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 requrement to work with older peers.
  26  *
  27  */
  28
  29 #include <asm/system.h>
  30 #include <asm/uaccess.h>
  31 #include <linux/types.h>
  32 #include <linux/capability.h>
  33 #include <linux/errno.h>
  34 #include <linux/timer.h>
  35 #include <linux/mm.h>
  36 #include <linux/kernel.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/stat.h>
  39 #include <linux/socket.h>
  40 #include <linux/in.h>
  41 #include <linux/inet.h>
  42 #include <linux/netdevice.h>
  43 #include <linux/inetdevice.h>
  44 #include <linux/igmp.h>
  45 #include <linux/proc_fs.h>
  46 #include <linux/seq_file.h>
  47 #include <linux/mroute.h>
  48 #include <linux/init.h>
  49 #include <linux/if_ether.h>
  50 #include <net/net_namespace.h>
  51 #include <net/ip.h>
  52 #include <net/protocol.h>
  53 #include <linux/skbuff.h>
  54 #include <net/route.h>
  55 #include <net/sock.h>
  56 #include <net/icmp.h>
  57 #include <net/udp.h>
  58 #include <net/raw.h>
  59 #include <linux/notifier.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/netfilter_ipv4.h>
  62 #include <net/ipip.h>
  63 #include <net/checksum.h>
  64 #include <net/netlink.h>
  65
  66 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
  67 #define CONFIG_IP_PIMSM 1
  68 #endif
  69
  70 static struct sock *mroute_socket;
  71
  72
  73 /* Big lock, protecting vif table, mrt cache and mroute socket state.
  74    Note that the changes are semaphored via rtnl_lock.
  75  */
  76
  77 static DEFINE_RWLOCK(mrt_lock);
  78
  79 /*
  80  *      Multicast router control variables
  81  */
  82
  83 static struct vif_device vif_table[MAXVIFS];            /* Devices              */
  84 static int maxvif;
  85
  86 #define VIF_EXISTS(idx) (vif_table[idx].dev != NULL)
  87
  88 static int mroute_do_assert;                            /* Set in PIM assert    */
  89 static int mroute_do_pim;
  90
  91 static struct mfc_cache *mfc_cache_array[MFC_LINES];    /* Forwarding cache     */
  92
  93 static struct mfc_cache *mfc_unres_queue;               /* Queue of unresolved entries */
  94 static atomic_t cache_resolve_queue_len;                /* Size of unresolved   */
  95
  96 /* Special spinlock for queue of unresolved entries */
  97 static DEFINE_SPINLOCK(mfc_unres_lock);
  98
  99 /* We return to original Alan's scheme. Hash table of resolved
 100    entries is changed only in process context and protected
 101    with weak lock mrt_lock. Queue of unresolved entries is protected
 102    with strong spinlock mfc_unres_lock.
 103
 104    In this case data path is free of exclusive locks at all.
 105  */
 106
 107 static struct kmem_cache *mrt_cachep __read_mostly;
 108
 109 static int ip_mr_forward(struct sk_buff *skb, struct mfc_cache *cache, int local);
 110 static int ipmr_cache_report(struct sk_buff *pkt, vifi_t vifi, int assert);
 111 static int ipmr_fill_mroute(struct sk_buff *skb, struct mfc_cache *c, struct rtmsg *rtm);
 112
 113 #ifdef CONFIG_IP_PIMSM_V2
 114 static struct net_protocol pim_protocol;
 115 #endif
 116
 117 static struct timer_list ipmr_expire_timer;
 118
 119 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 120
 121 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 122 {
 123         dev_close(dev);
 124
 125         dev = __dev_get_by_name(&init_net, "tunl0");
 126         if (dev) {
 127                 const struct net_device_ops *ops = dev->netdev_ops;
 128                 struct ifreq ifr;
 129                 struct ip_tunnel_parm p;
 130
 131                 memset(&p, 0, sizeof(p));
 132                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 133                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 134                 p.iph.version = 4;
 135                 p.iph.ihl = 5;
 136                 p.iph.protocol = IPPROTO_IPIP;
 137                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 138                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 139
 140                 if (ops->ndo_do_ioctl) {
 141                         mm_segment_t oldfs = get_fs();
 142
 143                         set_fs(KERNEL_DS);
 144                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 145                         set_fs(oldfs);
 146                 }
 147         }
 148 }
 149
 150 static
 151 struct net_device *ipmr_new_tunnel(struct vifctl *v)
 152 {
 153         struct net_device  *dev;
 154
 155         dev = __dev_get_by_name(&init_net, "tunl0");
 156
 157         if (dev) {
 158                 const struct net_device_ops *ops = dev->netdev_ops;
 159                 int err;
 160                 struct ifreq ifr;
 161                 struct ip_tunnel_parm p;
 162                 struct in_device  *in_dev;
 163
 164                 memset(&p, 0, sizeof(p));
 165                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 166                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 167                 p.iph.version = 4;
 168                 p.iph.ihl = 5;
 169                 p.iph.protocol = IPPROTO_IPIP;
 170                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 171                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 172
 173                 if (ops->ndo_do_ioctl) {
 174                         mm_segment_t oldfs = get_fs();
 175
 176                         set_fs(KERNEL_DS);
 177                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 178                         set_fs(oldfs);
 179                 } else
 180                         err = -EOPNOTSUPP;
 181
 182                 dev = NULL;
 183
 184                 if (err == 0 && (dev = __dev_get_by_name(&init_net, p.name)) != NULL) {
 185                         dev->flags |= IFF_MULTICAST;
 186
 187                         in_dev = __in_dev_get_rtnl(dev);
 188                         if (in_dev == NULL)
 189                                 goto failure;
 190
 191                         ipv4_devconf_setall(in_dev);
 192                         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 193
 194                         if (dev_open(dev))
 195                                 goto failure;
 196                         dev_hold(dev);
 197                 }
 198         }
 199         return dev;
 200
 201 failure:
 202         /* allow the register to be completed before unregistering. */
 203         rtnl_unlock();
 204         rtnl_lock();
 205
 206         unregister_netdevice(dev);
 207         return NULL;
 208 }
 209
 210 #ifdef CONFIG_IP_PIMSM
 211
 212 static int reg_vif_num = -1;
 213
 214 static int reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 215 {
 216         read_lock(&mrt_lock);
 217         dev->stats.tx_bytes += skb->len;
 218         dev->stats.tx_packets++;
 219         ipmr_cache_report(skb, reg_vif_num, IGMPMSG_WHOLEPKT);
 220         read_unlock(&mrt_lock);
 221         kfree_skb(skb);
 222         return 0;
 223 }
 224
 225 static void reg_vif_setup(struct net_device *dev)
 226 {
 227         dev->type               = ARPHRD_PIMREG;
 228         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 229         dev->flags              = IFF_NOARP;
 230         dev->hard_start_xmit    = reg_vif_xmit;
 231         dev->destructor         = free_netdev;
 232 }
 233
 234 static struct net_device *ipmr_reg_vif(void)
 235 {
 236         struct net_device *dev;
 237         struct in_device *in_dev;
 238
 239         dev = alloc_netdev(0, "pimreg", reg_vif_setup);
 240
 241         if (dev == NULL)
 242                 return NULL;
 243
 244         if (register_netdevice(dev)) {
 245                 free_netdev(dev);
 246                 return NULL;
 247         }
 248         dev->iflink = 0;
 249
 250         rcu_read_lock();
 251         if ((in_dev = __in_dev_get_rcu(dev)) == NULL) {
 252                 rcu_read_unlock();
 253                 goto failure;
 254         }
 255
 256         ipv4_devconf_setall(in_dev);
 257         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 258         rcu_read_unlock();
 259
 260         if (dev_open(dev))
 261                 goto failure;
 262
 263         dev_hold(dev);
 264
 265         return dev;
 266
 267 failure:
 268         /* allow the register to be completed before unregistering. */
 269         rtnl_unlock();
 270         rtnl_lock();
 271
 272         unregister_netdevice(dev);
 273         return NULL;
 274 }
 275 #endif
 276
 277 /*
 278  *      Delete a VIF entry
 279  *      @notify: Set to 1, if the caller is a notifier_call
 280  */
 281
 282 static int vif_delete(int vifi, int notify)
 283 {
 284         struct vif_device *v;
 285         struct net_device *dev;
 286         struct in_device *in_dev;
 287
 288         if (vifi < 0 || vifi >= maxvif)
 289                 return -EADDRNOTAVAIL;
 290
 291         v = &vif_table[vifi];
 292
 293         write_lock_bh(&mrt_lock);
 294         dev = v->dev;
 295         v->dev = NULL;
 296
 297         if (!dev) {
 298                 write_unlock_bh(&mrt_lock);
 299                 return -EADDRNOTAVAIL;
 300         }
 301
 302 #ifdef CONFIG_IP_PIMSM
 303         if (vifi == reg_vif_num)
 304                 reg_vif_num = -1;
 305 #endif
 306
 307         if (vifi+1 == maxvif) {
 308                 int tmp;
 309                 for (tmp=vifi-1; tmp>=0; tmp--) {
 310                         if (VIF_EXISTS(tmp))
 311                                 break;
 312                 }
 313                 maxvif = tmp+1;
 314         }
 315
 316         write_unlock_bh(&mrt_lock);
 317
 318         dev_set_allmulti(dev, -1);
 319
 320         if ((in_dev = __in_dev_get_rtnl(dev)) != NULL) {
 321                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 322                 ip_rt_multicast_event(in_dev);
 323         }
 324
 325         if (v->flags&(VIFF_TUNNEL|VIFF_REGISTER) && !notify)
 326                 unregister_netdevice(dev);
 327
 328         dev_put(dev);
 329         return 0;
 330 }
 331
 332 /* Destroy an unresolved cache entry, killing queued skbs
 333    and reporting error to netlink readers.
 334  */
 335
 336 static void ipmr_destroy_unres(struct mfc_cache *c)
 337 {
 338         struct sk_buff *skb;
 339         struct nlmsgerr *e;
 340
 341         atomic_dec(&cache_resolve_queue_len);
 342
 343         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
 344                 if (ip_hdr(skb)->version == 0) {
 345                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 346                         nlh->nlmsg_type = NLMSG_ERROR;
 347                         nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
 348                         skb_trim(skb, nlh->nlmsg_len);
 349                         e = NLMSG_DATA(nlh);
 350                         e->error = -ETIMEDOUT;
 351                         memset(&e->msg, 0, sizeof(e->msg));
 352
 353                         rtnl_unicast(skb, &init_net, NETLINK_CB(skb).pid);
 354                 } else
 355                         kfree_skb(skb);
 356         }
 357
 358         kmem_cache_free(mrt_cachep, c);
 359 }
 360
 361
 362 /* Single timer process for all the unresolved queue. */
 363
 364 static void ipmr_expire_process(unsigned long dummy)
 365 {
 366         unsigned long now;
 367         unsigned long expires;
 368         struct mfc_cache *c, **cp;
 369
 370         if (!spin_trylock(&mfc_unres_lock)) {
 371                 mod_timer(&ipmr_expire_timer, jiffies+HZ/10);
 372                 return;
 373         }
 374
 375         if (atomic_read(&cache_resolve_queue_len) == 0)
 376                 goto out;
 377
 378         now = jiffies;
 379         expires = 10*HZ;
 380         cp = &mfc_unres_queue;
 381
 382         while ((c=*cp) != NULL) {
 383                 if (time_after(c->mfc_un.unres.expires, now)) {
 384                         unsigned long interval = c->mfc_un.unres.expires - now;
 385                         if (interval < expires)
 386                                 expires = interval;
 387                         cp = &c->next;
 388                         continue;
 389                 }
 390
 391                 *cp = c->next;
 392
 393                 ipmr_destroy_unres(c);
 394         }
 395
 396         if (atomic_read(&cache_resolve_queue_len))
 397                 mod_timer(&ipmr_expire_timer, jiffies + expires);
 398
 399 out:
 400         spin_unlock(&mfc_unres_lock);
 401 }
 402
 403 /* Fill oifs list. It is called under write locked mrt_lock. */
 404
 405 static void ipmr_update_thresholds(struct mfc_cache *cache, unsigned char *ttls)
 406 {
 407         int vifi;
 408
 409         cache->mfc_un.res.minvif = MAXVIFS;
 410         cache->mfc_un.res.maxvif = 0;
 411         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 412
 413         for (vifi=0; vifi<maxvif; vifi++) {
 414                 if (VIF_EXISTS(vifi) && ttls[vifi] && ttls[vifi] < 255) {
 415                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 416                         if (cache->mfc_un.res.minvif > vifi)
 417                                 cache->mfc_un.res.minvif = vifi;
 418                         if (cache->mfc_un.res.maxvif <= vifi)
 419                                 cache->mfc_un.res.maxvif = vifi + 1;
 420                 }
 421         }
 422 }
 423
 424 static int vif_add(struct vifctl *vifc, int mrtsock)
 425 {
 426         int vifi = vifc->vifc_vifi;
 427         struct vif_device *v = &vif_table[vifi];
 428         struct net_device *dev;
 429         struct in_device *in_dev;
 430         int err;
 431
 432         /* Is vif busy ? */
 433         if (VIF_EXISTS(vifi))
 434                 return -EADDRINUSE;
 435
 436         switch (vifc->vifc_flags) {
 437 #ifdef CONFIG_IP_PIMSM
 438         case VIFF_REGISTER:
 439                 /*
 440                  * Special Purpose VIF in PIM
 441                  * All the packets will be sent to the daemon
 442                  */
 443                 if (reg_vif_num >= 0)
 444                         return -EADDRINUSE;
 445                 dev = ipmr_reg_vif();
 446                 if (!dev)
 447                         return -ENOBUFS;
 448                 err = dev_set_allmulti(dev, 1);
 449                 if (err) {
 450                         unregister_netdevice(dev);
 451                         dev_put(dev);
 452                         return err;
 453                 }
 454                 break;
 455 #endif
 456         case VIFF_TUNNEL:
 457                 dev = ipmr_new_tunnel(vifc);
 458                 if (!dev)
 459                         return -ENOBUFS;
 460                 err = dev_set_allmulti(dev, 1);
 461                 if (err) {
 462                         ipmr_del_tunnel(dev, vifc);
 463                         dev_put(dev);
 464                         return err;
 465                 }
 466                 break;
 467         case 0:
 468                 dev = ip_dev_find(&init_net, vifc->vifc_lcl_addr.s_addr);
 469                 if (!dev)
 470                         return -EADDRNOTAVAIL;
 471                 err = dev_set_allmulti(dev, 1);
 472                 if (err) {
 473                         dev_put(dev);
 474                         return err;
 475                 }
 476                 break;
 477         default:
 478                 return -EINVAL;
 479         }
 480
 481         if ((in_dev = __in_dev_get_rtnl(dev)) == NULL)
 482                 return -EADDRNOTAVAIL;
 483         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 484         ip_rt_multicast_event(in_dev);
 485
 486         /*
 487          *      Fill in the VIF structures
 488          */
 489         v->rate_limit = vifc->vifc_rate_limit;
 490         v->local = vifc->vifc_lcl_addr.s_addr;
 491         v->remote = vifc->vifc_rmt_addr.s_addr;
 492         v->flags = vifc->vifc_flags;
 493         if (!mrtsock)
 494                 v->flags |= VIFF_STATIC;
 495         v->threshold = vifc->vifc_threshold;
 496         v->bytes_in = 0;
 497         v->bytes_out = 0;
 498         v->pkt_in = 0;
 499         v->pkt_out = 0;
 500         v->link = dev->ifindex;
 501         if (v->flags&(VIFF_TUNNEL|VIFF_REGISTER))
 502                 v->link = dev->iflink;
 503
 504         /* And finish update writing critical data */
 505         write_lock_bh(&mrt_lock);
 506         v->dev = dev;
 507 #ifdef CONFIG_IP_PIMSM
 508         if (v->flags&VIFF_REGISTER)
 509                 reg_vif_num = vifi;
 510 #endif
 511         if (vifi+1 > maxvif)
 512                 maxvif = vifi+1;
 513         write_unlock_bh(&mrt_lock);
 514         return 0;
 515 }
 516
 517 static struct mfc_cache *ipmr_cache_find(__be32 origin, __be32 mcastgrp)
 518 {
 519         int line = MFC_HASH(mcastgrp, origin);
 520         struct mfc_cache *c;
 521
 522         for (c=mfc_cache_array[line]; c; c = c->next) {
 523                 if (c->mfc_origin==origin && c->mfc_mcastgrp==mcastgrp)
 524                         break;
 525         }
 526         return c;
 527 }
 528
 529 /*
 530  *      Allocate a multicast cache entry
 531  */
 532 static struct mfc_cache *ipmr_cache_alloc(void)
 533 {
 534         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 535         if (c == NULL)
 536                 return NULL;
 537         c->mfc_un.res.minvif = MAXVIFS;
 538         return c;
 539 }
 540
 541 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 542 {
 543         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 544         if (c == NULL)
 545                 return NULL;
 546         skb_queue_head_init(&c->mfc_un.unres.unresolved);
 547         c->mfc_un.unres.expires = jiffies + 10*HZ;
 548         return c;
 549 }
 550
 551 /*
 552  *      A cache entry has gone into a resolved state from queued
 553  */
 554
 555 static void ipmr_cache_resolve(struct mfc_cache *uc, struct mfc_cache *c)
 556 {
 557         struct sk_buff *skb;
 558         struct nlmsgerr *e;
 559
 560         /*
 561          *      Play the pending entries through our router
 562          */
 563
 564         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
 565                 if (ip_hdr(skb)->version == 0) {
 566                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 567
 568                         if (ipmr_fill_mroute(skb, c, NLMSG_DATA(nlh)) > 0) {
 569                                 nlh->nlmsg_len = (skb_tail_pointer(skb) -
 570                                                   (u8 *)nlh);
 571                         } else {
 572                                 nlh->nlmsg_type = NLMSG_ERROR;
 573                                 nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
 574                                 skb_trim(skb, nlh->nlmsg_len);
 575                                 e = NLMSG_DATA(nlh);
 576                                 e->error = -EMSGSIZE;
 577                                 memset(&e->msg, 0, sizeof(e->msg));
 578                         }
 579
 580                         rtnl_unicast(skb, &init_net, NETLINK_CB(skb).pid);
 581                 } else
 582                         ip_mr_forward(skb, c, 0);
 583         }
 584 }
 585
 586 /*
 587  *      Bounce a cache query up to mrouted. We could use netlink for this but mrouted
 588  *      expects the following bizarre scheme.
 589  *
 590  *      Called under mrt_lock.
 591  */
 592
 593 static int ipmr_cache_report(struct sk_buff *pkt, vifi_t vifi, int assert)
 594 {
 595         struct sk_buff *skb;
 596         const int ihl = ip_hdrlen(pkt);
 597         struct igmphdr *igmp;
 598         struct igmpmsg *msg;
 599         int ret;
 600
 601 #ifdef CONFIG_IP_PIMSM
 602         if (assert == IGMPMSG_WHOLEPKT)
 603                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
 604         else
 605 #endif
 606                 skb = alloc_skb(128, GFP_ATOMIC);
 607
 608         if (!skb)
 609                 return -ENOBUFS;
 610
 611 #ifdef CONFIG_IP_PIMSM
 612         if (assert == IGMPMSG_WHOLEPKT) {
 613                 /* Ugly, but we have no choice with this interface.
 614                    Duplicate old header, fix ihl, length etc.
 615                    And all this only to mangle msg->im_msgtype and
 616                    to set msg->im_mbz to "mbz" :-)
 617                  */
 618                 skb_push(skb, sizeof(struct iphdr));
 619                 skb_reset_network_header(skb);
 620                 skb_reset_transport_header(skb);
 621                 msg = (struct igmpmsg *)skb_network_header(skb);
 622                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
 623                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
 624                 msg->im_mbz = 0;
 625                 msg->im_vif = reg_vif_num;
 626                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
 627                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
 628                                              sizeof(struct iphdr));
 629         } else
 630 #endif
 631         {
 632
 633         /*
 634          *      Copy the IP header
 635          */
 636
 637         skb->network_header = skb->tail;
 638         skb_put(skb, ihl);
 639         skb_copy_to_linear_data(skb, pkt->data, ihl);
 640         ip_hdr(skb)->protocol = 0;                      /* Flag to the kernel this is a route add */
 641         msg = (struct igmpmsg *)skb_network_header(skb);
 642         msg->im_vif = vifi;
 643         skb->dst = dst_clone(pkt->dst);
 644
 645         /*
 646          *      Add our header
 647          */
 648
 649         igmp=(struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
 650         igmp->type      =
 651         msg->im_msgtype = assert;
 652         igmp->code      =       0;
 653         ip_hdr(skb)->tot_len = htons(skb->len);                 /* Fix the length */
 654         skb->transport_header = skb->network_header;
 655         }
 656
 657         if (mroute_socket == NULL) {
 658                 kfree_skb(skb);
 659                 return -EINVAL;
 660         }
 661
 662         /*
 663          *      Deliver to mrouted
 664          */
 665         if ((ret = sock_queue_rcv_skb(mroute_socket, skb))<0) {
 666                 if (net_ratelimit())
 667                         printk(KERN_WARNING "mroute: pending queue full, dropping entries.\n");
 668                 kfree_skb(skb);
 669         }
 670
 671         return ret;
 672 }
 673
 674 /*
 675  *      Queue a packet for resolution. It gets locked cache entry!
 676  */
 677
 678 static int
 679 ipmr_cache_unresolved(vifi_t vifi, struct sk_buff *skb)
 680 {
 681         int err;
 682         struct mfc_cache *c;
 683         const struct iphdr *iph = ip_hdr(skb);
 684
 685         spin_lock_bh(&mfc_unres_lock);
 686         for (c=mfc_unres_queue; c; c=c->next) {
 687                 if (c->mfc_mcastgrp == iph->daddr &&
 688                     c->mfc_origin == iph->saddr)
 689                         break;
 690         }
 691
 692         if (c == NULL) {
 693                 /*
 694                  *      Create a new entry if allowable
 695                  */
 696
 697                 if (atomic_read(&cache_resolve_queue_len) >= 10 ||
 698                     (c=ipmr_cache_alloc_unres())==NULL) {
 699                         spin_unlock_bh(&mfc_unres_lock);
 700
 701                         kfree_skb(skb);
 702                         return -ENOBUFS;
 703                 }
 704
 705                 /*
 706                  *      Fill in the new cache entry
 707                  */
 708                 c->mfc_parent   = -1;
 709                 c->mfc_origin   = iph->saddr;
 710                 c->mfc_mcastgrp = iph->daddr;
 711
 712                 /*
 713                  *      Reflect first query at mrouted.
 714                  */
 715                 if ((err = ipmr_cache_report(skb, vifi, IGMPMSG_NOCACHE))<0) {
 716                         /* If the report failed throw the cache entry
 717                            out - Brad Parker
 718                          */
 719                         spin_unlock_bh(&mfc_unres_lock);
 720
 721                         kmem_cache_free(mrt_cachep, c);
 722                         kfree_skb(skb);
 723                         return err;
 724                 }
 725
 726                 atomic_inc(&cache_resolve_queue_len);
 727                 c->next = mfc_unres_queue;
 728                 mfc_unres_queue = c;
 729
 730                 mod_timer(&ipmr_expire_timer, c->mfc_un.unres.expires);
 731         }
 732
 733         /*
 734          *      See if we can append the packet
 735          */
 736         if (c->mfc_un.unres.unresolved.qlen>3) {
 737                 kfree_skb(skb);
 738                 err = -ENOBUFS;
 739         } else {
 740                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
 741                 err = 0;
 742         }
 743
 744         spin_unlock_bh(&mfc_unres_lock);
 745         return err;
 746 }
 747
 748 /*
 749  *      MFC cache manipulation by user space mroute daemon
 750  */
 751
 752 static int ipmr_mfc_delete(struct mfcctl *mfc)
 753 {
 754         int line;
 755         struct mfc_cache *c, **cp;
 756
 757         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
 758
 759         for (cp=&mfc_cache_array[line]; (c=*cp) != NULL; cp = &c->next) {
 760                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
 761                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
 762                         write_lock_bh(&mrt_lock);
 763                         *cp = c->next;
 764                         write_unlock_bh(&mrt_lock);
 765
 766                         kmem_cache_free(mrt_cachep, c);
 767                         return 0;
 768                 }
 769         }
 770         return -ENOENT;
 771 }
 772
 773 static int ipmr_mfc_add(struct mfcctl *mfc, int mrtsock)
 774 {
 775         int line;
 776         struct mfc_cache *uc, *c, **cp;
 777
 778         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
 779
 780         for (cp=&mfc_cache_array[line]; (c=*cp) != NULL; cp = &c->next) {
 781                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
 782                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr)
 783                         break;
 784         }
 785
 786         if (c != NULL) {
 787                 write_lock_bh(&mrt_lock);
 788                 c->mfc_parent = mfc->mfcc_parent;
 789                 ipmr_update_thresholds(c, mfc->mfcc_ttls);
 790                 if (!mrtsock)
 791                         c->mfc_flags |= MFC_STATIC;
 792                 write_unlock_bh(&mrt_lock);
 793                 return 0;
 794         }
 795
 796         if (!ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
 797                 return -EINVAL;
 798
 799         c = ipmr_cache_alloc();
 800         if (c == NULL)
 801                 return -ENOMEM;
 802
 803         c->mfc_origin = mfc->mfcc_origin.s_addr;
 804         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
 805         c->mfc_parent = mfc->mfcc_parent;
 806         ipmr_update_thresholds(c, mfc->mfcc_ttls);
 807         if (!mrtsock)
 808                 c->mfc_flags |= MFC_STATIC;
 809
 810         write_lock_bh(&mrt_lock);
 811         c->next = mfc_cache_array[line];
 812         mfc_cache_array[line] = c;
 813         write_unlock_bh(&mrt_lock);
 814
 815         /*
 816          *      Check to see if we resolved a queued list. If so we
 817          *      need to send on the frames and tidy up.
 818          */
 819         spin_lock_bh(&mfc_unres_lock);
 820         for (cp = &mfc_unres_queue; (uc=*cp) != NULL;
 821              cp = &uc->next) {
 822                 if (uc->mfc_origin == c->mfc_origin &&
 823                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
 824                         *cp = uc->next;
 825                         if (atomic_dec_and_test(&cache_resolve_queue_len))
 826                                 del_timer(&ipmr_expire_timer);
 827                         break;
 828                 }
 829         }
 830         spin_unlock_bh(&mfc_unres_lock);
 831
 832         if (uc) {
 833                 ipmr_cache_resolve(uc, c);
 834                 kmem_cache_free(mrt_cachep, uc);
 835         }
 836         return 0;
 837 }
 838
 839 /*
 840  *      Close the multicast socket, and clear the vif tables etc
 841  */
 842
 843 static void mroute_clean_tables(struct sock *sk)
 844 {
 845         int i;
 846
 847         /*
 848          *      Shut down all active vif entries
 849          */
 850         for (i=0; i<maxvif; i++) {
 851                 if (!(vif_table[i].flags&VIFF_STATIC))
 852                         vif_delete(i, 0);
 853         }
 854
 855         /*
 856          *      Wipe the cache
 857          */
 858         for (i=0; i<MFC_LINES; i++) {
 859                 struct mfc_cache *c, **cp;
 860
 861                 cp = &mfc_cache_array[i];
 862                 while ((c = *cp) != NULL) {
 863                         if (c->mfc_flags&MFC_STATIC) {
 864                                 cp = &c->next;
 865                                 continue;
 866                         }
 867                         write_lock_bh(&mrt_lock);
 868                         *cp = c->next;
 869                         write_unlock_bh(&mrt_lock);
 870
 871                         kmem_cache_free(mrt_cachep, c);
 872                 }
 873         }
 874
 875         if (atomic_read(&cache_resolve_queue_len) != 0) {
 876                 struct mfc_cache *c;
 877
 878                 spin_lock_bh(&mfc_unres_lock);
 879                 while (mfc_unres_queue != NULL) {
 880                         c = mfc_unres_queue;
 881                         mfc_unres_queue = c->next;
 882                         spin_unlock_bh(&mfc_unres_lock);
 883
 884                         ipmr_destroy_unres(c);
 885
 886                         spin_lock_bh(&mfc_unres_lock);
 887                 }
 888                 spin_unlock_bh(&mfc_unres_lock);
 889         }
 890 }
 891
 892 static void mrtsock_destruct(struct sock *sk)
 893 {
 894         rtnl_lock();
 895         if (sk == mroute_socket) {
 896                 IPV4_DEVCONF_ALL(sock_net(sk), MC_FORWARDING)--;
 897
 898                 write_lock_bh(&mrt_lock);
 899                 mroute_socket = NULL;
 900                 write_unlock_bh(&mrt_lock);
 901
 902                 mroute_clean_tables(sk);
 903         }
 904         rtnl_unlock();
 905 }
 906
 907 /*
 908  *      Socket options and virtual interface manipulation. The whole
 909  *      virtual interface system is a complete heap, but unfortunately
 910  *      that's how BSD mrouted happens to think. Maybe one day with a proper
 911  *      MOSPF/PIM router set up we can clean this up.
 912  */
 913
 914 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval, int optlen)
 915 {
 916         int ret;
 917         struct vifctl vif;
 918         struct mfcctl mfc;
 919
 920         if (optname != MRT_INIT) {
 921                 if (sk != mroute_socket && !capable(CAP_NET_ADMIN))
 922                         return -EACCES;
 923         }
 924
 925         switch (optname) {
 926         case MRT_INIT:
 927                 if (sk->sk_type != SOCK_RAW ||
 928                     inet_sk(sk)->num != IPPROTO_IGMP)
 929                         return -EOPNOTSUPP;
 930                 if (optlen != sizeof(int))
 931                         return -ENOPROTOOPT;
 932
 933                 rtnl_lock();
 934                 if (mroute_socket) {
 935                         rtnl_unlock();
 936                         return -EADDRINUSE;
 937                 }
 938
 939                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
 940                 if (ret == 0) {
 941                         write_lock_bh(&mrt_lock);
 942                         mroute_socket = sk;
 943                         write_unlock_bh(&mrt_lock);
 944
 945                         IPV4_DEVCONF_ALL(sock_net(sk), MC_FORWARDING)++;
 946                 }
 947                 rtnl_unlock();
 948                 return ret;
 949         case MRT_DONE:
 950                 if (sk != mroute_socket)
 951                         return -EACCES;
 952                 return ip_ra_control(sk, 0, NULL);
 953         case MRT_ADD_VIF:
 954         case MRT_DEL_VIF:
 955                 if (optlen != sizeof(vif))
 956                         return -EINVAL;
 957                 if (copy_from_user(&vif, optval, sizeof(vif)))
 958                         return -EFAULT;
 959                 if (vif.vifc_vifi >= MAXVIFS)
 960                         return -ENFILE;
 961                 rtnl_lock();
 962                 if (optname == MRT_ADD_VIF) {
 963                         ret = vif_add(&vif, sk==mroute_socket);
 964                 } else {
 965                         ret = vif_delete(vif.vifc_vifi, 0);
 966                 }
 967                 rtnl_unlock();
 968                 return ret;
 969
 970                 /*
 971                  *      Manipulate the forwarding caches. These live
 972                  *      in a sort of kernel/user symbiosis.
 973                  */
 974         case MRT_ADD_MFC:
 975         case MRT_DEL_MFC:
 976                 if (optlen != sizeof(mfc))
 977                         return -EINVAL;
 978                 if (copy_from_user(&mfc, optval, sizeof(mfc)))
 979                         return -EFAULT;
 980                 rtnl_lock();
 981                 if (optname == MRT_DEL_MFC)
 982                         ret = ipmr_mfc_delete(&mfc);
 983                 else
 984                         ret = ipmr_mfc_add(&mfc, sk==mroute_socket);
 985                 rtnl_unlock();
 986                 return ret;
 987                 /*
 988                  *      Control PIM assert.
 989                  */
 990         case MRT_ASSERT:
 991         {
 992                 int v;
 993                 if (get_user(v,(int __user *)optval))
 994                         return -EFAULT;
 995                 mroute_do_assert=(v)?1:0;
 996                 return 0;
 997         }
 998 #ifdef CONFIG_IP_PIMSM
 999         case MRT_PIM:
1000         {
1001                 int v;
1002
1003                 if (get_user(v,(int __user *)optval))
1004                         return -EFAULT;
1005                 v = (v) ? 1 : 0;
1006
1007                 rtnl_lock();
1008                 ret = 0;
1009                 if (v != mroute_do_pim) {
1010                         mroute_do_pim = v;
1011                         mroute_do_assert = v;
1012 #ifdef CONFIG_IP_PIMSM_V2
1013                         if (mroute_do_pim)
1014                                 ret = inet_add_protocol(&pim_protocol,
1015                                                         IPPROTO_PIM);
1016                         else
1017                                 ret = inet_del_protocol(&pim_protocol,
1018                                                         IPPROTO_PIM);
1019                         if (ret < 0)
1020                                 ret = -EAGAIN;
1021 #endif
1022                 }
1023                 rtnl_unlock();
1024                 return ret;
1025         }
1026 #endif
1027         /*
1028          *      Spurious command, or MRT_VERSION which you cannot
1029          *      set.
1030          */
1031         default:
1032                 return -ENOPROTOOPT;
1033         }
1034 }
1035
1036 /*
1037  *      Getsock opt support for the multicast routing system.
1038  */
1039
1040 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1041 {
1042         int olr;
1043         int val;
1044
1045         if (optname != MRT_VERSION &&
1046 #ifdef CONFIG_IP_PIMSM
1047            optname!=MRT_PIM &&
1048 #endif
1049            optname!=MRT_ASSERT)
1050                 return -ENOPROTOOPT;
1051
1052         if (get_user(olr, optlen))
1053                 return -EFAULT;
1054
1055         olr = min_t(unsigned int, olr, sizeof(int));
1056         if (olr < 0)
1057                 return -EINVAL;
1058
1059         if (put_user(olr, optlen))
1060                 return -EFAULT;
1061         if (optname == MRT_VERSION)
1062                 val = 0x0305;
1063 #ifdef CONFIG_IP_PIMSM
1064         else if (optname == MRT_PIM)
1065                 val = mroute_do_pim;
1066 #endif
1067         else
1068                 val = mroute_do_assert;
1069         if (copy_to_user(optval, &val, olr))
1070                 return -EFAULT;
1071         return 0;
1072 }
1073
1074 /*
1075  *      The IP multicast ioctl support routines.
1076  */
1077
1078 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1079 {
1080         struct sioc_sg_req sr;
1081         struct sioc_vif_req vr;
1082         struct vif_device *vif;
1083         struct mfc_cache *c;
1084
1085         switch (cmd) {
1086         case SIOCGETVIFCNT:
1087                 if (copy_from_user(&vr, arg, sizeof(vr)))
1088                         return -EFAULT;
1089                 if (vr.vifi >= maxvif)
1090                         return -EINVAL;
1091                 read_lock(&mrt_lock);
1092                 vif=&vif_table[vr.vifi];
1093                 if (VIF_EXISTS(vr.vifi))        {
1094                         vr.icount = vif->pkt_in;
1095                         vr.ocount = vif->pkt_out;
1096                         vr.ibytes = vif->bytes_in;
1097                         vr.obytes = vif->bytes_out;
1098                         read_unlock(&mrt_lock);
1099
1100                         if (copy_to_user(arg, &vr, sizeof(vr)))
1101                                 return -EFAULT;
1102                         return 0;
1103                 }
1104                 read_unlock(&mrt_lock);
1105                 return -EADDRNOTAVAIL;
1106         case SIOCGETSGCNT:
1107                 if (copy_from_user(&sr, arg, sizeof(sr)))
1108                         return -EFAULT;
1109
1110                 read_lock(&mrt_lock);
1111                 c = ipmr_cache_find(sr.src.s_addr, sr.grp.s_addr);
1112                 if (c) {
1113                         sr.pktcnt = c->mfc_un.res.pkt;
1114                         sr.bytecnt = c->mfc_un.res.bytes;
1115                         sr.wrong_if = c->mfc_un.res.wrong_if;
1116                         read_unlock(&mrt_lock);
1117
1118                         if (copy_to_user(arg, &sr, sizeof(sr)))
1119                                 return -EFAULT;
1120                         return 0;
1121                 }
1122                 read_unlock(&mrt_lock);
1123                 return -EADDRNOTAVAIL;
1124         default:
1125                 return -ENOIOCTLCMD;
1126         }
1127 }
1128
1129
1130 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1131 {
1132         struct net_device *dev = ptr;
1133         struct vif_device *v;
1134         int ct;
1135
1136         if (!net_eq(dev_net(dev), &init_net))
1137                 return NOTIFY_DONE;
1138
1139         if (event != NETDEV_UNREGISTER)
1140                 return NOTIFY_DONE;
1141         v=&vif_table[0];
1142         for (ct=0; ct<maxvif; ct++,v++) {
1143                 if (v->dev == dev)
1144                         vif_delete(ct, 1);
1145         }
1146         return NOTIFY_DONE;
1147 }
1148
1149
1150 static struct notifier_block ip_mr_notifier = {
1151         .notifier_call = ipmr_device_event,
1152 };
1153
1154 /*
1155  *      Encapsulate a packet by attaching a valid IPIP header to it.
1156  *      This avoids tunnel drivers and other mess and gives us the speed so
1157  *      important for multicast video.
1158  */
1159
1160 static void ip_encap(struct sk_buff *skb, __be32 saddr, __be32 daddr)
1161 {
1162         struct iphdr *iph;
1163         struct iphdr *old_iph = ip_hdr(skb);
1164
1165         skb_push(skb, sizeof(struct iphdr));
1166         skb->transport_header = skb->network_header;
1167         skb_reset_network_header(skb);
1168         iph = ip_hdr(skb);
1169
1170         iph->version    =       4;
1171         iph->tos        =       old_iph->tos;
1172         iph->ttl        =       old_iph->ttl;
1173         iph->frag_off   =       0;
1174         iph->daddr      =       daddr;
1175         iph->saddr      =       saddr;
1176         iph->protocol   =       IPPROTO_IPIP;
1177         iph->ihl        =       5;
1178         iph->tot_len    =       htons(skb->len);
1179         ip_select_ident(iph, skb->dst, NULL);
1180         ip_send_check(iph);
1181
1182         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1183         nf_reset(skb);
1184 }
1185
1186 static inline int ipmr_forward_finish(struct sk_buff *skb)
1187 {
1188         struct ip_options * opt = &(IPCB(skb)->opt);
1189
1190         IP_INC_STATS_BH(dev_net(skb->dst->dev), IPSTATS_MIB_OUTFORWDATAGRAMS);
1191
1192         if (unlikely(opt->optlen))
1193                 ip_forward_options(skb);
1194
1195         return dst_output(skb);
1196 }
1197
1198 /*
1199  *      Processing handlers for ipmr_forward
1200  */
1201
1202 static void ipmr_queue_xmit(struct sk_buff *skb, struct mfc_cache *c, int vifi)
1203 {
1204         const struct iphdr *iph = ip_hdr(skb);
1205         struct vif_device *vif = &vif_table[vifi];
1206         struct net_device *dev;
1207         struct rtable *rt;
1208         int    encap = 0;
1209
1210         if (vif->dev == NULL)
1211                 goto out_free;
1212
1213 #ifdef CONFIG_IP_PIMSM
1214         if (vif->flags & VIFF_REGISTER) {
1215                 vif->pkt_out++;
1216                 vif->bytes_out += skb->len;
1217                 vif->dev->stats.tx_bytes += skb->len;
1218                 vif->dev->stats.tx_packets++;
1219                 ipmr_cache_report(skb, vifi, IGMPMSG_WHOLEPKT);
1220                 kfree_skb(skb);
1221                 return;
1222         }
1223 #endif
1224
1225         if (vif->flags&VIFF_TUNNEL) {
1226                 struct flowi fl = { .oif = vif->link,
1227                                     .nl_u = { .ip4_u =
1228                                               { .daddr = vif->remote,
1229                                                 .saddr = vif->local,
1230                                                 .tos = RT_TOS(iph->tos) } },
1231                                     .proto = IPPROTO_IPIP };
1232                 if (ip_route_output_key(&init_net, &rt, &fl))
1233                         goto out_free;
1234                 encap = sizeof(struct iphdr);
1235         } else {
1236                 struct flowi fl = { .oif = vif->link,
1237                                     .nl_u = { .ip4_u =
1238                                               { .daddr = iph->daddr,
1239                                                 .tos = RT_TOS(iph->tos) } },
1240                                     .proto = IPPROTO_IPIP };
1241                 if (ip_route_output_key(&init_net, &rt, &fl))
1242                         goto out_free;
1243         }
1244
1245         dev = rt->u.dst.dev;
1246
1247         if (skb->len+encap > dst_mtu(&rt->u.dst) && (ntohs(iph->frag_off) & IP_DF)) {
1248                 /* Do not fragment multicasts. Alas, IPv4 does not
1249                    allow to send ICMP, so that packets will disappear
1250                    to blackhole.
1251                  */
1252
1253                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
1254                 ip_rt_put(rt);
1255                 goto out_free;
1256         }
1257
1258         encap += LL_RESERVED_SPACE(dev) + rt->u.dst.header_len;
1259
1260         if (skb_cow(skb, encap)) {
1261                 ip_rt_put(rt);
1262                 goto out_free;
1263         }
1264
1265         vif->pkt_out++;
1266         vif->bytes_out += skb->len;
1267
1268         dst_release(skb->dst);
1269         skb->dst = &rt->u.dst;
1270         ip_decrease_ttl(ip_hdr(skb));
1271
1272         /* FIXME: forward and output firewalls used to be called here.
1273          * What do we do with netfilter? -- RR */
1274         if (vif->flags & VIFF_TUNNEL) {
1275                 ip_encap(skb, vif->local, vif->remote);
1276                 /* FIXME: extra output firewall step used to be here. --RR */
1277                 vif->dev->stats.tx_packets++;
1278                 vif->dev->stats.tx_bytes += skb->len;
1279         }
1280
1281         IPCB(skb)->flags |= IPSKB_FORWARDED;
1282
1283         /*
1284          * RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1285          * not only before forwarding, but after forwarding on all output
1286          * interfaces. It is clear, if mrouter runs a multicasting
1287          * program, it should receive packets not depending to what interface
1288          * program is joined.
1289          * If we will not make it, the program will have to join on all
1290          * interfaces. On the other hand, multihoming host (or router, but
1291          * not mrouter) cannot join to more than one interface - it will
1292          * result in receiving multiple packets.
1293          */
1294         NF_HOOK(PF_INET, NF_INET_FORWARD, skb, skb->dev, dev,
1295                 ipmr_forward_finish);
1296         return;
1297
1298 out_free:
1299         kfree_skb(skb);
1300         return;
1301 }
1302
1303 static int ipmr_find_vif(struct net_device *dev)
1304 {
1305         int ct;
1306         for (ct=maxvif-1; ct>=0; ct--) {
1307                 if (vif_table[ct].dev == dev)
1308                         break;
1309         }
1310         return ct;
1311 }
1312
1313 /* "local" means that we should preserve one skb (for local delivery) */
1314
1315 static int ip_mr_forward(struct sk_buff *skb, struct mfc_cache *cache, int local)
1316 {
1317         int psend = -1;
1318         int vif, ct;
1319
1320         vif = cache->mfc_parent;
1321         cache->mfc_un.res.pkt++;
1322         cache->mfc_un.res.bytes += skb->len;
1323
1324         /*
1325          * Wrong interface: drop packet and (maybe) send PIM assert.
1326          */
1327         if (vif_table[vif].dev != skb->dev) {
1328                 int true_vifi;
1329
1330                 if (skb->rtable->fl.iif == 0) {
1331                         /* It is our own packet, looped back.
1332                            Very complicated situation...
1333
1334                            The best workaround until routing daemons will be
1335                            fixed is not to redistribute packet, if it was
1336                            send through wrong interface. It means, that
1337                            multicast applications WILL NOT work for
1338                            (S,G), which have default multicast route pointing
1339                            to wrong oif. In any case, it is not a good
1340                            idea to use multicasting applications on router.
1341                          */
1342                         goto dont_forward;
1343                 }
1344
1345                 cache->mfc_un.res.wrong_if++;
1346                 true_vifi = ipmr_find_vif(skb->dev);
1347
1348                 if (true_vifi >= 0 && mroute_do_assert &&
1349                     /* pimsm uses asserts, when switching from RPT to SPT,
1350                        so that we cannot check that packet arrived on an oif.
1351                        It is bad, but otherwise we would need to move pretty
1352                        large chunk of pimd to kernel. Ough... --ANK
1353                      */
1354                     (mroute_do_pim || cache->mfc_un.res.ttls[true_vifi] < 255) &&
1355                     time_after(jiffies,
1356                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1357                         cache->mfc_un.res.last_assert = jiffies;
1358                         ipmr_cache_report(skb, true_vifi, IGMPMSG_WRONGVIF);
1359                 }
1360                 goto dont_forward;
1361         }
1362
1363         vif_table[vif].pkt_in++;
1364         vif_table[vif].bytes_in += skb->len;
1365
1366         /*
1367          *      Forward the frame
1368          */
1369         for (ct = cache->mfc_un.res.maxvif-1; ct >= cache->mfc_un.res.minvif; ct--) {
1370                 if (ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1371                         if (psend != -1) {
1372                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1373                                 if (skb2)
1374                                         ipmr_queue_xmit(skb2, cache, psend);
1375                         }
1376                         psend = ct;
1377                 }
1378         }
1379         if (psend != -1) {
1380                 if (local) {
1381                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1382                         if (skb2)
1383                                 ipmr_queue_xmit(skb2, cache, psend);
1384                 } else {
1385                         ipmr_queue_xmit(skb, cache, psend);
1386                         return 0;
1387                 }
1388         }
1389
1390 dont_forward:
1391         if (!local)
1392                 kfree_skb(skb);
1393         return 0;
1394 }
1395
1396
1397 /*
1398  *      Multicast packets for forwarding arrive here
1399  */
1400
1401 int ip_mr_input(struct sk_buff *skb)
1402 {
1403         struct mfc_cache *cache;
1404         int local = skb->rtable->rt_flags&RTCF_LOCAL;
1405
1406         /* Packet is looped back after forward, it should not be
1407            forwarded second time, but still can be delivered locally.
1408          */
1409         if (IPCB(skb)->flags&IPSKB_FORWARDED)
1410                 goto dont_forward;
1411
1412         if (!local) {
1413                     if (IPCB(skb)->opt.router_alert) {
1414                             if (ip_call_ra_chain(skb))
1415                                     return 0;
1416                     } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP){
1417                             /* IGMPv1 (and broken IGMPv2 implementations sort of
1418                                Cisco IOS <= 11.2(8)) do not put router alert
1419                                option to IGMP packets destined to routable
1420                                groups. It is very bad, because it means
1421                                that we can forward NO IGMP messages.
1422                              */
1423                             read_lock(&mrt_lock);
1424                             if (mroute_socket) {
1425                                     nf_reset(skb);
1426                                     raw_rcv(mroute_socket, skb);
1427                                     read_unlock(&mrt_lock);
1428                                     return 0;
1429                             }
1430                             read_unlock(&mrt_lock);
1431                     }
1432         }
1433
1434         read_lock(&mrt_lock);
1435         cache = ipmr_cache_find(ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1436
1437         /*
1438          *      No usable cache entry
1439          */
1440         if (cache == NULL) {
1441                 int vif;
1442
1443                 if (local) {
1444                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1445                         ip_local_deliver(skb);
1446                         if (skb2 == NULL) {
1447                                 read_unlock(&mrt_lock);
1448                                 return -ENOBUFS;
1449                         }
1450                         skb = skb2;
1451                 }
1452
1453                 vif = ipmr_find_vif(skb->dev);
1454                 if (vif >= 0) {
1455                         int err = ipmr_cache_unresolved(vif, skb);
1456                         read_unlock(&mrt_lock);
1457
1458                         return err;
1459                 }
1460                 read_unlock(&mrt_lock);
1461                 kfree_skb(skb);
1462                 return -ENODEV;
1463         }
1464
1465         ip_mr_forward(skb, cache, local);
1466
1467         read_unlock(&mrt_lock);
1468
1469         if (local)
1470                 return ip_local_deliver(skb);
1471
1472         return 0;
1473
1474 dont_forward:
1475         if (local)
1476                 return ip_local_deliver(skb);
1477         kfree_skb(skb);
1478         return 0;
1479 }
1480
1481 #ifdef CONFIG_IP_PIMSM_V1
1482 /*
1483  * Handle IGMP messages of PIMv1
1484  */
1485
1486 int pim_rcv_v1(struct sk_buff * skb)
1487 {
1488         struct igmphdr *pim;
1489         struct iphdr   *encap;
1490         struct net_device  *reg_dev = NULL;
1491
1492         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(*encap)))
1493                 goto drop;
1494
1495         pim = igmp_hdr(skb);
1496
1497         if (!mroute_do_pim ||
1498             skb->len < sizeof(*pim) + sizeof(*encap) ||
1499             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
1500                 goto drop;
1501
1502         encap = (struct iphdr *)(skb_transport_header(skb) +
1503                                  sizeof(struct igmphdr));
1504         /*
1505            Check that:
1506            a. packet is really destinted to a multicast group
1507            b. packet is not a NULL-REGISTER
1508            c. packet is not truncated
1509          */
1510         if (!ipv4_is_multicast(encap->daddr) ||
1511             encap->tot_len == 0 ||
1512             ntohs(encap->tot_len) + sizeof(*pim) > skb->len)
1513                 goto drop;
1514
1515         read_lock(&mrt_lock);
1516         if (reg_vif_num >= 0)
1517                 reg_dev = vif_table[reg_vif_num].dev;
1518         if (reg_dev)
1519                 dev_hold(reg_dev);
1520         read_unlock(&mrt_lock);
1521
1522         if (reg_dev == NULL)
1523                 goto drop;
1524
1525         skb->mac_header = skb->network_header;
1526         skb_pull(skb, (u8*)encap - skb->data);
1527         skb_reset_network_header(skb);
1528         skb->dev = reg_dev;
1529         skb->protocol = htons(ETH_P_IP);
1530         skb->ip_summed = 0;
1531         skb->pkt_type = PACKET_HOST;
1532         dst_release(skb->dst);
1533         skb->dst = NULL;
1534         reg_dev->stats.rx_bytes += skb->len;
1535         reg_dev->stats.rx_packets++;
1536         nf_reset(skb);
1537         netif_rx(skb);
1538         dev_put(reg_dev);
1539         return 0;
1540  drop:
1541         kfree_skb(skb);
1542         return 0;
1543 }
1544 #endif
1545
1546 #ifdef CONFIG_IP_PIMSM_V2
1547 static int pim_rcv(struct sk_buff * skb)
1548 {
1549         struct pimreghdr *pim;
1550         struct iphdr   *encap;
1551         struct net_device  *reg_dev = NULL;
1552
1553         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(*encap)))
1554                 goto drop;
1555
1556         pim = (struct pimreghdr *)skb_transport_header(skb);
1557         if (pim->type != ((PIM_VERSION<<4)|(PIM_REGISTER)) ||
1558             (pim->flags&PIM_NULL_REGISTER) ||
1559             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
1560              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
1561                 goto drop;
1562
1563         /* check if the inner packet is destined to mcast group */
1564         encap = (struct iphdr *)(skb_transport_header(skb) +
1565                                  sizeof(struct pimreghdr));
1566         if (!ipv4_is_multicast(encap->daddr) ||
1567             encap->tot_len == 0 ||
1568             ntohs(encap->tot_len) + sizeof(*pim) > skb->len)
1569                 goto drop;
1570
1571         read_lock(&mrt_lock);
1572         if (reg_vif_num >= 0)
1573                 reg_dev = vif_table[reg_vif_num].dev;
1574         if (reg_dev)
1575                 dev_hold(reg_dev);
1576         read_unlock(&mrt_lock);
1577
1578         if (reg_dev == NULL)
1579                 goto drop;
1580
1581         skb->mac_header = skb->network_header;
1582         skb_pull(skb, (u8*)encap - skb->data);
1583         skb_reset_network_header(skb);
1584         skb->dev = reg_dev;
1585         skb->protocol = htons(ETH_P_IP);
1586         skb->ip_summed = 0;
1587         skb->pkt_type = PACKET_HOST;
1588         dst_release(skb->dst);
1589         reg_dev->stats.rx_bytes += skb->len;
1590         reg_dev->stats.rx_packets++;
1591         skb->dst = NULL;
1592         nf_reset(skb);
1593         netif_rx(skb);
1594         dev_put(reg_dev);
1595         return 0;
1596  drop:
1597         kfree_skb(skb);
1598         return 0;
1599 }
1600 #endif
1601
1602 static int
1603 ipmr_fill_mroute(struct sk_buff *skb, struct mfc_cache *c, struct rtmsg *rtm)
1604 {
1605         int ct;
1606         struct rtnexthop *nhp;
1607         struct net_device *dev = vif_table[c->mfc_parent].dev;
1608         u8 *b = skb_tail_pointer(skb);
1609         struct rtattr *mp_head;
1610
1611         if (dev)
1612                 RTA_PUT(skb, RTA_IIF, 4, &dev->ifindex);
1613
1614         mp_head = (struct rtattr *)skb_put(skb, RTA_LENGTH(0));
1615
1616         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
1617                 if (c->mfc_un.res.ttls[ct] < 255) {
1618                         if (skb_tailroom(skb) < RTA_ALIGN(RTA_ALIGN(sizeof(*nhp)) + 4))
1619                                 goto rtattr_failure;
1620                         nhp = (struct rtnexthop *)skb_put(skb, RTA_ALIGN(sizeof(*nhp)));
1621                         nhp->rtnh_flags = 0;
1622                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
1623                         nhp->rtnh_ifindex = vif_table[ct].dev->ifindex;
1624                         nhp->rtnh_len = sizeof(*nhp);
1625                 }
1626         }
1627         mp_head->rta_type = RTA_MULTIPATH;
1628         mp_head->rta_len = skb_tail_pointer(skb) - (u8 *)mp_head;
1629         rtm->rtm_type = RTN_MULTICAST;
1630         return 1;
1631
1632 rtattr_failure:
1633         nlmsg_trim(skb, b);
1634         return -EMSGSIZE;
1635 }
1636
1637 int ipmr_get_route(struct sk_buff *skb, struct rtmsg *rtm, int nowait)
1638 {
1639         int err;
1640         struct mfc_cache *cache;
1641         struct rtable *rt = skb->rtable;
1642
1643         read_lock(&mrt_lock);
1644         cache = ipmr_cache_find(rt->rt_src, rt->rt_dst);
1645
1646         if (cache == NULL) {
1647                 struct sk_buff *skb2;
1648                 struct iphdr *iph;
1649                 struct net_device *dev;
1650                 int vif;
1651
1652                 if (nowait) {
1653                         read_unlock(&mrt_lock);
1654                         return -EAGAIN;
1655                 }
1656
1657                 dev = skb->dev;
1658                 if (dev == NULL || (vif = ipmr_find_vif(dev)) < 0) {
1659                         read_unlock(&mrt_lock);
1660                         return -ENODEV;
1661                 }
1662                 skb2 = skb_clone(skb, GFP_ATOMIC);
1663                 if (!skb2) {
1664                         read_unlock(&mrt_lock);
1665                         return -ENOMEM;
1666                 }
1667
1668                 skb_push(skb2, sizeof(struct iphdr));
1669                 skb_reset_network_header(skb2);
1670                 iph = ip_hdr(skb2);
1671                 iph->ihl = sizeof(struct iphdr) >> 2;
1672                 iph->saddr = rt->rt_src;
1673                 iph->daddr = rt->rt_dst;
1674                 iph->version = 0;
1675                 err = ipmr_cache_unresolved(vif, skb2);
1676                 read_unlock(&mrt_lock);
1677                 return err;
1678         }
1679
1680         if (!nowait && (rtm->rtm_flags&RTM_F_NOTIFY))
1681                 cache->mfc_flags |= MFC_NOTIFY;
1682         err = ipmr_fill_mroute(skb, cache, rtm);
1683         read_unlock(&mrt_lock);
1684         return err;
1685 }
1686
1687 #ifdef CONFIG_PROC_FS
1688 /*
1689  *      The /proc interfaces to multicast routing /proc/ip_mr_cache /proc/ip_mr_vif
1690  */
1691 struct ipmr_vif_iter {
1692         int ct;
1693 };
1694
1695 static struct vif_device *ipmr_vif_seq_idx(struct ipmr_vif_iter *iter,
1696                                            loff_t pos)
1697 {
1698         for (iter->ct = 0; iter->ct < maxvif; ++iter->ct) {
1699                 if (!VIF_EXISTS(iter->ct))
1700                         continue;
1701                 if (pos-- == 0)
1702                         return &vif_table[iter->ct];
1703         }
1704         return NULL;
1705 }
1706
1707 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
1708         __acquires(mrt_lock)
1709 {
1710         read_lock(&mrt_lock);
1711         return *pos ? ipmr_vif_seq_idx(seq->private, *pos - 1)
1712                 : SEQ_START_TOKEN;
1713 }
1714
1715 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1716 {
1717         struct ipmr_vif_iter *iter = seq->private;
1718
1719         ++*pos;
1720         if (v == SEQ_START_TOKEN)
1721                 return ipmr_vif_seq_idx(iter, 0);
1722
1723         while (++iter->ct < maxvif) {
1724                 if (!VIF_EXISTS(iter->ct))
1725                         continue;
1726                 return &vif_table[iter->ct];
1727         }
1728         return NULL;
1729 }
1730
1731 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
1732         __releases(mrt_lock)
1733 {
1734         read_unlock(&mrt_lock);
1735 }
1736
1737 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
1738 {
1739         if (v == SEQ_START_TOKEN) {
1740                 seq_puts(seq,
1741                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
1742         } else {
1743                 const struct vif_device *vif = v;
1744                 const char *name =  vif->dev ? vif->dev->name : "none";
1745
1746                 seq_printf(seq,
1747                            "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
1748                            vif - vif_table,
1749                            name, vif->bytes_in, vif->pkt_in,
1750                            vif->bytes_out, vif->pkt_out,
1751                            vif->flags, vif->local, vif->remote);
1752         }
1753         return 0;
1754 }
1755
1756 static const struct seq_operations ipmr_vif_seq_ops = {
1757         .start = ipmr_vif_seq_start,
1758         .next  = ipmr_vif_seq_next,
1759         .stop  = ipmr_vif_seq_stop,
1760         .show  = ipmr_vif_seq_show,
1761 };
1762
1763 static int ipmr_vif_open(struct inode *inode, struct file *file)
1764 {
1765         return seq_open_private(file, &ipmr_vif_seq_ops,
1766                         sizeof(struct ipmr_vif_iter));
1767 }
1768
1769 static const struct file_operations ipmr_vif_fops = {
1770         .owner   = THIS_MODULE,
1771         .open    = ipmr_vif_open,
1772         .read    = seq_read,
1773         .llseek  = seq_lseek,
1774         .release = seq_release_private,
1775 };
1776
1777 struct ipmr_mfc_iter {
1778         struct mfc_cache **cache;
1779         int ct;
1780 };
1781
1782
1783 static struct mfc_cache *ipmr_mfc_seq_idx(struct ipmr_mfc_iter *it, loff_t pos)
1784 {
1785         struct mfc_cache *mfc;
1786
1787         it->cache = mfc_cache_array;
1788         read_lock(&mrt_lock);
1789         for (it->ct = 0; it->ct < MFC_LINES; it->ct++)
1790                 for (mfc = mfc_cache_array[it->ct]; mfc; mfc = mfc->next)
1791                         if (pos-- == 0)
1792                                 return mfc;
1793         read_unlock(&mrt_lock);
1794
1795         it->cache = &mfc_unres_queue;
1796         spin_lock_bh(&mfc_unres_lock);
1797         for (mfc = mfc_unres_queue; mfc; mfc = mfc->next)
1798                 if (pos-- == 0)
1799                         return mfc;
1800         spin_unlock_bh(&mfc_unres_lock);
1801
1802         it->cache = NULL;
1803         return NULL;
1804 }
1805
1806
1807 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
1808 {
1809         struct ipmr_mfc_iter *it = seq->private;
1810         it->cache = NULL;
1811         it->ct = 0;
1812         return *pos ? ipmr_mfc_seq_idx(seq->private, *pos - 1)
1813                 : SEQ_START_TOKEN;
1814 }
1815
1816 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1817 {
1818         struct mfc_cache *mfc = v;
1819         struct ipmr_mfc_iter *it = seq->private;
1820
1821         ++*pos;
1822
1823         if (v == SEQ_START_TOKEN)
1824                 return ipmr_mfc_seq_idx(seq->private, 0);
1825
1826         if (mfc->next)
1827                 return mfc->next;
1828
1829         if (it->cache == &mfc_unres_queue)
1830                 goto end_of_list;
1831
1832         BUG_ON(it->cache != mfc_cache_array);
1833
1834         while (++it->ct < MFC_LINES) {
1835                 mfc = mfc_cache_array[it->ct];
1836                 if (mfc)
1837                         return mfc;
1838         }
1839
1840         /* exhausted cache_array, show unresolved */
1841         read_unlock(&mrt_lock);
1842         it->cache = &mfc_unres_queue;
1843         it->ct = 0;
1844
1845         spin_lock_bh(&mfc_unres_lock);
1846         mfc = mfc_unres_queue;
1847         if (mfc)
1848                 return mfc;
1849
1850  end_of_list:
1851         spin_unlock_bh(&mfc_unres_lock);
1852         it->cache = NULL;
1853
1854         return NULL;
1855 }
1856
1857 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
1858 {
1859         struct ipmr_mfc_iter *it = seq->private;
1860
1861         if (it->cache == &mfc_unres_queue)
1862                 spin_unlock_bh(&mfc_unres_lock);
1863         else if (it->cache == mfc_cache_array)
1864                 read_unlock(&mrt_lock);
1865 }
1866
1867 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
1868 {
1869         int n;
1870
1871         if (v == SEQ_START_TOKEN) {
1872                 seq_puts(seq,
1873                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
1874         } else {
1875                 const struct mfc_cache *mfc = v;
1876                 const struct ipmr_mfc_iter *it = seq->private;
1877
1878                 seq_printf(seq, "%08lX %08lX %-3d %8ld %8ld %8ld",
1879                            (unsigned long) mfc->mfc_mcastgrp,
1880                            (unsigned long) mfc->mfc_origin,
1881                            mfc->mfc_parent,
1882                            mfc->mfc_un.res.pkt,
1883                            mfc->mfc_un.res.bytes,
1884                            mfc->mfc_un.res.wrong_if);
1885
1886                 if (it->cache != &mfc_unres_queue) {
1887                         for (n = mfc->mfc_un.res.minvif;
1888                              n < mfc->mfc_un.res.maxvif; n++ ) {
1889                                 if (VIF_EXISTS(n)
1890                                    && mfc->mfc_un.res.ttls[n] < 255)
1891                                 seq_printf(seq,
1892                                            " %2d:%-3d",
1893                                            n, mfc->mfc_un.res.ttls[n]);
1894                         }
1895                 }
1896                 seq_putc(seq, '\n');
1897         }
1898         return 0;
1899 }
1900
1901 static const struct seq_operations ipmr_mfc_seq_ops = {
1902         .start = ipmr_mfc_seq_start,
1903         .next  = ipmr_mfc_seq_next,
1904         .stop  = ipmr_mfc_seq_stop,
1905         .show  = ipmr_mfc_seq_show,
1906 };
1907
1908 static int ipmr_mfc_open(struct inode *inode, struct file *file)
1909 {
1910         return seq_open_private(file, &ipmr_mfc_seq_ops,
1911                         sizeof(struct ipmr_mfc_iter));
1912 }
1913
1914 static const struct file_operations ipmr_mfc_fops = {
1915         .owner   = THIS_MODULE,
1916         .open    = ipmr_mfc_open,
1917         .read    = seq_read,
1918         .llseek  = seq_lseek,
1919         .release = seq_release_private,
1920 };
1921 #endif
1922
1923 #ifdef CONFIG_IP_PIMSM_V2
1924 static struct net_protocol pim_protocol = {
1925         .handler        =       pim_rcv,
1926 };
1927 #endif
1928
1929
1930 /*
1931  *      Setup for IP multicast routing
1932  */
1933
1934 int __init ip_mr_init(void)
1935 {
1936         int err;
1937
1938         mrt_cachep = kmem_cache_create("ip_mrt_cache",
1939                                        sizeof(struct mfc_cache),
1940                                        0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1941                                        NULL);
1942         if (!mrt_cachep)
1943                 return -ENOMEM;
1944
1945         setup_timer(&ipmr_expire_timer, ipmr_expire_process, 0);
1946         err = register_netdevice_notifier(&ip_mr_notifier);
1947         if (err)
1948                 goto reg_notif_fail;
1949 #ifdef CONFIG_PROC_FS
1950         err = -ENOMEM;
1951         if (!proc_net_fops_create(&init_net, "ip_mr_vif", 0, &ipmr_vif_fops))
1952                 goto proc_vif_fail;
1953         if (!proc_net_fops_create(&init_net, "ip_mr_cache", 0, &ipmr_mfc_fops))
1954                 goto proc_cache_fail;
1955 #endif
1956         return 0;
1957 #ifdef CONFIG_PROC_FS
1958 proc_cache_fail:
1959         proc_net_remove(&init_net, "ip_mr_vif");
1960 proc_vif_fail:
1961         unregister_netdevice_notifier(&ip_mr_notifier);
1962 #endif
1963 reg_notif_fail:
1964         del_timer(&ipmr_expire_timer);
1965         kmem_cache_destroy(mrt_cachep);
1966         return err;
1967 }