net/ipv4/ipmr.c

   1 /*
   2  *      IP multicast routing support for mrouted 3.6/3.8
   3  *
   4  *              (c) 1995 Alan Cox, <alan@lxorguk.ukuu.org.uk>
   5  *        Linux Consultancy and Custom Driver Development
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  *
  12  *      Fixes:
  13  *      Michael Chastain        :       Incorrect size of copying.
  14  *      Alan Cox                :       Added the cache manager code
  15  *      Alan Cox                :       Fixed the clone/copy bug and device race.
  16  *      Mike McLagan            :       Routing by source
  17  *      Malcolm Beattie         :       Buffer handling fixes.
  18  *      Alexey Kuznetsov        :       Double buffer free and other fixes.
  19  *      SVR Anand               :       Fixed several multicast bugs and problems.
  20  *      Alexey Kuznetsov        :       Status, optimisations and more.
  21  *      Brad Parker             :       Better behaviour on mrouted upcall
  22  *                                      overflow.
  23  *      Carlos Picoto           :       PIMv1 Support
  24  *      Pavlin Ivanov Radoslavov:       PIMv2 Registers must checksum only PIM header
  25  *                                      Relax this requirement to work with older peers.
  26  *
  27  */
  28
  29 #include <linux/uaccess.h>
  30 #include <linux/types.h>
  31 #include <linux/cache.h>
  32 #include <linux/capability.h>
  33 #include <linux/errno.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/icmp.h>
  56 #include <net/udp.h>
  57 #include <net/raw.h>
  58 #include <linux/notifier.h>
  59 #include <linux/if_arp.h>
  60 #include <linux/netfilter_ipv4.h>
  61 #include <linux/compat.h>
  62 #include <linux/export.h>
  63 #include <linux/rhashtable.h>
  64 #include <net/ip_tunnels.h>
  65 #include <net/checksum.h>
  66 #include <net/netlink.h>
  67 #include <net/fib_rules.h>
  68 #include <linux/netconf.h>
  69 #include <net/nexthop.h>
  70 #include <net/switchdev.h>
  71
  72 #include <linux/nospec.h>
  73
  74 struct ipmr_rule {
  75         struct fib_rule         common;
  76 };
  77
  78 struct ipmr_result {
  79         struct mr_table         *mrt;
  80 };
  81
  82 /* Big lock, protecting vif table, mrt cache and mroute socket state.
  83  * Note that the changes are semaphored via rtnl_lock.
  84  */
  85
  86 static DEFINE_RWLOCK(mrt_lock);
  87
  88 /* Multicast router control variables */
  89
  90 /* Special spinlock for queue of unresolved entries */
  91 static DEFINE_SPINLOCK(mfc_unres_lock);
  92
  93 /* We return to original Alan's scheme. Hash table of resolved
  94  * entries is changed only in process context and protected
  95  * with weak lock mrt_lock. Queue of unresolved entries is protected
  96  * with strong spinlock mfc_unres_lock.
  97  *
  98  * In this case data path is free of exclusive locks at all.
  99  */
 100
 101 static struct kmem_cache *mrt_cachep __ro_after_init;
 102
 103 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 104 static void ipmr_free_table(struct mr_table *mrt);
 105
 106 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
 107                           struct net_device *dev, struct sk_buff *skb,
 108                           struct mfc_cache *cache, int local);
 109 static int ipmr_cache_report(struct mr_table *mrt,
 110                              struct sk_buff *pkt, vifi_t vifi, int assert);
 111 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
 112                                  int cmd);
 113 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt);
 114 static void mroute_clean_tables(struct mr_table *mrt, bool all);
 115 static void ipmr_expire_process(struct timer_list *t);
 116
 117 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 118 #define ipmr_for_each_table(mrt, net) \
 119         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
 120
 121 static struct mr_table *ipmr_mr_table_iter(struct net *net,
 122                                            struct mr_table *mrt)
 123 {
 124         struct mr_table *ret;
 125
 126         if (!mrt)
 127                 ret = list_entry_rcu(net->ipv4.mr_tables.next,
 128                                      struct mr_table, list);
 129         else
 130                 ret = list_entry_rcu(mrt->list.next,
 131                                      struct mr_table, list);
 132
 133         if (&ret->list == &net->ipv4.mr_tables)
 134                 return NULL;
 135         return ret;
 136 }
 137
 138 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 139 {
 140         struct mr_table *mrt;
 141
 142         ipmr_for_each_table(mrt, net) {
 143                 if (mrt->id == id)
 144                         return mrt;
 145         }
 146         return NULL;
 147 }
 148
 149 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 150                            struct mr_table **mrt)
 151 {
 152         int err;
 153         struct ipmr_result res;
 154         struct fib_lookup_arg arg = {
 155                 .result = &res,
 156                 .flags = FIB_LOOKUP_NOREF,
 157         };
 158
 159         /* update flow if oif or iif point to device enslaved to l3mdev */
 160         l3mdev_update_flow(net, flowi4_to_flowi(flp4));
 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         arg->table = fib_rule_get_table(rule, arg);
 189
 190         mrt = ipmr_get_table(rule->fr_net, arg->table);
 191         if (!mrt)
 192                 return -EAGAIN;
 193         res->mrt = mrt;
 194         return 0;
 195 }
 196
 197 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 198 {
 199         return 1;
 200 }
 201
 202 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 203         FRA_GENERIC_POLICY,
 204 };
 205
 206 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 207                                struct fib_rule_hdr *frh, struct nlattr **tb,
 208                                struct netlink_ext_ack *extack)
 209 {
 210         return 0;
 211 }
 212
 213 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 214                              struct nlattr **tb)
 215 {
 216         return 1;
 217 }
 218
 219 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 220                           struct fib_rule_hdr *frh)
 221 {
 222         frh->dst_len = 0;
 223         frh->src_len = 0;
 224         frh->tos     = 0;
 225         return 0;
 226 }
 227
 228 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
 229         .family         = RTNL_FAMILY_IPMR,
 230         .rule_size      = sizeof(struct ipmr_rule),
 231         .addr_size      = sizeof(u32),
 232         .action         = ipmr_rule_action,
 233         .match          = ipmr_rule_match,
 234         .configure      = ipmr_rule_configure,
 235         .compare        = ipmr_rule_compare,
 236         .fill           = ipmr_rule_fill,
 237         .nlgroup        = RTNLGRP_IPV4_RULE,
 238         .policy         = ipmr_rule_policy,
 239         .owner          = THIS_MODULE,
 240 };
 241
 242 static int __net_init ipmr_rules_init(struct net *net)
 243 {
 244         struct fib_rules_ops *ops;
 245         struct mr_table *mrt;
 246         int err;
 247
 248         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 249         if (IS_ERR(ops))
 250                 return PTR_ERR(ops);
 251
 252         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 253
 254         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 255         if (IS_ERR(mrt)) {
 256                 err = PTR_ERR(mrt);
 257                 goto err1;
 258         }
 259
 260         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 261         if (err < 0)
 262                 goto err2;
 263
 264         net->ipv4.mr_rules_ops = ops;
 265         return 0;
 266
 267 err2:
 268         rtnl_lock();
 269         ipmr_free_table(mrt);
 270         rtnl_unlock();
 271 err1:
 272         fib_rules_unregister(ops);
 273         return err;
 274 }
 275
 276 static void __net_exit ipmr_rules_exit(struct net *net)
 277 {
 278         struct mr_table *mrt, *next;
 279
 280         rtnl_lock();
 281         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 282                 list_del(&mrt->list);
 283                 ipmr_free_table(mrt);
 284         }
 285         fib_rules_unregister(net->ipv4.mr_rules_ops);
 286         rtnl_unlock();
 287 }
 288
 289 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb)
 290 {
 291         return fib_rules_dump(net, nb, RTNL_FAMILY_IPMR);
 292 }
 293
 294 static unsigned int ipmr_rules_seq_read(struct net *net)
 295 {
 296         return fib_rules_seq_read(net, RTNL_FAMILY_IPMR);
 297 }
 298
 299 bool ipmr_rule_default(const struct fib_rule *rule)
 300 {
 301         return fib_rule_matchall(rule) && rule->table == RT_TABLE_DEFAULT;
 302 }
 303 EXPORT_SYMBOL(ipmr_rule_default);
 304 #else
 305 #define ipmr_for_each_table(mrt, net) \
 306         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 307
 308 static struct mr_table *ipmr_mr_table_iter(struct net *net,
 309                                            struct mr_table *mrt)
 310 {
 311         if (!mrt)
 312                 return net->ipv4.mrt;
 313         return NULL;
 314 }
 315
 316 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 317 {
 318         return net->ipv4.mrt;
 319 }
 320
 321 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 322                            struct mr_table **mrt)
 323 {
 324         *mrt = net->ipv4.mrt;
 325         return 0;
 326 }
 327
 328 static int __net_init ipmr_rules_init(struct net *net)
 329 {
 330         struct mr_table *mrt;
 331
 332         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 333         if (IS_ERR(mrt))
 334                 return PTR_ERR(mrt);
 335         net->ipv4.mrt = mrt;
 336         return 0;
 337 }
 338
 339 static void __net_exit ipmr_rules_exit(struct net *net)
 340 {
 341         rtnl_lock();
 342         ipmr_free_table(net->ipv4.mrt);
 343         net->ipv4.mrt = NULL;
 344         rtnl_unlock();
 345 }
 346
 347 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb)
 348 {
 349         return 0;
 350 }
 351
 352 static unsigned int ipmr_rules_seq_read(struct net *net)
 353 {
 354         return 0;
 355 }
 356
 357 bool ipmr_rule_default(const struct fib_rule *rule)
 358 {
 359         return true;
 360 }
 361 EXPORT_SYMBOL(ipmr_rule_default);
 362 #endif
 363
 364 static inline int ipmr_hash_cmp(struct rhashtable_compare_arg *arg,
 365                                 const void *ptr)
 366 {
 367         const struct mfc_cache_cmp_arg *cmparg = arg->key;
 368         struct mfc_cache *c = (struct mfc_cache *)ptr;
 369
 370         return cmparg->mfc_mcastgrp != c->mfc_mcastgrp ||
 371                cmparg->mfc_origin != c->mfc_origin;
 372 }
 373
 374 static const struct rhashtable_params ipmr_rht_params = {
 375         .head_offset = offsetof(struct mr_mfc, mnode),
 376         .key_offset = offsetof(struct mfc_cache, cmparg),
 377         .key_len = sizeof(struct mfc_cache_cmp_arg),
 378         .nelem_hint = 3,
 379         .locks_mul = 1,
 380         .obj_cmpfn = ipmr_hash_cmp,
 381         .automatic_shrinking = true,
 382 };
 383
 384 static void ipmr_new_table_set(struct mr_table *mrt,
 385                                struct net *net)
 386 {
 387 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 388         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 389 #endif
 390 }
 391
 392 static struct mfc_cache_cmp_arg ipmr_mr_table_ops_cmparg_any = {
 393         .mfc_mcastgrp = htonl(INADDR_ANY),
 394         .mfc_origin = htonl(INADDR_ANY),
 395 };
 396
 397 static struct mr_table_ops ipmr_mr_table_ops = {
 398         .rht_params = &ipmr_rht_params,
 399         .cmparg_any = &ipmr_mr_table_ops_cmparg_any,
 400 };
 401
 402 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 403 {
 404         struct mr_table *mrt;
 405
 406         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 407         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 408                 return ERR_PTR(-EINVAL);
 409
 410         mrt = ipmr_get_table(net, id);
 411         if (mrt)
 412                 return mrt;
 413
 414         return mr_table_alloc(net, id, &ipmr_mr_table_ops,
 415                               ipmr_expire_process, ipmr_new_table_set);
 416 }
 417
 418 static void ipmr_free_table(struct mr_table *mrt)
 419 {
 420         del_timer_sync(&mrt->ipmr_expire_timer);
 421         mroute_clean_tables(mrt, true);
 422         rhltable_destroy(&mrt->mfc_hash);
 423         kfree(mrt);
 424 }
 425
 426 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 427
 428 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 429 {
 430         struct net *net = dev_net(dev);
 431
 432         dev_close(dev);
 433
 434         dev = __dev_get_by_name(net, "tunl0");
 435         if (dev) {
 436                 const struct net_device_ops *ops = dev->netdev_ops;
 437                 struct ifreq ifr;
 438                 struct ip_tunnel_parm p;
 439
 440                 memset(&p, 0, sizeof(p));
 441                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 442                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 443                 p.iph.version = 4;
 444                 p.iph.ihl = 5;
 445                 p.iph.protocol = IPPROTO_IPIP;
 446                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 447                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 448
 449                 if (ops->ndo_do_ioctl) {
 450                         mm_segment_t oldfs = get_fs();
 451
 452                         set_fs(KERNEL_DS);
 453                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 454                         set_fs(oldfs);
 455                 }
 456         }
 457 }
 458
 459 /* Initialize ipmr pimreg/tunnel in_device */
 460 static bool ipmr_init_vif_indev(const struct net_device *dev)
 461 {
 462         struct in_device *in_dev;
 463
 464         ASSERT_RTNL();
 465
 466         in_dev = __in_dev_get_rtnl(dev);
 467         if (!in_dev)
 468                 return false;
 469         ipv4_devconf_setall(in_dev);
 470         neigh_parms_data_state_setall(in_dev->arp_parms);
 471         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 472
 473         return true;
 474 }
 475
 476 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 477 {
 478         struct net_device  *dev;
 479
 480         dev = __dev_get_by_name(net, "tunl0");
 481
 482         if (dev) {
 483                 const struct net_device_ops *ops = dev->netdev_ops;
 484                 int err;
 485                 struct ifreq ifr;
 486                 struct ip_tunnel_parm p;
 487
 488                 memset(&p, 0, sizeof(p));
 489                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 490                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 491                 p.iph.version = 4;
 492                 p.iph.ihl = 5;
 493                 p.iph.protocol = IPPROTO_IPIP;
 494                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 495                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 496
 497                 if (ops->ndo_do_ioctl) {
 498                         mm_segment_t oldfs = get_fs();
 499
 500                         set_fs(KERNEL_DS);
 501                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 502                         set_fs(oldfs);
 503                 } else {
 504                         err = -EOPNOTSUPP;
 505                 }
 506                 dev = NULL;
 507
 508                 if (err == 0 &&
 509                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 510                         dev->flags |= IFF_MULTICAST;
 511                         if (!ipmr_init_vif_indev(dev))
 512                                 goto failure;
 513                         if (dev_open(dev))
 514                                 goto failure;
 515                         dev_hold(dev);
 516                 }
 517         }
 518         return dev;
 519
 520 failure:
 521         unregister_netdevice(dev);
 522         return NULL;
 523 }
 524
 525 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 526 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 527 {
 528         struct net *net = dev_net(dev);
 529         struct mr_table *mrt;
 530         struct flowi4 fl4 = {
 531                 .flowi4_oif     = dev->ifindex,
 532                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 533                 .flowi4_mark    = skb->mark,
 534         };
 535         int err;
 536
 537         err = ipmr_fib_lookup(net, &fl4, &mrt);
 538         if (err < 0) {
 539                 kfree_skb(skb);
 540                 return err;
 541         }
 542
 543         read_lock(&mrt_lock);
 544         dev->stats.tx_bytes += skb->len;
 545         dev->stats.tx_packets++;
 546         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 547         read_unlock(&mrt_lock);
 548         kfree_skb(skb);
 549         return NETDEV_TX_OK;
 550 }
 551
 552 static int reg_vif_get_iflink(const struct net_device *dev)
 553 {
 554         return 0;
 555 }
 556
 557 static const struct net_device_ops reg_vif_netdev_ops = {
 558         .ndo_start_xmit = reg_vif_xmit,
 559         .ndo_get_iflink = reg_vif_get_iflink,
 560 };
 561
 562 static void reg_vif_setup(struct net_device *dev)
 563 {
 564         dev->type               = ARPHRD_PIMREG;
 565         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 566         dev->flags              = IFF_NOARP;
 567         dev->netdev_ops         = &reg_vif_netdev_ops;
 568         dev->needs_free_netdev  = true;
 569         dev->features           |= NETIF_F_NETNS_LOCAL;
 570 }
 571
 572 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 573 {
 574         struct net_device *dev;
 575         char name[IFNAMSIZ];
 576
 577         if (mrt->id == RT_TABLE_DEFAULT)
 578                 sprintf(name, "pimreg");
 579         else
 580                 sprintf(name, "pimreg%u", mrt->id);
 581
 582         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 583
 584         if (!dev)
 585                 return NULL;
 586
 587         dev_net_set(dev, net);
 588
 589         if (register_netdevice(dev)) {
 590                 free_netdev(dev);
 591                 return NULL;
 592         }
 593
 594         if (!ipmr_init_vif_indev(dev))
 595                 goto failure;
 596         if (dev_open(dev))
 597                 goto failure;
 598
 599         dev_hold(dev);
 600
 601         return dev;
 602
 603 failure:
 604         unregister_netdevice(dev);
 605         return NULL;
 606 }
 607
 608 /* called with rcu_read_lock() */
 609 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 610                      unsigned int pimlen)
 611 {
 612         struct net_device *reg_dev = NULL;
 613         struct iphdr *encap;
 614
 615         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 616         /* Check that:
 617          * a. packet is really sent to a multicast group
 618          * b. packet is not a NULL-REGISTER
 619          * c. packet is not truncated
 620          */
 621         if (!ipv4_is_multicast(encap->daddr) ||
 622             encap->tot_len == 0 ||
 623             ntohs(encap->tot_len) + pimlen > skb->len)
 624                 return 1;
 625
 626         read_lock(&mrt_lock);
 627         if (mrt->mroute_reg_vif_num >= 0)
 628                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 629         read_unlock(&mrt_lock);
 630
 631         if (!reg_dev)
 632                 return 1;
 633
 634         skb->mac_header = skb->network_header;
 635         skb_pull(skb, (u8 *)encap - skb->data);
 636         skb_reset_network_header(skb);
 637         skb->protocol = htons(ETH_P_IP);
 638         skb->ip_summed = CHECKSUM_NONE;
 639
 640         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 641
 642         netif_rx(skb);
 643
 644         return NET_RX_SUCCESS;
 645 }
 646 #else
 647 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 648 {
 649         return NULL;
 650 }
 651 #endif
 652
 653 static int call_ipmr_vif_entry_notifiers(struct net *net,
 654                                          enum fib_event_type event_type,
 655                                          struct vif_device *vif,
 656                                          vifi_t vif_index, u32 tb_id)
 657 {
 658         return mr_call_vif_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 659                                      vif, vif_index, tb_id,
 660                                      &net->ipv4.ipmr_seq);
 661 }
 662
 663 static int call_ipmr_mfc_entry_notifiers(struct net *net,
 664                                          enum fib_event_type event_type,
 665                                          struct mfc_cache *mfc, u32 tb_id)
 666 {
 667         return mr_call_mfc_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 668                                      &mfc->_c, tb_id, &net->ipv4.ipmr_seq);
 669 }
 670
 671 /**
 672  *      vif_delete - Delete a VIF entry
 673  *      @notify: Set to 1, if the caller is a notifier_call
 674  */
 675 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 676                       struct list_head *head)
 677 {
 678         struct net *net = read_pnet(&mrt->net);
 679         struct vif_device *v;
 680         struct net_device *dev;
 681         struct in_device *in_dev;
 682
 683         if (vifi < 0 || vifi >= mrt->maxvif)
 684                 return -EADDRNOTAVAIL;
 685
 686         v = &mrt->vif_table[vifi];
 687
 688         if (VIF_EXISTS(mrt, vifi))
 689                 call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_DEL, v, vifi,
 690                                               mrt->id);
 691
 692         write_lock_bh(&mrt_lock);
 693         dev = v->dev;
 694         v->dev = NULL;
 695
 696         if (!dev) {
 697                 write_unlock_bh(&mrt_lock);
 698                 return -EADDRNOTAVAIL;
 699         }
 700
 701         if (vifi == mrt->mroute_reg_vif_num)
 702                 mrt->mroute_reg_vif_num = -1;
 703
 704         if (vifi + 1 == mrt->maxvif) {
 705                 int tmp;
 706
 707                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 708                         if (VIF_EXISTS(mrt, tmp))
 709                                 break;
 710                 }
 711                 mrt->maxvif = tmp+1;
 712         }
 713
 714         write_unlock_bh(&mrt_lock);
 715
 716         dev_set_allmulti(dev, -1);
 717
 718         in_dev = __in_dev_get_rtnl(dev);
 719         if (in_dev) {
 720                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 721                 inet_netconf_notify_devconf(dev_net(dev), RTM_NEWNETCONF,
 722                                             NETCONFA_MC_FORWARDING,
 723                                             dev->ifindex, &in_dev->cnf);
 724                 ip_rt_multicast_event(in_dev);
 725         }
 726
 727         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 728                 unregister_netdevice_queue(dev, head);
 729
 730         dev_put(dev);
 731         return 0;
 732 }
 733
 734 static void ipmr_cache_free_rcu(struct rcu_head *head)
 735 {
 736         struct mr_mfc *c = container_of(head, struct mr_mfc, rcu);
 737
 738         kmem_cache_free(mrt_cachep, (struct mfc_cache *)c);
 739 }
 740
 741 static void ipmr_cache_free(struct mfc_cache *c)
 742 {
 743         call_rcu(&c->_c.rcu, ipmr_cache_free_rcu);
 744 }
 745
 746 /* Destroy an unresolved cache entry, killing queued skbs
 747  * and reporting error to netlink readers.
 748  */
 749 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 750 {
 751         struct net *net = read_pnet(&mrt->net);
 752         struct sk_buff *skb;
 753         struct nlmsgerr *e;
 754
 755         atomic_dec(&mrt->cache_resolve_queue_len);
 756
 757         while ((skb = skb_dequeue(&c->_c.mfc_un.unres.unresolved))) {
 758                 if (ip_hdr(skb)->version == 0) {
 759                         struct nlmsghdr *nlh = skb_pull(skb,
 760                                                         sizeof(struct iphdr));
 761                         nlh->nlmsg_type = NLMSG_ERROR;
 762                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 763                         skb_trim(skb, nlh->nlmsg_len);
 764                         e = nlmsg_data(nlh);
 765                         e->error = -ETIMEDOUT;
 766                         memset(&e->msg, 0, sizeof(e->msg));
 767
 768                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 769                 } else {
 770                         kfree_skb(skb);
 771                 }
 772         }
 773
 774         ipmr_cache_free(c);
 775 }
 776
 777 /* Timer process for the unresolved queue. */
 778 static void ipmr_expire_process(struct timer_list *t)
 779 {
 780         struct mr_table *mrt = from_timer(mrt, t, ipmr_expire_timer);
 781         struct mr_mfc *c, *next;
 782         unsigned long expires;
 783         unsigned long now;
 784
 785         if (!spin_trylock(&mfc_unres_lock)) {
 786                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 787                 return;
 788         }
 789
 790         if (list_empty(&mrt->mfc_unres_queue))
 791                 goto out;
 792
 793         now = jiffies;
 794         expires = 10*HZ;
 795
 796         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 797                 if (time_after(c->mfc_un.unres.expires, now)) {
 798                         unsigned long interval = c->mfc_un.unres.expires - now;
 799                         if (interval < expires)
 800                                 expires = interval;
 801                         continue;
 802                 }
 803
 804                 list_del(&c->list);
 805                 mroute_netlink_event(mrt, (struct mfc_cache *)c, RTM_DELROUTE);
 806                 ipmr_destroy_unres(mrt, (struct mfc_cache *)c);
 807         }
 808
 809         if (!list_empty(&mrt->mfc_unres_queue))
 810                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 811
 812 out:
 813         spin_unlock(&mfc_unres_lock);
 814 }
 815
 816 /* Fill oifs list. It is called under write locked mrt_lock. */
 817 static void ipmr_update_thresholds(struct mr_table *mrt, struct mr_mfc *cache,
 818                                    unsigned char *ttls)
 819 {
 820         int vifi;
 821
 822         cache->mfc_un.res.minvif = MAXVIFS;
 823         cache->mfc_un.res.maxvif = 0;
 824         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 825
 826         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 827                 if (VIF_EXISTS(mrt, vifi) &&
 828                     ttls[vifi] && ttls[vifi] < 255) {
 829                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 830                         if (cache->mfc_un.res.minvif > vifi)
 831                                 cache->mfc_un.res.minvif = vifi;
 832                         if (cache->mfc_un.res.maxvif <= vifi)
 833                                 cache->mfc_un.res.maxvif = vifi + 1;
 834                 }
 835         }
 836         cache->mfc_un.res.lastuse = jiffies;
 837 }
 838
 839 static int vif_add(struct net *net, struct mr_table *mrt,
 840                    struct vifctl *vifc, int mrtsock)
 841 {
 842         int vifi = vifc->vifc_vifi;
 843         struct switchdev_attr attr = {
 844                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
 845         };
 846         struct vif_device *v = &mrt->vif_table[vifi];
 847         struct net_device *dev;
 848         struct in_device *in_dev;
 849         int err;
 850
 851         /* Is vif busy ? */
 852         if (VIF_EXISTS(mrt, vifi))
 853                 return -EADDRINUSE;
 854
 855         switch (vifc->vifc_flags) {
 856         case VIFF_REGISTER:
 857                 if (!ipmr_pimsm_enabled())
 858                         return -EINVAL;
 859                 /* Special Purpose VIF in PIM
 860                  * All the packets will be sent to the daemon
 861                  */
 862                 if (mrt->mroute_reg_vif_num >= 0)
 863                         return -EADDRINUSE;
 864                 dev = ipmr_reg_vif(net, mrt);
 865                 if (!dev)
 866                         return -ENOBUFS;
 867                 err = dev_set_allmulti(dev, 1);
 868                 if (err) {
 869                         unregister_netdevice(dev);
 870                         dev_put(dev);
 871                         return err;
 872                 }
 873                 break;
 874         case VIFF_TUNNEL:
 875                 dev = ipmr_new_tunnel(net, vifc);
 876                 if (!dev)
 877                         return -ENOBUFS;
 878                 err = dev_set_allmulti(dev, 1);
 879                 if (err) {
 880                         ipmr_del_tunnel(dev, vifc);
 881                         dev_put(dev);
 882                         return err;
 883                 }
 884                 break;
 885         case VIFF_USE_IFINDEX:
 886         case 0:
 887                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 888                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 889                         if (dev && !__in_dev_get_rtnl(dev)) {
 890                                 dev_put(dev);
 891                                 return -EADDRNOTAVAIL;
 892                         }
 893                 } else {
 894                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 895                 }
 896                 if (!dev)
 897                         return -EADDRNOTAVAIL;
 898                 err = dev_set_allmulti(dev, 1);
 899                 if (err) {
 900                         dev_put(dev);
 901                         return err;
 902                 }
 903                 break;
 904         default:
 905                 return -EINVAL;
 906         }
 907
 908         in_dev = __in_dev_get_rtnl(dev);
 909         if (!in_dev) {
 910                 dev_put(dev);
 911                 return -EADDRNOTAVAIL;
 912         }
 913         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 914         inet_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_MC_FORWARDING,
 915                                     dev->ifindex, &in_dev->cnf);
 916         ip_rt_multicast_event(in_dev);
 917
 918         /* Fill in the VIF structures */
 919         vif_device_init(v, dev, vifc->vifc_rate_limit,
 920                         vifc->vifc_threshold,
 921                         vifc->vifc_flags | (!mrtsock ? VIFF_STATIC : 0),
 922                         (VIFF_TUNNEL | VIFF_REGISTER));
 923
 924         attr.orig_dev = dev;
 925         if (!switchdev_port_attr_get(dev, &attr)) {
 926                 memcpy(v->dev_parent_id.id, attr.u.ppid.id, attr.u.ppid.id_len);
 927                 v->dev_parent_id.id_len = attr.u.ppid.id_len;
 928         } else {
 929                 v->dev_parent_id.id_len = 0;
 930         }
 931
 932         v->local = vifc->vifc_lcl_addr.s_addr;
 933         v->remote = vifc->vifc_rmt_addr.s_addr;
 934
 935         /* And finish update writing critical data */
 936         write_lock_bh(&mrt_lock);
 937         v->dev = dev;
 938         if (v->flags & VIFF_REGISTER)
 939                 mrt->mroute_reg_vif_num = vifi;
 940         if (vifi+1 > mrt->maxvif)
 941                 mrt->maxvif = vifi+1;
 942         write_unlock_bh(&mrt_lock);
 943         call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_ADD, v, vifi, mrt->id);
 944         return 0;
 945 }
 946
 947 /* called with rcu_read_lock() */
 948 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 949                                          __be32 origin,
 950                                          __be32 mcastgrp)
 951 {
 952         struct mfc_cache_cmp_arg arg = {
 953                         .mfc_mcastgrp = mcastgrp,
 954                         .mfc_origin = origin
 955         };
 956
 957         return mr_mfc_find(mrt, &arg);
 958 }
 959
 960 /* Look for a (*,G) entry */
 961 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 962                                              __be32 mcastgrp, int vifi)
 963 {
 964         struct mfc_cache_cmp_arg arg = {
 965                         .mfc_mcastgrp = mcastgrp,
 966                         .mfc_origin = htonl(INADDR_ANY)
 967         };
 968
 969         if (mcastgrp == htonl(INADDR_ANY))
 970                 return mr_mfc_find_any_parent(mrt, vifi);
 971         return mr_mfc_find_any(mrt, vifi, &arg);
 972 }
 973
 974 /* Look for a (S,G,iif) entry if parent != -1 */
 975 static struct mfc_cache *ipmr_cache_find_parent(struct mr_table *mrt,
 976                                                 __be32 origin, __be32 mcastgrp,
 977                                                 int parent)
 978 {
 979         struct mfc_cache_cmp_arg arg = {
 980                         .mfc_mcastgrp = mcastgrp,
 981                         .mfc_origin = origin,
 982         };
 983
 984         return mr_mfc_find_parent(mrt, &arg, parent);
 985 }
 986
 987 /* Allocate a multicast cache entry */
 988 static struct mfc_cache *ipmr_cache_alloc(void)
 989 {
 990         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 991
 992         if (c) {
 993                 c->_c.mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 994                 c->_c.mfc_un.res.minvif = MAXVIFS;
 995                 c->_c.free = ipmr_cache_free_rcu;
 996                 refcount_set(&c->_c.mfc_un.res.refcount, 1);
 997         }
 998         return c;
 999 }
1000
1001 static struct mfc_cache *ipmr_cache_alloc_unres(void)
1002 {
1003         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
1004
1005         if (c) {
1006                 skb_queue_head_init(&c->_c.mfc_un.unres.unresolved);
1007                 c->_c.mfc_un.unres.expires = jiffies + 10 * HZ;
1008         }
1009         return c;
1010 }
1011
1012 /* A cache entry has gone into a resolved state from queued */
1013 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
1014                                struct mfc_cache *uc, struct mfc_cache *c)
1015 {
1016         struct sk_buff *skb;
1017         struct nlmsgerr *e;
1018
1019         /* Play the pending entries through our router */
1020         while ((skb = __skb_dequeue(&uc->_c.mfc_un.unres.unresolved))) {
1021                 if (ip_hdr(skb)->version == 0) {
1022                         struct nlmsghdr *nlh = skb_pull(skb,
1023                                                         sizeof(struct iphdr));
1024
1025                         if (mr_fill_mroute(mrt, skb, &c->_c,
1026                                            nlmsg_data(nlh)) > 0) {
1027                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
1028                                                  (u8 *)nlh;
1029                         } else {
1030                                 nlh->nlmsg_type = NLMSG_ERROR;
1031                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
1032                                 skb_trim(skb, nlh->nlmsg_len);
1033                                 e = nlmsg_data(nlh);
1034                                 e->error = -EMSGSIZE;
1035                                 memset(&e->msg, 0, sizeof(e->msg));
1036                         }
1037
1038                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
1039                 } else {
1040                         ip_mr_forward(net, mrt, skb->dev, skb, c, 0);
1041                 }
1042         }
1043 }
1044
1045 /* Bounce a cache query up to mrouted and netlink.
1046  *
1047  * Called under mrt_lock.
1048  */
1049 static int ipmr_cache_report(struct mr_table *mrt,
1050                              struct sk_buff *pkt, vifi_t vifi, int assert)
1051 {
1052         const int ihl = ip_hdrlen(pkt);
1053         struct sock *mroute_sk;
1054         struct igmphdr *igmp;
1055         struct igmpmsg *msg;
1056         struct sk_buff *skb;
1057         int ret;
1058
1059         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE)
1060                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
1061         else
1062                 skb = alloc_skb(128, GFP_ATOMIC);
1063
1064         if (!skb)
1065                 return -ENOBUFS;
1066
1067         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE) {
1068                 /* Ugly, but we have no choice with this interface.
1069                  * Duplicate old header, fix ihl, length etc.
1070                  * And all this only to mangle msg->im_msgtype and
1071                  * to set msg->im_mbz to "mbz" :-)
1072                  */
1073                 skb_push(skb, sizeof(struct iphdr));
1074                 skb_reset_network_header(skb);
1075                 skb_reset_transport_header(skb);
1076                 msg = (struct igmpmsg *)skb_network_header(skb);
1077                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
1078                 msg->im_msgtype = assert;
1079                 msg->im_mbz = 0;
1080                 if (assert == IGMPMSG_WRVIFWHOLE)
1081                         msg->im_vif = vifi;
1082                 else
1083                         msg->im_vif = mrt->mroute_reg_vif_num;
1084                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
1085                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
1086                                              sizeof(struct iphdr));
1087         } else {
1088                 /* Copy the IP header */
1089                 skb_set_network_header(skb, skb->len);
1090                 skb_put(skb, ihl);
1091                 skb_copy_to_linear_data(skb, pkt->data, ihl);
1092                 /* Flag to the kernel this is a route add */
1093                 ip_hdr(skb)->protocol = 0;
1094                 msg = (struct igmpmsg *)skb_network_header(skb);
1095                 msg->im_vif = vifi;
1096                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1097                 /* Add our header */
1098                 igmp = skb_put(skb, sizeof(struct igmphdr));
1099                 igmp->type = assert;
1100                 msg->im_msgtype = assert;
1101                 igmp->code = 0;
1102                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1103                 skb->transport_header = skb->network_header;
1104         }
1105
1106         rcu_read_lock();
1107         mroute_sk = rcu_dereference(mrt->mroute_sk);
1108         if (!mroute_sk) {
1109                 rcu_read_unlock();
1110                 kfree_skb(skb);
1111                 return -EINVAL;
1112         }
1113
1114         igmpmsg_netlink_event(mrt, skb);
1115
1116         /* Deliver to mrouted */
1117         ret = sock_queue_rcv_skb(mroute_sk, skb);
1118         rcu_read_unlock();
1119         if (ret < 0) {
1120                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1121                 kfree_skb(skb);
1122         }
1123
1124         return ret;
1125 }
1126
1127 /* Queue a packet for resolution. It gets locked cache entry! */
1128 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1129                                  struct sk_buff *skb, struct net_device *dev)
1130 {
1131         const struct iphdr *iph = ip_hdr(skb);
1132         struct mfc_cache *c;
1133         bool found = false;
1134         int err;
1135
1136         spin_lock_bh(&mfc_unres_lock);
1137         list_for_each_entry(c, &mrt->mfc_unres_queue, _c.list) {
1138                 if (c->mfc_mcastgrp == iph->daddr &&
1139                     c->mfc_origin == iph->saddr) {
1140                         found = true;
1141                         break;
1142                 }
1143         }
1144
1145         if (!found) {
1146                 /* Create a new entry if allowable */
1147                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1148                     (c = ipmr_cache_alloc_unres()) == NULL) {
1149                         spin_unlock_bh(&mfc_unres_lock);
1150
1151                         kfree_skb(skb);
1152                         return -ENOBUFS;
1153                 }
1154
1155                 /* Fill in the new cache entry */
1156                 c->_c.mfc_parent = -1;
1157                 c->mfc_origin   = iph->saddr;
1158                 c->mfc_mcastgrp = iph->daddr;
1159
1160                 /* Reflect first query at mrouted. */
1161                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1162
1163                 if (err < 0) {
1164                         /* If the report failed throw the cache entry
1165                            out - Brad Parker
1166                          */
1167                         spin_unlock_bh(&mfc_unres_lock);
1168
1169                         ipmr_cache_free(c);
1170                         kfree_skb(skb);
1171                         return err;
1172                 }
1173
1174                 atomic_inc(&mrt->cache_resolve_queue_len);
1175                 list_add(&c->_c.list, &mrt->mfc_unres_queue);
1176                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1177
1178                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1179                         mod_timer(&mrt->ipmr_expire_timer,
1180                                   c->_c.mfc_un.unres.expires);
1181         }
1182
1183         /* See if we can append the packet */
1184         if (c->_c.mfc_un.unres.unresolved.qlen > 3) {
1185                 kfree_skb(skb);
1186                 err = -ENOBUFS;
1187         } else {
1188                 if (dev) {
1189                         skb->dev = dev;
1190                         skb->skb_iif = dev->ifindex;
1191                 }
1192                 skb_queue_tail(&c->_c.mfc_un.unres.unresolved, skb);
1193                 err = 0;
1194         }
1195
1196         spin_unlock_bh(&mfc_unres_lock);
1197         return err;
1198 }
1199
1200 /* MFC cache manipulation by user space mroute daemon */
1201
1202 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1203 {
1204         struct net *net = read_pnet(&mrt->net);
1205         struct mfc_cache *c;
1206
1207         /* The entries are added/deleted only under RTNL */
1208         rcu_read_lock();
1209         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1210                                    mfc->mfcc_mcastgrp.s_addr, parent);
1211         rcu_read_unlock();
1212         if (!c)
1213                 return -ENOENT;
1214         rhltable_remove(&mrt->mfc_hash, &c->_c.mnode, ipmr_rht_params);
1215         list_del_rcu(&c->_c.list);
1216         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, c, mrt->id);
1217         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1218         mr_cache_put(&c->_c);
1219
1220         return 0;
1221 }
1222
1223 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1224                         struct mfcctl *mfc, int mrtsock, int parent)
1225 {
1226         struct mfc_cache *uc, *c;
1227         struct mr_mfc *_uc;
1228         bool found;
1229         int ret;
1230
1231         if (mfc->mfcc_parent >= MAXVIFS)
1232                 return -ENFILE;
1233
1234         /* The entries are added/deleted only under RTNL */
1235         rcu_read_lock();
1236         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1237                                    mfc->mfcc_mcastgrp.s_addr, parent);
1238         rcu_read_unlock();
1239         if (c) {
1240                 write_lock_bh(&mrt_lock);
1241                 c->_c.mfc_parent = mfc->mfcc_parent;
1242                 ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1243                 if (!mrtsock)
1244                         c->_c.mfc_flags |= MFC_STATIC;
1245                 write_unlock_bh(&mrt_lock);
1246                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, c,
1247                                               mrt->id);
1248                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1249                 return 0;
1250         }
1251
1252         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1253             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1254                 return -EINVAL;
1255
1256         c = ipmr_cache_alloc();
1257         if (!c)
1258                 return -ENOMEM;
1259
1260         c->mfc_origin = mfc->mfcc_origin.s_addr;
1261         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1262         c->_c.mfc_parent = mfc->mfcc_parent;
1263         ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1264         if (!mrtsock)
1265                 c->_c.mfc_flags |= MFC_STATIC;
1266
1267         ret = rhltable_insert_key(&mrt->mfc_hash, &c->cmparg, &c->_c.mnode,
1268                                   ipmr_rht_params);
1269         if (ret) {
1270                 pr_err("ipmr: rhtable insert error %d\n", ret);
1271                 ipmr_cache_free(c);
1272                 return ret;
1273         }
1274         list_add_tail_rcu(&c->_c.list, &mrt->mfc_cache_list);
1275         /* Check to see if we resolved a queued list. If so we
1276          * need to send on the frames and tidy up.
1277          */
1278         found = false;
1279         spin_lock_bh(&mfc_unres_lock);
1280         list_for_each_entry(_uc, &mrt->mfc_unres_queue, list) {
1281                 uc = (struct mfc_cache *)_uc;
1282                 if (uc->mfc_origin == c->mfc_origin &&
1283                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1284                         list_del(&_uc->list);
1285                         atomic_dec(&mrt->cache_resolve_queue_len);
1286                         found = true;
1287                         break;
1288                 }
1289         }
1290         if (list_empty(&mrt->mfc_unres_queue))
1291                 del_timer(&mrt->ipmr_expire_timer);
1292         spin_unlock_bh(&mfc_unres_lock);
1293
1294         if (found) {
1295                 ipmr_cache_resolve(net, mrt, uc, c);
1296                 ipmr_cache_free(uc);
1297         }
1298         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, c, mrt->id);
1299         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1300         return 0;
1301 }
1302
1303 /* Close the multicast socket, and clear the vif tables etc */
1304 static void mroute_clean_tables(struct mr_table *mrt, bool all)
1305 {
1306         struct net *net = read_pnet(&mrt->net);
1307         struct mr_mfc *c, *tmp;
1308         struct mfc_cache *cache;
1309         LIST_HEAD(list);
1310         int i;
1311
1312         /* Shut down all active vif entries */
1313         for (i = 0; i < mrt->maxvif; i++) {
1314                 if (!all && (mrt->vif_table[i].flags & VIFF_STATIC))
1315                         continue;
1316                 vif_delete(mrt, i, 0, &list);
1317         }
1318         unregister_netdevice_many(&list);
1319
1320         /* Wipe the cache */
1321         list_for_each_entry_safe(c, tmp, &mrt->mfc_cache_list, list) {
1322                 if (!all && (c->mfc_flags & MFC_STATIC))
1323                         continue;
1324                 rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1325                 list_del_rcu(&c->list);
1326                 cache = (struct mfc_cache *)c;
1327                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, cache,
1328                                               mrt->id);
1329                 mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1330                 mr_cache_put(c);
1331         }
1332
1333         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1334                 spin_lock_bh(&mfc_unres_lock);
1335                 list_for_each_entry_safe(c, tmp, &mrt->mfc_unres_queue, list) {
1336                         list_del(&c->list);
1337                         cache = (struct mfc_cache *)c;
1338                         mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1339                         ipmr_destroy_unres(mrt, cache);
1340                 }
1341                 spin_unlock_bh(&mfc_unres_lock);
1342         }
1343 }
1344
1345 /* called from ip_ra_control(), before an RCU grace period,
1346  * we dont need to call synchronize_rcu() here
1347  */
1348 static void mrtsock_destruct(struct sock *sk)
1349 {
1350         struct net *net = sock_net(sk);
1351         struct mr_table *mrt;
1352
1353         rtnl_lock();
1354         ipmr_for_each_table(mrt, net) {
1355                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1356                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1357                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1358                                                     NETCONFA_MC_FORWARDING,
1359                                                     NETCONFA_IFINDEX_ALL,
1360                                                     net->ipv4.devconf_all);
1361                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1362                         mroute_clean_tables(mrt, false);
1363                 }
1364         }
1365         rtnl_unlock();
1366 }
1367
1368 /* Socket options and virtual interface manipulation. The whole
1369  * virtual interface system is a complete heap, but unfortunately
1370  * that's how BSD mrouted happens to think. Maybe one day with a proper
1371  * MOSPF/PIM router set up we can clean this up.
1372  */
1373
1374 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval,
1375                          unsigned int optlen)
1376 {
1377         struct net *net = sock_net(sk);
1378         int val, ret = 0, parent = 0;
1379         struct mr_table *mrt;
1380         struct vifctl vif;
1381         struct mfcctl mfc;
1382         bool do_wrvifwhole;
1383         u32 uval;
1384
1385         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1386         rtnl_lock();
1387         if (sk->sk_type != SOCK_RAW ||
1388             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1389                 ret = -EOPNOTSUPP;
1390                 goto out_unlock;
1391         }
1392
1393         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1394         if (!mrt) {
1395                 ret = -ENOENT;
1396                 goto out_unlock;
1397         }
1398         if (optname != MRT_INIT) {
1399                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1400                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1401                         ret = -EACCES;
1402                         goto out_unlock;
1403                 }
1404         }
1405
1406         switch (optname) {
1407         case MRT_INIT:
1408                 if (optlen != sizeof(int)) {
1409                         ret = -EINVAL;
1410                         break;
1411                 }
1412                 if (rtnl_dereference(mrt->mroute_sk)) {
1413                         ret = -EADDRINUSE;
1414                         break;
1415                 }
1416
1417                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1418                 if (ret == 0) {
1419                         rcu_assign_pointer(mrt->mroute_sk, sk);
1420                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1421                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1422                                                     NETCONFA_MC_FORWARDING,
1423                                                     NETCONFA_IFINDEX_ALL,
1424                                                     net->ipv4.devconf_all);
1425                 }
1426                 break;
1427         case MRT_DONE:
1428                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1429                         ret = -EACCES;
1430                 } else {
1431                         /* We need to unlock here because mrtsock_destruct takes
1432                          * care of rtnl itself and we can't change that due to
1433                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1434                          */
1435                         rtnl_unlock();
1436                         ret = ip_ra_control(sk, 0, NULL);
1437                         goto out;
1438                 }
1439                 break;
1440         case MRT_ADD_VIF:
1441         case MRT_DEL_VIF:
1442                 if (optlen != sizeof(vif)) {
1443                         ret = -EINVAL;
1444                         break;
1445                 }
1446                 if (copy_from_user(&vif, optval, sizeof(vif))) {
1447                         ret = -EFAULT;
1448                         break;
1449                 }
1450                 if (vif.vifc_vifi >= MAXVIFS) {
1451                         ret = -ENFILE;
1452                         break;
1453                 }
1454                 if (optname == MRT_ADD_VIF) {
1455                         ret = vif_add(net, mrt, &vif,
1456                                       sk == rtnl_dereference(mrt->mroute_sk));
1457                 } else {
1458                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1459                 }
1460                 break;
1461         /* Manipulate the forwarding caches. These live
1462          * in a sort of kernel/user symbiosis.
1463          */
1464         case MRT_ADD_MFC:
1465         case MRT_DEL_MFC:
1466                 parent = -1;
1467                 /* fall through */
1468         case MRT_ADD_MFC_PROXY:
1469         case MRT_DEL_MFC_PROXY:
1470                 if (optlen != sizeof(mfc)) {
1471                         ret = -EINVAL;
1472                         break;
1473                 }
1474                 if (copy_from_user(&mfc, optval, sizeof(mfc))) {
1475                         ret = -EFAULT;
1476                         break;
1477                 }
1478                 if (parent == 0)
1479                         parent = mfc.mfcc_parent;
1480                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1481                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1482                 else
1483                         ret = ipmr_mfc_add(net, mrt, &mfc,
1484                                            sk == rtnl_dereference(mrt->mroute_sk),
1485                                            parent);
1486                 break;
1487         /* Control PIM assert. */
1488         case MRT_ASSERT:
1489                 if (optlen != sizeof(val)) {
1490                         ret = -EINVAL;
1491                         break;
1492                 }
1493                 if (get_user(val, (int __user *)optval)) {
1494                         ret = -EFAULT;
1495                         break;
1496                 }
1497                 mrt->mroute_do_assert = val;
1498                 break;
1499         case MRT_PIM:
1500                 if (!ipmr_pimsm_enabled()) {
1501                         ret = -ENOPROTOOPT;
1502                         break;
1503                 }
1504                 if (optlen != sizeof(val)) {
1505                         ret = -EINVAL;
1506                         break;
1507                 }
1508                 if (get_user(val, (int __user *)optval)) {
1509                         ret = -EFAULT;
1510                         break;
1511                 }
1512
1513                 do_wrvifwhole = (val == IGMPMSG_WRVIFWHOLE);
1514                 val = !!val;
1515                 if (val != mrt->mroute_do_pim) {
1516                         mrt->mroute_do_pim = val;
1517                         mrt->mroute_do_assert = val;
1518                         mrt->mroute_do_wrvifwhole = do_wrvifwhole;
1519                 }
1520                 break;
1521         case MRT_TABLE:
1522                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1523                         ret = -ENOPROTOOPT;
1524                         break;
1525                 }
1526                 if (optlen != sizeof(uval)) {
1527                         ret = -EINVAL;
1528                         break;
1529                 }
1530                 if (get_user(uval, (u32 __user *)optval)) {
1531                         ret = -EFAULT;
1532                         break;
1533                 }
1534
1535                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1536                         ret = -EBUSY;
1537                 } else {
1538                         mrt = ipmr_new_table(net, uval);
1539                         if (IS_ERR(mrt))
1540                                 ret = PTR_ERR(mrt);
1541                         else
1542                                 raw_sk(sk)->ipmr_table = uval;
1543                 }
1544                 break;
1545         /* Spurious command, or MRT_VERSION which you cannot set. */
1546         default:
1547                 ret = -ENOPROTOOPT;
1548         }
1549 out_unlock:
1550         rtnl_unlock();
1551 out:
1552         return ret;
1553 }
1554
1555 /* Getsock opt support for the multicast routing system. */
1556 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1557 {
1558         int olr;
1559         int val;
1560         struct net *net = sock_net(sk);
1561         struct mr_table *mrt;
1562
1563         if (sk->sk_type != SOCK_RAW ||
1564             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1565                 return -EOPNOTSUPP;
1566
1567         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1568         if (!mrt)
1569                 return -ENOENT;
1570
1571         switch (optname) {
1572         case MRT_VERSION:
1573                 val = 0x0305;
1574                 break;
1575         case MRT_PIM:
1576                 if (!ipmr_pimsm_enabled())
1577                         return -ENOPROTOOPT;
1578                 val = mrt->mroute_do_pim;
1579                 break;
1580         case MRT_ASSERT:
1581                 val = mrt->mroute_do_assert;
1582                 break;
1583         default:
1584                 return -ENOPROTOOPT;
1585         }
1586
1587         if (get_user(olr, optlen))
1588                 return -EFAULT;
1589         olr = min_t(unsigned int, olr, sizeof(int));
1590         if (olr < 0)
1591                 return -EINVAL;
1592         if (put_user(olr, optlen))
1593                 return -EFAULT;
1594         if (copy_to_user(optval, &val, olr))
1595                 return -EFAULT;
1596         return 0;
1597 }
1598
1599 /* The IP multicast ioctl support routines. */
1600 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1601 {
1602         struct sioc_sg_req sr;
1603         struct sioc_vif_req vr;
1604         struct vif_device *vif;
1605         struct mfc_cache *c;
1606         struct net *net = sock_net(sk);
1607         struct mr_table *mrt;
1608
1609         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1610         if (!mrt)
1611                 return -ENOENT;
1612
1613         switch (cmd) {
1614         case SIOCGETVIFCNT:
1615                 if (copy_from_user(&vr, arg, sizeof(vr)))
1616                         return -EFAULT;
1617                 if (vr.vifi >= mrt->maxvif)
1618                         return -EINVAL;
1619                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1620                 read_lock(&mrt_lock);
1621                 vif = &mrt->vif_table[vr.vifi];
1622                 if (VIF_EXISTS(mrt, vr.vifi)) {
1623                         vr.icount = vif->pkt_in;
1624                         vr.ocount = vif->pkt_out;
1625                         vr.ibytes = vif->bytes_in;
1626                         vr.obytes = vif->bytes_out;
1627                         read_unlock(&mrt_lock);
1628
1629                         if (copy_to_user(arg, &vr, sizeof(vr)))
1630                                 return -EFAULT;
1631                         return 0;
1632                 }
1633                 read_unlock(&mrt_lock);
1634                 return -EADDRNOTAVAIL;
1635         case SIOCGETSGCNT:
1636                 if (copy_from_user(&sr, arg, sizeof(sr)))
1637                         return -EFAULT;
1638
1639                 rcu_read_lock();
1640                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1641                 if (c) {
1642                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1643                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1644                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1645                         rcu_read_unlock();
1646
1647                         if (copy_to_user(arg, &sr, sizeof(sr)))
1648                                 return -EFAULT;
1649                         return 0;
1650                 }
1651                 rcu_read_unlock();
1652                 return -EADDRNOTAVAIL;
1653         default:
1654                 return -ENOIOCTLCMD;
1655         }
1656 }
1657
1658 #ifdef CONFIG_COMPAT
1659 struct compat_sioc_sg_req {
1660         struct in_addr src;
1661         struct in_addr grp;
1662         compat_ulong_t pktcnt;
1663         compat_ulong_t bytecnt;
1664         compat_ulong_t wrong_if;
1665 };
1666
1667 struct compat_sioc_vif_req {
1668         vifi_t  vifi;           /* Which iface */
1669         compat_ulong_t icount;
1670         compat_ulong_t ocount;
1671         compat_ulong_t ibytes;
1672         compat_ulong_t obytes;
1673 };
1674
1675 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1676 {
1677         struct compat_sioc_sg_req sr;
1678         struct compat_sioc_vif_req vr;
1679         struct vif_device *vif;
1680         struct mfc_cache *c;
1681         struct net *net = sock_net(sk);
1682         struct mr_table *mrt;
1683
1684         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1685         if (!mrt)
1686                 return -ENOENT;
1687
1688         switch (cmd) {
1689         case SIOCGETVIFCNT:
1690                 if (copy_from_user(&vr, arg, sizeof(vr)))
1691                         return -EFAULT;
1692                 if (vr.vifi >= mrt->maxvif)
1693                         return -EINVAL;
1694                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1695                 read_lock(&mrt_lock);
1696                 vif = &mrt->vif_table[vr.vifi];
1697                 if (VIF_EXISTS(mrt, vr.vifi)) {
1698                         vr.icount = vif->pkt_in;
1699                         vr.ocount = vif->pkt_out;
1700                         vr.ibytes = vif->bytes_in;
1701                         vr.obytes = vif->bytes_out;
1702                         read_unlock(&mrt_lock);
1703
1704                         if (copy_to_user(arg, &vr, sizeof(vr)))
1705                                 return -EFAULT;
1706                         return 0;
1707                 }
1708                 read_unlock(&mrt_lock);
1709                 return -EADDRNOTAVAIL;
1710         case SIOCGETSGCNT:
1711                 if (copy_from_user(&sr, arg, sizeof(sr)))
1712                         return -EFAULT;
1713
1714                 rcu_read_lock();
1715                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1716                 if (c) {
1717                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1718                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1719                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1720                         rcu_read_unlock();
1721
1722                         if (copy_to_user(arg, &sr, sizeof(sr)))
1723                                 return -EFAULT;
1724                         return 0;
1725                 }
1726                 rcu_read_unlock();
1727                 return -EADDRNOTAVAIL;
1728         default:
1729                 return -ENOIOCTLCMD;
1730         }
1731 }
1732 #endif
1733
1734 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1735 {
1736         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1737         struct net *net = dev_net(dev);
1738         struct mr_table *mrt;
1739         struct vif_device *v;
1740         int ct;
1741
1742         if (event != NETDEV_UNREGISTER)
1743                 return NOTIFY_DONE;
1744
1745         ipmr_for_each_table(mrt, net) {
1746                 v = &mrt->vif_table[0];
1747                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1748                         if (v->dev == dev)
1749                                 vif_delete(mrt, ct, 1, NULL);
1750                 }
1751         }
1752         return NOTIFY_DONE;
1753 }
1754
1755 static struct notifier_block ip_mr_notifier = {
1756         .notifier_call = ipmr_device_event,
1757 };
1758
1759 /* Encapsulate a packet by attaching a valid IPIP header to it.
1760  * This avoids tunnel drivers and other mess and gives us the speed so
1761  * important for multicast video.
1762  */
1763 static void ip_encap(struct net *net, struct sk_buff *skb,
1764                      __be32 saddr, __be32 daddr)
1765 {
1766         struct iphdr *iph;
1767         const struct iphdr *old_iph = ip_hdr(skb);
1768
1769         skb_push(skb, sizeof(struct iphdr));
1770         skb->transport_header = skb->network_header;
1771         skb_reset_network_header(skb);
1772         iph = ip_hdr(skb);
1773
1774         iph->version    =       4;
1775         iph->tos        =       old_iph->tos;
1776         iph->ttl        =       old_iph->ttl;
1777         iph->frag_off   =       0;
1778         iph->daddr      =       daddr;
1779         iph->saddr      =       saddr;
1780         iph->protocol   =       IPPROTO_IPIP;
1781         iph->ihl        =       5;
1782         iph->tot_len    =       htons(skb->len);
1783         ip_select_ident(net, skb, NULL);
1784         ip_send_check(iph);
1785
1786         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1787         nf_reset(skb);
1788 }
1789
1790 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1791                                       struct sk_buff *skb)
1792 {
1793         struct ip_options *opt = &(IPCB(skb)->opt);
1794
1795         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1796         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1797
1798         if (unlikely(opt->optlen))
1799                 ip_forward_options(skb);
1800
1801         return dst_output(net, sk, skb);
1802 }
1803
1804 #ifdef CONFIG_NET_SWITCHDEV
1805 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1806                                    int in_vifi, int out_vifi)
1807 {
1808         struct vif_device *out_vif = &mrt->vif_table[out_vifi];
1809         struct vif_device *in_vif = &mrt->vif_table[in_vifi];
1810
1811         if (!skb->offload_mr_fwd_mark)
1812                 return false;
1813         if (!out_vif->dev_parent_id.id_len || !in_vif->dev_parent_id.id_len)
1814                 return false;
1815         return netdev_phys_item_id_same(&out_vif->dev_parent_id,
1816                                         &in_vif->dev_parent_id);
1817 }
1818 #else
1819 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1820                                    int in_vifi, int out_vifi)
1821 {
1822         return false;
1823 }
1824 #endif
1825
1826 /* Processing handlers for ipmr_forward */
1827
1828 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1829                             int in_vifi, struct sk_buff *skb,
1830                             struct mfc_cache *c, int vifi)
1831 {
1832         const struct iphdr *iph = ip_hdr(skb);
1833         struct vif_device *vif = &mrt->vif_table[vifi];
1834         struct net_device *dev;
1835         struct rtable *rt;
1836         struct flowi4 fl4;
1837         int    encap = 0;
1838
1839         if (!vif->dev)
1840                 goto out_free;
1841
1842         if (vif->flags & VIFF_REGISTER) {
1843                 vif->pkt_out++;
1844                 vif->bytes_out += skb->len;
1845                 vif->dev->stats.tx_bytes += skb->len;
1846                 vif->dev->stats.tx_packets++;
1847                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1848                 goto out_free;
1849         }
1850
1851         if (ipmr_forward_offloaded(skb, mrt, in_vifi, vifi))
1852                 goto out_free;
1853
1854         if (vif->flags & VIFF_TUNNEL) {
1855                 rt = ip_route_output_ports(net, &fl4, NULL,
1856                                            vif->remote, vif->local,
1857                                            0, 0,
1858                                            IPPROTO_IPIP,
1859                                            RT_TOS(iph->tos), vif->link);
1860                 if (IS_ERR(rt))
1861                         goto out_free;
1862                 encap = sizeof(struct iphdr);
1863         } else {
1864                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1865                                            0, 0,
1866                                            IPPROTO_IPIP,
1867                                            RT_TOS(iph->tos), vif->link);
1868                 if (IS_ERR(rt))
1869                         goto out_free;
1870         }
1871
1872         dev = rt->dst.dev;
1873
1874         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1875                 /* Do not fragment multicasts. Alas, IPv4 does not
1876                  * allow to send ICMP, so that packets will disappear
1877                  * to blackhole.
1878                  */
1879                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1880                 ip_rt_put(rt);
1881                 goto out_free;
1882         }
1883
1884         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1885
1886         if (skb_cow(skb, encap)) {
1887                 ip_rt_put(rt);
1888                 goto out_free;
1889         }
1890
1891         vif->pkt_out++;
1892         vif->bytes_out += skb->len;
1893
1894         skb_dst_drop(skb);
1895         skb_dst_set(skb, &rt->dst);
1896         ip_decrease_ttl(ip_hdr(skb));
1897
1898         /* FIXME: forward and output firewalls used to be called here.
1899          * What do we do with netfilter? -- RR
1900          */
1901         if (vif->flags & VIFF_TUNNEL) {
1902                 ip_encap(net, skb, vif->local, vif->remote);
1903                 /* FIXME: extra output firewall step used to be here. --RR */
1904                 vif->dev->stats.tx_packets++;
1905                 vif->dev->stats.tx_bytes += skb->len;
1906         }
1907
1908         IPCB(skb)->flags |= IPSKB_FORWARDED;
1909
1910         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1911          * not only before forwarding, but after forwarding on all output
1912          * interfaces. It is clear, if mrouter runs a multicasting
1913          * program, it should receive packets not depending to what interface
1914          * program is joined.
1915          * If we will not make it, the program will have to join on all
1916          * interfaces. On the other hand, multihoming host (or router, but
1917          * not mrouter) cannot join to more than one interface - it will
1918          * result in receiving multiple packets.
1919          */
1920         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1921                 net, NULL, skb, skb->dev, dev,
1922                 ipmr_forward_finish);
1923         return;
1924
1925 out_free:
1926         kfree_skb(skb);
1927 }
1928
1929 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1930 {
1931         int ct;
1932
1933         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1934                 if (mrt->vif_table[ct].dev == dev)
1935                         break;
1936         }
1937         return ct;
1938 }
1939
1940 /* "local" means that we should preserve one skb (for local delivery) */
1941 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1942                           struct net_device *dev, struct sk_buff *skb,
1943                           struct mfc_cache *c, int local)
1944 {
1945         int true_vifi = ipmr_find_vif(mrt, dev);
1946         int psend = -1;
1947         int vif, ct;
1948
1949         vif = c->_c.mfc_parent;
1950         c->_c.mfc_un.res.pkt++;
1951         c->_c.mfc_un.res.bytes += skb->len;
1952         c->_c.mfc_un.res.lastuse = jiffies;
1953
1954         if (c->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1955                 struct mfc_cache *cache_proxy;
1956
1957                 /* For an (*,G) entry, we only check that the incomming
1958                  * interface is part of the static tree.
1959                  */
1960                 cache_proxy = mr_mfc_find_any_parent(mrt, vif);
1961                 if (cache_proxy &&
1962                     cache_proxy->_c.mfc_un.res.ttls[true_vifi] < 255)
1963                         goto forward;
1964         }
1965
1966         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1967         if (mrt->vif_table[vif].dev != dev) {
1968                 if (rt_is_output_route(skb_rtable(skb))) {
1969                         /* It is our own packet, looped back.
1970                          * Very complicated situation...
1971                          *
1972                          * The best workaround until routing daemons will be
1973                          * fixed is not to redistribute packet, if it was
1974                          * send through wrong interface. It means, that
1975                          * multicast applications WILL NOT work for
1976                          * (S,G), which have default multicast route pointing
1977                          * to wrong oif. In any case, it is not a good
1978                          * idea to use multicasting applications on router.
1979                          */
1980                         goto dont_forward;
1981                 }
1982
1983                 c->_c.mfc_un.res.wrong_if++;
1984
1985                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1986                     /* pimsm uses asserts, when switching from RPT to SPT,
1987                      * so that we cannot check that packet arrived on an oif.
1988                      * It is bad, but otherwise we would need to move pretty
1989                      * large chunk of pimd to kernel. Ough... --ANK
1990                      */
1991                     (mrt->mroute_do_pim ||
1992                      c->_c.mfc_un.res.ttls[true_vifi] < 255) &&
1993                     time_after(jiffies,
1994                                c->_c.mfc_un.res.last_assert +
1995                                MFC_ASSERT_THRESH)) {
1996                         c->_c.mfc_un.res.last_assert = jiffies;
1997                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1998                         if (mrt->mroute_do_wrvifwhole)
1999                                 ipmr_cache_report(mrt, skb, true_vifi,
2000                                                   IGMPMSG_WRVIFWHOLE);
2001                 }
2002                 goto dont_forward;
2003         }
2004
2005 forward:
2006         mrt->vif_table[vif].pkt_in++;
2007         mrt->vif_table[vif].bytes_in += skb->len;
2008
2009         /* Forward the frame */
2010         if (c->mfc_origin == htonl(INADDR_ANY) &&
2011             c->mfc_mcastgrp == htonl(INADDR_ANY)) {
2012                 if (true_vifi >= 0 &&
2013                     true_vifi != c->_c.mfc_parent &&
2014                     ip_hdr(skb)->ttl >
2015                                 c->_c.mfc_un.res.ttls[c->_c.mfc_parent]) {
2016                         /* It's an (*,*) entry and the packet is not coming from
2017                          * the upstream: forward the packet to the upstream
2018                          * only.
2019                          */
2020                         psend = c->_c.mfc_parent;
2021                         goto last_forward;
2022                 }
2023                 goto dont_forward;
2024         }
2025         for (ct = c->_c.mfc_un.res.maxvif - 1;
2026              ct >= c->_c.mfc_un.res.minvif; ct--) {
2027                 /* For (*,G) entry, don't forward to the incoming interface */
2028                 if ((c->mfc_origin != htonl(INADDR_ANY) ||
2029                      ct != true_vifi) &&
2030                     ip_hdr(skb)->ttl > c->_c.mfc_un.res.ttls[ct]) {
2031                         if (psend != -1) {
2032                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2033
2034                                 if (skb2)
2035                                         ipmr_queue_xmit(net, mrt, true_vifi,
2036                                                         skb2, c, psend);
2037                         }
2038                         psend = ct;
2039                 }
2040         }
2041 last_forward:
2042         if (psend != -1) {
2043                 if (local) {
2044                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2045
2046                         if (skb2)
2047                                 ipmr_queue_xmit(net, mrt, true_vifi, skb2,
2048                                                 c, psend);
2049                 } else {
2050                         ipmr_queue_xmit(net, mrt, true_vifi, skb, c, psend);
2051                         return;
2052                 }
2053         }
2054
2055 dont_forward:
2056         if (!local)
2057                 kfree_skb(skb);
2058 }
2059
2060 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
2061 {
2062         struct rtable *rt = skb_rtable(skb);
2063         struct iphdr *iph = ip_hdr(skb);
2064         struct flowi4 fl4 = {
2065                 .daddr = iph->daddr,
2066                 .saddr = iph->saddr,
2067                 .flowi4_tos = RT_TOS(iph->tos),
2068                 .flowi4_oif = (rt_is_output_route(rt) ?
2069                                skb->dev->ifindex : 0),
2070                 .flowi4_iif = (rt_is_output_route(rt) ?
2071                                LOOPBACK_IFINDEX :
2072                                skb->dev->ifindex),
2073                 .flowi4_mark = skb->mark,
2074         };
2075         struct mr_table *mrt;
2076         int err;
2077
2078         err = ipmr_fib_lookup(net, &fl4, &mrt);
2079         if (err)
2080                 return ERR_PTR(err);
2081         return mrt;
2082 }
2083
2084 /* Multicast packets for forwarding arrive here
2085  * Called with rcu_read_lock();
2086  */
2087 int ip_mr_input(struct sk_buff *skb)
2088 {
2089         struct mfc_cache *cache;
2090         struct net *net = dev_net(skb->dev);
2091         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
2092         struct mr_table *mrt;
2093         struct net_device *dev;
2094
2095         /* skb->dev passed in is the loX master dev for vrfs.
2096          * As there are no vifs associated with loopback devices,
2097          * get the proper interface that does have a vif associated with it.
2098          */
2099         dev = skb->dev;
2100         if (netif_is_l3_master(skb->dev)) {
2101                 dev = dev_get_by_index_rcu(net, IPCB(skb)->iif);
2102                 if (!dev) {
2103                         kfree_skb(skb);
2104                         return -ENODEV;
2105                 }
2106         }
2107
2108         /* Packet is looped back after forward, it should not be
2109          * forwarded second time, but still can be delivered locally.
2110          */
2111         if (IPCB(skb)->flags & IPSKB_FORWARDED)
2112                 goto dont_forward;
2113
2114         mrt = ipmr_rt_fib_lookup(net, skb);
2115         if (IS_ERR(mrt)) {
2116                 kfree_skb(skb);
2117                 return PTR_ERR(mrt);
2118         }
2119         if (!local) {
2120                 if (IPCB(skb)->opt.router_alert) {
2121                         if (ip_call_ra_chain(skb))
2122                                 return 0;
2123                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
2124                         /* IGMPv1 (and broken IGMPv2 implementations sort of
2125                          * Cisco IOS <= 11.2(8)) do not put router alert
2126                          * option to IGMP packets destined to routable
2127                          * groups. It is very bad, because it means
2128                          * that we can forward NO IGMP messages.
2129                          */
2130                         struct sock *mroute_sk;
2131
2132                         mroute_sk = rcu_dereference(mrt->mroute_sk);
2133                         if (mroute_sk) {
2134                                 nf_reset(skb);
2135                                 raw_rcv(mroute_sk, skb);
2136                                 return 0;
2137                         }
2138                     }
2139         }
2140
2141         /* already under rcu_read_lock() */
2142         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
2143         if (!cache) {
2144                 int vif = ipmr_find_vif(mrt, dev);
2145
2146                 if (vif >= 0)
2147                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2148                                                     vif);
2149         }
2150
2151         /* No usable cache entry */
2152         if (!cache) {
2153                 int vif;
2154
2155                 if (local) {
2156                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2157                         ip_local_deliver(skb);
2158                         if (!skb2)
2159                                 return -ENOBUFS;
2160                         skb = skb2;
2161                 }
2162
2163                 read_lock(&mrt_lock);
2164                 vif = ipmr_find_vif(mrt, dev);
2165                 if (vif >= 0) {
2166                         int err2 = ipmr_cache_unresolved(mrt, vif, skb, dev);
2167                         read_unlock(&mrt_lock);
2168
2169                         return err2;
2170                 }
2171                 read_unlock(&mrt_lock);
2172                 kfree_skb(skb);
2173                 return -ENODEV;
2174         }
2175
2176         read_lock(&mrt_lock);
2177         ip_mr_forward(net, mrt, dev, skb, cache, local);
2178         read_unlock(&mrt_lock);
2179
2180         if (local)
2181                 return ip_local_deliver(skb);
2182
2183         return 0;
2184
2185 dont_forward:
2186         if (local)
2187                 return ip_local_deliver(skb);
2188         kfree_skb(skb);
2189         return 0;
2190 }
2191
2192 #ifdef CONFIG_IP_PIMSM_V1
2193 /* Handle IGMP messages of PIMv1 */
2194 int pim_rcv_v1(struct sk_buff *skb)
2195 {
2196         struct igmphdr *pim;
2197         struct net *net = dev_net(skb->dev);
2198         struct mr_table *mrt;
2199
2200         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2201                 goto drop;
2202
2203         pim = igmp_hdr(skb);
2204
2205         mrt = ipmr_rt_fib_lookup(net, skb);
2206         if (IS_ERR(mrt))
2207                 goto drop;
2208         if (!mrt->mroute_do_pim ||
2209             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2210                 goto drop;
2211
2212         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2213 drop:
2214                 kfree_skb(skb);
2215         }
2216         return 0;
2217 }
2218 #endif
2219
2220 #ifdef CONFIG_IP_PIMSM_V2
2221 static int pim_rcv(struct sk_buff *skb)
2222 {
2223         struct pimreghdr *pim;
2224         struct net *net = dev_net(skb->dev);
2225         struct mr_table *mrt;
2226
2227         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2228                 goto drop;
2229
2230         pim = (struct pimreghdr *)skb_transport_header(skb);
2231         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2232             (pim->flags & PIM_NULL_REGISTER) ||
2233             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2234              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2235                 goto drop;
2236
2237         mrt = ipmr_rt_fib_lookup(net, skb);
2238         if (IS_ERR(mrt))
2239                 goto drop;
2240         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2241 drop:
2242                 kfree_skb(skb);
2243         }
2244         return 0;
2245 }
2246 #endif
2247
2248 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2249                    __be32 saddr, __be32 daddr,
2250                    struct rtmsg *rtm, u32 portid)
2251 {
2252         struct mfc_cache *cache;
2253         struct mr_table *mrt;
2254         int err;
2255
2256         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2257         if (!mrt)
2258                 return -ENOENT;
2259
2260         rcu_read_lock();
2261         cache = ipmr_cache_find(mrt, saddr, daddr);
2262         if (!cache && skb->dev) {
2263                 int vif = ipmr_find_vif(mrt, skb->dev);
2264
2265                 if (vif >= 0)
2266                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2267         }
2268         if (!cache) {
2269                 struct sk_buff *skb2;
2270                 struct iphdr *iph;
2271                 struct net_device *dev;
2272                 int vif = -1;
2273
2274                 dev = skb->dev;
2275                 read_lock(&mrt_lock);
2276                 if (dev)
2277                         vif = ipmr_find_vif(mrt, dev);
2278                 if (vif < 0) {
2279                         read_unlock(&mrt_lock);
2280                         rcu_read_unlock();
2281                         return -ENODEV;
2282                 }
2283
2284                 skb2 = skb_realloc_headroom(skb, sizeof(struct iphdr));
2285                 if (!skb2) {
2286                         read_unlock(&mrt_lock);
2287                         rcu_read_unlock();
2288                         return -ENOMEM;
2289                 }
2290
2291                 NETLINK_CB(skb2).portid = portid;
2292                 skb_push(skb2, sizeof(struct iphdr));
2293                 skb_reset_network_header(skb2);
2294                 iph = ip_hdr(skb2);
2295                 iph->ihl = sizeof(struct iphdr) >> 2;
2296                 iph->saddr = saddr;
2297                 iph->daddr = daddr;
2298                 iph->version = 0;
2299                 err = ipmr_cache_unresolved(mrt, vif, skb2, dev);
2300                 read_unlock(&mrt_lock);
2301                 rcu_read_unlock();
2302                 return err;
2303         }
2304
2305         read_lock(&mrt_lock);
2306         err = mr_fill_mroute(mrt, skb, &cache->_c, rtm);
2307         read_unlock(&mrt_lock);
2308         rcu_read_unlock();
2309         return err;
2310 }
2311
2312 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2313                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2314                             int flags)
2315 {
2316         struct nlmsghdr *nlh;
2317         struct rtmsg *rtm;
2318         int err;
2319
2320         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2321         if (!nlh)
2322                 return -EMSGSIZE;
2323
2324         rtm = nlmsg_data(nlh);
2325         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2326         rtm->rtm_dst_len  = 32;
2327         rtm->rtm_src_len  = 32;
2328         rtm->rtm_tos      = 0;
2329         rtm->rtm_table    = mrt->id;
2330         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2331                 goto nla_put_failure;
2332         rtm->rtm_type     = RTN_MULTICAST;
2333         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2334         if (c->_c.mfc_flags & MFC_STATIC)
2335                 rtm->rtm_protocol = RTPROT_STATIC;
2336         else
2337                 rtm->rtm_protocol = RTPROT_MROUTED;
2338         rtm->rtm_flags    = 0;
2339
2340         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2341             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2342                 goto nla_put_failure;
2343         err = mr_fill_mroute(mrt, skb, &c->_c, rtm);
2344         /* do not break the dump if cache is unresolved */
2345         if (err < 0 && err != -ENOENT)
2346                 goto nla_put_failure;
2347
2348         nlmsg_end(skb, nlh);
2349         return 0;
2350
2351 nla_put_failure:
2352         nlmsg_cancel(skb, nlh);
2353         return -EMSGSIZE;
2354 }
2355
2356 static int _ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2357                              u32 portid, u32 seq, struct mr_mfc *c, int cmd,
2358                              int flags)
2359 {
2360         return ipmr_fill_mroute(mrt, skb, portid, seq, (struct mfc_cache *)c,
2361                                 cmd, flags);
2362 }
2363
2364 static size_t mroute_msgsize(bool unresolved, int maxvif)
2365 {
2366         size_t len =
2367                 NLMSG_ALIGN(sizeof(struct rtmsg))
2368                 + nla_total_size(4)     /* RTA_TABLE */
2369                 + nla_total_size(4)     /* RTA_SRC */
2370                 + nla_total_size(4)     /* RTA_DST */
2371                 ;
2372
2373         if (!unresolved)
2374                 len = len
2375                       + nla_total_size(4)       /* RTA_IIF */
2376                       + nla_total_size(0)       /* RTA_MULTIPATH */
2377                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2378                                                 /* RTA_MFC_STATS */
2379                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2380                 ;
2381
2382         return len;
2383 }
2384
2385 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2386                                  int cmd)
2387 {
2388         struct net *net = read_pnet(&mrt->net);
2389         struct sk_buff *skb;
2390         int err = -ENOBUFS;
2391
2392         skb = nlmsg_new(mroute_msgsize(mfc->_c.mfc_parent >= MAXVIFS,
2393                                        mrt->maxvif),
2394                         GFP_ATOMIC);
2395         if (!skb)
2396                 goto errout;
2397
2398         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2399         if (err < 0)
2400                 goto errout;
2401
2402         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2403         return;
2404
2405 errout:
2406         kfree_skb(skb);
2407         if (err < 0)
2408                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2409 }
2410
2411 static size_t igmpmsg_netlink_msgsize(size_t payloadlen)
2412 {
2413         size_t len =
2414                 NLMSG_ALIGN(sizeof(struct rtgenmsg))
2415                 + nla_total_size(1)     /* IPMRA_CREPORT_MSGTYPE */
2416                 + nla_total_size(4)     /* IPMRA_CREPORT_VIF_ID */
2417                 + nla_total_size(4)     /* IPMRA_CREPORT_SRC_ADDR */
2418                 + nla_total_size(4)     /* IPMRA_CREPORT_DST_ADDR */
2419                                         /* IPMRA_CREPORT_PKT */
2420                 + nla_total_size(payloadlen)
2421                 ;
2422
2423         return len;
2424 }
2425
2426 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt)
2427 {
2428         struct net *net = read_pnet(&mrt->net);
2429         struct nlmsghdr *nlh;
2430         struct rtgenmsg *rtgenm;
2431         struct igmpmsg *msg;
2432         struct sk_buff *skb;
2433         struct nlattr *nla;
2434         int payloadlen;
2435
2436         payloadlen = pkt->len - sizeof(struct igmpmsg);
2437         msg = (struct igmpmsg *)skb_network_header(pkt);
2438
2439         skb = nlmsg_new(igmpmsg_netlink_msgsize(payloadlen), GFP_ATOMIC);
2440         if (!skb)
2441                 goto errout;
2442
2443         nlh = nlmsg_put(skb, 0, 0, RTM_NEWCACHEREPORT,
2444                         sizeof(struct rtgenmsg), 0);
2445         if (!nlh)
2446                 goto errout;
2447         rtgenm = nlmsg_data(nlh);
2448         rtgenm->rtgen_family = RTNL_FAMILY_IPMR;
2449         if (nla_put_u8(skb, IPMRA_CREPORT_MSGTYPE, msg->im_msgtype) ||
2450             nla_put_u32(skb, IPMRA_CREPORT_VIF_ID, msg->im_vif) ||
2451             nla_put_in_addr(skb, IPMRA_CREPORT_SRC_ADDR,
2452                             msg->im_src.s_addr) ||
2453             nla_put_in_addr(skb, IPMRA_CREPORT_DST_ADDR,
2454                             msg->im_dst.s_addr))
2455                 goto nla_put_failure;
2456
2457         nla = nla_reserve(skb, IPMRA_CREPORT_PKT, payloadlen);
2458         if (!nla || skb_copy_bits(pkt, sizeof(struct igmpmsg),
2459                                   nla_data(nla), payloadlen))
2460                 goto nla_put_failure;
2461
2462         nlmsg_end(skb, nlh);
2463
2464         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE_R, NULL, GFP_ATOMIC);
2465         return;
2466
2467 nla_put_failure:
2468         nlmsg_cancel(skb, nlh);
2469 errout:
2470         kfree_skb(skb);
2471         rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE_R, -ENOBUFS);
2472 }
2473
2474 static int ipmr_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2475                              struct netlink_ext_ack *extack)
2476 {
2477         struct net *net = sock_net(in_skb->sk);
2478         struct nlattr *tb[RTA_MAX + 1];
2479         struct sk_buff *skb = NULL;
2480         struct mfc_cache *cache;
2481         struct mr_table *mrt;
2482         struct rtmsg *rtm;
2483         __be32 src, grp;
2484         u32 tableid;
2485         int err;
2486
2487         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX,
2488                           rtm_ipv4_policy, extack);
2489         if (err < 0)
2490                 goto errout;
2491
2492         rtm = nlmsg_data(nlh);
2493
2494         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2495         grp = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2496         tableid = tb[RTA_TABLE] ? nla_get_u32(tb[RTA_TABLE]) : 0;
2497
2498         mrt = ipmr_get_table(net, tableid ? tableid : RT_TABLE_DEFAULT);
2499         if (!mrt) {
2500                 err = -ENOENT;
2501                 goto errout_free;
2502         }
2503
2504         /* entries are added/deleted only under RTNL */
2505         rcu_read_lock();
2506         cache = ipmr_cache_find(mrt, src, grp);
2507         rcu_read_unlock();
2508         if (!cache) {
2509                 err = -ENOENT;
2510                 goto errout_free;
2511         }
2512
2513         skb = nlmsg_new(mroute_msgsize(false, mrt->maxvif), GFP_KERNEL);
2514         if (!skb) {
2515                 err = -ENOBUFS;
2516                 goto errout_free;
2517         }
2518
2519         err = ipmr_fill_mroute(mrt, skb, NETLINK_CB(in_skb).portid,
2520                                nlh->nlmsg_seq, cache,
2521                                RTM_NEWROUTE, 0);
2522         if (err < 0)
2523                 goto errout_free;
2524
2525         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2526
2527 errout:
2528         return err;
2529
2530 errout_free:
2531         kfree_skb(skb);
2532         goto errout;
2533 }
2534
2535 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2536 {
2537         return mr_rtm_dumproute(skb, cb, ipmr_mr_table_iter,
2538                                 _ipmr_fill_mroute, &mfc_unres_lock);
2539 }
2540
2541 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2542         [RTA_SRC]       = { .type = NLA_U32 },
2543         [RTA_DST]       = { .type = NLA_U32 },
2544         [RTA_IIF]       = { .type = NLA_U32 },
2545         [RTA_TABLE]     = { .type = NLA_U32 },
2546         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2547 };
2548
2549 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2550 {
2551         switch (rtm_protocol) {
2552         case RTPROT_STATIC:
2553         case RTPROT_MROUTED:
2554                 return true;
2555         }
2556         return false;
2557 }
2558
2559 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2560 {
2561         struct rtnexthop *rtnh = nla_data(nla);
2562         int remaining = nla_len(nla), vifi = 0;
2563
2564         while (rtnh_ok(rtnh, remaining)) {
2565                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2566                 if (++vifi == MAXVIFS)
2567                         break;
2568                 rtnh = rtnh_next(rtnh, &remaining);
2569         }
2570
2571         return remaining > 0 ? -EINVAL : vifi;
2572 }
2573
2574 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2575 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2576                             struct mfcctl *mfcc, int *mrtsock,
2577                             struct mr_table **mrtret,
2578                             struct netlink_ext_ack *extack)
2579 {
2580         struct net_device *dev = NULL;
2581         u32 tblid = RT_TABLE_DEFAULT;
2582         struct mr_table *mrt;
2583         struct nlattr *attr;
2584         struct rtmsg *rtm;
2585         int ret, rem;
2586
2587         ret = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipmr_policy,
2588                              extack);
2589         if (ret < 0)
2590                 goto out;
2591         rtm = nlmsg_data(nlh);
2592
2593         ret = -EINVAL;
2594         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2595             rtm->rtm_type != RTN_MULTICAST ||
2596             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2597             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2598                 goto out;
2599
2600         memset(mfcc, 0, sizeof(*mfcc));
2601         mfcc->mfcc_parent = -1;
2602         ret = 0;
2603         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2604                 switch (nla_type(attr)) {
2605                 case RTA_SRC:
2606                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2607                         break;
2608                 case RTA_DST:
2609                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2610                         break;
2611                 case RTA_IIF:
2612                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2613                         if (!dev) {
2614                                 ret = -ENODEV;
2615                                 goto out;
2616                         }
2617                         break;
2618                 case RTA_MULTIPATH:
2619                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2620                                 ret = -EINVAL;
2621                                 goto out;
2622                         }
2623                         break;
2624                 case RTA_PREFSRC:
2625                         ret = 1;
2626                         break;
2627                 case RTA_TABLE:
2628                         tblid = nla_get_u32(attr);
2629                         break;
2630                 }
2631         }
2632         mrt = ipmr_get_table(net, tblid);
2633         if (!mrt) {
2634                 ret = -ENOENT;
2635                 goto out;
2636         }
2637         *mrtret = mrt;
2638         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2639         if (dev)
2640                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2641
2642 out:
2643         return ret;
2644 }
2645
2646 /* takes care of both newroute and delroute */
2647 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,
2648                           struct netlink_ext_ack *extack)
2649 {
2650         struct net *net = sock_net(skb->sk);
2651         int ret, mrtsock, parent;
2652         struct mr_table *tbl;
2653         struct mfcctl mfcc;
2654
2655         mrtsock = 0;
2656         tbl = NULL;
2657         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl, extack);
2658         if (ret < 0)
2659                 return ret;
2660
2661         parent = ret ? mfcc.mfcc_parent : -1;
2662         if (nlh->nlmsg_type == RTM_NEWROUTE)
2663                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2664         else
2665                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2666 }
2667
2668 static bool ipmr_fill_table(struct mr_table *mrt, struct sk_buff *skb)
2669 {
2670         u32 queue_len = atomic_read(&mrt->cache_resolve_queue_len);
2671
2672         if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) ||
2673             nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
2674             nla_put_s32(skb, IPMRA_TABLE_MROUTE_REG_VIF_NUM,
2675                         mrt->mroute_reg_vif_num) ||
2676             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_ASSERT,
2677                        mrt->mroute_do_assert) ||
2678             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_PIM, mrt->mroute_do_pim) ||
2679             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_WRVIFWHOLE,
2680                        mrt->mroute_do_wrvifwhole))
2681                 return false;
2682
2683         return true;
2684 }
2685
2686 static bool ipmr_fill_vif(struct mr_table *mrt, u32 vifid, struct sk_buff *skb)
2687 {
2688         struct nlattr *vif_nest;
2689         struct vif_device *vif;
2690
2691         /* if the VIF doesn't exist just continue */
2692         if (!VIF_EXISTS(mrt, vifid))
2693                 return true;
2694
2695         vif = &mrt->vif_table[vifid];
2696         vif_nest = nla_nest_start(skb, IPMRA_VIF);
2697         if (!vif_nest)
2698                 return false;
2699         if (nla_put_u32(skb, IPMRA_VIFA_IFINDEX, vif->dev->ifindex) ||
2700             nla_put_u32(skb, IPMRA_VIFA_VIF_ID, vifid) ||
2701             nla_put_u16(skb, IPMRA_VIFA_FLAGS, vif->flags) ||
2702             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_IN, vif->bytes_in,
2703                               IPMRA_VIFA_PAD) ||
2704             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_OUT, vif->bytes_out,
2705                               IPMRA_VIFA_PAD) ||
2706             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_IN, vif->pkt_in,
2707                               IPMRA_VIFA_PAD) ||
2708             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_OUT, vif->pkt_out,
2709                               IPMRA_VIFA_PAD) ||
2710             nla_put_be32(skb, IPMRA_VIFA_LOCAL_ADDR, vif->local) ||
2711             nla_put_be32(skb, IPMRA_VIFA_REMOTE_ADDR, vif->remote)) {
2712                 nla_nest_cancel(skb, vif_nest);
2713                 return false;
2714         }
2715         nla_nest_end(skb, vif_nest);
2716
2717         return true;
2718 }
2719
2720 static int ipmr_rtm_dumplink(struct sk_buff *skb, struct netlink_callback *cb)
2721 {
2722         struct net *net = sock_net(skb->sk);
2723         struct nlmsghdr *nlh = NULL;
2724         unsigned int t = 0, s_t;
2725         unsigned int e = 0, s_e;
2726         struct mr_table *mrt;
2727
2728         s_t = cb->args[0];
2729         s_e = cb->args[1];
2730
2731         ipmr_for_each_table(mrt, net) {
2732                 struct nlattr *vifs, *af;
2733                 struct ifinfomsg *hdr;
2734                 u32 i;
2735
2736                 if (t < s_t)
2737                         goto skip_table;
2738                 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2739                                 cb->nlh->nlmsg_seq, RTM_NEWLINK,
2740                                 sizeof(*hdr), NLM_F_MULTI);
2741                 if (!nlh)
2742                         break;
2743
2744                 hdr = nlmsg_data(nlh);
2745                 memset(hdr, 0, sizeof(*hdr));
2746                 hdr->ifi_family = RTNL_FAMILY_IPMR;
2747
2748                 af = nla_nest_start(skb, IFLA_AF_SPEC);
2749                 if (!af) {
2750                         nlmsg_cancel(skb, nlh);
2751                         goto out;
2752                 }
2753
2754                 if (!ipmr_fill_table(mrt, skb)) {
2755                         nlmsg_cancel(skb, nlh);
2756                         goto out;
2757                 }
2758
2759                 vifs = nla_nest_start(skb, IPMRA_TABLE_VIFS);
2760                 if (!vifs) {
2761                         nla_nest_end(skb, af);
2762                         nlmsg_end(skb, nlh);
2763                         goto out;
2764                 }
2765                 for (i = 0; i < mrt->maxvif; i++) {
2766                         if (e < s_e)
2767                                 goto skip_entry;
2768                         if (!ipmr_fill_vif(mrt, i, skb)) {
2769                                 nla_nest_end(skb, vifs);
2770                                 nla_nest_end(skb, af);
2771                                 nlmsg_end(skb, nlh);
2772                                 goto out;
2773                         }
2774 skip_entry:
2775                         e++;
2776                 }
2777                 s_e = 0;
2778                 e = 0;
2779                 nla_nest_end(skb, vifs);
2780                 nla_nest_end(skb, af);
2781                 nlmsg_end(skb, nlh);
2782 skip_table:
2783                 t++;
2784         }
2785
2786 out:
2787         cb->args[1] = e;
2788         cb->args[0] = t;
2789
2790         return skb->len;
2791 }
2792
2793 #ifdef CONFIG_PROC_FS
2794 /* The /proc interfaces to multicast routing :
2795  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2796  */
2797
2798 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2799         __acquires(mrt_lock)
2800 {
2801         struct mr_vif_iter *iter = seq->private;
2802         struct net *net = seq_file_net(seq);
2803         struct mr_table *mrt;
2804
2805         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2806         if (!mrt)
2807                 return ERR_PTR(-ENOENT);
2808
2809         iter->mrt = mrt;
2810
2811         read_lock(&mrt_lock);
2812         return mr_vif_seq_start(seq, pos);
2813 }
2814
2815 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2816         __releases(mrt_lock)
2817 {
2818         read_unlock(&mrt_lock);
2819 }
2820
2821 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2822 {
2823         struct mr_vif_iter *iter = seq->private;
2824         struct mr_table *mrt = iter->mrt;
2825
2826         if (v == SEQ_START_TOKEN) {
2827                 seq_puts(seq,
2828                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2829         } else {
2830                 const struct vif_device *vif = v;
2831                 const char *name =  vif->dev ?
2832                                     vif->dev->name : "none";
2833
2834                 seq_printf(seq,
2835                            "%2td %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2836                            vif - mrt->vif_table,
2837                            name, vif->bytes_in, vif->pkt_in,
2838                            vif->bytes_out, vif->pkt_out,
2839                            vif->flags, vif->local, vif->remote);
2840         }
2841         return 0;
2842 }
2843
2844 static const struct seq_operations ipmr_vif_seq_ops = {
2845         .start = ipmr_vif_seq_start,
2846         .next  = mr_vif_seq_next,
2847         .stop  = ipmr_vif_seq_stop,
2848         .show  = ipmr_vif_seq_show,
2849 };
2850
2851 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2852 {
2853         struct net *net = seq_file_net(seq);
2854         struct mr_table *mrt;
2855
2856         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2857         if (!mrt)
2858                 return ERR_PTR(-ENOENT);
2859
2860         return mr_mfc_seq_start(seq, pos, mrt, &mfc_unres_lock);
2861 }
2862
2863 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2864 {
2865         int n;
2866
2867         if (v == SEQ_START_TOKEN) {
2868                 seq_puts(seq,
2869                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2870         } else {
2871                 const struct mfc_cache *mfc = v;
2872                 const struct mr_mfc_iter *it = seq->private;
2873                 const struct mr_table *mrt = it->mrt;
2874
2875                 seq_printf(seq, "%08X %08X %-3hd",
2876                            (__force u32) mfc->mfc_mcastgrp,
2877                            (__force u32) mfc->mfc_origin,
2878                            mfc->_c.mfc_parent);
2879
2880                 if (it->cache != &mrt->mfc_unres_queue) {
2881                         seq_printf(seq, " %8lu %8lu %8lu",
2882                                    mfc->_c.mfc_un.res.pkt,
2883                                    mfc->_c.mfc_un.res.bytes,
2884                                    mfc->_c.mfc_un.res.wrong_if);
2885                         for (n = mfc->_c.mfc_un.res.minvif;
2886                              n < mfc->_c.mfc_un.res.maxvif; n++) {
2887                                 if (VIF_EXISTS(mrt, n) &&
2888                                     mfc->_c.mfc_un.res.ttls[n] < 255)
2889                                         seq_printf(seq,
2890                                            " %2d:%-3d",
2891                                            n, mfc->_c.mfc_un.res.ttls[n]);
2892                         }
2893                 } else {
2894                         /* unresolved mfc_caches don't contain
2895                          * pkt, bytes and wrong_if values
2896                          */
2897                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2898                 }
2899                 seq_putc(seq, '\n');
2900         }
2901         return 0;
2902 }
2903
2904 static const struct seq_operations ipmr_mfc_seq_ops = {
2905         .start = ipmr_mfc_seq_start,
2906         .next  = mr_mfc_seq_next,
2907         .stop  = mr_mfc_seq_stop,
2908         .show  = ipmr_mfc_seq_show,
2909 };
2910 #endif
2911
2912 #ifdef CONFIG_IP_PIMSM_V2
2913 static const struct net_protocol pim_protocol = {
2914         .handler        =       pim_rcv,
2915         .netns_ok       =       1,
2916 };
2917 #endif
2918
2919 static unsigned int ipmr_seq_read(struct net *net)
2920 {
2921         ASSERT_RTNL();
2922
2923         return net->ipv4.ipmr_seq + ipmr_rules_seq_read(net);
2924 }
2925
2926 static int ipmr_dump(struct net *net, struct notifier_block *nb)
2927 {
2928         return mr_dump(net, nb, RTNL_FAMILY_IPMR, ipmr_rules_dump,
2929                        ipmr_mr_table_iter, &mrt_lock);
2930 }
2931
2932 static const struct fib_notifier_ops ipmr_notifier_ops_template = {
2933         .family         = RTNL_FAMILY_IPMR,
2934         .fib_seq_read   = ipmr_seq_read,
2935         .fib_dump       = ipmr_dump,
2936         .owner          = THIS_MODULE,
2937 };
2938
2939 static int __net_init ipmr_notifier_init(struct net *net)
2940 {
2941         struct fib_notifier_ops *ops;
2942
2943         net->ipv4.ipmr_seq = 0;
2944
2945         ops = fib_notifier_ops_register(&ipmr_notifier_ops_template, net);
2946         if (IS_ERR(ops))
2947                 return PTR_ERR(ops);
2948         net->ipv4.ipmr_notifier_ops = ops;
2949
2950         return 0;
2951 }
2952
2953 static void __net_exit ipmr_notifier_exit(struct net *net)
2954 {
2955         fib_notifier_ops_unregister(net->ipv4.ipmr_notifier_ops);
2956         net->ipv4.ipmr_notifier_ops = NULL;
2957 }
2958
2959 /* Setup for IP multicast routing */
2960 static int __net_init ipmr_net_init(struct net *net)
2961 {
2962         int err;
2963
2964         err = ipmr_notifier_init(net);
2965         if (err)
2966                 goto ipmr_notifier_fail;
2967
2968         err = ipmr_rules_init(net);
2969         if (err < 0)
2970                 goto ipmr_rules_fail;
2971
2972 #ifdef CONFIG_PROC_FS
2973         err = -ENOMEM;
2974         if (!proc_create_net("ip_mr_vif", 0, net->proc_net, &ipmr_vif_seq_ops,
2975                         sizeof(struct mr_vif_iter)))
2976                 goto proc_vif_fail;
2977         if (!proc_create_net("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_seq_ops,
2978                         sizeof(struct mr_mfc_iter)))
2979                 goto proc_cache_fail;
2980 #endif
2981         return 0;
2982
2983 #ifdef CONFIG_PROC_FS
2984 proc_cache_fail:
2985         remove_proc_entry("ip_mr_vif", net->proc_net);
2986 proc_vif_fail:
2987         ipmr_rules_exit(net);
2988 #endif
2989 ipmr_rules_fail:
2990         ipmr_notifier_exit(net);
2991 ipmr_notifier_fail:
2992         return err;
2993 }
2994
2995 static void __net_exit ipmr_net_exit(struct net *net)
2996 {
2997 #ifdef CONFIG_PROC_FS
2998         remove_proc_entry("ip_mr_cache", net->proc_net);
2999         remove_proc_entry("ip_mr_vif", net->proc_net);
3000 #endif
3001         ipmr_notifier_exit(net);
3002         ipmr_rules_exit(net);
3003 }
3004
3005 static struct pernet_operations ipmr_net_ops = {
3006         .init = ipmr_net_init,
3007         .exit = ipmr_net_exit,
3008 };
3009
3010 int __init ip_mr_init(void)
3011 {
3012         int err;
3013
3014         mrt_cachep = kmem_cache_create("ip_mrt_cache",
3015                                        sizeof(struct mfc_cache),
3016                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
3017                                        NULL);
3018
3019         err = register_pernet_subsys(&ipmr_net_ops);
3020         if (err)
3021                 goto reg_pernet_fail;
3022
3023         err = register_netdevice_notifier(&ip_mr_notifier);
3024         if (err)
3025                 goto reg_notif_fail;
3026 #ifdef CONFIG_IP_PIMSM_V2
3027         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
3028                 pr_err("%s: can't add PIM protocol\n", __func__);
3029                 err = -EAGAIN;
3030                 goto add_proto_fail;
3031         }
3032 #endif
3033         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
3034                       ipmr_rtm_getroute, ipmr_rtm_dumproute, 0);
3035         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
3036                       ipmr_rtm_route, NULL, 0);
3037         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
3038                       ipmr_rtm_route, NULL, 0);
3039
3040         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETLINK,
3041                       NULL, ipmr_rtm_dumplink, 0);
3042         return 0;
3043
3044 #ifdef CONFIG_IP_PIMSM_V2
3045 add_proto_fail:
3046         unregister_netdevice_notifier(&ip_mr_notifier);
3047 #endif
3048 reg_notif_fail:
3049         unregister_pernet_subsys(&ipmr_net_ops);
3050 reg_pernet_fail:
3051         kmem_cache_destroy(mrt_cachep);
3052         return err;
3053 }