1;
1777              ct >= cache->mfc_un.res.minvif; ct--) {
1778                 if (ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1779                         if (psend != -1) {
1780                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1781
1782                                 if (skb2)
1783                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1784                                                         psend);
1785                         }
1786                         psend = ct;
1787                 }
1788         }
1789         if (psend != -1) {
1790                 if (local) {
1791                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1792
1793                         if (skb2)
1794                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1795                 } else {
1796                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1797                         return 0;
1798                 }
1799         }
1800
1801 dont_forward:
1802         if (!local)
1803                 kfree_skb(skb);
1804         return 0;
1805 }
1806
1807 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1808 {
1809         struct rtable *rt = skb_rtable(skb);
1810         struct iphdr *iph = ip_hdr(skb);
1811         struct flowi4 fl4 = {
1812                 .daddr = iph->daddr,
1813                 .saddr = iph->saddr,
1814                 .flowi4_tos = RT_TOS(iph->tos),
1815                 .flowi4_oif = rt->rt_oif,
1816                 .flowi4_iif = rt->rt_iif,
1817                 .flowi4_mark = rt->rt_mark,
1818         };
1819         struct mr_table *mrt;
1820         int err;
1821
1822         err = ipmr_fib_lookup(net, &fl4, &mrt);
1823         if (err)
1824                 return ERR_PTR(err);
1825         return mrt;
1826 }
1827
1828 /*
1829  *      Multicast packets for forwarding arrive here
1830  *      Called with rcu_read_lock();
1831  */
1832
1833 int ip_mr_input(struct sk_buff *skb)
1834 {
1835         struct mfc_cache *cache;
1836         struct net *net = dev_net(skb->dev);
1837         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1838         struct mr_table *mrt;
1839
1840         /* Packet is looped back after forward, it should not be
1841          * forwarded second time, but still can be delivered locally.
1842          */
1843         if (IPCB(skb)->flags & IPSKB_FORWARDED)
1844                 goto dont_forward;
1845
1846         mrt = ipmr_rt_fib_lookup(net, skb);
1847         if (IS_ERR(mrt)) {
1848                 kfree_skb(skb);
1849                 return PTR_ERR(mrt);
1850         }
1851         if (!local) {
1852                 if (IPCB(skb)->opt.router_alert) {
1853                         if (ip_call_ra_chain(skb))
1854                                 return 0;
1855                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1856                         /* IGMPv1 (and broken IGMPv2 implementations sort of
1857                          * Cisco IOS <= 11.2(8)) do not put router alert
1858                          * option to IGMP packets destined to routable
1859                          * groups. It is very bad, because it means
1860                          * that we can forward NO IGMP messages.
1861                          */
1862                         struct sock *mroute_sk;
1863
1864                         mroute_sk = rcu_dereference(mrt->mroute_sk);
1865                         if (mroute_sk) {
1866                                 nf_reset(skb);
1867                                 raw_rcv(mroute_sk, skb);
1868                                 return 0;
1869                         }
1870                     }
1871         }
1872
1873         /* already under rcu_read_lock() */
1874         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1875
1876         /*
1877          *      No usable cache entry
1878          */
1879         if (cache == NULL) {
1880                 int vif;
1881
1882                 if (local) {
1883                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1884                         ip_local_deliver(skb);
1885                         if (skb2 == NULL)
1886                                 return -ENOBUFS;
1887                         skb = skb2;
1888                 }
1889
1890                 read_lock(&mrt_lock);
1891                 vif = ipmr_find_vif(mrt, skb->dev);
1892                 if (vif >= 0) {
1893                         int err2 = ipmr_cache_unresolved(mrt, vif, skb);
1894                         read_unlock(&mrt_lock);
1895
1896                         return err2;
1897                 }
1898                 read_unlock(&mrt_lock);
1899                 kfree_skb(skb);
1900                 return -ENODEV;
1901         }
1902
1903         read_lock(&mrt_lock);
1904         ip_mr_forward(net, mrt, skb, cache, local);
1905         read_unlock(&mrt_lock);
1906
1907         if (local)
1908                 return ip_local_deliver(skb);
1909
1910         return 0;
1911
1912 dont_forward:
1913         if (local)
1914                 return ip_local_deliver(skb);
1915         kfree_skb(skb);
1916         return 0;
1917 }
1918
1919 #ifdef CONFIG_IP_PIMSM
1920 /* called with rcu_read_lock() */
1921 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
1922                      unsigned int pimlen)
1923 {
1924         struct net_device *reg_dev = NULL;
1925         struct iphdr *encap;
1926
1927         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
1928         /*
1929          * Check that:
1930          * a. packet is really sent to a multicast group
1931          * b. packet is not a NULL-REGISTER
1932          * c. packet is not truncated
1933          */
1934         if (!ipv4_is_multicast(encap->daddr) ||
1935             encap->tot_len == 0 ||
1936             ntohs(encap->tot_len) + pimlen > skb->len)
1937                 return 1;
1938
1939         read_lock(&mrt_lock);
1940         if (mrt->mroute_reg_vif_num >= 0)
1941                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
1942         read_unlock(&mrt_lock);
1943
1944         if (reg_dev == NULL)
1945                 return 1;
1946
1947         skb->mac_header = skb->network_header;
1948         skb_pull(skb, (u8 *)encap - skb->data);
1949         skb_reset_network_header(skb);
1950         skb->protocol = htons(ETH_P_IP);
1951         skb->ip_summed = CHECKSUM_NONE;
1952         skb->pkt_type = PACKET_HOST;
1953
1954         skb_tunnel_rx(skb, reg_dev);
1955
1956         netif_rx(skb);
1957
1958         return NET_RX_SUCCESS;
1959 }
1960 #endif
1961
1962 #ifdef CONFIG_IP_PIMSM_V1
1963 /*
1964  * Handle IGMP messages of PIMv1
1965  */
1966
1967 int pim_rcv_v1(struct sk_buff *skb)
1968 {
1969         struct igmphdr *pim;
1970         struct net *net = dev_net(skb->dev);
1971         struct mr_table *mrt;
1972
1973         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
1974                 goto drop;
1975
1976         pim = igmp_hdr(skb);
1977
1978         mrt = ipmr_rt_fib_lookup(net, skb);
1979         if (IS_ERR(mrt))
1980                 goto drop;
1981         if (!mrt->mroute_do_pim ||
1982             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
1983                 goto drop;
1984
1985         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
1986 drop:
1987                 kfree_skb(skb);
1988         }
1989         return 0;
1990 }
1991 #endif
1992
1993 #ifdef CONFIG_IP_PIMSM_V2
1994 static int pim_rcv(struct sk_buff *skb)
1995 {
1996         struct pimreghdr *pim;
1997         struct net *net = dev_net(skb->dev);
1998         struct mr_table *mrt;
1999
2000         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2001                 goto drop;
2002
2003         pim = (struct pimreghdr *)skb_transport_header(skb);
2004         if (pim->type != ((PIM_VERSION << 4) | (PIM_REGISTER)) ||
2005             (pim->flags & PIM_NULL_REGISTER) ||
2006             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2007              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2008                 goto drop;
2009
2010         mrt = ipmr_rt_fib_lookup(net, skb);
2011         if (IS_ERR(mrt))
2012                 goto drop;
2013         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2014 drop:
2015                 kfree_skb(skb);
2016         }
2017         return 0;
2018 }
2019 #endif
2020
2021 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2022                               struct mfc_cache *c, struct rtmsg *rtm)
2023 {
2024         int ct;
2025         struct rtnexthop *nhp;
2026         u8 *b = skb_tail_pointer(skb);
2027         struct rtattr *mp_head;
2028
2029         /* If cache is unresolved, don't try to parse IIF and OIF */
2030         if (c->mfc_parent >= MAXVIFS)
2031                 return -ENOENT;
2032
2033         if (VIF_EXISTS(mrt, c->mfc_parent))
2034                 RTA_PUT(skb, RTA_IIF, 4, &mrt->vif_table[c->mfc_parent].dev->ifindex);
2035
2036         mp_head = (struct rtattr *)skb_put(skb, RTA_LENGTH(0));
2037
2038         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2039                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2040                         if (skb_tailroom(skb) < RTA_ALIGN(RTA_ALIGN(sizeof(*nhp)) + 4))
2041                                 goto rtattr_failure;
2042                         nhp = (struct rtnexthop *)skb_put(skb, RTA_ALIGN(sizeof(*nhp)));
2043                         nhp->rtnh_flags = 0;
2044                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2045                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2046                         nhp->rtnh_len = sizeof(*nhp);
2047                 }
2048         }
2049         mp_head->rta_type = RTA_MULTIPATH;
2050         mp_head->rta_len = skb_tail_pointer(skb) - (u8 *)mp_head;
2051         rtm->rtm_type = RTN_MULTICAST;
2052         return 1;
2053
2054 rtattr_failure:
2055         nlmsg_trim(skb, b);
2056         return -EMSGSIZE;
2057 }
2058
2059 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2060                    __be32 saddr, __be32 daddr,
2061                    struct rtmsg *rtm, int nowait)
2062 {
2063         struct mfc_cache *cache;
2064         struct mr_table *mrt;
2065         int err;
2066
2067         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2068         if (mrt == NULL)
2069                 return -ENOENT;
2070
2071         rcu_read_lock();
2072         cache = ipmr_cache_find(mrt, saddr, daddr);
2073
2074         if (cache == NULL) {
2075                 struct sk_buff *skb2;
2076                 struct iphdr *iph;
2077                 struct net_device *dev;
2078                 int vif = -1;
2079
2080                 if (nowait) {
2081                         rcu_read_unlock();
2082                         return -EAGAIN;
2083                 }
2084
2085                 dev = skb->dev;
2086                 read_lock(&mrt_lock);
2087                 if (dev)
2088                         vif = ipmr_find_vif(mrt, dev);
2089                 if (vif < 0) {
2090                         read_unlock(&mrt_lock);
2091                         rcu_read_unlock();
2092                         return -ENODEV;
2093                 }
2094                 skb2 = skb_clone(skb, GFP_ATOMIC);
2095                 if (!skb2) {
2096                         read_unlock(&mrt_lock);
2097                         rcu_read_unlock();
2098                         return -ENOMEM;
2099                 }
2100
2101                 skb_push(skb2, sizeof(struct iphdr));
2102                 skb_reset_network_header(skb2);
2103                 iph = ip_hdr(skb2);
2104                 iph->ihl = sizeof(struct iphdr) >> 2;
2105                 iph->saddr = saddr;
2106                 iph->daddr = daddr;
2107                 iph->version = 0;
2108                 err = ipmr_cache_unresolved(mrt, vif, skb2);
2109                 read_unlock(&mrt_lock);
2110                 rcu_read_unlock();
2111                 return err;
2112         }
2113
2114         read_lock(&mrt_lock);
2115         if (!nowait && (rtm->rtm_flags & RTM_F_NOTIFY))
2116                 cache->mfc_flags |= MFC_NOTIFY;
2117         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2118         read_unlock(&mrt_lock);
2119         rcu_read_unlock();
2120         return err;
2121 }
2122
2123 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2124                             u32 pid, u32 seq, struct mfc_cache *c)
2125 {
2126         struct nlmsghdr *nlh;
2127         struct rtmsg *rtm;
2128
2129         nlh = nlmsg_put(skb, pid, seq, RTM_NEWROUTE, sizeof(*rtm), NLM_F_MULTI);
2130         if (nlh == NULL)
2131                 return -EMSGSIZE;
2132
2133         rtm = nlmsg_data(nlh);
2134         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2135         rtm->rtm_dst_len  = 32;
2136         rtm->rtm_src_len  = 32;
2137         rtm->rtm_tos      = 0;
2138         rtm->rtm_table    = mrt->id;
2139         NLA_PUT_U32(skb, RTA_TABLE, mrt->id);
2140         rtm->rtm_type     = RTN_MULTICAST;
2141         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2142         rtm->rtm_protocol = RTPROT_UNSPEC;
2143         rtm->rtm_flags    = 0;
2144
2145         NLA_PUT_BE32(skb, RTA_SRC, c->mfc_origin);
2146         NLA_PUT_BE32(skb, RTA_DST, c->mfc_mcastgrp);
2147
2148         if (__ipmr_fill_mroute(mrt, skb, c, rtm) < 0)
2149                 goto nla_put_failure;
2150
2151         return nlmsg_end(skb, nlh);
2152
2153 nla_put_failure:
2154         nlmsg_cancel(skb, nlh);
2155         return -EMSGSIZE;
2156 }
2157
2158 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2159 {
2160         struct net *net = sock_net(skb->sk);
2161         struct mr_table *mrt;
2162         struct mfc_cache *mfc;
2163         unsigned int t = 0, s_t;
2164         unsigned int h = 0, s_h;
2165         unsigned int e = 0, s_e;
2166
2167         s_t = cb->args[0];
2168         s_h = cb->args[1];
2169         s_e = cb->args[2];
2170
2171         rcu_read_lock();
2172         ipmr_for_each_table(mrt, net) {
2173                 if (t < s_t)
2174                         goto next_table;
2175                 if (t > s_t)
2176                         s_h = 0;
2177                 for (h = s_h; h < MFC_LINES; h++) {
2178                         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2179                                 if (e < s_e)
2180                                         goto next_entry;
2181                                 if (ipmr_fill_mroute(mrt, skb,
2182                                                      NETLINK_CB(cb->skb).pid,
2183                                                      cb->nlh->nlmsg_seq,
2184                                                      mfc) < 0)
2185                                         goto done;
2186 next_entry:
2187                                 e++;
2188                         }
2189                         e = s_e = 0;
2190                 }
2191                 s_h = 0;
2192 next_table:
2193                 t++;
2194         }
2195 done:
2196         rcu_read_unlock();
2197
2198         cb->args[2] = e;
2199         cb->args[1] = h;
2200         cb->args[0] = t;
2201
2202         return skb->len;
2203 }
2204
2205 #ifdef CONFIG_PROC_FS
2206 /*
2207  *      The /proc interfaces to multicast routing :
2208  *      /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2209  */
2210 struct ipmr_vif_iter {
2211         struct seq_net_private p;
2212         struct mr_table *mrt;
2213         int ct;
2214 };
2215
2216 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2217                                            struct ipmr_vif_iter *iter,
2218                                            loff_t pos)
2219 {
2220         struct mr_table *mrt = iter->mrt;
2221
2222         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2223                 if (!VIF_EXISTS(mrt, iter->ct))
2224                         continue;
2225                 if (pos-- == 0)
2226                         return &mrt->vif_table[iter->ct];
2227         }
2228         return NULL;
2229 }
2230
2231 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2232         __acquires(mrt_lock)
2233 {
2234         struct ipmr_vif_iter *iter = seq->private;
2235         struct net *net = seq_file_net(seq);
2236         struct mr_table *mrt;
2237
2238         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2239         if (mrt == NULL)
2240                 return ERR_PTR(-ENOENT);
2241
2242         iter->mrt = mrt;
2243
2244         read_lock(&mrt_lock);
2245         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2246                 : SEQ_START_TOKEN;
2247 }
2248
2249 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2250 {
2251         struct ipmr_vif_iter *iter = seq->private;
2252         struct net *net = seq_file_net(seq);
2253         struct mr_table *mrt = iter->mrt;
2254
2255         ++*pos;
2256         if (v == SEQ_START_TOKEN)
2257                 return ipmr_vif_seq_idx(net, iter, 0);
2258
2259         while (++iter->ct < mrt->maxvif) {
2260                 if (!VIF_EXISTS(mrt, iter->ct))
2261                         continue;
2262                 return &mrt->vif_table[iter->ct];
2263         }
2264         return NULL;
2265 }
2266
2267 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2268         __releases(mrt_lock)
2269 {
2270         read_unlock(&mrt_lock);
2271 }
2272
2273 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2274 {
2275         struct ipmr_vif_iter *iter = seq->private;
2276         struct mr_table *mrt = iter->mrt;
2277
2278         if (v == SEQ_START_TOKEN) {
2279                 seq_puts(seq,
2280                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2281         } else {
2282                 const struct vif_device *vif = v;
2283                 const char *name =  vif->dev ? vif->dev->name : "none";
2284
2285                 seq_printf(seq,
2286                            "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2287                            vif - mrt->vif_table,
2288                            name, vif->bytes_in, vif->pkt_in,
2289                            vif->bytes_out, vif->pkt_out,
2290                            vif->flags, vif->local, vif->remote);
2291         }
2292         return 0;
2293 }
2294
2295 static const struct seq_operations ipmr_vif_seq_ops = {
2296         .start = ipmr_vif_seq_start,
2297         .next  = ipmr_vif_seq_next,
2298         .stop  = ipmr_vif_seq_stop,
2299         .show  = ipmr_vif_seq_show,
2300 };
2301
2302 static int ipmr_vif_open(struct inode *inode, struct file *file)
2303 {
2304         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2305                             sizeof(struct ipmr_vif_iter));
2306 }
2307
2308 static const struct file_operations ipmr_vif_fops = {
2309         .owner   = THIS_MODULE,
2310         .open    = ipmr_vif_open,
2311         .read    = seq_read,
2312         .llseek  = seq_lseek,
2313         .release = seq_release_net,
2314 };
2315
2316 struct ipmr_mfc_iter {
2317         struct seq_net_private p;
2318         struct mr_table *mrt;
2319         struct list_head *cache;
2320         int ct;
2321 };
2322
2323
2324 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2325                                           struct ipmr_mfc_iter *it, loff_t pos)
2326 {
2327         struct mr_table *mrt = it->mrt;
2328         struct mfc_cache *mfc;
2329
2330         rcu_read_lock();
2331         for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2332                 it->cache = &mrt->mfc_cache_array[it->ct];
2333                 list_for_each_entry_rcu(mfc, it->cache, list)
2334                         if (pos-- == 0)
2335                                 return mfc;
2336         }
2337         rcu_read_unlock();
2338
2339         spin_lock_bh(&mfc_unres_lock);
2340         it->cache = &mrt->mfc_unres_queue;
2341         list_for_each_entry(mfc, it->cache, list)
2342                 if (pos-- == 0)
2343                         return mfc;
2344         spin_unlock_bh(&mfc_unres_lock);
2345
2346         it->cache = NULL;
2347         return NULL;
2348 }
2349
2350
2351 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2352 {
2353         struct ipmr_mfc_iter *it = seq->private;
2354         struct net *net = seq_file_net(seq);
2355         struct mr_table *mrt;
2356
2357         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2358         if (mrt == NULL)
2359                 return ERR_PTR(-ENOENT);
2360
2361         it->mrt = mrt;
2362         it->cache = NULL;
2363         it->ct = 0;
2364         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2365                 : SEQ_START_TOKEN;
2366 }
2367
2368 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2369 {
2370         struct mfc_cache *mfc = v;
2371         struct ipmr_mfc_iter *it = seq->private;
2372         struct net *net = seq_file_net(seq);
2373         struct mr_table *mrt = it->mrt;
2374
2375         ++*pos;
2376
2377         if (v == SEQ_START_TOKEN)
2378                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2379
2380         if (mfc->list.next != it->cache)
2381                 return list_entry(mfc->list.next, struct mfc_cache, list);
2382
2383         if (it->cache == &mrt->mfc_unres_queue)
2384                 goto end_of_list;
2385
2386         BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2387
2388         while (++it->ct < MFC_LINES) {
2389                 it->cache = &mrt->mfc_cache_array[it->ct];
2390                 if (list_empty(it->cache))
2391                         continue;
2392                 return list_first_entry(it->cache, struct mfc_cache, list);
2393         }
2394
2395         /* exhausted cache_array, show unresolved */
2396         rcu_read_unlock();
2397         it->cache = &mrt->mfc_unres_queue;
2398         it->ct = 0;
2399
2400         spin_lock_bh(&mfc_unres_lock);
2401         if (!list_empty(it->cache))
2402                 return list_first_entry(it->cache, struct mfc_cache, list);
2403
2404 end_of_list:
2405         spin_unlock_bh(&mfc_unres_lock);
2406         it->cache = NULL;
2407
2408         return NULL;
2409 }
2410
2411 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2412 {
2413         struct ipmr_mfc_iter *it = seq->private;
2414         struct mr_table *mrt = it->mrt;
2415
2416         if (it->cache == &mrt->mfc_unres_queue)
2417                 spin_unlock_bh(&mfc_unres_lock);
2418         else if (it->cache == &mrt->mfc_cache_array[it->ct])
2419                 rcu_read_unlock();
2420 }
2421
2422 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2423 {
2424         int n;
2425
2426         if (v == SEQ_START_TOKEN) {
2427                 seq_puts(seq,
2428                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2429         } else {
2430                 const struct mfc_cache *mfc = v;
2431                 const struct ipmr_mfc_iter *it = seq->private;
2432                 const struct mr_table *mrt = it->mrt;
2433
2434                 seq_printf(seq, "%08X %08X %-3hd",
2435                            (__force u32) mfc->mfc_mcastgrp,
2436                            (__force u32) mfc->mfc_origin,
2437                            mfc->mfc_parent);
2438
2439                 if (it->cache != &mrt->mfc_unres_queue) {
2440                         seq_printf(seq, " %8lu %8lu %8lu",
2441                                    mfc->mfc_un.res.pkt,
2442                                    mfc->mfc_un.res.bytes,
2443                                    mfc->mfc_un.res.wrong_if);
2444                         for (n = mfc->mfc_un.res.minvif;
2445                              n < mfc->mfc_un.res.maxvif; n++) {
2446                                 if (VIF_EXISTS(mrt, n) &&
2447                                     mfc->mfc_un.res.ttls[n] < 255)
2448                                         seq_printf(seq,
2449                                            " %2d:%-3d",
2450                                            n, mfc->mfc_un.res.ttls[n]);
2451                         }
2452                 } else {
2453                         /* unresolved mfc_caches don't contain
2454                          * pkt, bytes and wrong_if values
2455                          */
2456                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2457                 }
2458                 seq_putc(seq, '\n');
2459         }
2460         return 0;
2461 }
2462
2463 static const struct seq_operations ipmr_mfc_seq_ops = {
2464         .start = ipmr_mfc_seq_start,
2465         .next  = ipmr_mfc_seq_next,
2466         .stop  = ipmr_mfc_seq_stop,
2467         .show  = ipmr_mfc_seq_show,
2468 };
2469
2470 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2471 {
2472         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2473                             sizeof(struct ipmr_mfc_iter));
2474 }
2475
2476 static const struct file_operations ipmr_mfc_fops = {
2477         .owner   = THIS_MODULE,
2478         .open    = ipmr_mfc_open,
2479         .read    = seq_read,
2480         .llseek  = seq_lseek,
2481         .release = seq_release_net,
2482 };
2483 #endif
2484
2485 #ifdef CONFIG_IP_PIMSM_V2
2486 static const struct net_protocol pim_protocol = {
2487         .handler        =       pim_rcv,
2488         .netns_ok       =       1,
2489 };
2490 #endif
2491
2492
2493 /*
2494  *      Setup for IP multicast routing
2495  */
2496 static int __net_init ipmr_net_init(struct net *net)
2497 {
2498         int err;
2499
2500         err = ipmr_rules_init(net);
2501         if (err < 0)
2502                 goto fail;
2503
2504 #ifdef CONFIG_PROC_FS
2505         err = -ENOMEM;
2506         if (!proc_net_fops_create(net, "ip_mr_vif", 0, &ipmr_vif_fops))
2507                 goto proc_vif_fail;
2508         if (!proc_net_fops_create(net, "ip_mr_cache", 0, &ipmr_mfc_fops))
2509                 goto proc_cache_fail;
2510 #endif
2511         return 0;
2512
2513 #ifdef CONFIG_PROC_FS
2514 proc_cache_fail:
2515         proc_net_remove(net, "ip_mr_vif");
2516 proc_vif_fail:
2517         ipmr_rules_exit(net);
2518 #endif
2519 fail:
2520         return err;
2521 }
2522
2523 static void __net_exit ipmr_net_exit(struct net *net)
2524 {
2525 #ifdef CONFIG_PROC_FS
2526         proc_net_remove(net, "ip_mr_cache");
2527         proc_net_remove(net, "ip_mr_vif");
2528 #endif
2529         ipmr_rules_exit(net);
2530 }
2531
2532 static struct pernet_operations ipmr_net_ops = {
2533         .init = ipmr_net_init,
2534         .exit = ipmr_net_exit,
2535 };
2536
2537 int __init ip_mr_init(void)
2538 {
2539         int err;
2540
2541         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2542                                        sizeof(struct mfc_cache),
2543                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2544                                        NULL);
2545         if (!mrt_cachep)
2546                 return -ENOMEM;
2547
2548         err = register_pernet_subsys(&ipmr_net_ops);
2549         if (err)
2550                 goto reg_pernet_fail;
2551
2552         err = register_netdevice_notifier(&ip_mr_notifier);
2553         if (err)
2554                 goto reg_notif_fail;
2555 #ifdef CONFIG_IP_PIMSM_V2
2556         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2557                 printk(KERN_ERR "ip_mr_init: can't add PIM protocol\n");
2558                 err = -EAGAIN;
2559                 goto add_proto_fail;
2560         }
2561 #endif
2562         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
2563                       NULL, ipmr_rtm_dumproute, NULL);
2564         return 0;
2565
2566 #ifdef CONFIG_IP_PIMSM_V2
2567 add_proto_fail:
2568         unregister_netdevice_notifier(&ip_mr_notifier);
2569 #endif
2570 reg_notif_fail:
2571         unregister_pernet_subsys(&ipmr_net_ops);
2572 reg_pernet_fail:
2573         kmem_cache_destroy(mrt_cachep);
2574         return err;
2575 }