net/ipv4/ipmr.c

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