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 <net/ip_tunnels.h>
  64 #include <net/checksum.h>
  65 #include <net/netlink.h>
  66 #include <net/fib_rules.h>
  67 #include <linux/netconf.h>
  68 #include <net/nexthop.h>
  69 #include <net/switchdev.h>
  70
  71 struct ipmr_rule {
  72         struct fib_rule         common;
  73 };
  74
  75 struct ipmr_result {
  76         struct mr_table         *mrt;
  77 };
  78
  79 /* Big lock, protecting vif table, mrt cache and mroute socket state.
  80  * Note that the changes are semaphored via rtnl_lock.
  81  */
  82
  83 static DEFINE_RWLOCK(mrt_lock);
  84
  85 /* Multicast router control variables */
  86
  87 /* Special spinlock for queue of unresolved entries */
  88 static DEFINE_SPINLOCK(mfc_unres_lock);
  89
  90 /* We return to original Alan's scheme. Hash table of resolved
  91  * entries is changed only in process context and protected
  92  * with weak lock mrt_lock. Queue of unresolved entries is protected
  93  * with strong spinlock mfc_unres_lock.
  94  *
  95  * In this case data path is free of exclusive locks at all.
  96  */
  97
  98 static struct kmem_cache *mrt_cachep __ro_after_init;
  99
 100 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 101 static void ipmr_free_table(struct mr_table *mrt);
 102
 103 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
 104                           struct net_device *dev, struct sk_buff *skb,
 105                           struct mfc_cache *cache, int local);
 106 static int ipmr_cache_report(struct mr_table *mrt,
 107                              struct sk_buff *pkt, vifi_t vifi, int assert);
 108 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
 109                                  int cmd);
 110 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt);
 111 static void mroute_clean_tables(struct mr_table *mrt, bool all);
 112 static void ipmr_expire_process(struct timer_list *t);
 113
 114 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 115 #define ipmr_for_each_table(mrt, net) \
 116         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
 117
 118 static struct mr_table *ipmr_mr_table_iter(struct net *net,
 119                                            struct mr_table *mrt)
 120 {
 121         struct mr_table *ret;
 122
 123         if (!mrt)
 124                 ret = list_entry_rcu(net->ipv4.mr_tables.next,
 125                                      struct mr_table, list);
 126         else
 127                 ret = list_entry_rcu(mrt->list.next,
 128                                      struct mr_table, list);
 129
 130         if (&ret->list == &net->ipv4.mr_tables)
 131                 return NULL;
 132         return ret;
 133 }
 134
 135 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 136 {
 137         struct mr_table *mrt;
 138
 139         ipmr_for_each_table(mrt, net) {
 140                 if (mrt->id == id)
 141                         return mrt;
 142         }
 143         return NULL;
 144 }
 145
 146 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 147                            struct mr_table **mrt)
 148 {
 149         int err;
 150         struct ipmr_result res;
 151         struct fib_lookup_arg arg = {
 152                 .result = &res,
 153                 .flags = FIB_LOOKUP_NOREF,
 154         };
 155
 156         /* update flow if oif or iif point to device enslaved to l3mdev */
 157         l3mdev_update_flow(net, flowi4_to_flowi(flp4));
 158
 159         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 160                                flowi4_to_flowi(flp4), 0, &arg);
 161         if (err < 0)
 162                 return err;
 163         *mrt = res.mrt;
 164         return 0;
 165 }
 166
 167 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 168                             int flags, struct fib_lookup_arg *arg)
 169 {
 170         struct ipmr_result *res = arg->result;
 171         struct mr_table *mrt;
 172
 173         switch (rule->action) {
 174         case FR_ACT_TO_TBL:
 175                 break;
 176         case FR_ACT_UNREACHABLE:
 177                 return -ENETUNREACH;
 178         case FR_ACT_PROHIBIT:
 179                 return -EACCES;
 180         case FR_ACT_BLACKHOLE:
 181         default:
 182                 return -EINVAL;
 183         }
 184
 185         arg->table = fib_rule_get_table(rule, arg);
 186
 187         mrt = ipmr_get_table(rule->fr_net, arg->table);
 188         if (!mrt)
 189                 return -EAGAIN;
 190         res->mrt = mrt;
 191         return 0;
 192 }
 193
 194 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 195 {
 196         return 1;
 197 }
 198
 199 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 200         FRA_GENERIC_POLICY,
 201 };
 202
 203 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 204                                struct fib_rule_hdr *frh, struct nlattr **tb,
 205                                struct netlink_ext_ack *extack)
 206 {
 207         return 0;
 208 }
 209
 210 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 211                              struct nlattr **tb)
 212 {
 213         return 1;
 214 }
 215
 216 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 217                           struct fib_rule_hdr *frh)
 218 {
 219         frh->dst_len = 0;
 220         frh->src_len = 0;
 221         frh->tos     = 0;
 222         return 0;
 223 }
 224
 225 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
 226         .family         = RTNL_FAMILY_IPMR,
 227         .rule_size      = sizeof(struct ipmr_rule),
 228         .addr_size      = sizeof(u32),
 229         .action         = ipmr_rule_action,
 230         .match          = ipmr_rule_match,
 231         .configure      = ipmr_rule_configure,
 232         .compare        = ipmr_rule_compare,
 233         .fill           = ipmr_rule_fill,
 234         .nlgroup        = RTNLGRP_IPV4_RULE,
 235         .policy         = ipmr_rule_policy,
 236         .owner          = THIS_MODULE,
 237 };
 238
 239 static int __net_init ipmr_rules_init(struct net *net)
 240 {
 241         struct fib_rules_ops *ops;
 242         struct mr_table *mrt;
 243         int err;
 244
 245         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 246         if (IS_ERR(ops))
 247                 return PTR_ERR(ops);
 248
 249         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 250
 251         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 252         if (IS_ERR(mrt)) {
 253                 err = PTR_ERR(mrt);
 254                 goto err1;
 255         }
 256
 257         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 258         if (err < 0)
 259                 goto err2;
 260
 261         net->ipv4.mr_rules_ops = ops;
 262         return 0;
 263
 264 err2:
 265         ipmr_free_table(mrt);
 266 err1:
 267         fib_rules_unregister(ops);
 268         return err;
 269 }
 270
 271 static void __net_exit ipmr_rules_exit(struct net *net)
 272 {
 273         struct mr_table *mrt, *next;
 274
 275         rtnl_lock();
 276         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 277                 list_del(&mrt->list);
 278                 ipmr_free_table(mrt);
 279         }
 280         fib_rules_unregister(net->ipv4.mr_rules_ops);
 281         rtnl_unlock();
 282 }
 283
 284 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb)
 285 {
 286         return fib_rules_dump(net, nb, RTNL_FAMILY_IPMR);
 287 }
 288
 289 static unsigned int ipmr_rules_seq_read(struct net *net)
 290 {
 291         return fib_rules_seq_read(net, RTNL_FAMILY_IPMR);
 292 }
 293
 294 bool ipmr_rule_default(const struct fib_rule *rule)
 295 {
 296         return fib_rule_matchall(rule) && rule->table == RT_TABLE_DEFAULT;
 297 }
 298 EXPORT_SYMBOL(ipmr_rule_default);
 299 #else
 300 #define ipmr_for_each_table(mrt, net) \
 301         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 302
 303 static struct mr_table *ipmr_mr_table_iter(struct net *net,
 304                                            struct mr_table *mrt)
 305 {
 306         if (!mrt)
 307                 return net->ipv4.mrt;
 308         return NULL;
 309 }
 310
 311 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 312 {
 313         return net->ipv4.mrt;
 314 }
 315
 316 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 317                            struct mr_table **mrt)
 318 {
 319         *mrt = net->ipv4.mrt;
 320         return 0;
 321 }
 322
 323 static int __net_init ipmr_rules_init(struct net *net)
 324 {
 325         struct mr_table *mrt;
 326
 327         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 328         if (IS_ERR(mrt))
 329                 return PTR_ERR(mrt);
 330         net->ipv4.mrt = mrt;
 331         return 0;
 332 }
 333
 334 static void __net_exit ipmr_rules_exit(struct net *net)
 335 {
 336         rtnl_lock();
 337         ipmr_free_table(net->ipv4.mrt);
 338         net->ipv4.mrt = NULL;
 339         rtnl_unlock();
 340 }
 341
 342 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb)
 343 {
 344         return 0;
 345 }
 346
 347 static unsigned int ipmr_rules_seq_read(struct net *net)
 348 {
 349         return 0;
 350 }
 351
 352 bool ipmr_rule_default(const struct fib_rule *rule)
 353 {
 354         return true;
 355 }
 356 EXPORT_SYMBOL(ipmr_rule_default);
 357 #endif
 358
 359 static inline int ipmr_hash_cmp(struct rhashtable_compare_arg *arg,
 360                                 const void *ptr)
 361 {
 362         const struct mfc_cache_cmp_arg *cmparg = arg->key;
 363         struct mfc_cache *c = (struct mfc_cache *)ptr;
 364
 365         return cmparg->mfc_mcastgrp != c->mfc_mcastgrp ||
 366                cmparg->mfc_origin != c->mfc_origin;
 367 }
 368
 369 static const struct rhashtable_params ipmr_rht_params = {
 370         .head_offset = offsetof(struct mr_mfc, mnode),
 371         .key_offset = offsetof(struct mfc_cache, cmparg),
 372         .key_len = sizeof(struct mfc_cache_cmp_arg),
 373         .nelem_hint = 3,
 374         .locks_mul = 1,
 375         .obj_cmpfn = ipmr_hash_cmp,
 376         .automatic_shrinking = true,
 377 };
 378
 379 static void ipmr_new_table_set(struct mr_table *mrt,
 380                                struct net *net)
 381 {
 382 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 383         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 384 #endif
 385 }
 386
 387 static struct mfc_cache_cmp_arg ipmr_mr_table_ops_cmparg_any = {
 388         .mfc_mcastgrp = htonl(INADDR_ANY),
 389         .mfc_origin = htonl(INADDR_ANY),
 390 };
 391
 392 static struct mr_table_ops ipmr_mr_table_ops = {
 393         .rht_params = &ipmr_rht_params,
 394         .cmparg_any = &ipmr_mr_table_ops_cmparg_any,
 395 };
 396
 397 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 398 {
 399         struct mr_table *mrt;
 400
 401         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 402         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 403                 return ERR_PTR(-EINVAL);
 404
 405         mrt = ipmr_get_table(net, id);
 406         if (mrt)
 407                 return mrt;
 408
 409         return mr_table_alloc(net, id, &ipmr_mr_table_ops,
 410                               ipmr_expire_process, ipmr_new_table_set);
 411 }
 412
 413 static void ipmr_free_table(struct mr_table *mrt)
 414 {
 415         del_timer_sync(&mrt->ipmr_expire_timer);
 416         mroute_clean_tables(mrt, true);
 417         rhltable_destroy(&mrt->mfc_hash);
 418         kfree(mrt);
 419 }
 420
 421 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 422
 423 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 424 {
 425         struct net *net = dev_net(dev);
 426
 427         dev_close(dev);
 428
 429         dev = __dev_get_by_name(net, "tunl0");
 430         if (dev) {
 431                 const struct net_device_ops *ops = dev->netdev_ops;
 432                 struct ifreq ifr;
 433                 struct ip_tunnel_parm p;
 434
 435                 memset(&p, 0, sizeof(p));
 436                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 437                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 438                 p.iph.version = 4;
 439                 p.iph.ihl = 5;
 440                 p.iph.protocol = IPPROTO_IPIP;
 441                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 442                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 443
 444                 if (ops->ndo_do_ioctl) {
 445                         mm_segment_t oldfs = get_fs();
 446
 447                         set_fs(KERNEL_DS);
 448                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 449                         set_fs(oldfs);
 450                 }
 451         }
 452 }
 453
 454 /* Initialize ipmr pimreg/tunnel in_device */
 455 static bool ipmr_init_vif_indev(const struct net_device *dev)
 456 {
 457         struct in_device *in_dev;
 458
 459         ASSERT_RTNL();
 460
 461         in_dev = __in_dev_get_rtnl(dev);
 462         if (!in_dev)
 463                 return false;
 464         ipv4_devconf_setall(in_dev);
 465         neigh_parms_data_state_setall(in_dev->arp_parms);
 466         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 467
 468         return true;
 469 }
 470
 471 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 472 {
 473         struct net_device  *dev;
 474
 475         dev = __dev_get_by_name(net, "tunl0");
 476
 477         if (dev) {
 478                 const struct net_device_ops *ops = dev->netdev_ops;
 479                 int err;
 480                 struct ifreq ifr;
 481                 struct ip_tunnel_parm p;
 482
 483                 memset(&p, 0, sizeof(p));
 484                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 485                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 486                 p.iph.version = 4;
 487                 p.iph.ihl = 5;
 488                 p.iph.protocol = IPPROTO_IPIP;
 489                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 490                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 491
 492                 if (ops->ndo_do_ioctl) {
 493                         mm_segment_t oldfs = get_fs();
 494
 495                         set_fs(KERNEL_DS);
 496                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 497                         set_fs(oldfs);
 498                 } else {
 499                         err = -EOPNOTSUPP;
 500                 }
 501                 dev = NULL;
 502
 503                 if (err == 0 &&
 504                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 505                         dev->flags |= IFF_MULTICAST;
 506                         if (!ipmr_init_vif_indev(dev))
 507                                 goto failure;
 508                         if (dev_open(dev))
 509                                 goto failure;
 510                         dev_hold(dev);
 511                 }
 512         }
 513         return dev;
 514
 515 failure:
 516         unregister_netdevice(dev);
 517         return NULL;
 518 }
 519
 520 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 521 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 522 {
 523         struct net *net = dev_net(dev);
 524         struct mr_table *mrt;
 525         struct flowi4 fl4 = {
 526                 .flowi4_oif     = dev->ifindex,
 527                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 528                 .flowi4_mark    = skb->mark,
 529         };
 530         int err;
 531
 532         err = ipmr_fib_lookup(net, &fl4, &mrt);
 533         if (err < 0) {
 534                 kfree_skb(skb);
 535                 return err;
 536         }
 537
 538         read_lock(&mrt_lock);
 539         dev->stats.tx_bytes += skb->len;
 540         dev->stats.tx_packets++;
 541         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 542         read_unlock(&mrt_lock);
 543         kfree_skb(skb);
 544         return NETDEV_TX_OK;
 545 }
 546
 547 static int reg_vif_get_iflink(const struct net_device *dev)
 548 {
 549         return 0;
 550 }
 551
 552 static const struct net_device_ops reg_vif_netdev_ops = {
 553         .ndo_start_xmit = reg_vif_xmit,
 554         .ndo_get_iflink = reg_vif_get_iflink,
 555 };
 556
 557 static void reg_vif_setup(struct net_device *dev)
 558 {
 559         dev->type               = ARPHRD_PIMREG;
 560         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 561         dev->flags              = IFF_NOARP;
 562         dev->netdev_ops         = &reg_vif_netdev_ops;
 563         dev->needs_free_netdev  = true;
 564         dev->features           |= NETIF_F_NETNS_LOCAL;
 565 }
 566
 567 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 568 {
 569         struct net_device *dev;
 570         char name[IFNAMSIZ];
 571
 572         if (mrt->id == RT_TABLE_DEFAULT)
 573                 sprintf(name, "pimreg");
 574         else
 575                 sprintf(name, "pimreg%u", mrt->id);
 576
 577         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 578
 579         if (!dev)
 580                 return NULL;
 581
 582         dev_net_set(dev, net);
 583
 584         if (register_netdevice(dev)) {
 585                 free_netdev(dev);
 586                 return NULL;
 587         }
 588
 589         if (!ipmr_init_vif_indev(dev))
 590                 goto failure;
 591         if (dev_open(dev))
 592                 goto failure;
 593
 594         dev_hold(dev);
 595
 596         return dev;
 597
 598 failure:
 599         unregister_netdevice(dev);
 600         return NULL;
 601 }
 602
 603 /* called with rcu_read_lock() */
 604 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 605                      unsigned int pimlen)
 606 {
 607         struct net_device *reg_dev = NULL;
 608         struct iphdr *encap;
 609
 610         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 611         /* Check that:
 612          * a. packet is really sent to a multicast group
 613          * b. packet is not a NULL-REGISTER
 614          * c. packet is not truncated
 615          */
 616         if (!ipv4_is_multicast(encap->daddr) ||
 617             encap->tot_len == 0 ||
 618             ntohs(encap->tot_len) + pimlen > skb->len)
 619                 return 1;
 620
 621         read_lock(&mrt_lock);
 622         if (mrt->mroute_reg_vif_num >= 0)
 623                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 624         read_unlock(&mrt_lock);
 625
 626         if (!reg_dev)
 627                 return 1;
 628
 629         skb->mac_header = skb->network_header;
 630         skb_pull(skb, (u8 *)encap - skb->data);
 631         skb_reset_network_header(skb);
 632         skb->protocol = htons(ETH_P_IP);
 633         skb->ip_summed = CHECKSUM_NONE;
 634
 635         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 636
 637         netif_rx(skb);
 638
 639         return NET_RX_SUCCESS;
 640 }
 641 #else
 642 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 643 {
 644         return NULL;
 645 }
 646 #endif
 647
 648 static int call_ipmr_vif_entry_notifiers(struct net *net,
 649                                          enum fib_event_type event_type,
 650                                          struct vif_device *vif,
 651                                          vifi_t vif_index, u32 tb_id)
 652 {
 653         return mr_call_vif_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 654                                      vif, vif_index, tb_id,
 655                                      &net->ipv4.ipmr_seq);
 656 }
 657
 658 static int call_ipmr_mfc_entry_notifiers(struct net *net,
 659                                          enum fib_event_type event_type,
 660                                          struct mfc_cache *mfc, u32 tb_id)
 661 {
 662         return mr_call_mfc_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 663                                      &mfc->_c, tb_id, &net->ipv4.ipmr_seq);
 664 }
 665
 666 /**
 667  *      vif_delete - Delete a VIF entry
 668  *      @notify: Set to 1, if the caller is a notifier_call
 669  */
 670 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 671                       struct list_head *head)
 672 {
 673         struct net *net = read_pnet(&mrt->net);
 674         struct vif_device *v;
 675         struct net_device *dev;
 676         struct in_device *in_dev;
 677
 678         if (vifi < 0 || vifi >= mrt->maxvif)
 679                 return -EADDRNOTAVAIL;
 680
 681         v = &mrt->vif_table[vifi];
 682
 683         if (VIF_EXISTS(mrt, vifi))
 684                 call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_DEL, v, vifi,
 685                                               mrt->id);
 686
 687         write_lock_bh(&mrt_lock);
 688         dev = v->dev;
 689         v->dev = NULL;
 690
 691         if (!dev) {
 692                 write_unlock_bh(&mrt_lock);
 693                 return -EADDRNOTAVAIL;
 694         }
 695
 696         if (vifi == mrt->mroute_reg_vif_num)
 697                 mrt->mroute_reg_vif_num = -1;
 698
 699         if (vifi + 1 == mrt->maxvif) {
 700                 int tmp;
 701
 702                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 703                         if (VIF_EXISTS(mrt, tmp))
 704                                 break;
 705                 }
 706                 mrt->maxvif = tmp+1;
 707         }
 708
 709         write_unlock_bh(&mrt_lock);
 710
 711         dev_set_allmulti(dev, -1);
 712
 713         in_dev = __in_dev_get_rtnl(dev);
 714         if (in_dev) {
 715                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 716                 inet_netconf_notify_devconf(dev_net(dev), RTM_NEWNETCONF,
 717                                             NETCONFA_MC_FORWARDING,
 718                                             dev->ifindex, &in_dev->cnf);
 719                 ip_rt_multicast_event(in_dev);
 720         }
 721
 722         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 723                 unregister_netdevice_queue(dev, head);
 724
 725         dev_put(dev);
 726         return 0;
 727 }
 728
 729 static void ipmr_cache_free_rcu(struct rcu_head *head)
 730 {
 731         struct mr_mfc *c = container_of(head, struct mr_mfc, rcu);
 732
 733         kmem_cache_free(mrt_cachep, (struct mfc_cache *)c);
 734 }
 735
 736 static void ipmr_cache_free(struct mfc_cache *c)
 737 {
 738         call_rcu(&c->_c.rcu, ipmr_cache_free_rcu);
 739 }
 740
 741 /* Destroy an unresolved cache entry, killing queued skbs
 742  * and reporting error to netlink readers.
 743  */
 744 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 745 {
 746         struct net *net = read_pnet(&mrt->net);
 747         struct sk_buff *skb;
 748         struct nlmsgerr *e;
 749
 750         atomic_dec(&mrt->cache_resolve_queue_len);
 751
 752         while ((skb = skb_dequeue(&c->_c.mfc_un.unres.unresolved))) {
 753                 if (ip_hdr(skb)->version == 0) {
 754                         struct nlmsghdr *nlh = skb_pull(skb,
 755                                                         sizeof(struct iphdr));
 756                         nlh->nlmsg_type = NLMSG_ERROR;
 757                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 758                         skb_trim(skb, nlh->nlmsg_len);
 759                         e = nlmsg_data(nlh);
 760                         e->error = -ETIMEDOUT;
 761                         memset(&e->msg, 0, sizeof(e->msg));
 762
 763                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 764                 } else {
 765                         kfree_skb(skb);
 766                 }
 767         }
 768
 769         ipmr_cache_free(c);
 770 }
 771
 772 /* Timer process for the unresolved queue. */
 773 static void ipmr_expire_process(struct timer_list *t)
 774 {
 775         struct mr_table *mrt = from_timer(mrt, t, ipmr_expire_timer);
 776         struct mr_mfc *c, *next;
 777         unsigned long expires;
 778         unsigned long now;
 779
 780         if (!spin_trylock(&mfc_unres_lock)) {
 781                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 782                 return;
 783         }
 784
 785         if (list_empty(&mrt->mfc_unres_queue))
 786                 goto out;
 787
 788         now = jiffies;
 789         expires = 10*HZ;
 790
 791         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 792                 if (time_after(c->mfc_un.unres.expires, now)) {
 793                         unsigned long interval = c->mfc_un.unres.expires - now;
 794                         if (interval < expires)
 795                                 expires = interval;
 796                         continue;
 797                 }
 798
 799                 list_del(&c->list);
 800                 mroute_netlink_event(mrt, (struct mfc_cache *)c, RTM_DELROUTE);
 801                 ipmr_destroy_unres(mrt, (struct mfc_cache *)c);
 802         }
 803
 804         if (!list_empty(&mrt->mfc_unres_queue))
 805                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 806
 807 out:
 808         spin_unlock(&mfc_unres_lock);
 809 }
 810
 811 /* Fill oifs list. It is called under write locked mrt_lock. */
 812 static void ipmr_update_thresholds(struct mr_table *mrt, struct mr_mfc *cache,
 813                                    unsigned char *ttls)
 814 {
 815         int vifi;
 816
 817         cache->mfc_un.res.minvif = MAXVIFS;
 818         cache->mfc_un.res.maxvif = 0;
 819         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 820
 821         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 822                 if (VIF_EXISTS(mrt, vifi) &&
 823                     ttls[vifi] && ttls[vifi] < 255) {
 824                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 825                         if (cache->mfc_un.res.minvif > vifi)
 826                                 cache->mfc_un.res.minvif = vifi;
 827                         if (cache->mfc_un.res.maxvif <= vifi)
 828                                 cache->mfc_un.res.maxvif = vifi + 1;
 829                 }
 830         }
 831         cache->mfc_un.res.lastuse = jiffies;
 832 }
 833
 834 static int vif_add(struct net *net, struct mr_table *mrt,
 835                    struct vifctl *vifc, int mrtsock)
 836 {
 837         int vifi = vifc->vifc_vifi;
 838         struct switchdev_attr attr = {
 839                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
 840         };
 841         struct vif_device *v = &mrt->vif_table[vifi];
 842         struct net_device *dev;
 843         struct in_device *in_dev;
 844         int err;
 845
 846         /* Is vif busy ? */
 847         if (VIF_EXISTS(mrt, vifi))
 848                 return -EADDRINUSE;
 849
 850         switch (vifc->vifc_flags) {
 851         case VIFF_REGISTER:
 852                 if (!ipmr_pimsm_enabled())
 853                         return -EINVAL;
 854                 /* Special Purpose VIF in PIM
 855                  * All the packets will be sent to the daemon
 856                  */
 857                 if (mrt->mroute_reg_vif_num >= 0)
 858                         return -EADDRINUSE;
 859                 dev = ipmr_reg_vif(net, mrt);
 860                 if (!dev)
 861                         return -ENOBUFS;
 862                 err = dev_set_allmulti(dev, 1);
 863                 if (err) {
 864                         unregister_netdevice(dev);
 865                         dev_put(dev);
 866                         return err;
 867                 }
 868                 break;
 869         case VIFF_TUNNEL:
 870                 dev = ipmr_new_tunnel(net, vifc);
 871                 if (!dev)
 872                         return -ENOBUFS;
 873                 err = dev_set_allmulti(dev, 1);
 874                 if (err) {
 875                         ipmr_del_tunnel(dev, vifc);
 876                         dev_put(dev);
 877                         return err;
 878                 }
 879                 break;
 880         case VIFF_USE_IFINDEX:
 881         case 0:
 882                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 883                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 884                         if (dev && !__in_dev_get_rtnl(dev)) {
 885                                 dev_put(dev);
 886                                 return -EADDRNOTAVAIL;
 887                         }
 888                 } else {
 889                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 890                 }
 891                 if (!dev)
 892                         return -EADDRNOTAVAIL;
 893                 err = dev_set_allmulti(dev, 1);
 894                 if (err) {
 895                         dev_put(dev);
 896                         return err;
 897                 }
 898                 break;
 899         default:
 900                 return -EINVAL;
 901         }
 902
 903         in_dev = __in_dev_get_rtnl(dev);
 904         if (!in_dev) {
 905                 dev_put(dev);
 906                 return -EADDRNOTAVAIL;
 907         }
 908         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 909         inet_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_MC_FORWARDING,
 910                                     dev->ifindex, &in_dev->cnf);
 911         ip_rt_multicast_event(in_dev);
 912
 913         /* Fill in the VIF structures */
 914         vif_device_init(v, dev, vifc->vifc_rate_limit,
 915                         vifc->vifc_threshold,
 916                         vifc->vifc_flags | (!mrtsock ? VIFF_STATIC : 0),
 917                         (VIFF_TUNNEL | VIFF_REGISTER));
 918
 919         attr.orig_dev = dev;
 920         if (!switchdev_port_attr_get(dev, &attr)) {
 921                 memcpy(v->dev_parent_id.id, attr.u.ppid.id, attr.u.ppid.id_len);
 922                 v->dev_parent_id.id_len = attr.u.ppid.id_len;
 923         } else {
 924                 v->dev_parent_id.id_len = 0;
 925         }
 926
 927         v->local = vifc->vifc_lcl_addr.s_addr;
 928         v->remote = vifc->vifc_rmt_addr.s_addr;
 929
 930         /* And finish update writing critical data */
 931         write_lock_bh(&mrt_lock);
 932         v->dev = dev;
 933         if (v->flags & VIFF_REGISTER)
 934                 mrt->mroute_reg_vif_num = vifi;
 935         if (vifi+1 > mrt->maxvif)
 936                 mrt->maxvif = vifi+1;
 937         write_unlock_bh(&mrt_lock);
 938         call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_ADD, v, vifi, mrt->id);
 939         return 0;
 940 }
 941
 942 /* called with rcu_read_lock() */
 943 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 944                                          __be32 origin,
 945                                          __be32 mcastgrp)
 946 {
 947         struct mfc_cache_cmp_arg arg = {
 948                         .mfc_mcastgrp = mcastgrp,
 949                         .mfc_origin = origin
 950         };
 951
 952         return mr_mfc_find(mrt, &arg);
 953 }
 954
 955 /* Look for a (*,G) entry */
 956 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 957                                              __be32 mcastgrp, int vifi)
 958 {
 959         struct mfc_cache_cmp_arg arg = {
 960                         .mfc_mcastgrp = mcastgrp,
 961                         .mfc_origin = htonl(INADDR_ANY)
 962         };
 963
 964         if (mcastgrp == htonl(INADDR_ANY))
 965                 return mr_mfc_find_any_parent(mrt, vifi);
 966         return mr_mfc_find_any(mrt, vifi, &arg);
 967 }
 968
 969 /* Look for a (S,G,iif) entry if parent != -1 */
 970 static struct mfc_cache *ipmr_cache_find_parent(struct mr_table *mrt,
 971                                                 __be32 origin, __be32 mcastgrp,
 972                                                 int parent)
 973 {
 974         struct mfc_cache_cmp_arg arg = {
 975                         .mfc_mcastgrp = mcastgrp,
 976                         .mfc_origin = origin,
 977         };
 978
 979         return mr_mfc_find_parent(mrt, &arg, parent);
 980 }
 981
 982 /* Allocate a multicast cache entry */
 983 static struct mfc_cache *ipmr_cache_alloc(void)
 984 {
 985         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 986
 987         if (c) {
 988                 c->_c.mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 989                 c->_c.mfc_un.res.minvif = MAXVIFS;
 990                 c->_c.free = ipmr_cache_free_rcu;
 991                 refcount_set(&c->_c.mfc_un.res.refcount, 1);
 992         }
 993         return c;
 994 }
 995
 996 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 997 {
 998         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 999
1000         if (c) {
1001                 skb_queue_head_init(&c->_c.mfc_un.unres.unresolved);
1002                 c->_c.mfc_un.unres.expires = jiffies + 10 * HZ;
1003         }
1004         return c;
1005 }
1006
1007 /* A cache entry has gone into a resolved state from queued */
1008 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
1009                                struct mfc_cache *uc, struct mfc_cache *c)
1010 {
1011         struct sk_buff *skb;
1012         struct nlmsgerr *e;
1013
1014         /* Play the pending entries through our router */
1015         while ((skb = __skb_dequeue(&uc->_c.mfc_un.unres.unresolved))) {
1016                 if (ip_hdr(skb)->version == 0) {
1017                         struct nlmsghdr *nlh = skb_pull(skb,
1018                                                         sizeof(struct iphdr));
1019
1020                         if (mr_fill_mroute(mrt, skb, &c->_c,
1021                                            nlmsg_data(nlh)) > 0) {
1022                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
1023                                                  (u8 *)nlh;
1024                         } else {
1025                                 nlh->nlmsg_type = NLMSG_ERROR;
1026                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
1027                                 skb_trim(skb, nlh->nlmsg_len);
1028                                 e = nlmsg_data(nlh);
1029                                 e->error = -EMSGSIZE;
1030                                 memset(&e->msg, 0, sizeof(e->msg));
1031                         }
1032
1033                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
1034                 } else {
1035                         ip_mr_forward(net, mrt, skb->dev, skb, c, 0);
1036                 }
1037         }
1038 }
1039
1040 /* Bounce a cache query up to mrouted and netlink.
1041  *
1042  * Called under mrt_lock.
1043  */
1044 static int ipmr_cache_report(struct mr_table *mrt,
1045                              struct sk_buff *pkt, vifi_t vifi, int assert)
1046 {
1047         const int ihl = ip_hdrlen(pkt);
1048         struct sock *mroute_sk;
1049         struct igmphdr *igmp;
1050         struct igmpmsg *msg;
1051         struct sk_buff *skb;
1052         int ret;
1053
1054         if (assert == IGMPMSG_WHOLEPKT)
1055                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
1056         else
1057                 skb = alloc_skb(128, GFP_ATOMIC);
1058
1059         if (!skb)
1060                 return -ENOBUFS;
1061
1062         if (assert == IGMPMSG_WHOLEPKT) {
1063                 /* Ugly, but we have no choice with this interface.
1064                  * Duplicate old header, fix ihl, length etc.
1065                  * And all this only to mangle msg->im_msgtype and
1066                  * to set msg->im_mbz to "mbz" :-)
1067                  */
1068                 skb_push(skb, sizeof(struct iphdr));
1069                 skb_reset_network_header(skb);
1070                 skb_reset_transport_header(skb);
1071                 msg = (struct igmpmsg *)skb_network_header(skb);
1072                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
1073                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
1074                 msg->im_mbz = 0;
1075                 msg->im_vif = mrt->mroute_reg_vif_num;
1076                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
1077                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
1078                                              sizeof(struct iphdr));
1079         } else {
1080                 /* Copy the IP header */
1081                 skb_set_network_header(skb, skb->len);
1082                 skb_put(skb, ihl);
1083                 skb_copy_to_linear_data(skb, pkt->data, ihl);
1084                 /* Flag to the kernel this is a route add */
1085                 ip_hdr(skb)->protocol = 0;
1086                 msg = (struct igmpmsg *)skb_network_header(skb);
1087                 msg->im_vif = vifi;
1088                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1089                 /* Add our header */
1090                 igmp = skb_put(skb, sizeof(struct igmphdr));
1091                 igmp->type = assert;
1092                 msg->im_msgtype = assert;
1093                 igmp->code = 0;
1094                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1095                 skb->transport_header = skb->network_header;
1096         }
1097
1098         rcu_read_lock();
1099         mroute_sk = rcu_dereference(mrt->mroute_sk);
1100         if (!mroute_sk) {
1101                 rcu_read_unlock();
1102                 kfree_skb(skb);
1103                 return -EINVAL;
1104         }
1105
1106         igmpmsg_netlink_event(mrt, skb);
1107
1108         /* Deliver to mrouted */
1109         ret = sock_queue_rcv_skb(mroute_sk, skb);
1110         rcu_read_unlock();
1111         if (ret < 0) {
1112                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1113                 kfree_skb(skb);
1114         }
1115
1116         return ret;
1117 }
1118
1119 /* Queue a packet for resolution. It gets locked cache entry! */
1120 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1121                                  struct sk_buff *skb, struct net_device *dev)
1122 {
1123         const struct iphdr *iph = ip_hdr(skb);
1124         struct mfc_cache *c;
1125         bool found = false;
1126         int err;
1127
1128         spin_lock_bh(&mfc_unres_lock);
1129         list_for_each_entry(c, &mrt->mfc_unres_queue, _c.list) {
1130                 if (c->mfc_mcastgrp == iph->daddr &&
1131                     c->mfc_origin == iph->saddr) {
1132                         found = true;
1133                         break;
1134                 }
1135         }
1136
1137         if (!found) {
1138                 /* Create a new entry if allowable */
1139                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1140                     (c = ipmr_cache_alloc_unres()) == NULL) {
1141                         spin_unlock_bh(&mfc_unres_lock);
1142
1143                         kfree_skb(skb);
1144                         return -ENOBUFS;
1145                 }
1146
1147                 /* Fill in the new cache entry */
1148                 c->_c.mfc_parent = -1;
1149                 c->mfc_origin   = iph->saddr;
1150                 c->mfc_mcastgrp = iph->daddr;
1151
1152                 /* Reflect first query at mrouted. */
1153                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1154
1155                 if (err < 0) {
1156                         /* If the report failed throw the cache entry
1157                            out - Brad Parker
1158                          */
1159                         spin_unlock_bh(&mfc_unres_lock);
1160
1161                         ipmr_cache_free(c);
1162                         kfree_skb(skb);
1163                         return err;
1164                 }
1165
1166                 atomic_inc(&mrt->cache_resolve_queue_len);
1167                 list_add(&c->_c.list, &mrt->mfc_unres_queue);
1168                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1169
1170                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1171                         mod_timer(&mrt->ipmr_expire_timer,
1172                                   c->_c.mfc_un.unres.expires);
1173         }
1174
1175         /* See if we can append the packet */
1176         if (c->_c.mfc_un.unres.unresolved.qlen > 3) {
1177                 kfree_skb(skb);
1178                 err = -ENOBUFS;
1179         } else {
1180                 if (dev) {
1181                         skb->dev = dev;
1182                         skb->skb_iif = dev->ifindex;
1183                 }
1184                 skb_queue_tail(&c->_c.mfc_un.unres.unresolved, skb);
1185                 err = 0;
1186         }
1187
1188         spin_unlock_bh(&mfc_unres_lock);
1189         return err;
1190 }
1191
1192 /* MFC cache manipulation by user space mroute daemon */
1193
1194 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1195 {
1196         struct net *net = read_pnet(&mrt->net);
1197         struct mfc_cache *c;
1198
1199         /* The entries are added/deleted only under RTNL */
1200         rcu_read_lock();
1201         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1202                                    mfc->mfcc_mcastgrp.s_addr, parent);
1203         rcu_read_unlock();
1204         if (!c)
1205                 return -ENOENT;
1206         rhltable_remove(&mrt->mfc_hash, &c->_c.mnode, ipmr_rht_params);
1207         list_del_rcu(&c->_c.list);
1208         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, c, mrt->id);
1209         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1210         mr_cache_put(&c->_c);
1211
1212         return 0;
1213 }
1214
1215 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1216                         struct mfcctl *mfc, int mrtsock, int parent)
1217 {
1218         struct mfc_cache *uc, *c;
1219         struct mr_mfc *_uc;
1220         bool found;
1221         int ret;
1222
1223         if (mfc->mfcc_parent >= MAXVIFS)
1224                 return -ENFILE;
1225
1226         /* The entries are added/deleted only under RTNL */
1227         rcu_read_lock();
1228         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1229                                    mfc->mfcc_mcastgrp.s_addr, parent);
1230         rcu_read_unlock();
1231         if (c) {
1232                 write_lock_bh(&mrt_lock);
1233                 c->_c.mfc_parent = mfc->mfcc_parent;
1234                 ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1235                 if (!mrtsock)
1236                         c->_c.mfc_flags |= MFC_STATIC;
1237                 write_unlock_bh(&mrt_lock);
1238                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, c,
1239                                               mrt->id);
1240                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1241                 return 0;
1242         }
1243
1244         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1245             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1246                 return -EINVAL;
1247
1248         c = ipmr_cache_alloc();
1249         if (!c)
1250                 return -ENOMEM;
1251
1252         c->mfc_origin = mfc->mfcc_origin.s_addr;
1253         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1254         c->_c.mfc_parent = mfc->mfcc_parent;
1255         ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1256         if (!mrtsock)
1257                 c->_c.mfc_flags |= MFC_STATIC;
1258
1259         ret = rhltable_insert_key(&mrt->mfc_hash, &c->cmparg, &c->_c.mnode,
1260                                   ipmr_rht_params);
1261         if (ret) {
1262                 pr_err("ipmr: rhtable insert error %d\n", ret);
1263                 ipmr_cache_free(c);
1264                 return ret;
1265         }
1266         list_add_tail_rcu(&c->_c.list, &mrt->mfc_cache_list);
1267         /* Check to see if we resolved a queued list. If so we
1268          * need to send on the frames and tidy up.
1269          */
1270         found = false;
1271         spin_lock_bh(&mfc_unres_lock);
1272         list_for_each_entry(_uc, &mrt->mfc_unres_queue, list) {
1273                 uc = (struct mfc_cache *)_uc;
1274                 if (uc->mfc_origin == c->mfc_origin &&
1275                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1276                         list_del(&_uc->list);
1277                         atomic_dec(&mrt->cache_resolve_queue_len);
1278                         found = true;
1279                         break;
1280                 }
1281         }
1282         if (list_empty(&mrt->mfc_unres_queue))
1283                 del_timer(&mrt->ipmr_expire_timer);
1284         spin_unlock_bh(&mfc_unres_lock);
1285
1286         if (found) {
1287                 ipmr_cache_resolve(net, mrt, uc, c);
1288                 ipmr_cache_free(uc);
1289         }
1290         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, c, mrt->id);
1291         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1292         return 0;
1293 }
1294
1295 /* Close the multicast socket, and clear the vif tables etc */
1296 static void mroute_clean_tables(struct mr_table *mrt, bool all)
1297 {
1298         struct net *net = read_pnet(&mrt->net);
1299         struct mr_mfc *c, *tmp;
1300         struct mfc_cache *cache;
1301         LIST_HEAD(list);
1302         int i;
1303
1304         /* Shut down all active vif entries */
1305         for (i = 0; i < mrt->maxvif; i++) {
1306                 if (!all && (mrt->vif_table[i].flags & VIFF_STATIC))
1307                         continue;
1308                 vif_delete(mrt, i, 0, &list);
1309         }
1310         unregister_netdevice_many(&list);
1311
1312         /* Wipe the cache */
1313         list_for_each_entry_safe(c, tmp, &mrt->mfc_cache_list, list) {
1314                 if (!all && (c->mfc_flags & MFC_STATIC))
1315                         continue;
1316                 rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1317                 list_del_rcu(&c->list);
1318                 cache = (struct mfc_cache *)c;
1319                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, cache,
1320                                               mrt->id);
1321                 mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1322                 mr_cache_put(c);
1323         }
1324
1325         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1326                 spin_lock_bh(&mfc_unres_lock);
1327                 list_for_each_entry_safe(c, tmp, &mrt->mfc_unres_queue, list) {
1328                         list_del(&c->list);
1329                         cache = (struct mfc_cache *)c;
1330                         mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1331                         ipmr_destroy_unres(mrt, cache);
1332                 }
1333                 spin_unlock_bh(&mfc_unres_lock);
1334         }
1335 }
1336
1337 /* called from ip_ra_control(), before an RCU grace period,
1338  * we dont need to call synchronize_rcu() here
1339  */
1340 static void mrtsock_destruct(struct sock *sk)
1341 {
1342         struct net *net = sock_net(sk);
1343         struct mr_table *mrt;
1344
1345         rtnl_lock();
1346         ipmr_for_each_table(mrt, net) {
1347                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1348                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1349                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1350                                                     NETCONFA_MC_FORWARDING,
1351                                                     NETCONFA_IFINDEX_ALL,
1352                                                     net->ipv4.devconf_all);
1353                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1354                         mroute_clean_tables(mrt, false);
1355                 }
1356         }
1357         rtnl_unlock();
1358 }
1359
1360 /* Socket options and virtual interface manipulation. The whole
1361  * virtual interface system is a complete heap, but unfortunately
1362  * that's how BSD mrouted happens to think. Maybe one day with a proper
1363  * MOSPF/PIM router set up we can clean this up.
1364  */
1365
1366 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval,
1367                          unsigned int optlen)
1368 {
1369         struct net *net = sock_net(sk);
1370         int val, ret = 0, parent = 0;
1371         struct mr_table *mrt;
1372         struct vifctl vif;
1373         struct mfcctl mfc;
1374         u32 uval;
1375
1376         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1377         rtnl_lock();
1378         if (sk->sk_type != SOCK_RAW ||
1379             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1380                 ret = -EOPNOTSUPP;
1381                 goto out_unlock;
1382         }
1383
1384         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1385         if (!mrt) {
1386                 ret = -ENOENT;
1387                 goto out_unlock;
1388         }
1389         if (optname != MRT_INIT) {
1390                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1391                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1392                         ret = -EACCES;
1393                         goto out_unlock;
1394                 }
1395         }
1396
1397         switch (optname) {
1398         case MRT_INIT:
1399                 if (optlen != sizeof(int)) {
1400                         ret = -EINVAL;
1401                         break;
1402                 }
1403                 if (rtnl_dereference(mrt->mroute_sk)) {
1404                         ret = -EADDRINUSE;
1405                         break;
1406                 }
1407
1408                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1409                 if (ret == 0) {
1410                         rcu_assign_pointer(mrt->mroute_sk, sk);
1411                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1412                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1413                                                     NETCONFA_MC_FORWARDING,
1414                                                     NETCONFA_IFINDEX_ALL,
1415                                                     net->ipv4.devconf_all);
1416                 }
1417                 break;
1418         case MRT_DONE:
1419                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1420                         ret = -EACCES;
1421                 } else {
1422                         /* We need to unlock here because mrtsock_destruct takes
1423                          * care of rtnl itself and we can't change that due to
1424                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1425                          */
1426                         rtnl_unlock();
1427                         ret = ip_ra_control(sk, 0, NULL);
1428                         goto out;
1429                 }
1430                 break;
1431         case MRT_ADD_VIF:
1432         case MRT_DEL_VIF:
1433                 if (optlen != sizeof(vif)) {
1434                         ret = -EINVAL;
1435                         break;
1436                 }
1437                 if (copy_from_user(&vif, optval, sizeof(vif))) {
1438                         ret = -EFAULT;
1439                         break;
1440                 }
1441                 if (vif.vifc_vifi >= MAXVIFS) {
1442                         ret = -ENFILE;
1443                         break;
1444                 }
1445                 if (optname == MRT_ADD_VIF) {
1446                         ret = vif_add(net, mrt, &vif,
1447                                       sk == rtnl_dereference(mrt->mroute_sk));
1448                 } else {
1449                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1450                 }
1451                 break;
1452         /* Manipulate the forwarding caches. These live
1453          * in a sort of kernel/user symbiosis.
1454          */
1455         case MRT_ADD_MFC:
1456         case MRT_DEL_MFC:
1457                 parent = -1;
1458                 /* fall through */
1459         case MRT_ADD_MFC_PROXY:
1460         case MRT_DEL_MFC_PROXY:
1461                 if (optlen != sizeof(mfc)) {
1462                         ret = -EINVAL;
1463                         break;
1464                 }
1465                 if (copy_from_user(&mfc, optval, sizeof(mfc))) {
1466                         ret = -EFAULT;
1467                         break;
1468                 }
1469                 if (parent == 0)
1470                         parent = mfc.mfcc_parent;
1471                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1472                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1473                 else
1474                         ret = ipmr_mfc_add(net, mrt, &mfc,
1475                                            sk == rtnl_dereference(mrt->mroute_sk),
1476                                            parent);
1477                 break;
1478         /* Control PIM assert. */
1479         case MRT_ASSERT:
1480                 if (optlen != sizeof(val)) {
1481                         ret = -EINVAL;
1482                         break;
1483                 }
1484                 if (get_user(val, (int __user *)optval)) {
1485                         ret = -EFAULT;
1486                         break;
1487                 }
1488                 mrt->mroute_do_assert = val;
1489                 break;
1490         case MRT_PIM:
1491                 if (!ipmr_pimsm_enabled()) {
1492                         ret = -ENOPROTOOPT;
1493                         break;
1494                 }
1495                 if (optlen != sizeof(val)) {
1496                         ret = -EINVAL;
1497                         break;
1498                 }
1499                 if (get_user(val, (int __user *)optval)) {
1500                         ret = -EFAULT;
1501                         break;
1502                 }
1503
1504                 val = !!val;
1505                 if (val != mrt->mroute_do_pim) {
1506                         mrt->mroute_do_pim = val;
1507                         mrt->mroute_do_assert = val;
1508                 }
1509                 break;
1510         case MRT_TABLE:
1511                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1512                         ret = -ENOPROTOOPT;
1513                         break;
1514                 }
1515                 if (optlen != sizeof(uval)) {
1516                         ret = -EINVAL;
1517                         break;
1518                 }
1519                 if (get_user(uval, (u32 __user *)optval)) {
1520                         ret = -EFAULT;
1521                         break;
1522                 }
1523
1524                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1525                         ret = -EBUSY;
1526                 } else {
1527                         mrt = ipmr_new_table(net, uval);
1528                         if (IS_ERR(mrt))
1529                                 ret = PTR_ERR(mrt);
1530                         else
1531                                 raw_sk(sk)->ipmr_table = uval;
1532                 }
1533                 break;
1534         /* Spurious command, or MRT_VERSION which you cannot set. */
1535         default:
1536                 ret = -ENOPROTOOPT;
1537         }
1538 out_unlock:
1539         rtnl_unlock();
1540 out:
1541         return ret;
1542 }
1543
1544 /* Getsock opt support for the multicast routing system. */
1545 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1546 {
1547         int olr;
1548         int val;
1549         struct net *net = sock_net(sk);
1550         struct mr_table *mrt;
1551
1552         if (sk->sk_type != SOCK_RAW ||
1553             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1554                 return -EOPNOTSUPP;
1555
1556         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1557         if (!mrt)
1558                 return -ENOENT;
1559
1560         switch (optname) {
1561         case MRT_VERSION:
1562                 val = 0x0305;
1563                 break;
1564         case MRT_PIM:
1565                 if (!ipmr_pimsm_enabled())
1566                         return -ENOPROTOOPT;
1567                 val = mrt->mroute_do_pim;
1568                 break;
1569         case MRT_ASSERT:
1570                 val = mrt->mroute_do_assert;
1571                 break;
1572         default:
1573                 return -ENOPROTOOPT;
1574         }
1575
1576         if (get_user(olr, optlen))
1577                 return -EFAULT;
1578         olr = min_t(unsigned int, olr, sizeof(int));
1579         if (olr < 0)
1580                 return -EINVAL;
1581         if (put_user(olr, optlen))
1582                 return -EFAULT;
1583         if (copy_to_user(optval, &val, olr))
1584                 return -EFAULT;
1585         return 0;
1586 }
1587
1588 /* The IP multicast ioctl support routines. */
1589 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1590 {
1591         struct sioc_sg_req sr;
1592         struct sioc_vif_req vr;
1593         struct vif_device *vif;
1594         struct mfc_cache *c;
1595         struct net *net = sock_net(sk);
1596         struct mr_table *mrt;
1597
1598         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1599         if (!mrt)
1600                 return -ENOENT;
1601
1602         switch (cmd) {
1603         case SIOCGETVIFCNT:
1604                 if (copy_from_user(&vr, arg, sizeof(vr)))
1605                         return -EFAULT;
1606                 if (vr.vifi >= mrt->maxvif)
1607                         return -EINVAL;
1608                 read_lock(&mrt_lock);
1609                 vif = &mrt->vif_table[vr.vifi];
1610                 if (VIF_EXISTS(mrt, vr.vifi)) {
1611                         vr.icount = vif->pkt_in;
1612                         vr.ocount = vif->pkt_out;
1613                         vr.ibytes = vif->bytes_in;
1614                         vr.obytes = vif->bytes_out;
1615                         read_unlock(&mrt_lock);
1616
1617                         if (copy_to_user(arg, &vr, sizeof(vr)))
1618                                 return -EFAULT;
1619                         return 0;
1620                 }
1621                 read_unlock(&mrt_lock);
1622                 return -EADDRNOTAVAIL;
1623         case SIOCGETSGCNT:
1624                 if (copy_from_user(&sr, arg, sizeof(sr)))
1625                         return -EFAULT;
1626
1627                 rcu_read_lock();
1628                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1629                 if (c) {
1630                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1631                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1632                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1633                         rcu_read_unlock();
1634
1635                         if (copy_to_user(arg, &sr, sizeof(sr)))
1636                                 return -EFAULT;
1637                         return 0;
1638                 }
1639                 rcu_read_unlock();
1640                 return -EADDRNOTAVAIL;
1641         default:
1642                 return -ENOIOCTLCMD;
1643         }
1644 }
1645
1646 #ifdef CONFIG_COMPAT
1647 struct compat_sioc_sg_req {
1648         struct in_addr src;
1649         struct in_addr grp;
1650         compat_ulong_t pktcnt;
1651         compat_ulong_t bytecnt;
1652         compat_ulong_t wrong_if;
1653 };
1654
1655 struct compat_sioc_vif_req {
1656         vifi_t  vifi;           /* Which iface */
1657         compat_ulong_t icount;
1658         compat_ulong_t ocount;
1659         compat_ulong_t ibytes;
1660         compat_ulong_t obytes;
1661 };
1662
1663 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1664 {
1665         struct compat_sioc_sg_req sr;
1666         struct compat_sioc_vif_req vr;
1667         struct vif_device *vif;
1668         struct mfc_cache *c;
1669         struct net *net = sock_net(sk);
1670         struct mr_table *mrt;
1671
1672         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1673         if (!mrt)
1674                 return -ENOENT;
1675
1676         switch (cmd) {
1677         case SIOCGETVIFCNT:
1678                 if (copy_from_user(&vr, arg, sizeof(vr)))
1679                         return -EFAULT;
1680                 if (vr.vifi >= mrt->maxvif)
1681                         return -EINVAL;
1682                 read_lock(&mrt_lock);
1683                 vif = &mrt->vif_table[vr.vifi];
1684                 if (VIF_EXISTS(mrt, vr.vifi)) {
1685                         vr.icount = vif->pkt_in;
1686                         vr.ocount = vif->pkt_out;
1687                         vr.ibytes = vif->bytes_in;
1688                         vr.obytes = vif->bytes_out;
1689                         read_unlock(&mrt_lock);
1690
1691                         if (copy_to_user(arg, &vr, sizeof(vr)))
1692                                 return -EFAULT;
1693                         return 0;
1694                 }
1695                 read_unlock(&mrt_lock);
1696                 return -EADDRNOTAVAIL;
1697         case SIOCGETSGCNT:
1698                 if (copy_from_user(&sr, arg, sizeof(sr)))
1699                         return -EFAULT;
1700
1701                 rcu_read_lock();
1702                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1703                 if (c) {
1704                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1705                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1706                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1707                         rcu_read_unlock();
1708
1709                         if (copy_to_user(arg, &sr, sizeof(sr)))
1710                                 return -EFAULT;
1711                         return 0;
1712                 }
1713                 rcu_read_unlock();
1714                 return -EADDRNOTAVAIL;
1715         default:
1716                 return -ENOIOCTLCMD;
1717         }
1718 }
1719 #endif
1720
1721 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1722 {
1723         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1724         struct net *net = dev_net(dev);
1725         struct mr_table *mrt;
1726         struct vif_device *v;
1727         int ct;
1728
1729         if (event != NETDEV_UNREGISTER)
1730                 return NOTIFY_DONE;
1731
1732         ipmr_for_each_table(mrt, net) {
1733                 v = &mrt->vif_table[0];
1734                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1735                         if (v->dev == dev)
1736                                 vif_delete(mrt, ct, 1, NULL);
1737                 }
1738         }
1739         return NOTIFY_DONE;
1740 }
1741
1742 static struct notifier_block ip_mr_notifier = {
1743         .notifier_call = ipmr_device_event,
1744 };
1745
1746 /* Encapsulate a packet by attaching a valid IPIP header to it.
1747  * This avoids tunnel drivers and other mess and gives us the speed so
1748  * important for multicast video.
1749  */
1750 static void ip_encap(struct net *net, struct sk_buff *skb,
1751                      __be32 saddr, __be32 daddr)
1752 {
1753         struct iphdr *iph;
1754         const struct iphdr *old_iph = ip_hdr(skb);
1755
1756         skb_push(skb, sizeof(struct iphdr));
1757         skb->transport_header = skb->network_header;
1758         skb_reset_network_header(skb);
1759         iph = ip_hdr(skb);
1760
1761         iph->version    =       4;
1762         iph->tos        =       old_iph->tos;
1763         iph->ttl        =       old_iph->ttl;
1764         iph->frag_off   =       0;
1765         iph->daddr      =       daddr;
1766         iph->saddr      =       saddr;
1767         iph->protocol   =       IPPROTO_IPIP;
1768         iph->ihl        =       5;
1769         iph->tot_len    =       htons(skb->len);
1770         ip_select_ident(net, skb, NULL);
1771         ip_send_check(iph);
1772
1773         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1774         nf_reset(skb);
1775 }
1776
1777 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1778                                       struct sk_buff *skb)
1779 {
1780         struct ip_options *opt = &(IPCB(skb)->opt);
1781
1782         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1783         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1784
1785         if (unlikely(opt->optlen))
1786                 ip_forward_options(skb);
1787
1788         return dst_output(net, sk, skb);
1789 }
1790
1791 #ifdef CONFIG_NET_SWITCHDEV
1792 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1793                                    int in_vifi, int out_vifi)
1794 {
1795         struct vif_device *out_vif = &mrt->vif_table[out_vifi];
1796         struct vif_device *in_vif = &mrt->vif_table[in_vifi];
1797
1798         if (!skb->offload_mr_fwd_mark)
1799                 return false;
1800         if (!out_vif->dev_parent_id.id_len || !in_vif->dev_parent_id.id_len)
1801                 return false;
1802         return netdev_phys_item_id_same(&out_vif->dev_parent_id,
1803                                         &in_vif->dev_parent_id);
1804 }
1805 #else
1806 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1807                                    int in_vifi, int out_vifi)
1808 {
1809         return false;
1810 }
1811 #endif
1812
1813 /* Processing handlers for ipmr_forward */
1814
1815 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1816                             int in_vifi, struct sk_buff *skb,
1817                             struct mfc_cache *c, int vifi)
1818 {
1819         const struct iphdr *iph = ip_hdr(skb);
1820         struct vif_device *vif = &mrt->vif_table[vifi];
1821         struct net_device *dev;
1822         struct rtable *rt;
1823         struct flowi4 fl4;
1824         int    encap = 0;
1825
1826         if (!vif->dev)
1827                 goto out_free;
1828
1829         if (vif->flags & VIFF_REGISTER) {
1830                 vif->pkt_out++;
1831                 vif->bytes_out += skb->len;
1832                 vif->dev->stats.tx_bytes += skb->len;
1833                 vif->dev->stats.tx_packets++;
1834                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1835                 goto out_free;
1836         }
1837
1838         if (ipmr_forward_offloaded(skb, mrt, in_vifi, vifi))
1839                 goto out_free;
1840
1841         if (vif->flags & VIFF_TUNNEL) {
1842                 rt = ip_route_output_ports(net, &fl4, NULL,
1843                                            vif->remote, vif->local,
1844                                            0, 0,
1845                                            IPPROTO_IPIP,
1846                                            RT_TOS(iph->tos), vif->link);
1847                 if (IS_ERR(rt))
1848                         goto out_free;
1849                 encap = sizeof(struct iphdr);
1850         } else {
1851                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1852                                            0, 0,
1853                                            IPPROTO_IPIP,
1854                                            RT_TOS(iph->tos), vif->link);
1855                 if (IS_ERR(rt))
1856                         goto out_free;
1857         }
1858
1859         dev = rt->dst.dev;
1860
1861         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1862                 /* Do not fragment multicasts. Alas, IPv4 does not
1863                  * allow to send ICMP, so that packets will disappear
1864                  * to blackhole.
1865                  */
1866                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1867                 ip_rt_put(rt);
1868                 goto out_free;
1869         }
1870
1871         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1872
1873         if (skb_cow(skb, encap)) {
1874                 ip_rt_put(rt);
1875                 goto out_free;
1876         }
1877
1878         vif->pkt_out++;
1879         vif->bytes_out += skb->len;
1880
1881         skb_dst_drop(skb);
1882         skb_dst_set(skb, &rt->dst);
1883         ip_decrease_ttl(ip_hdr(skb));
1884
1885         /* FIXME: forward and output firewalls used to be called here.
1886          * What do we do with netfilter? -- RR
1887          */
1888         if (vif->flags & VIFF_TUNNEL) {
1889                 ip_encap(net, skb, vif->local, vif->remote);
1890                 /* FIXME: extra output firewall step used to be here. --RR */
1891                 vif->dev->stats.tx_packets++;
1892                 vif->dev->stats.tx_bytes += skb->len;
1893         }
1894
1895         IPCB(skb)->flags |= IPSKB_FORWARDED;
1896
1897         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1898          * not only before forwarding, but after forwarding on all output
1899          * interfaces. It is clear, if mrouter runs a multicasting
1900          * program, it should receive packets not depending to what interface
1901          * program is joined.
1902          * If we will not make it, the program will have to join on all
1903          * interfaces. On the other hand, multihoming host (or router, but
1904          * not mrouter) cannot join to more than one interface - it will
1905          * result in receiving multiple packets.
1906          */
1907         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1908                 net, NULL, skb, skb->dev, dev,
1909                 ipmr_forward_finish);
1910         return;
1911
1912 out_free:
1913         kfree_skb(skb);
1914 }
1915
1916 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1917 {
1918         int ct;
1919
1920         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1921                 if (mrt->vif_table[ct].dev == dev)
1922                         break;
1923         }
1924         return ct;
1925 }
1926
1927 /* "local" means that we should preserve one skb (for local delivery) */
1928 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1929                           struct net_device *dev, struct sk_buff *skb,
1930                           struct mfc_cache *c, int local)
1931 {
1932         int true_vifi = ipmr_find_vif(mrt, dev);
1933         int psend = -1;
1934         int vif, ct;
1935
1936         vif = c->_c.mfc_parent;
1937         c->_c.mfc_un.res.pkt++;
1938         c->_c.mfc_un.res.bytes += skb->len;
1939         c->_c.mfc_un.res.lastuse = jiffies;
1940
1941         if (c->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1942                 struct mfc_cache *cache_proxy;
1943
1944                 /* For an (*,G) entry, we only check that the incomming
1945                  * interface is part of the static tree.
1946                  */
1947                 cache_proxy = mr_mfc_find_any_parent(mrt, vif);
1948                 if (cache_proxy &&
1949                     cache_proxy->_c.mfc_un.res.ttls[true_vifi] < 255)
1950                         goto forward;
1951         }
1952
1953         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1954         if (mrt->vif_table[vif].dev != dev) {
1955                 if (rt_is_output_route(skb_rtable(skb))) {
1956                         /* It is our own packet, looped back.
1957                          * Very complicated situation...
1958                          *
1959                          * The best workaround until routing daemons will be
1960                          * fixed is not to redistribute packet, if it was
1961                          * send through wrong interface. It means, that
1962                          * multicast applications WILL NOT work for
1963                          * (S,G), which have default multicast route pointing
1964                          * to wrong oif. In any case, it is not a good
1965                          * idea to use multicasting applications on router.
1966                          */
1967                         goto dont_forward;
1968                 }
1969
1970                 c->_c.mfc_un.res.wrong_if++;
1971
1972                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1973                     /* pimsm uses asserts, when switching from RPT to SPT,
1974                      * so that we cannot check that packet arrived on an oif.
1975                      * It is bad, but otherwise we would need to move pretty
1976                      * large chunk of pimd to kernel. Ough... --ANK
1977                      */
1978                     (mrt->mroute_do_pim ||
1979                      c->_c.mfc_un.res.ttls[true_vifi] < 255) &&
1980                     time_after(jiffies,
1981                                c->_c.mfc_un.res.last_assert +
1982                                MFC_ASSERT_THRESH)) {
1983                         c->_c.mfc_un.res.last_assert = jiffies;
1984                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1985                 }
1986                 goto dont_forward;
1987         }
1988
1989 forward:
1990         mrt->vif_table[vif].pkt_in++;
1991         mrt->vif_table[vif].bytes_in += skb->len;
1992
1993         /* Forward the frame */
1994         if (c->mfc_origin == htonl(INADDR_ANY) &&
1995             c->mfc_mcastgrp == htonl(INADDR_ANY)) {
1996                 if (true_vifi >= 0 &&
1997                     true_vifi != c->_c.mfc_parent &&
1998                     ip_hdr(skb)->ttl >
1999                                 c->_c.mfc_un.res.ttls[c->_c.mfc_parent]) {
2000                         /* It's an (*,*) entry and the packet is not coming from
2001                          * the upstream: forward the packet to the upstream
2002                          * only.
2003                          */
2004                         psend = c->_c.mfc_parent;
2005                         goto last_forward;
2006                 }
2007                 goto dont_forward;
2008         }
2009         for (ct = c->_c.mfc_un.res.maxvif - 1;
2010              ct >= c->_c.mfc_un.res.minvif; ct--) {
2011                 /* For (*,G) entry, don't forward to the incoming interface */
2012                 if ((c->mfc_origin != htonl(INADDR_ANY) ||
2013                      ct != true_vifi) &&
2014                     ip_hdr(skb)->ttl > c->_c.mfc_un.res.ttls[ct]) {
2015                         if (psend != -1) {
2016                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2017
2018                                 if (skb2)
2019                                         ipmr_queue_xmit(net, mrt, true_vifi,
2020                                                         skb2, c, psend);
2021                         }
2022                         psend = ct;
2023                 }
2024         }
2025 last_forward:
2026         if (psend != -1) {
2027                 if (local) {
2028                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2029
2030                         if (skb2)
2031                                 ipmr_queue_xmit(net, mrt, true_vifi, skb2,
2032                                                 c, psend);
2033                 } else {
2034                         ipmr_queue_xmit(net, mrt, true_vifi, skb, c, psend);
2035                         return;
2036                 }
2037         }
2038
2039 dont_forward:
2040         if (!local)
2041                 kfree_skb(skb);
2042 }
2043
2044 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
2045 {
2046         struct rtable *rt = skb_rtable(skb);
2047         struct iphdr *iph = ip_hdr(skb);
2048         struct flowi4 fl4 = {
2049                 .daddr = iph->daddr,
2050                 .saddr = iph->saddr,
2051                 .flowi4_tos = RT_TOS(iph->tos),
2052                 .flowi4_oif = (rt_is_output_route(rt) ?
2053                                skb->dev->ifindex : 0),
2054                 .flowi4_iif = (rt_is_output_route(rt) ?
2055                                LOOPBACK_IFINDEX :
2056                                skb->dev->ifindex),
2057                 .flowi4_mark = skb->mark,
2058         };
2059         struct mr_table *mrt;
2060         int err;
2061
2062         err = ipmr_fib_lookup(net, &fl4, &mrt);
2063         if (err)
2064                 return ERR_PTR(err);
2065         return mrt;
2066 }
2067
2068 /* Multicast packets for forwarding arrive here
2069  * Called with rcu_read_lock();
2070  */
2071 int ip_mr_input(struct sk_buff *skb)
2072 {
2073         struct mfc_cache *cache;
2074         struct net *net = dev_net(skb->dev);
2075         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
2076         struct mr_table *mrt;
2077         struct net_device *dev;
2078
2079         /* skb->dev passed in is the loX master dev for vrfs.
2080          * As there are no vifs associated with loopback devices,
2081          * get the proper interface that does have a vif associated with it.
2082          */
2083         dev = skb->dev;
2084         if (netif_is_l3_master(skb->dev)) {
2085                 dev = dev_get_by_index_rcu(net, IPCB(skb)->iif);
2086                 if (!dev) {
2087                         kfree_skb(skb);
2088                         return -ENODEV;
2089                 }
2090         }
2091
2092         /* Packet is looped back after forward, it should not be
2093          * forwarded second time, but still can be delivered locally.
2094          */
2095         if (IPCB(skb)->flags & IPSKB_FORWARDED)
2096                 goto dont_forward;
2097
2098         mrt = ipmr_rt_fib_lookup(net, skb);
2099         if (IS_ERR(mrt)) {
2100                 kfree_skb(skb);
2101                 return PTR_ERR(mrt);
2102         }
2103         if (!local) {
2104                 if (IPCB(skb)->opt.router_alert) {
2105                         if (ip_call_ra_chain(skb))
2106                                 return 0;
2107                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
2108                         /* IGMPv1 (and broken IGMPv2 implementations sort of
2109                          * Cisco IOS <= 11.2(8)) do not put router alert
2110                          * option to IGMP packets destined to routable
2111                          * groups. It is very bad, because it means
2112                          * that we can forward NO IGMP messages.
2113                          */
2114                         struct sock *mroute_sk;
2115
2116                         mroute_sk = rcu_dereference(mrt->mroute_sk);
2117                         if (mroute_sk) {
2118                                 nf_reset(skb);
2119                                 raw_rcv(mroute_sk, skb);
2120                                 return 0;
2121                         }
2122                     }
2123         }
2124
2125         /* already under rcu_read_lock() */
2126         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
2127         if (!cache) {
2128                 int vif = ipmr_find_vif(mrt, dev);
2129
2130                 if (vif >= 0)
2131                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2132                                                     vif);
2133         }
2134
2135         /* No usable cache entry */
2136         if (!cache) {
2137                 int vif;
2138
2139                 if (local) {
2140                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2141                         ip_local_deliver(skb);
2142                         if (!skb2)
2143                                 return -ENOBUFS;
2144                         skb = skb2;
2145                 }
2146
2147                 read_lock(&mrt_lock);
2148                 vif = ipmr_find_vif(mrt, dev);
2149                 if (vif >= 0) {
2150                         int err2 = ipmr_cache_unresolved(mrt, vif, skb, dev);
2151                         read_unlock(&mrt_lock);
2152
2153                         return err2;
2154                 }
2155                 read_unlock(&mrt_lock);
2156                 kfree_skb(skb);
2157                 return -ENODEV;
2158         }
2159
2160         read_lock(&mrt_lock);
2161         ip_mr_forward(net, mrt, dev, skb, cache, local);
2162         read_unlock(&mrt_lock);
2163
2164         if (local)
2165                 return ip_local_deliver(skb);
2166
2167         return 0;
2168
2169 dont_forward:
2170         if (local)
2171                 return ip_local_deliver(skb);
2172         kfree_skb(skb);
2173         return 0;
2174 }
2175
2176 #ifdef CONFIG_IP_PIMSM_V1
2177 /* Handle IGMP messages of PIMv1 */
2178 int pim_rcv_v1(struct sk_buff *skb)
2179 {
2180         struct igmphdr *pim;
2181         struct net *net = dev_net(skb->dev);
2182         struct mr_table *mrt;
2183
2184         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2185                 goto drop;
2186
2187         pim = igmp_hdr(skb);
2188
2189         mrt = ipmr_rt_fib_lookup(net, skb);
2190         if (IS_ERR(mrt))
2191                 goto drop;
2192         if (!mrt->mroute_do_pim ||
2193             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2194                 goto drop;
2195
2196         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2197 drop:
2198                 kfree_skb(skb);
2199         }
2200         return 0;
2201 }
2202 #endif
2203
2204 #ifdef CONFIG_IP_PIMSM_V2
2205 static int pim_rcv(struct sk_buff *skb)
2206 {
2207         struct pimreghdr *pim;
2208         struct net *net = dev_net(skb->dev);
2209         struct mr_table *mrt;
2210
2211         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2212                 goto drop;
2213
2214         pim = (struct pimreghdr *)skb_transport_header(skb);
2215         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2216             (pim->flags & PIM_NULL_REGISTER) ||
2217             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2218              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2219                 goto drop;
2220
2221         mrt = ipmr_rt_fib_lookup(net, skb);
2222         if (IS_ERR(mrt))
2223                 goto drop;
2224         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2225 drop:
2226                 kfree_skb(skb);
2227         }
2228         return 0;
2229 }
2230 #endif
2231
2232 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2233                    __be32 saddr, __be32 daddr,
2234                    struct rtmsg *rtm, u32 portid)
2235 {
2236         struct mfc_cache *cache;
2237         struct mr_table *mrt;
2238         int err;
2239
2240         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2241         if (!mrt)
2242                 return -ENOENT;
2243
2244         rcu_read_lock();
2245         cache = ipmr_cache_find(mrt, saddr, daddr);
2246         if (!cache && skb->dev) {
2247                 int vif = ipmr_find_vif(mrt, skb->dev);
2248
2249                 if (vif >= 0)
2250                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2251         }
2252         if (!cache) {
2253                 struct sk_buff *skb2;
2254                 struct iphdr *iph;
2255                 struct net_device *dev;
2256                 int vif = -1;
2257
2258                 dev = skb->dev;
2259                 read_lock(&mrt_lock);
2260                 if (dev)
2261                         vif = ipmr_find_vif(mrt, dev);
2262                 if (vif < 0) {
2263                         read_unlock(&mrt_lock);
2264                         rcu_read_unlock();
2265                         return -ENODEV;
2266                 }
2267                 skb2 = skb_clone(skb, GFP_ATOMIC);
2268                 if (!skb2) {
2269                         read_unlock(&mrt_lock);
2270                         rcu_read_unlock();
2271                         return -ENOMEM;
2272                 }
2273
2274                 NETLINK_CB(skb2).portid = portid;
2275                 skb_push(skb2, sizeof(struct iphdr));
2276                 skb_reset_network_header(skb2);
2277                 iph = ip_hdr(skb2);
2278                 iph->ihl = sizeof(struct iphdr) >> 2;
2279                 iph->saddr = saddr;
2280                 iph->daddr = daddr;
2281                 iph->version = 0;
2282                 err = ipmr_cache_unresolved(mrt, vif, skb2, dev);
2283                 read_unlock(&mrt_lock);
2284                 rcu_read_unlock();
2285                 return err;
2286         }
2287
2288         read_lock(&mrt_lock);
2289         err = mr_fill_mroute(mrt, skb, &cache->_c, rtm);
2290         read_unlock(&mrt_lock);
2291         rcu_read_unlock();
2292         return err;
2293 }
2294
2295 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2296                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2297                             int flags)
2298 {
2299         struct nlmsghdr *nlh;
2300         struct rtmsg *rtm;
2301         int err;
2302
2303         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2304         if (!nlh)
2305                 return -EMSGSIZE;
2306
2307         rtm = nlmsg_data(nlh);
2308         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2309         rtm->rtm_dst_len  = 32;
2310         rtm->rtm_src_len  = 32;
2311         rtm->rtm_tos      = 0;
2312         rtm->rtm_table    = mrt->id;
2313         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2314                 goto nla_put_failure;
2315         rtm->rtm_type     = RTN_MULTICAST;
2316         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2317         if (c->_c.mfc_flags & MFC_STATIC)
2318                 rtm->rtm_protocol = RTPROT_STATIC;
2319         else
2320                 rtm->rtm_protocol = RTPROT_MROUTED;
2321         rtm->rtm_flags    = 0;
2322
2323         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2324             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2325                 goto nla_put_failure;
2326         err = mr_fill_mroute(mrt, skb, &c->_c, rtm);
2327         /* do not break the dump if cache is unresolved */
2328         if (err < 0 && err != -ENOENT)
2329                 goto nla_put_failure;
2330
2331         nlmsg_end(skb, nlh);
2332         return 0;
2333
2334 nla_put_failure:
2335         nlmsg_cancel(skb, nlh);
2336         return -EMSGSIZE;
2337 }
2338
2339 static int _ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2340                              u32 portid, u32 seq, struct mr_mfc *c, int cmd,
2341                              int flags)
2342 {
2343         return ipmr_fill_mroute(mrt, skb, portid, seq, (struct mfc_cache *)c,
2344                                 cmd, flags);
2345 }
2346
2347 static size_t mroute_msgsize(bool unresolved, int maxvif)
2348 {
2349         size_t len =
2350                 NLMSG_ALIGN(sizeof(struct rtmsg))
2351                 + nla_total_size(4)     /* RTA_TABLE */
2352                 + nla_total_size(4)     /* RTA_SRC */
2353                 + nla_total_size(4)     /* RTA_DST */
2354                 ;
2355
2356         if (!unresolved)
2357                 len = len
2358                       + nla_total_size(4)       /* RTA_IIF */
2359                       + nla_total_size(0)       /* RTA_MULTIPATH */
2360                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2361                                                 /* RTA_MFC_STATS */
2362                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2363                 ;
2364
2365         return len;
2366 }
2367
2368 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2369                                  int cmd)
2370 {
2371         struct net *net = read_pnet(&mrt->net);
2372         struct sk_buff *skb;
2373         int err = -ENOBUFS;
2374
2375         skb = nlmsg_new(mroute_msgsize(mfc->_c.mfc_parent >= MAXVIFS,
2376                                        mrt->maxvif),
2377                         GFP_ATOMIC);
2378         if (!skb)
2379                 goto errout;
2380
2381         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2382         if (err < 0)
2383                 goto errout;
2384
2385         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2386         return;
2387
2388 errout:
2389         kfree_skb(skb);
2390         if (err < 0)
2391                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2392 }
2393
2394 static size_t igmpmsg_netlink_msgsize(size_t payloadlen)
2395 {
2396         size_t len =
2397                 NLMSG_ALIGN(sizeof(struct rtgenmsg))
2398                 + nla_total_size(1)     /* IPMRA_CREPORT_MSGTYPE */
2399                 + nla_total_size(4)     /* IPMRA_CREPORT_VIF_ID */
2400                 + nla_total_size(4)     /* IPMRA_CREPORT_SRC_ADDR */
2401                 + nla_total_size(4)     /* IPMRA_CREPORT_DST_ADDR */
2402                                         /* IPMRA_CREPORT_PKT */
2403                 + nla_total_size(payloadlen)
2404                 ;
2405
2406         return len;
2407 }
2408
2409 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt)
2410 {
2411         struct net *net = read_pnet(&mrt->net);
2412         struct nlmsghdr *nlh;
2413         struct rtgenmsg *rtgenm;
2414         struct igmpmsg *msg;
2415         struct sk_buff *skb;
2416         struct nlattr *nla;
2417         int payloadlen;
2418
2419         payloadlen = pkt->len - sizeof(struct igmpmsg);
2420         msg = (struct igmpmsg *)skb_network_header(pkt);
2421
2422         skb = nlmsg_new(igmpmsg_netlink_msgsize(payloadlen), GFP_ATOMIC);
2423         if (!skb)
2424                 goto errout;
2425
2426         nlh = nlmsg_put(skb, 0, 0, RTM_NEWCACHEREPORT,
2427                         sizeof(struct rtgenmsg), 0);
2428         if (!nlh)
2429                 goto errout;
2430         rtgenm = nlmsg_data(nlh);
2431         rtgenm->rtgen_family = RTNL_FAMILY_IPMR;
2432         if (nla_put_u8(skb, IPMRA_CREPORT_MSGTYPE, msg->im_msgtype) ||
2433             nla_put_u32(skb, IPMRA_CREPORT_VIF_ID, msg->im_vif) ||
2434             nla_put_in_addr(skb, IPMRA_CREPORT_SRC_ADDR,
2435                             msg->im_src.s_addr) ||
2436             nla_put_in_addr(skb, IPMRA_CREPORT_DST_ADDR,
2437                             msg->im_dst.s_addr))
2438                 goto nla_put_failure;
2439
2440         nla = nla_reserve(skb, IPMRA_CREPORT_PKT, payloadlen);
2441         if (!nla || skb_copy_bits(pkt, sizeof(struct igmpmsg),
2442                                   nla_data(nla), payloadlen))
2443                 goto nla_put_failure;
2444
2445         nlmsg_end(skb, nlh);
2446
2447         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE_R, NULL, GFP_ATOMIC);
2448         return;
2449
2450 nla_put_failure:
2451         nlmsg_cancel(skb, nlh);
2452 errout:
2453         kfree_skb(skb);
2454         rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE_R, -ENOBUFS);
2455 }
2456
2457 static int ipmr_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2458                              struct netlink_ext_ack *extack)
2459 {
2460         struct net *net = sock_net(in_skb->sk);
2461         struct nlattr *tb[RTA_MAX + 1];
2462         struct sk_buff *skb = NULL;
2463         struct mfc_cache *cache;
2464         struct mr_table *mrt;
2465         struct rtmsg *rtm;
2466         __be32 src, grp;
2467         u32 tableid;
2468         int err;
2469
2470         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX,
2471                           rtm_ipv4_policy, extack);
2472         if (err < 0)
2473                 goto errout;
2474
2475         rtm = nlmsg_data(nlh);
2476
2477         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2478         grp = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2479         tableid = tb[RTA_TABLE] ? nla_get_u32(tb[RTA_TABLE]) : 0;
2480
2481         mrt = ipmr_get_table(net, tableid ? tableid : RT_TABLE_DEFAULT);
2482         if (!mrt) {
2483                 err = -ENOENT;
2484                 goto errout_free;
2485         }
2486
2487         /* entries are added/deleted only under RTNL */
2488         rcu_read_lock();
2489         cache = ipmr_cache_find(mrt, src, grp);
2490         rcu_read_unlock();
2491         if (!cache) {
2492                 err = -ENOENT;
2493                 goto errout_free;
2494         }
2495
2496         skb = nlmsg_new(mroute_msgsize(false, mrt->maxvif), GFP_KERNEL);
2497         if (!skb) {
2498                 err = -ENOBUFS;
2499                 goto errout_free;
2500         }
2501
2502         err = ipmr_fill_mroute(mrt, skb, NETLINK_CB(in_skb).portid,
2503                                nlh->nlmsg_seq, cache,
2504                                RTM_NEWROUTE, 0);
2505         if (err < 0)
2506                 goto errout_free;
2507
2508         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2509
2510 errout:
2511         return err;
2512
2513 errout_free:
2514         kfree_skb(skb);
2515         goto errout;
2516 }
2517
2518 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2519 {
2520         return mr_rtm_dumproute(skb, cb, ipmr_mr_table_iter,
2521                                 _ipmr_fill_mroute, &mfc_unres_lock);
2522 }
2523
2524 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2525         [RTA_SRC]       = { .type = NLA_U32 },
2526         [RTA_DST]       = { .type = NLA_U32 },
2527         [RTA_IIF]       = { .type = NLA_U32 },
2528         [RTA_TABLE]     = { .type = NLA_U32 },
2529         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2530 };
2531
2532 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2533 {
2534         switch (rtm_protocol) {
2535         case RTPROT_STATIC:
2536         case RTPROT_MROUTED:
2537                 return true;
2538         }
2539         return false;
2540 }
2541
2542 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2543 {
2544         struct rtnexthop *rtnh = nla_data(nla);
2545         int remaining = nla_len(nla), vifi = 0;
2546
2547         while (rtnh_ok(rtnh, remaining)) {
2548                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2549                 if (++vifi == MAXVIFS)
2550                         break;
2551                 rtnh = rtnh_next(rtnh, &remaining);
2552         }
2553
2554         return remaining > 0 ? -EINVAL : vifi;
2555 }
2556
2557 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2558 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2559                             struct mfcctl *mfcc, int *mrtsock,
2560                             struct mr_table **mrtret,
2561                             struct netlink_ext_ack *extack)
2562 {
2563         struct net_device *dev = NULL;
2564         u32 tblid = RT_TABLE_DEFAULT;
2565         struct mr_table *mrt;
2566         struct nlattr *attr;
2567         struct rtmsg *rtm;
2568         int ret, rem;
2569
2570         ret = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipmr_policy,
2571                              extack);
2572         if (ret < 0)
2573                 goto out;
2574         rtm = nlmsg_data(nlh);
2575
2576         ret = -EINVAL;
2577         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2578             rtm->rtm_type != RTN_MULTICAST ||
2579             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2580             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2581                 goto out;
2582
2583         memset(mfcc, 0, sizeof(*mfcc));
2584         mfcc->mfcc_parent = -1;
2585         ret = 0;
2586         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2587                 switch (nla_type(attr)) {
2588                 case RTA_SRC:
2589                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2590                         break;
2591                 case RTA_DST:
2592                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2593                         break;
2594                 case RTA_IIF:
2595                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2596                         if (!dev) {
2597                                 ret = -ENODEV;
2598                                 goto out;
2599                         }
2600                         break;
2601                 case RTA_MULTIPATH:
2602                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2603                                 ret = -EINVAL;
2604                                 goto out;
2605                         }
2606                         break;
2607                 case RTA_PREFSRC:
2608                         ret = 1;
2609                         break;
2610                 case RTA_TABLE:
2611                         tblid = nla_get_u32(attr);
2612                         break;
2613                 }
2614         }
2615         mrt = ipmr_get_table(net, tblid);
2616         if (!mrt) {
2617                 ret = -ENOENT;
2618                 goto out;
2619         }
2620         *mrtret = mrt;
2621         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2622         if (dev)
2623                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2624
2625 out:
2626         return ret;
2627 }
2628
2629 /* takes care of both newroute and delroute */
2630 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,
2631                           struct netlink_ext_ack *extack)
2632 {
2633         struct net *net = sock_net(skb->sk);
2634         int ret, mrtsock, parent;
2635         struct mr_table *tbl;
2636         struct mfcctl mfcc;
2637
2638         mrtsock = 0;
2639         tbl = NULL;
2640         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl, extack);
2641         if (ret < 0)
2642                 return ret;
2643
2644         parent = ret ? mfcc.mfcc_parent : -1;
2645         if (nlh->nlmsg_type == RTM_NEWROUTE)
2646                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2647         else
2648                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2649 }
2650
2651 static bool ipmr_fill_table(struct mr_table *mrt, struct sk_buff *skb)
2652 {
2653         u32 queue_len = atomic_read(&mrt->cache_resolve_queue_len);
2654
2655         if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) ||
2656             nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
2657             nla_put_s32(skb, IPMRA_TABLE_MROUTE_REG_VIF_NUM,
2658                         mrt->mroute_reg_vif_num) ||
2659             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_ASSERT,
2660                        mrt->mroute_do_assert) ||
2661             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_PIM, mrt->mroute_do_pim))
2662                 return false;
2663
2664         return true;
2665 }
2666
2667 static bool ipmr_fill_vif(struct mr_table *mrt, u32 vifid, struct sk_buff *skb)
2668 {
2669         struct nlattr *vif_nest;
2670         struct vif_device *vif;
2671
2672         /* if the VIF doesn't exist just continue */
2673         if (!VIF_EXISTS(mrt, vifid))
2674                 return true;
2675
2676         vif = &mrt->vif_table[vifid];
2677         vif_nest = nla_nest_start(skb, IPMRA_VIF);
2678         if (!vif_nest)
2679                 return false;
2680         if (nla_put_u32(skb, IPMRA_VIFA_IFINDEX, vif->dev->ifindex) ||
2681             nla_put_u32(skb, IPMRA_VIFA_VIF_ID, vifid) ||
2682             nla_put_u16(skb, IPMRA_VIFA_FLAGS, vif->flags) ||
2683             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_IN, vif->bytes_in,
2684                               IPMRA_VIFA_PAD) ||
2685             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_OUT, vif->bytes_out,
2686                               IPMRA_VIFA_PAD) ||
2687             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_IN, vif->pkt_in,
2688                               IPMRA_VIFA_PAD) ||
2689             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_OUT, vif->pkt_out,
2690                               IPMRA_VIFA_PAD) ||
2691             nla_put_be32(skb, IPMRA_VIFA_LOCAL_ADDR, vif->local) ||
2692             nla_put_be32(skb, IPMRA_VIFA_REMOTE_ADDR, vif->remote)) {
2693                 nla_nest_cancel(skb, vif_nest);
2694                 return false;
2695         }
2696         nla_nest_end(skb, vif_nest);
2697
2698         return true;
2699 }
2700
2701 static int ipmr_rtm_dumplink(struct sk_buff *skb, struct netlink_callback *cb)
2702 {
2703         struct net *net = sock_net(skb->sk);
2704         struct nlmsghdr *nlh = NULL;
2705         unsigned int t = 0, s_t;
2706         unsigned int e = 0, s_e;
2707         struct mr_table *mrt;
2708
2709         s_t = cb->args[0];
2710         s_e = cb->args[1];
2711
2712         ipmr_for_each_table(mrt, net) {
2713                 struct nlattr *vifs, *af;
2714                 struct ifinfomsg *hdr;
2715                 u32 i;
2716
2717                 if (t < s_t)
2718                         goto skip_table;
2719                 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2720                                 cb->nlh->nlmsg_seq, RTM_NEWLINK,
2721                                 sizeof(*hdr), NLM_F_MULTI);
2722                 if (!nlh)
2723                         break;
2724
2725                 hdr = nlmsg_data(nlh);
2726                 memset(hdr, 0, sizeof(*hdr));
2727                 hdr->ifi_family = RTNL_FAMILY_IPMR;
2728
2729                 af = nla_nest_start(skb, IFLA_AF_SPEC);
2730                 if (!af) {
2731                         nlmsg_cancel(skb, nlh);
2732                         goto out;
2733                 }
2734
2735                 if (!ipmr_fill_table(mrt, skb)) {
2736                         nlmsg_cancel(skb, nlh);
2737                         goto out;
2738                 }
2739
2740                 vifs = nla_nest_start(skb, IPMRA_TABLE_VIFS);
2741                 if (!vifs) {
2742                         nla_nest_end(skb, af);
2743                         nlmsg_end(skb, nlh);
2744                         goto out;
2745                 }
2746                 for (i = 0; i < mrt->maxvif; i++) {
2747                         if (e < s_e)
2748                                 goto skip_entry;
2749                         if (!ipmr_fill_vif(mrt, i, skb)) {
2750                                 nla_nest_end(skb, vifs);
2751                                 nla_nest_end(skb, af);
2752                                 nlmsg_end(skb, nlh);
2753                                 goto out;
2754                         }
2755 skip_entry:
2756                         e++;
2757                 }
2758                 s_e = 0;
2759                 e = 0;
2760                 nla_nest_end(skb, vifs);
2761                 nla_nest_end(skb, af);
2762                 nlmsg_end(skb, nlh);
2763 skip_table:
2764                 t++;
2765         }
2766
2767 out:
2768         cb->args[1] = e;
2769         cb->args[0] = t;
2770
2771         return skb->len;
2772 }
2773
2774 #ifdef CONFIG_PROC_FS
2775 /* The /proc interfaces to multicast routing :
2776  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2777  */
2778
2779 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2780         __acquires(mrt_lock)
2781 {
2782         struct mr_vif_iter *iter = seq->private;
2783         struct net *net = seq_file_net(seq);
2784         struct mr_table *mrt;
2785
2786         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2787         if (!mrt)
2788                 return ERR_PTR(-ENOENT);
2789
2790         iter->mrt = mrt;
2791
2792         read_lock(&mrt_lock);
2793         return mr_vif_seq_start(seq, pos);
2794 }
2795
2796 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2797         __releases(mrt_lock)
2798 {
2799         read_unlock(&mrt_lock);
2800 }
2801
2802 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2803 {
2804         struct mr_vif_iter *iter = seq->private;
2805         struct mr_table *mrt = iter->mrt;
2806
2807         if (v == SEQ_START_TOKEN) {
2808                 seq_puts(seq,
2809                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2810         } else {
2811                 const struct vif_device *vif = v;
2812                 const char *name =  vif->dev ?
2813                                     vif->dev->name : "none";
2814
2815                 seq_printf(seq,
2816                            "%2td %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2817                            vif - mrt->vif_table,
2818                            name, vif->bytes_in, vif->pkt_in,
2819                            vif->bytes_out, vif->pkt_out,
2820                            vif->flags, vif->local, vif->remote);
2821         }
2822         return 0;
2823 }
2824
2825 static const struct seq_operations ipmr_vif_seq_ops = {
2826         .start = ipmr_vif_seq_start,
2827         .next  = mr_vif_seq_next,
2828         .stop  = ipmr_vif_seq_stop,
2829         .show  = ipmr_vif_seq_show,
2830 };
2831
2832 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2833 {
2834         struct net *net = seq_file_net(seq);
2835         struct mr_table *mrt;
2836
2837         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2838         if (!mrt)
2839                 return ERR_PTR(-ENOENT);
2840
2841         return mr_mfc_seq_start(seq, pos, mrt, &mfc_unres_lock);
2842 }
2843
2844 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2845 {
2846         int n;
2847
2848         if (v == SEQ_START_TOKEN) {
2849                 seq_puts(seq,
2850                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2851         } else {
2852                 const struct mfc_cache *mfc = v;
2853                 const struct mr_mfc_iter *it = seq->private;
2854                 const struct mr_table *mrt = it->mrt;
2855
2856                 seq_printf(seq, "%08X %08X %-3hd",
2857                            (__force u32) mfc->mfc_mcastgrp,
2858                            (__force u32) mfc->mfc_origin,
2859                            mfc->_c.mfc_parent);
2860
2861                 if (it->cache != &mrt->mfc_unres_queue) {
2862                         seq_printf(seq, " %8lu %8lu %8lu",
2863                                    mfc->_c.mfc_un.res.pkt,
2864                                    mfc->_c.mfc_un.res.bytes,
2865                                    mfc->_c.mfc_un.res.wrong_if);
2866                         for (n = mfc->_c.mfc_un.res.minvif;
2867                              n < mfc->_c.mfc_un.res.maxvif; n++) {
2868                                 if (VIF_EXISTS(mrt, n) &&
2869                                     mfc->_c.mfc_un.res.ttls[n] < 255)
2870                                         seq_printf(seq,
2871                                            " %2d:%-3d",
2872                                            n, mfc->_c.mfc_un.res.ttls[n]);
2873                         }
2874                 } else {
2875                         /* unresolved mfc_caches don't contain
2876                          * pkt, bytes and wrong_if values
2877                          */
2878                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2879                 }
2880                 seq_putc(seq, '\n');
2881         }
2882         return 0;
2883 }
2884
2885 static const struct seq_operations ipmr_mfc_seq_ops = {
2886         .start = ipmr_mfc_seq_start,
2887         .next  = mr_mfc_seq_next,
2888         .stop  = mr_mfc_seq_stop,
2889         .show  = ipmr_mfc_seq_show,
2890 };
2891 #endif
2892
2893 #ifdef CONFIG_IP_PIMSM_V2
2894 static const struct net_protocol pim_protocol = {
2895         .handler        =       pim_rcv,
2896         .netns_ok       =       1,
2897 };
2898 #endif
2899
2900 static unsigned int ipmr_seq_read(struct net *net)
2901 {
2902         ASSERT_RTNL();
2903
2904         return net->ipv4.ipmr_seq + ipmr_rules_seq_read(net);
2905 }
2906
2907 static int ipmr_dump(struct net *net, struct notifier_block *nb)
2908 {
2909         return mr_dump(net, nb, RTNL_FAMILY_IPMR, ipmr_rules_dump,
2910                        ipmr_mr_table_iter, &mrt_lock);
2911 }
2912
2913 static const struct fib_notifier_ops ipmr_notifier_ops_template = {
2914         .family         = RTNL_FAMILY_IPMR,
2915         .fib_seq_read   = ipmr_seq_read,
2916         .fib_dump       = ipmr_dump,
2917         .owner          = THIS_MODULE,
2918 };
2919
2920 static int __net_init ipmr_notifier_init(struct net *net)
2921 {
2922         struct fib_notifier_ops *ops;
2923
2924         net->ipv4.ipmr_seq = 0;
2925
2926         ops = fib_notifier_ops_register(&ipmr_notifier_ops_template, net);
2927         if (IS_ERR(ops))
2928                 return PTR_ERR(ops);
2929         net->ipv4.ipmr_notifier_ops = ops;
2930
2931         return 0;
2932 }
2933
2934 static void __net_exit ipmr_notifier_exit(struct net *net)
2935 {
2936         fib_notifier_ops_unregister(net->ipv4.ipmr_notifier_ops);
2937         net->ipv4.ipmr_notifier_ops = NULL;
2938 }
2939
2940 /* Setup for IP multicast routing */
2941 static int __net_init ipmr_net_init(struct net *net)
2942 {
2943         int err;
2944
2945         err = ipmr_notifier_init(net);
2946         if (err)
2947                 goto ipmr_notifier_fail;
2948
2949         err = ipmr_rules_init(net);
2950         if (err < 0)
2951                 goto ipmr_rules_fail;
2952
2953 #ifdef CONFIG_PROC_FS
2954         err = -ENOMEM;
2955         if (!proc_create_net("ip_mr_vif", 0, net->proc_net, &ipmr_vif_seq_ops,
2956                         sizeof(struct mr_vif_iter)))
2957                 goto proc_vif_fail;
2958         if (!proc_create_net("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_seq_ops,
2959                         sizeof(struct mr_mfc_iter)))
2960                 goto proc_cache_fail;
2961 #endif
2962         return 0;
2963
2964 #ifdef CONFIG_PROC_FS
2965 proc_cache_fail:
2966         remove_proc_entry("ip_mr_vif", net->proc_net);
2967 proc_vif_fail:
2968         ipmr_rules_exit(net);
2969 #endif
2970 ipmr_rules_fail:
2971         ipmr_notifier_exit(net);
2972 ipmr_notifier_fail:
2973         return err;
2974 }
2975
2976 static void __net_exit ipmr_net_exit(struct net *net)
2977 {
2978 #ifdef CONFIG_PROC_FS
2979         remove_proc_entry("ip_mr_cache", net->proc_net);
2980         remove_proc_entry("ip_mr_vif", net->proc_net);
2981 #endif
2982         ipmr_notifier_exit(net);
2983         ipmr_rules_exit(net);
2984 }
2985
2986 static struct pernet_operations ipmr_net_ops = {
2987         .init = ipmr_net_init,
2988         .exit = ipmr_net_exit,
2989 };
2990
2991 int __init ip_mr_init(void)
2992 {
2993         int err;
2994
2995         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2996                                        sizeof(struct mfc_cache),
2997                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2998                                        NULL);
2999
3000         err = register_pernet_subsys(&ipmr_net_ops);
3001         if (err)
3002                 goto reg_pernet_fail;
3003
3004         err = register_netdevice_notifier(&ip_mr_notifier);
3005         if (err)
3006                 goto reg_notif_fail;
3007 #ifdef CONFIG_IP_PIMSM_V2
3008         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
3009                 pr_err("%s: can't add PIM protocol\n", __func__);
3010                 err = -EAGAIN;
3011                 goto add_proto_fail;
3012         }
3013 #endif
3014         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
3015                       ipmr_rtm_getroute, ipmr_rtm_dumproute, 0);
3016         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
3017                       ipmr_rtm_route, NULL, 0);
3018         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
3019                       ipmr_rtm_route, NULL, 0);
3020
3021         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETLINK,
3022                       NULL, ipmr_rtm_dumplink, 0);
3023         return 0;
3024
3025 #ifdef CONFIG_IP_PIMSM_V2
3026 add_proto_fail:
3027         unregister_netdevice_notifier(&ip_mr_notifier);
3028 #endif
3029 reg_notif_fail:
3030         unregister_pernet_subsys(&ipmr_net_ops);
3031 reg_pernet_fail:
3032         kmem_cache_destroy(mrt_cachep);
3033         return err;
3034 }