net/ipv4/route.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              ROUTE - implementation of the IP router.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <[email protected]>
  10  *              Alan Cox, <[email protected]>
  11  *              Linus Torvalds, <[email protected]>
  12  *              Alexey Kuznetsov, <[email protected]>
  13  *
  14  * Fixes:
  15  *              Alan Cox        :       Verify area fixes.
  16  *              Alan Cox        :       cli() protects routing changes
  17  *              Rui Oliveira    :       ICMP routing table updates
  18  *              ([email protected])      Routing table insertion and update
  19  *              Linus Torvalds  :       Rewrote bits to be sensible
  20  *              Alan Cox        :       Added BSD route gw semantics
  21  *              Alan Cox        :       Super /proc >4K
  22  *              Alan Cox        :       MTU in route table
  23  *              Alan Cox        :       MSS actually. Also added the window
  24  *                                      clamper.
  25  *              Sam Lantinga    :       Fixed route matching in rt_del()
  26  *              Alan Cox        :       Routing cache support.
  27  *              Alan Cox        :       Removed compatibility cruft.
  28  *              Alan Cox        :       RTF_REJECT support.
  29  *              Alan Cox        :       TCP irtt support.
  30  *              Jonathan Naylor :       Added Metric support.
  31  *      Miquel van Smoorenburg  :       BSD API fixes.
  32  *      Miquel van Smoorenburg  :       Metrics.
  33  *              Alan Cox        :       Use __u32 properly
  34  *              Alan Cox        :       Aligned routing errors more closely with BSD
  35  *                                      our system is still very different.
  36  *              Alan Cox        :       Faster /proc handling
  37  *      Alexey Kuznetsov        :       Massive rework to support tree based routing,
  38  *                                      routing caches and better behaviour.
  39  *
  40  *              Olaf Erb        :       irtt wasn't being copied right.
  41  *              Bjorn Ekwall    :       Kerneld route support.
  42  *              Alan Cox        :       Multicast fixed (I hope)
  43  *              Pavel Krauz     :       Limited broadcast fixed
  44  *              Mike McLagan    :       Routing by source
  45  *      Alexey Kuznetsov        :       End of old history. Split to fib.c and
  46  *                                      route.c and rewritten from scratch.
  47  *              Andi Kleen      :       Load-limit warning messages.
  48  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  49  *      Vitaly E. Lavrov        :       Race condition in ip_route_input_slow.
  50  *      Tobias Ringstrom        :       Uninitialized res.type in ip_route_output_slow.
  51  *      Vladimir V. Ivanov      :       IP rule info (flowid) is really useful.
  52  *              Marc Boucher    :       routing by fwmark
  53  *      Robert Olsson           :       Added rt_cache statistics
  54  *      Arnaldo C. Melo         :       Convert proc stuff to seq_file
  55  *      Eric Dumazet            :       hashed spinlocks and rt_check_expire() fixes.
  56  *      Ilia Sotnikov           :       Ignore TOS on PMTUD and Redirect
  57  *      Ilia Sotnikov           :       Removed TOS from hash calculations
  58  *
  59  *              This program is free software; you can redistribute it and/or
  60  *              modify it under the terms of the GNU General Public License
  61  *              as published by the Free Software Foundation; either version
  62  *              2 of the License, or (at your option) any later version.
  63  */
  64
  65 #define pr_fmt(fmt) "IPv4: " fmt
  66
  67 #include <linux/module.h>
  68 #include <asm/uaccess.h>
  69 #include <linux/bitops.h>
  70 #include <linux/types.h>
  71 #include <linux/kernel.h>
  72 #include <linux/mm.h>
  73 #include <linux/string.h>
  74 #include <linux/socket.h>
  75 #include <linux/sockios.h>
  76 #include <linux/errno.h>
  77 #include <linux/in.h>
  78 #include <linux/inet.h>
  79 #include <linux/netdevice.h>
  80 #include <linux/proc_fs.h>
  81 #include <linux/init.h>
  82 #include <linux/skbuff.h>
  83 #include <linux/inetdevice.h>
  84 #include <linux/igmp.h>
  85 #include <linux/pkt_sched.h>
  86 #include <linux/mroute.h>
  87 #include <linux/netfilter_ipv4.h>
  88 #include <linux/random.h>
  89 #include <linux/rcupdate.h>
  90 #include <linux/times.h>
  91 #include <linux/slab.h>
  92 #include <net/dst.h>
  93 #include <net/net_namespace.h>
  94 #include <net/protocol.h>
  95 #include <net/ip.h>
  96 #include <net/route.h>
  97 #include <net/inetpeer.h>
  98 #include <net/sock.h>
  99 #include <net/ip_fib.h>
 100 #include <net/arp.h>
 101 #include <net/tcp.h>
 102 #include <net/icmp.h>
 103 #include <net/xfrm.h>
 104 #include <net/netevent.h>
 105 #include <net/rtnetlink.h>
 106 #ifdef CONFIG_SYSCTL
 107 #include <linux/sysctl.h>
 108 #include <linux/kmemleak.h>
 109 #endif
 110 #include <net/secure_seq.h>
 111
 112 #define RT_FL_TOS(oldflp4) \
 113         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
 114
 115 #define IP_MAX_MTU      0xFFF0
 116
 117 #define RT_GC_TIMEOUT (300*HZ)
 118
 119 static int ip_rt_max_size;
 120 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
 121 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
 122 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
 123 static int ip_rt_redirect_number __read_mostly  = 9;
 124 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 125 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 126 static int ip_rt_error_cost __read_mostly       = HZ;
 127 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 128 static int ip_rt_gc_elasticity __read_mostly    = 8;
 129 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
 130 static int ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
 131 static int ip_rt_min_advmss __read_mostly       = 256;
 132
 133 /*
 134  *      Interface to generic destination cache.
 135  */
 136
 137 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 138 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 139 static unsigned int      ipv4_mtu(const struct dst_entry *dst);
 140 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 141 static void              ipv4_link_failure(struct sk_buff *skb);
 142 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 143                                            struct sk_buff *skb, u32 mtu);
 144 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 145                                         struct sk_buff *skb);
 146 static void             ipv4_dst_destroy(struct dst_entry *dst);
 147
 148 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 149                             int how)
 150 {
 151 }
 152
 153 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 154 {
 155         WARN_ON(1);
 156         return NULL;
 157 }
 158
 159 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 160                                            struct sk_buff *skb,
 161                                            const void *daddr);
 162
 163 static struct dst_ops ipv4_dst_ops = {
 164         .family =               AF_INET,
 165         .protocol =             cpu_to_be16(ETH_P_IP),
 166         .check =                ipv4_dst_check,
 167         .default_advmss =       ipv4_default_advmss,
 168         .mtu =                  ipv4_mtu,
 169         .cow_metrics =          ipv4_cow_metrics,
 170         .destroy =              ipv4_dst_destroy,
 171         .ifdown =               ipv4_dst_ifdown,
 172         .negative_advice =      ipv4_negative_advice,
 173         .link_failure =         ipv4_link_failure,
 174         .update_pmtu =          ip_rt_update_pmtu,
 175         .redirect =             ip_do_redirect,
 176         .local_out =            __ip_local_out,
 177         .neigh_lookup =         ipv4_neigh_lookup,
 178 };
 179
 180 #define ECN_OR_COST(class)      TC_PRIO_##class
 181
 182 const __u8 ip_tos2prio[16] = {
 183         TC_PRIO_BESTEFFORT,
 184         ECN_OR_COST(BESTEFFORT),
 185         TC_PRIO_BESTEFFORT,
 186         ECN_OR_COST(BESTEFFORT),
 187         TC_PRIO_BULK,
 188         ECN_OR_COST(BULK),
 189         TC_PRIO_BULK,
 190         ECN_OR_COST(BULK),
 191         TC_PRIO_INTERACTIVE,
 192         ECN_OR_COST(INTERACTIVE),
 193         TC_PRIO_INTERACTIVE,
 194         ECN_OR_COST(INTERACTIVE),
 195         TC_PRIO_INTERACTIVE_BULK,
 196         ECN_OR_COST(INTERACTIVE_BULK),
 197         TC_PRIO_INTERACTIVE_BULK,
 198         ECN_OR_COST(INTERACTIVE_BULK)
 199 };
 200 EXPORT_SYMBOL(ip_tos2prio);
 201
 202 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 203 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
 204
 205 static inline int rt_genid(struct net *net)
 206 {
 207         return atomic_read(&net->ipv4.rt_genid);
 208 }
 209
 210 #ifdef CONFIG_PROC_FS
 211 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 212 {
 213         if (*pos)
 214                 return NULL;
 215         return SEQ_START_TOKEN;
 216 }
 217
 218 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 219 {
 220         ++*pos;
 221         return NULL;
 222 }
 223
 224 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 225 {
 226 }
 227
 228 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 229 {
 230         if (v == SEQ_START_TOKEN)
 231                 seq_printf(seq, "%-127s\n",
 232                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 233                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 234                            "HHUptod\tSpecDst");
 235         return 0;
 236 }
 237
 238 static const struct seq_operations rt_cache_seq_ops = {
 239         .start  = rt_cache_seq_start,
 240         .next   = rt_cache_seq_next,
 241         .stop   = rt_cache_seq_stop,
 242         .show   = rt_cache_seq_show,
 243 };
 244
 245 static int rt_cache_seq_open(struct inode *inode, struct file *file)
 246 {
 247         return seq_open(file, &rt_cache_seq_ops);
 248 }
 249
 250 static const struct file_operations rt_cache_seq_fops = {
 251         .owner   = THIS_MODULE,
 252         .open    = rt_cache_seq_open,
 253         .read    = seq_read,
 254         .llseek  = seq_lseek,
 255         .release = seq_release,
 256 };
 257
 258
 259 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 260 {
 261         int cpu;
 262
 263         if (*pos == 0)
 264                 return SEQ_START_TOKEN;
 265
 266         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 267                 if (!cpu_possible(cpu))
 268                         continue;
 269                 *pos = cpu+1;
 270                 return &per_cpu(rt_cache_stat, cpu);
 271         }
 272         return NULL;
 273 }
 274
 275 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 276 {
 277         int cpu;
 278
 279         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 280                 if (!cpu_possible(cpu))
 281                         continue;
 282                 *pos = cpu+1;
 283                 return &per_cpu(rt_cache_stat, cpu);
 284         }
 285         return NULL;
 286
 287 }
 288
 289 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 290 {
 291
 292 }
 293
 294 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 295 {
 296         struct rt_cache_stat *st = v;
 297
 298         if (v == SEQ_START_TOKEN) {
 299                 seq_printf(seq, "entries  in_hit in_slow_tot in_slow_mc in_no_route in_brd in_martian_dst in_martian_src  out_hit out_slow_tot out_slow_mc  gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
 300                 return 0;
 301         }
 302
 303         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 304                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 305                    dst_entries_get_slow(&ipv4_dst_ops),
 306                    st->in_hit,
 307                    st->in_slow_tot,
 308                    st->in_slow_mc,
 309                    st->in_no_route,
 310                    st->in_brd,
 311                    st->in_martian_dst,
 312                    st->in_martian_src,
 313
 314                    st->out_hit,
 315                    st->out_slow_tot,
 316                    st->out_slow_mc,
 317
 318                    st->gc_total,
 319                    st->gc_ignored,
 320                    st->gc_goal_miss,
 321                    st->gc_dst_overflow,
 322                    st->in_hlist_search,
 323                    st->out_hlist_search
 324                 );
 325         return 0;
 326 }
 327
 328 static const struct seq_operations rt_cpu_seq_ops = {
 329         .start  = rt_cpu_seq_start,
 330         .next   = rt_cpu_seq_next,
 331         .stop   = rt_cpu_seq_stop,
 332         .show   = rt_cpu_seq_show,
 333 };
 334
 335
 336 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
 337 {
 338         return seq_open(file, &rt_cpu_seq_ops);
 339 }
 340
 341 static const struct file_operations rt_cpu_seq_fops = {
 342         .owner   = THIS_MODULE,
 343         .open    = rt_cpu_seq_open,
 344         .read    = seq_read,
 345         .llseek  = seq_lseek,
 346         .release = seq_release,
 347 };
 348
 349 #ifdef CONFIG_IP_ROUTE_CLASSID
 350 static int rt_acct_proc_show(struct seq_file *m, void *v)
 351 {
 352         struct ip_rt_acct *dst, *src;
 353         unsigned int i, j;
 354
 355         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 356         if (!dst)
 357                 return -ENOMEM;
 358
 359         for_each_possible_cpu(i) {
 360                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 361                 for (j = 0; j < 256; j++) {
 362                         dst[j].o_bytes   += src[j].o_bytes;
 363                         dst[j].o_packets += src[j].o_packets;
 364                         dst[j].i_bytes   += src[j].i_bytes;
 365                         dst[j].i_packets += src[j].i_packets;
 366                 }
 367         }
 368
 369         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 370         kfree(dst);
 371         return 0;
 372 }
 373
 374 static int rt_acct_proc_open(struct inode *inode, struct file *file)
 375 {
 376         return single_open(file, rt_acct_proc_show, NULL);
 377 }
 378
 379 static const struct file_operations rt_acct_proc_fops = {
 380         .owner          = THIS_MODULE,
 381         .open           = rt_acct_proc_open,
 382         .read           = seq_read,
 383         .llseek         = seq_lseek,
 384         .release        = single_release,
 385 };
 386 #endif
 387
 388 static int __net_init ip_rt_do_proc_init(struct net *net)
 389 {
 390         struct proc_dir_entry *pde;
 391
 392         pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
 393                         &rt_cache_seq_fops);
 394         if (!pde)
 395                 goto err1;
 396
 397         pde = proc_create("rt_cache", S_IRUGO,
 398                           net->proc_net_stat, &rt_cpu_seq_fops);
 399         if (!pde)
 400                 goto err2;
 401
 402 #ifdef CONFIG_IP_ROUTE_CLASSID
 403         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
 404         if (!pde)
 405                 goto err3;
 406 #endif
 407         return 0;
 408
 409 #ifdef CONFIG_IP_ROUTE_CLASSID
 410 err3:
 411         remove_proc_entry("rt_cache", net->proc_net_stat);
 412 #endif
 413 err2:
 414         remove_proc_entry("rt_cache", net->proc_net);
 415 err1:
 416         return -ENOMEM;
 417 }
 418
 419 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 420 {
 421         remove_proc_entry("rt_cache", net->proc_net_stat);
 422         remove_proc_entry("rt_cache", net->proc_net);
 423 #ifdef CONFIG_IP_ROUTE_CLASSID
 424         remove_proc_entry("rt_acct", net->proc_net);
 425 #endif
 426 }
 427
 428 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 429         .init = ip_rt_do_proc_init,
 430         .exit = ip_rt_do_proc_exit,
 431 };
 432
 433 static int __init ip_rt_proc_init(void)
 434 {
 435         return register_pernet_subsys(&ip_rt_proc_ops);
 436 }
 437
 438 #else
 439 static inline int ip_rt_proc_init(void)
 440 {
 441         return 0;
 442 }
 443 #endif /* CONFIG_PROC_FS */
 444
 445 static inline bool rt_is_expired(const struct rtable *rth)
 446 {
 447         return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
 448 }
 449
 450 /*
 451  * Perturbation of rt_genid by a small quantity [1..256]
 452  * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
 453  * many times (2^24) without giving recent rt_genid.
 454  * Jenkins hash is strong enough that litle changes of rt_genid are OK.
 455  */
 456 static void rt_cache_invalidate(struct net *net)
 457 {
 458         unsigned char shuffle;
 459
 460         get_random_bytes(&shuffle, sizeof(shuffle));
 461         atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
 462 }
 463
 464 void rt_cache_flush(struct net *net)
 465 {
 466         rt_cache_invalidate(net);
 467 }
 468
 469 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 470                                            struct sk_buff *skb,
 471                                            const void *daddr)
 472 {
 473         struct net_device *dev = dst->dev;
 474         const __be32 *pkey = daddr;
 475         const struct rtable *rt;
 476         struct neighbour *n;
 477
 478         rt = (const struct rtable *) dst;
 479         if (rt->rt_gateway)
 480                 pkey = (const __be32 *) &rt->rt_gateway;
 481         else if (skb)
 482                 pkey = &ip_hdr(skb)->daddr;
 483
 484         n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
 485         if (n)
 486                 return n;
 487         return neigh_create(&arp_tbl, pkey, dev);
 488 }
 489
 490 /*
 491  * Peer allocation may fail only in serious out-of-memory conditions.  However
 492  * we still can generate some output.
 493  * Random ID selection looks a bit dangerous because we have no chances to
 494  * select ID being unique in a reasonable period of time.
 495  * But broken packet identifier may be better than no packet at all.
 496  */
 497 static void ip_select_fb_ident(struct iphdr *iph)
 498 {
 499         static DEFINE_SPINLOCK(ip_fb_id_lock);
 500         static u32 ip_fallback_id;
 501         u32 salt;
 502
 503         spin_lock_bh(&ip_fb_id_lock);
 504         salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
 505         iph->id = htons(salt & 0xFFFF);
 506         ip_fallback_id = salt;
 507         spin_unlock_bh(&ip_fb_id_lock);
 508 }
 509
 510 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
 511 {
 512         struct net *net = dev_net(dst->dev);
 513         struct inet_peer *peer;
 514
 515         peer = inet_getpeer_v4(net->ipv4.peers, iph->daddr, 1);
 516         if (peer) {
 517                 iph->id = htons(inet_getid(peer, more));
 518                 inet_putpeer(peer);
 519                 return;
 520         }
 521
 522         ip_select_fb_ident(iph);
 523 }
 524 EXPORT_SYMBOL(__ip_select_ident);
 525
 526 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 527                              const struct iphdr *iph,
 528                              int oif, u8 tos,
 529                              u8 prot, u32 mark, int flow_flags)
 530 {
 531         if (sk) {
 532                 const struct inet_sock *inet = inet_sk(sk);
 533
 534                 oif = sk->sk_bound_dev_if;
 535                 mark = sk->sk_mark;
 536                 tos = RT_CONN_FLAGS(sk);
 537                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 538         }
 539         flowi4_init_output(fl4, oif, mark, tos,
 540                            RT_SCOPE_UNIVERSE, prot,
 541                            flow_flags,
 542                            iph->daddr, iph->saddr, 0, 0);
 543 }
 544
 545 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 546                                const struct sock *sk)
 547 {
 548         const struct iphdr *iph = ip_hdr(skb);
 549         int oif = skb->dev->ifindex;
 550         u8 tos = RT_TOS(iph->tos);
 551         u8 prot = iph->protocol;
 552         u32 mark = skb->mark;
 553
 554         __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
 555 }
 556
 557 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 558 {
 559         const struct inet_sock *inet = inet_sk(sk);
 560         const struct ip_options_rcu *inet_opt;
 561         __be32 daddr = inet->inet_daddr;
 562
 563         rcu_read_lock();
 564         inet_opt = rcu_dereference(inet->inet_opt);
 565         if (inet_opt && inet_opt->opt.srr)
 566                 daddr = inet_opt->opt.faddr;
 567         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 568                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 569                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 570                            inet_sk_flowi_flags(sk),
 571                            daddr, inet->inet_saddr, 0, 0);
 572         rcu_read_unlock();
 573 }
 574
 575 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 576                                  const struct sk_buff *skb)
 577 {
 578         if (skb)
 579                 build_skb_flow_key(fl4, skb, sk);
 580         else
 581                 build_sk_flow_key(fl4, sk);
 582 }
 583
 584 static inline void rt_free(struct rtable *rt)
 585 {
 586         call_rcu(&rt->dst.rcu_head, dst_rcu_free);
 587 }
 588
 589 static DEFINE_SPINLOCK(fnhe_lock);
 590
 591 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
 592 {
 593         struct fib_nh_exception *fnhe, *oldest;
 594         struct rtable *orig;
 595
 596         oldest = rcu_dereference(hash->chain);
 597         for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
 598              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 599                 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
 600                         oldest = fnhe;
 601         }
 602         orig = rcu_dereference(oldest->fnhe_rth);
 603         if (orig) {
 604                 RCU_INIT_POINTER(oldest->fnhe_rth, NULL);
 605                 rt_free(orig);
 606         }
 607         return oldest;
 608 }
 609
 610 static inline u32 fnhe_hashfun(__be32 daddr)
 611 {
 612         u32 hval;
 613
 614         hval = (__force u32) daddr;
 615         hval ^= (hval >> 11) ^ (hval >> 22);
 616
 617         return hval & (FNHE_HASH_SIZE - 1);
 618 }
 619
 620 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
 621                                   u32 pmtu, unsigned long expires)
 622 {
 623         struct fnhe_hash_bucket *hash;
 624         struct fib_nh_exception *fnhe;
 625         int depth;
 626         u32 hval = fnhe_hashfun(daddr);
 627
 628         spin_lock_bh(&fnhe_lock);
 629
 630         hash = nh->nh_exceptions;
 631         if (!hash) {
 632                 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
 633                 if (!hash)
 634                         goto out_unlock;
 635                 nh->nh_exceptions = hash;
 636         }
 637
 638         hash += hval;
 639
 640         depth = 0;
 641         for (fnhe = rcu_dereference(hash->chain); fnhe;
 642              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 643                 if (fnhe->fnhe_daddr == daddr)
 644                         break;
 645                 depth++;
 646         }
 647
 648         if (fnhe) {
 649                 if (gw)
 650                         fnhe->fnhe_gw = gw;
 651                 if (pmtu) {
 652                         fnhe->fnhe_pmtu = pmtu;
 653                         fnhe->fnhe_expires = expires;
 654                 }
 655         } else {
 656                 if (depth > FNHE_RECLAIM_DEPTH)
 657                         fnhe = fnhe_oldest(hash);
 658                 else {
 659                         fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 660                         if (!fnhe)
 661                                 goto out_unlock;
 662
 663                         fnhe->fnhe_next = hash->chain;
 664                         rcu_assign_pointer(hash->chain, fnhe);
 665                 }
 666                 fnhe->fnhe_daddr = daddr;
 667                 fnhe->fnhe_gw = gw;
 668                 fnhe->fnhe_pmtu = pmtu;
 669                 fnhe->fnhe_expires = expires;
 670         }
 671
 672         fnhe->fnhe_stamp = jiffies;
 673
 674 out_unlock:
 675         spin_unlock_bh(&fnhe_lock);
 676         return;
 677 }
 678
 679 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 680                              bool kill_route)
 681 {
 682         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 683         __be32 old_gw = ip_hdr(skb)->saddr;
 684         struct net_device *dev = skb->dev;
 685         struct in_device *in_dev;
 686         struct fib_result res;
 687         struct neighbour *n;
 688         struct net *net;
 689
 690         switch (icmp_hdr(skb)->code & 7) {
 691         case ICMP_REDIR_NET:
 692         case ICMP_REDIR_NETTOS:
 693         case ICMP_REDIR_HOST:
 694         case ICMP_REDIR_HOSTTOS:
 695                 break;
 696
 697         default:
 698                 return;
 699         }
 700
 701         if (rt->rt_gateway != old_gw)
 702                 return;
 703
 704         in_dev = __in_dev_get_rcu(dev);
 705         if (!in_dev)
 706                 return;
 707
 708         net = dev_net(dev);
 709         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 710             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 711             ipv4_is_zeronet(new_gw))
 712                 goto reject_redirect;
 713
 714         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 715                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 716                         goto reject_redirect;
 717                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 718                         goto reject_redirect;
 719         } else {
 720                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 721                         goto reject_redirect;
 722         }
 723
 724         n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
 725         if (n) {
 726                 if (!(n->nud_state & NUD_VALID)) {
 727                         neigh_event_send(n, NULL);
 728                 } else {
 729                         if (fib_lookup(net, fl4, &res) == 0) {
 730                                 struct fib_nh *nh = &FIB_RES_NH(res);
 731
 732                                 update_or_create_fnhe(nh, fl4->daddr, new_gw,
 733                                                       0, 0);
 734                         }
 735                         if (kill_route)
 736                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 737                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 738                 }
 739                 neigh_release(n);
 740         }
 741         return;
 742
 743 reject_redirect:
 744 #ifdef CONFIG_IP_ROUTE_VERBOSE
 745         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 746                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 747                 __be32 daddr = iph->daddr;
 748                 __be32 saddr = iph->saddr;
 749
 750                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 751                                      "  Advised path = %pI4 -> %pI4\n",
 752                                      &old_gw, dev->name, &new_gw,
 753                                      &saddr, &daddr);
 754         }
 755 #endif
 756         ;
 757 }
 758
 759 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 760 {
 761         struct rtable *rt;
 762         struct flowi4 fl4;
 763
 764         rt = (struct rtable *) dst;
 765
 766         ip_rt_build_flow_key(&fl4, sk, skb);
 767         __ip_do_redirect(rt, skb, &fl4, true);
 768 }
 769
 770 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 771 {
 772         struct rtable *rt = (struct rtable *)dst;
 773         struct dst_entry *ret = dst;
 774
 775         if (rt) {
 776                 if (dst->obsolete > 0) {
 777                         ip_rt_put(rt);
 778                         ret = NULL;
 779                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 780                            rt->dst.expires) {
 781                         ip_rt_put(rt);
 782                         ret = NULL;
 783                 }
 784         }
 785         return ret;
 786 }
 787
 788 /*
 789  * Algorithm:
 790  *      1. The first ip_rt_redirect_number redirects are sent
 791  *         with exponential backoff, then we stop sending them at all,
 792  *         assuming that the host ignores our redirects.
 793  *      2. If we did not see packets requiring redirects
 794  *         during ip_rt_redirect_silence, we assume that the host
 795  *         forgot redirected route and start to send redirects again.
 796  *
 797  * This algorithm is much cheaper and more intelligent than dumb load limiting
 798  * in icmp.c.
 799  *
 800  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 801  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 802  */
 803
 804 void ip_rt_send_redirect(struct sk_buff *skb)
 805 {
 806         struct rtable *rt = skb_rtable(skb);
 807         struct in_device *in_dev;
 808         struct inet_peer *peer;
 809         struct net *net;
 810         int log_martians;
 811
 812         rcu_read_lock();
 813         in_dev = __in_dev_get_rcu(rt->dst.dev);
 814         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 815                 rcu_read_unlock();
 816                 return;
 817         }
 818         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 819         rcu_read_unlock();
 820
 821         net = dev_net(rt->dst.dev);
 822         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
 823         if (!peer) {
 824                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
 825                 return;
 826         }
 827
 828         /* No redirected packets during ip_rt_redirect_silence;
 829          * reset the algorithm.
 830          */
 831         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
 832                 peer->rate_tokens = 0;
 833
 834         /* Too many ignored redirects; do not send anything
 835          * set dst.rate_last to the last seen redirected packet.
 836          */
 837         if (peer->rate_tokens >= ip_rt_redirect_number) {
 838                 peer->rate_last = jiffies;
 839                 goto out_put_peer;
 840         }
 841
 842         /* Check for load limit; set rate_last to the latest sent
 843          * redirect.
 844          */
 845         if (peer->rate_tokens == 0 ||
 846             time_after(jiffies,
 847                        (peer->rate_last +
 848                         (ip_rt_redirect_load << peer->rate_tokens)))) {
 849                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
 850                 peer->rate_last = jiffies;
 851                 ++peer->rate_tokens;
 852 #ifdef CONFIG_IP_ROUTE_VERBOSE
 853                 if (log_martians &&
 854                     peer->rate_tokens == ip_rt_redirect_number)
 855                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 856                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 857                                              &ip_hdr(skb)->daddr, &rt->rt_gateway);
 858 #endif
 859         }
 860 out_put_peer:
 861         inet_putpeer(peer);
 862 }
 863
 864 static int ip_error(struct sk_buff *skb)
 865 {
 866         struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
 867         struct rtable *rt = skb_rtable(skb);
 868         struct inet_peer *peer;
 869         unsigned long now;
 870         struct net *net;
 871         bool send;
 872         int code;
 873
 874         net = dev_net(rt->dst.dev);
 875         if (!IN_DEV_FORWARD(in_dev)) {
 876                 switch (rt->dst.error) {
 877                 case EHOSTUNREACH:
 878                         IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
 879                         break;
 880
 881                 case ENETUNREACH:
 882                         IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
 883                         break;
 884                 }
 885                 goto out;
 886         }
 887
 888         switch (rt->dst.error) {
 889         case EINVAL:
 890         default:
 891                 goto out;
 892         case EHOSTUNREACH:
 893                 code = ICMP_HOST_UNREACH;
 894                 break;
 895         case ENETUNREACH:
 896                 code = ICMP_NET_UNREACH;
 897                 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
 898                 break;
 899         case EACCES:
 900                 code = ICMP_PKT_FILTERED;
 901                 break;
 902         }
 903
 904         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
 905
 906         send = true;
 907         if (peer) {
 908                 now = jiffies;
 909                 peer->rate_tokens += now - peer->rate_last;
 910                 if (peer->rate_tokens > ip_rt_error_burst)
 911                         peer->rate_tokens = ip_rt_error_burst;
 912                 peer->rate_last = now;
 913                 if (peer->rate_tokens >= ip_rt_error_cost)
 914                         peer->rate_tokens -= ip_rt_error_cost;
 915                 else
 916                         send = false;
 917                 inet_putpeer(peer);
 918         }
 919         if (send)
 920                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
 921
 922 out:    kfree_skb(skb);
 923         return 0;
 924 }
 925
 926 static u32 __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
 927 {
 928         struct fib_result res;
 929
 930         if (mtu < ip_rt_min_pmtu)
 931                 mtu = ip_rt_min_pmtu;
 932
 933         rcu_read_lock();
 934         if (fib_lookup(dev_net(rt->dst.dev), fl4, &res) == 0) {
 935                 struct fib_nh *nh = &FIB_RES_NH(res);
 936
 937                 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
 938                                       jiffies + ip_rt_mtu_expires);
 939         }
 940         rcu_read_unlock();
 941         return mtu;
 942 }
 943
 944 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 945                               struct sk_buff *skb, u32 mtu)
 946 {
 947         struct rtable *rt = (struct rtable *) dst;
 948         struct flowi4 fl4;
 949
 950         ip_rt_build_flow_key(&fl4, sk, skb);
 951         mtu = __ip_rt_update_pmtu(rt, &fl4, mtu);
 952
 953         if (!rt->rt_pmtu) {
 954                 dst->obsolete = DST_OBSOLETE_KILL;
 955         } else {
 956                 rt->rt_pmtu = mtu;
 957                 rt->dst.expires = max(1UL, jiffies + ip_rt_mtu_expires);
 958         }
 959 }
 960
 961 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
 962                       int oif, u32 mark, u8 protocol, int flow_flags)
 963 {
 964         const struct iphdr *iph = (const struct iphdr *) skb->data;
 965         struct flowi4 fl4;
 966         struct rtable *rt;
 967
 968         __build_flow_key(&fl4, NULL, iph, oif,
 969                          RT_TOS(iph->tos), protocol, mark, flow_flags);
 970         rt = __ip_route_output_key(net, &fl4);
 971         if (!IS_ERR(rt)) {
 972                 __ip_rt_update_pmtu(rt, &fl4, mtu);
 973                 ip_rt_put(rt);
 974         }
 975 }
 976 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
 977
 978 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
 979 {
 980         const struct iphdr *iph = (const struct iphdr *) skb->data;
 981         struct flowi4 fl4;
 982         struct rtable *rt;
 983
 984         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
 985         rt = __ip_route_output_key(sock_net(sk), &fl4);
 986         if (!IS_ERR(rt)) {
 987                 __ip_rt_update_pmtu(rt, &fl4, mtu);
 988                 ip_rt_put(rt);
 989         }
 990 }
 991 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
 992
 993 void ipv4_redirect(struct sk_buff *skb, struct net *net,
 994                    int oif, u32 mark, u8 protocol, int flow_flags)
 995 {
 996         const struct iphdr *iph = (const struct iphdr *) skb->data;
 997         struct flowi4 fl4;
 998         struct rtable *rt;
 999
1000         __build_flow_key(&fl4, NULL, iph, oif,
1001                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1002         rt = __ip_route_output_key(net, &fl4);
1003         if (!IS_ERR(rt)) {
1004                 __ip_do_redirect(rt, skb, &fl4, false);
1005                 ip_rt_put(rt);
1006         }
1007 }
1008 EXPORT_SYMBOL_GPL(ipv4_redirect);
1009
1010 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1011 {
1012         const struct iphdr *iph = (const struct iphdr *) skb->data;
1013         struct flowi4 fl4;
1014         struct rtable *rt;
1015
1016         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1017         rt = __ip_route_output_key(sock_net(sk), &fl4);
1018         if (!IS_ERR(rt)) {
1019                 __ip_do_redirect(rt, skb, &fl4, false);
1020                 ip_rt_put(rt);
1021         }
1022 }
1023 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1024
1025 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1026 {
1027         struct rtable *rt = (struct rtable *) dst;
1028
1029         /* All IPV4 dsts are created with ->obsolete set to the value
1030          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1031          * into this function always.
1032          *
1033          * When a PMTU/redirect information update invalidates a
1034          * route, this is indicated by setting obsolete to
1035          * DST_OBSOLETE_KILL.
1036          */
1037         if (dst->obsolete == DST_OBSOLETE_KILL || rt_is_expired(rt))
1038                 return NULL;
1039         return dst;
1040 }
1041
1042 static void ipv4_link_failure(struct sk_buff *skb)
1043 {
1044         struct rtable *rt;
1045
1046         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1047
1048         rt = skb_rtable(skb);
1049         if (rt)
1050                 dst_set_expires(&rt->dst, 0);
1051 }
1052
1053 static int ip_rt_bug(struct sk_buff *skb)
1054 {
1055         pr_debug("%s: %pI4 -> %pI4, %s\n",
1056                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1057                  skb->dev ? skb->dev->name : "?");
1058         kfree_skb(skb);
1059         WARN_ON(1);
1060         return 0;
1061 }
1062
1063 /*
1064    We do not cache source address of outgoing interface,
1065    because it is used only by IP RR, TS and SRR options,
1066    so that it out of fast path.
1067
1068    BTW remember: "addr" is allowed to be not aligned
1069    in IP options!
1070  */
1071
1072 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1073 {
1074         __be32 src;
1075
1076         if (rt_is_output_route(rt))
1077                 src = ip_hdr(skb)->saddr;
1078         else {
1079                 struct fib_result res;
1080                 struct flowi4 fl4;
1081                 struct iphdr *iph;
1082
1083                 iph = ip_hdr(skb);
1084
1085                 memset(&fl4, 0, sizeof(fl4));
1086                 fl4.daddr = iph->daddr;
1087                 fl4.saddr = iph->saddr;
1088                 fl4.flowi4_tos = RT_TOS(iph->tos);
1089                 fl4.flowi4_oif = rt->dst.dev->ifindex;
1090                 fl4.flowi4_iif = skb->dev->ifindex;
1091                 fl4.flowi4_mark = skb->mark;
1092
1093                 rcu_read_lock();
1094                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1095                         src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1096                 else
1097                         src = inet_select_addr(rt->dst.dev,
1098                                                rt_nexthop(rt, iph->daddr),
1099                                                RT_SCOPE_UNIVERSE);
1100                 rcu_read_unlock();
1101         }
1102         memcpy(addr, &src, 4);
1103 }
1104
1105 #ifdef CONFIG_IP_ROUTE_CLASSID
1106 static void set_class_tag(struct rtable *rt, u32 tag)
1107 {
1108         if (!(rt->dst.tclassid & 0xFFFF))
1109                 rt->dst.tclassid |= tag & 0xFFFF;
1110         if (!(rt->dst.tclassid & 0xFFFF0000))
1111                 rt->dst.tclassid |= tag & 0xFFFF0000;
1112 }
1113 #endif
1114
1115 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1116 {
1117         unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1118
1119         if (advmss == 0) {
1120                 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1121                                ip_rt_min_advmss);
1122                 if (advmss > 65535 - 40)
1123                         advmss = 65535 - 40;
1124         }
1125         return advmss;
1126 }
1127
1128 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1129 {
1130         const struct rtable *rt = (const struct rtable *) dst;
1131         unsigned int mtu = rt->rt_pmtu;
1132
1133         if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1134                 mtu = dst_metric_raw(dst, RTAX_MTU);
1135
1136         if (mtu && rt_is_output_route(rt))
1137                 return mtu;
1138
1139         mtu = dst->dev->mtu;
1140
1141         if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1142                 if (rt->rt_gateway && mtu > 576)
1143                         mtu = 576;
1144         }
1145
1146         if (mtu > IP_MAX_MTU)
1147                 mtu = IP_MAX_MTU;
1148
1149         return mtu;
1150 }
1151
1152 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1153 {
1154         struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1155         struct fib_nh_exception *fnhe;
1156         u32 hval;
1157
1158         if (!hash)
1159                 return NULL;
1160
1161         hval = fnhe_hashfun(daddr);
1162
1163         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1164              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1165                 if (fnhe->fnhe_daddr == daddr)
1166                         return fnhe;
1167         }
1168         return NULL;
1169 }
1170
1171 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1172                               __be32 daddr)
1173 {
1174         bool ret = false;
1175
1176         spin_lock_bh(&fnhe_lock);
1177
1178         if (daddr == fnhe->fnhe_daddr) {
1179                 struct rtable *orig;
1180
1181                 if (fnhe->fnhe_pmtu) {
1182                         unsigned long expires = fnhe->fnhe_expires;
1183                         unsigned long diff = expires - jiffies;
1184
1185                         if (time_before(jiffies, expires)) {
1186                                 rt->rt_pmtu = fnhe->fnhe_pmtu;
1187                                 dst_set_expires(&rt->dst, diff);
1188                         }
1189                 }
1190                 if (fnhe->fnhe_gw) {
1191                         rt->rt_flags |= RTCF_REDIRECTED;
1192                         rt->rt_gateway = fnhe->fnhe_gw;
1193                 }
1194
1195                 orig = rcu_dereference(fnhe->fnhe_rth);
1196                 rcu_assign_pointer(fnhe->fnhe_rth, rt);
1197                 if (orig)
1198                         rt_free(orig);
1199
1200                 fnhe->fnhe_stamp = jiffies;
1201                 ret = true;
1202         } else {
1203                 /* Routes we intend to cache in nexthop exception have
1204                  * the DST_NOCACHE bit clear.  However, if we are
1205                  * unsuccessful at storing this route into the cache
1206                  * we really need to set it.
1207                  */
1208                 rt->dst.flags |= DST_NOCACHE;
1209         }
1210         spin_unlock_bh(&fnhe_lock);
1211
1212         return ret;
1213 }
1214
1215 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1216 {
1217         struct rtable *orig, *prev, **p;
1218         bool ret = true;
1219
1220         if (rt_is_input_route(rt)) {
1221                 p = (struct rtable **)&nh->nh_rth_input;
1222         } else {
1223                 if (!nh->nh_pcpu_rth_output)
1224                         goto nocache;
1225                 p = (struct rtable **)__this_cpu_ptr(nh->nh_pcpu_rth_output);
1226         }
1227         orig = *p;
1228
1229         prev = cmpxchg(p, orig, rt);
1230         if (prev == orig) {
1231                 if (orig)
1232                         rt_free(orig);
1233         } else {
1234                 /* Routes we intend to cache in the FIB nexthop have
1235                  * the DST_NOCACHE bit clear.  However, if we are
1236                  * unsuccessful at storing this route into the cache
1237                  * we really need to set it.
1238                  */
1239 nocache:
1240                 rt->dst.flags |= DST_NOCACHE;
1241                 ret = false;
1242         }
1243
1244         return ret;
1245 }
1246
1247 static DEFINE_SPINLOCK(rt_uncached_lock);
1248 static LIST_HEAD(rt_uncached_list);
1249
1250 static void rt_add_uncached_list(struct rtable *rt)
1251 {
1252         spin_lock_bh(&rt_uncached_lock);
1253         list_add_tail(&rt->rt_uncached, &rt_uncached_list);
1254         spin_unlock_bh(&rt_uncached_lock);
1255 }
1256
1257 static void ipv4_dst_destroy(struct dst_entry *dst)
1258 {
1259         struct rtable *rt = (struct rtable *) dst;
1260
1261         if (!list_empty(&rt->rt_uncached)) {
1262                 spin_lock_bh(&rt_uncached_lock);
1263                 list_del(&rt->rt_uncached);
1264                 spin_unlock_bh(&rt_uncached_lock);
1265         }
1266 }
1267
1268 void rt_flush_dev(struct net_device *dev)
1269 {
1270         if (!list_empty(&rt_uncached_list)) {
1271                 struct net *net = dev_net(dev);
1272                 struct rtable *rt;
1273
1274                 spin_lock_bh(&rt_uncached_lock);
1275                 list_for_each_entry(rt, &rt_uncached_list, rt_uncached) {
1276                         if (rt->dst.dev != dev)
1277                                 continue;
1278                         rt->dst.dev = net->loopback_dev;
1279                         dev_hold(rt->dst.dev);
1280                         dev_put(dev);
1281                 }
1282                 spin_unlock_bh(&rt_uncached_lock);
1283         }
1284 }
1285
1286 static bool rt_cache_valid(const struct rtable *rt)
1287 {
1288         return  rt &&
1289                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1290                 !rt_is_expired(rt);
1291 }
1292
1293 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1294                            const struct fib_result *res,
1295                            struct fib_nh_exception *fnhe,
1296                            struct fib_info *fi, u16 type, u32 itag)
1297 {
1298         bool cached = false;
1299
1300         if (fi) {
1301                 struct fib_nh *nh = &FIB_RES_NH(*res);
1302
1303                 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK)
1304                         rt->rt_gateway = nh->nh_gw;
1305                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1306 #ifdef CONFIG_IP_ROUTE_CLASSID
1307                 rt->dst.tclassid = nh->nh_tclassid;
1308 #endif
1309                 if (unlikely(fnhe))
1310                         cached = rt_bind_exception(rt, fnhe, daddr);
1311                 else if (!(rt->dst.flags & DST_NOCACHE))
1312                         cached = rt_cache_route(nh, rt);
1313         }
1314         if (unlikely(!cached))
1315                 rt_add_uncached_list(rt);
1316
1317 #ifdef CONFIG_IP_ROUTE_CLASSID
1318 #ifdef CONFIG_IP_MULTIPLE_TABLES
1319         set_class_tag(rt, res->tclassid);
1320 #endif
1321         set_class_tag(rt, itag);
1322 #endif
1323 }
1324
1325 static struct rtable *rt_dst_alloc(struct net_device *dev,
1326                                    bool nopolicy, bool noxfrm, bool will_cache)
1327 {
1328         return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1329                          (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1330                          (nopolicy ? DST_NOPOLICY : 0) |
1331                          (noxfrm ? DST_NOXFRM : 0));
1332 }
1333
1334 /* called in rcu_read_lock() section */
1335 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1336                                 u8 tos, struct net_device *dev, int our)
1337 {
1338         struct rtable *rth;
1339         struct in_device *in_dev = __in_dev_get_rcu(dev);
1340         u32 itag = 0;
1341         int err;
1342
1343         /* Primary sanity checks. */
1344
1345         if (in_dev == NULL)
1346                 return -EINVAL;
1347
1348         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1349             skb->protocol != htons(ETH_P_IP))
1350                 goto e_inval;
1351
1352         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1353                 if (ipv4_is_loopback(saddr))
1354                         goto e_inval;
1355
1356         if (ipv4_is_zeronet(saddr)) {
1357                 if (!ipv4_is_local_multicast(daddr))
1358                         goto e_inval;
1359         } else {
1360                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1361                                           in_dev, &itag);
1362                 if (err < 0)
1363                         goto e_err;
1364         }
1365         rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1366                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1367         if (!rth)
1368                 goto e_nobufs;
1369
1370 #ifdef CONFIG_IP_ROUTE_CLASSID
1371         rth->dst.tclassid = itag;
1372 #endif
1373         rth->dst.output = ip_rt_bug;
1374
1375         rth->rt_genid   = rt_genid(dev_net(dev));
1376         rth->rt_flags   = RTCF_MULTICAST;
1377         rth->rt_type    = RTN_MULTICAST;
1378         rth->rt_is_input= 1;
1379         rth->rt_iif     = 0;
1380         rth->rt_pmtu    = 0;
1381         rth->rt_gateway = 0;
1382         INIT_LIST_HEAD(&rth->rt_uncached);
1383         if (our) {
1384                 rth->dst.input= ip_local_deliver;
1385                 rth->rt_flags |= RTCF_LOCAL;
1386         }
1387
1388 #ifdef CONFIG_IP_MROUTE
1389         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1390                 rth->dst.input = ip_mr_input;
1391 #endif
1392         RT_CACHE_STAT_INC(in_slow_mc);
1393
1394         skb_dst_set(skb, &rth->dst);
1395         return 0;
1396
1397 e_nobufs:
1398         return -ENOBUFS;
1399 e_inval:
1400         return -EINVAL;
1401 e_err:
1402         return err;
1403 }
1404
1405
1406 static void ip_handle_martian_source(struct net_device *dev,
1407                                      struct in_device *in_dev,
1408                                      struct sk_buff *skb,
1409                                      __be32 daddr,
1410                                      __be32 saddr)
1411 {
1412         RT_CACHE_STAT_INC(in_martian_src);
1413 #ifdef CONFIG_IP_ROUTE_VERBOSE
1414         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1415                 /*
1416                  *      RFC1812 recommendation, if source is martian,
1417                  *      the only hint is MAC header.
1418                  */
1419                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1420                         &daddr, &saddr, dev->name);
1421                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1422                         print_hex_dump(KERN_WARNING, "ll header: ",
1423                                        DUMP_PREFIX_OFFSET, 16, 1,
1424                                        skb_mac_header(skb),
1425                                        dev->hard_header_len, true);
1426                 }
1427         }
1428 #endif
1429 }
1430
1431 /* called in rcu_read_lock() section */
1432 static int __mkroute_input(struct sk_buff *skb,
1433                            const struct fib_result *res,
1434                            struct in_device *in_dev,
1435                            __be32 daddr, __be32 saddr, u32 tos)
1436 {
1437         struct rtable *rth;
1438         int err;
1439         struct in_device *out_dev;
1440         unsigned int flags = 0;
1441         bool do_cache;
1442         u32 itag;
1443
1444         /* get a working reference to the output device */
1445         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1446         if (out_dev == NULL) {
1447                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1448                 return -EINVAL;
1449         }
1450
1451
1452         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1453                                   in_dev->dev, in_dev, &itag);
1454         if (err < 0) {
1455                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1456                                          saddr);
1457
1458                 goto cleanup;
1459         }
1460
1461         if (out_dev == in_dev && err &&
1462             (IN_DEV_SHARED_MEDIA(out_dev) ||
1463              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1464                 flags |= RTCF_DOREDIRECT;
1465
1466         if (skb->protocol != htons(ETH_P_IP)) {
1467                 /* Not IP (i.e. ARP). Do not create route, if it is
1468                  * invalid for proxy arp. DNAT routes are always valid.
1469                  *
1470                  * Proxy arp feature have been extended to allow, ARP
1471                  * replies back to the same interface, to support
1472                  * Private VLAN switch technologies. See arp.c.
1473                  */
1474                 if (out_dev == in_dev &&
1475                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1476                         err = -EINVAL;
1477                         goto cleanup;
1478                 }
1479         }
1480
1481         do_cache = false;
1482         if (res->fi) {
1483                 if (!itag) {
1484                         rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1485                         if (rt_cache_valid(rth)) {
1486                                 skb_dst_set_noref(skb, &rth->dst);
1487                                 goto out;
1488                         }
1489                         do_cache = true;
1490                 }
1491         }
1492
1493         rth = rt_dst_alloc(out_dev->dev,
1494                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1495                            IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1496         if (!rth) {
1497                 err = -ENOBUFS;
1498                 goto cleanup;
1499         }
1500
1501         rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
1502         rth->rt_flags = flags;
1503         rth->rt_type = res->type;
1504         rth->rt_is_input = 1;
1505         rth->rt_iif     = 0;
1506         rth->rt_pmtu    = 0;
1507         rth->rt_gateway = 0;
1508         INIT_LIST_HEAD(&rth->rt_uncached);
1509
1510         rth->dst.input = ip_forward;
1511         rth->dst.output = ip_output;
1512
1513         rt_set_nexthop(rth, daddr, res, NULL, res->fi, res->type, itag);
1514         skb_dst_set(skb, &rth->dst);
1515 out:
1516         err = 0;
1517  cleanup:
1518         return err;
1519 }
1520
1521 static int ip_mkroute_input(struct sk_buff *skb,
1522                             struct fib_result *res,
1523                             const struct flowi4 *fl4,
1524                             struct in_device *in_dev,
1525                             __be32 daddr, __be32 saddr, u32 tos)
1526 {
1527 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1528         if (res->fi && res->fi->fib_nhs > 1)
1529                 fib_select_multipath(res);
1530 #endif
1531
1532         /* create a routing cache entry */
1533         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1534 }
1535
1536 /*
1537  *      NOTE. We drop all the packets that has local source
1538  *      addresses, because every properly looped back packet
1539  *      must have correct destination already attached by output routine.
1540  *
1541  *      Such approach solves two big problems:
1542  *      1. Not simplex devices are handled properly.
1543  *      2. IP spoofing attempts are filtered with 100% of guarantee.
1544  *      called with rcu_read_lock()
1545  */
1546
1547 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1548                                u8 tos, struct net_device *dev)
1549 {
1550         struct fib_result res;
1551         struct in_device *in_dev = __in_dev_get_rcu(dev);
1552         struct flowi4   fl4;
1553         unsigned int    flags = 0;
1554         u32             itag = 0;
1555         struct rtable   *rth;
1556         int             err = -EINVAL;
1557         struct net    *net = dev_net(dev);
1558         bool do_cache;
1559
1560         /* IP on this device is disabled. */
1561
1562         if (!in_dev)
1563                 goto out;
1564
1565         /* Check for the most weird martians, which can be not detected
1566            by fib_lookup.
1567          */
1568
1569         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1570                 goto martian_source;
1571
1572         res.fi = NULL;
1573         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1574                 goto brd_input;
1575
1576         /* Accept zero addresses only to limited broadcast;
1577          * I even do not know to fix it or not. Waiting for complains :-)
1578          */
1579         if (ipv4_is_zeronet(saddr))
1580                 goto martian_source;
1581
1582         if (ipv4_is_zeronet(daddr))
1583                 goto martian_destination;
1584
1585         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1586          * and call it once if daddr or/and saddr are loopback addresses
1587          */
1588         if (ipv4_is_loopback(daddr)) {
1589                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1590                         goto martian_destination;
1591         } else if (ipv4_is_loopback(saddr)) {
1592                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1593                         goto martian_source;
1594         }
1595
1596         /*
1597          *      Now we are ready to route packet.
1598          */
1599         fl4.flowi4_oif = 0;
1600         fl4.flowi4_iif = dev->ifindex;
1601         fl4.flowi4_mark = skb->mark;
1602         fl4.flowi4_tos = tos;
1603         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1604         fl4.daddr = daddr;
1605         fl4.saddr = saddr;
1606         err = fib_lookup(net, &fl4, &res);
1607         if (err != 0)
1608                 goto no_route;
1609
1610         RT_CACHE_STAT_INC(in_slow_tot);
1611
1612         if (res.type == RTN_BROADCAST)
1613                 goto brd_input;
1614
1615         if (res.type == RTN_LOCAL) {
1616                 err = fib_validate_source(skb, saddr, daddr, tos,
1617                                           LOOPBACK_IFINDEX,
1618                                           dev, in_dev, &itag);
1619                 if (err < 0)
1620                         goto martian_source_keep_err;
1621                 goto local_input;
1622         }
1623
1624         if (!IN_DEV_FORWARD(in_dev))
1625                 goto no_route;
1626         if (res.type != RTN_UNICAST)
1627                 goto martian_destination;
1628
1629         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1630 out:    return err;
1631
1632 brd_input:
1633         if (skb->protocol != htons(ETH_P_IP))
1634                 goto e_inval;
1635
1636         if (!ipv4_is_zeronet(saddr)) {
1637                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1638                                           in_dev, &itag);
1639                 if (err < 0)
1640                         goto martian_source_keep_err;
1641         }
1642         flags |= RTCF_BROADCAST;
1643         res.type = RTN_BROADCAST;
1644         RT_CACHE_STAT_INC(in_brd);
1645
1646 local_input:
1647         do_cache = false;
1648         if (res.fi) {
1649                 if (!itag) {
1650                         rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1651                         if (rt_cache_valid(rth)) {
1652                                 skb_dst_set_noref(skb, &rth->dst);
1653                                 err = 0;
1654                                 goto out;
1655                         }
1656                         do_cache = true;
1657                 }
1658         }
1659
1660         rth = rt_dst_alloc(net->loopback_dev,
1661                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1662         if (!rth)
1663                 goto e_nobufs;
1664
1665         rth->dst.input= ip_local_deliver;
1666         rth->dst.output= ip_rt_bug;
1667 #ifdef CONFIG_IP_ROUTE_CLASSID
1668         rth->dst.tclassid = itag;
1669 #endif
1670
1671         rth->rt_genid = rt_genid(net);
1672         rth->rt_flags   = flags|RTCF_LOCAL;
1673         rth->rt_type    = res.type;
1674         rth->rt_is_input = 1;
1675         rth->rt_iif     = 0;
1676         rth->rt_pmtu    = 0;
1677         rth->rt_gateway = 0;
1678         INIT_LIST_HEAD(&rth->rt_uncached);
1679         if (res.type == RTN_UNREACHABLE) {
1680                 rth->dst.input= ip_error;
1681                 rth->dst.error= -err;
1682                 rth->rt_flags   &= ~RTCF_LOCAL;
1683         }
1684         if (do_cache)
1685                 rt_cache_route(&FIB_RES_NH(res), rth);
1686         skb_dst_set(skb, &rth->dst);
1687         err = 0;
1688         goto out;
1689
1690 no_route:
1691         RT_CACHE_STAT_INC(in_no_route);
1692         res.type = RTN_UNREACHABLE;
1693         if (err == -ESRCH)
1694                 err = -ENETUNREACH;
1695         goto local_input;
1696
1697         /*
1698          *      Do not cache martian addresses: they should be logged (RFC1812)
1699          */
1700 martian_destination:
1701         RT_CACHE_STAT_INC(in_martian_dst);
1702 #ifdef CONFIG_IP_ROUTE_VERBOSE
1703         if (IN_DEV_LOG_MARTIANS(in_dev))
1704                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1705                                      &daddr, &saddr, dev->name);
1706 #endif
1707
1708 e_inval:
1709         err = -EINVAL;
1710         goto out;
1711
1712 e_nobufs:
1713         err = -ENOBUFS;
1714         goto out;
1715
1716 martian_source:
1717         err = -EINVAL;
1718 martian_source_keep_err:
1719         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1720         goto out;
1721 }
1722
1723 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1724                          u8 tos, struct net_device *dev)
1725 {
1726         int res;
1727
1728         rcu_read_lock();
1729
1730         /* Multicast recognition logic is moved from route cache to here.
1731            The problem was that too many Ethernet cards have broken/missing
1732            hardware multicast filters :-( As result the host on multicasting
1733            network acquires a lot of useless route cache entries, sort of
1734            SDR messages from all the world. Now we try to get rid of them.
1735            Really, provided software IP multicast filter is organized
1736            reasonably (at least, hashed), it does not result in a slowdown
1737            comparing with route cache reject entries.
1738            Note, that multicast routers are not affected, because
1739            route cache entry is created eventually.
1740          */
1741         if (ipv4_is_multicast(daddr)) {
1742                 struct in_device *in_dev = __in_dev_get_rcu(dev);
1743
1744                 if (in_dev) {
1745                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1746                                                   ip_hdr(skb)->protocol);
1747                         if (our
1748 #ifdef CONFIG_IP_MROUTE
1749                                 ||
1750                             (!ipv4_is_local_multicast(daddr) &&
1751                              IN_DEV_MFORWARD(in_dev))
1752 #endif
1753                            ) {
1754                                 int res = ip_route_input_mc(skb, daddr, saddr,
1755                                                             tos, dev, our);
1756                                 rcu_read_unlock();
1757                                 return res;
1758                         }
1759                 }
1760                 rcu_read_unlock();
1761                 return -EINVAL;
1762         }
1763         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1764         rcu_read_unlock();
1765         return res;
1766 }
1767 EXPORT_SYMBOL(ip_route_input_noref);
1768
1769 /* called with rcu_read_lock() */
1770 static struct rtable *__mkroute_output(const struct fib_result *res,
1771                                        const struct flowi4 *fl4, int orig_oif,
1772                                        struct net_device *dev_out,
1773                                        unsigned int flags)
1774 {
1775         struct fib_info *fi = res->fi;
1776         struct fib_nh_exception *fnhe;
1777         struct in_device *in_dev;
1778         u16 type = res->type;
1779         struct rtable *rth;
1780
1781         in_dev = __in_dev_get_rcu(dev_out);
1782         if (!in_dev)
1783                 return ERR_PTR(-EINVAL);
1784
1785         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1786                 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1787                         return ERR_PTR(-EINVAL);
1788
1789         if (ipv4_is_lbcast(fl4->daddr))
1790                 type = RTN_BROADCAST;
1791         else if (ipv4_is_multicast(fl4->daddr))
1792                 type = RTN_MULTICAST;
1793         else if (ipv4_is_zeronet(fl4->daddr))
1794                 return ERR_PTR(-EINVAL);
1795
1796         if (dev_out->flags & IFF_LOOPBACK)
1797                 flags |= RTCF_LOCAL;
1798
1799         if (type == RTN_BROADCAST) {
1800                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1801                 fi = NULL;
1802         } else if (type == RTN_MULTICAST) {
1803                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1804                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1805                                      fl4->flowi4_proto))
1806                         flags &= ~RTCF_LOCAL;
1807                 /* If multicast route do not exist use
1808                  * default one, but do not gateway in this case.
1809                  * Yes, it is hack.
1810                  */
1811                 if (fi && res->prefixlen < 4)
1812                         fi = NULL;
1813         }
1814
1815         fnhe = NULL;
1816         if (fi) {
1817                 struct rtable __rcu **prth;
1818
1819                 fnhe = find_exception(&FIB_RES_NH(*res), fl4->daddr);
1820                 if (fnhe)
1821                         prth = &fnhe->fnhe_rth;
1822                 else
1823                         prth = __this_cpu_ptr(FIB_RES_NH(*res).nh_pcpu_rth_output);
1824                 rth = rcu_dereference(*prth);
1825                 if (rt_cache_valid(rth)) {
1826                         dst_hold(&rth->dst);
1827                         return rth;
1828                 }
1829         }
1830         rth = rt_dst_alloc(dev_out,
1831                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1832                            IN_DEV_CONF_GET(in_dev, NOXFRM),
1833                            fi);
1834         if (!rth)
1835                 return ERR_PTR(-ENOBUFS);
1836
1837         rth->dst.output = ip_output;
1838
1839         rth->rt_genid = rt_genid(dev_net(dev_out));
1840         rth->rt_flags   = flags;
1841         rth->rt_type    = type;
1842         rth->rt_is_input = 0;
1843         rth->rt_iif     = orig_oif ? : 0;
1844         rth->rt_pmtu    = 0;
1845         rth->rt_gateway = 0;
1846         INIT_LIST_HEAD(&rth->rt_uncached);
1847
1848         RT_CACHE_STAT_INC(out_slow_tot);
1849
1850         if (flags & RTCF_LOCAL)
1851                 rth->dst.input = ip_local_deliver;
1852         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1853                 if (flags & RTCF_LOCAL &&
1854                     !(dev_out->flags & IFF_LOOPBACK)) {
1855                         rth->dst.output = ip_mc_output;
1856                         RT_CACHE_STAT_INC(out_slow_mc);
1857                 }
1858 #ifdef CONFIG_IP_MROUTE
1859                 if (type == RTN_MULTICAST) {
1860                         if (IN_DEV_MFORWARD(in_dev) &&
1861                             !ipv4_is_local_multicast(fl4->daddr)) {
1862                                 rth->dst.input = ip_mr_input;
1863                                 rth->dst.output = ip_mc_output;
1864                         }
1865                 }
1866 #endif
1867         }
1868
1869         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
1870
1871         return rth;
1872 }
1873
1874 /*
1875  * Major route resolver routine.
1876  */
1877
1878 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
1879 {
1880         struct net_device *dev_out = NULL;
1881         __u8 tos = RT_FL_TOS(fl4);
1882         unsigned int flags = 0;
1883         struct fib_result res;
1884         struct rtable *rth;
1885         int orig_oif;
1886
1887         res.tclassid    = 0;
1888         res.fi          = NULL;
1889         res.table       = NULL;
1890
1891         orig_oif = fl4->flowi4_oif;
1892
1893         fl4->flowi4_iif = LOOPBACK_IFINDEX;
1894         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
1895         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
1896                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
1897
1898         rcu_read_lock();
1899         if (fl4->saddr) {
1900                 rth = ERR_PTR(-EINVAL);
1901                 if (ipv4_is_multicast(fl4->saddr) ||
1902                     ipv4_is_lbcast(fl4->saddr) ||
1903                     ipv4_is_zeronet(fl4->saddr))
1904                         goto out;
1905
1906                 /* I removed check for oif == dev_out->oif here.
1907                    It was wrong for two reasons:
1908                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
1909                       is assigned to multiple interfaces.
1910                    2. Moreover, we are allowed to send packets with saddr
1911                       of another iface. --ANK
1912                  */
1913
1914                 if (fl4->flowi4_oif == 0 &&
1915                     (ipv4_is_multicast(fl4->daddr) ||
1916                      ipv4_is_lbcast(fl4->daddr))) {
1917                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1918                         dev_out = __ip_dev_find(net, fl4->saddr, false);
1919                         if (dev_out == NULL)
1920                                 goto out;
1921
1922                         /* Special hack: user can direct multicasts
1923                            and limited broadcast via necessary interface
1924                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
1925                            This hack is not just for fun, it allows
1926                            vic,vat and friends to work.
1927                            They bind socket to loopback, set ttl to zero
1928                            and expect that it will work.
1929                            From the viewpoint of routing cache they are broken,
1930                            because we are not allowed to build multicast path
1931                            with loopback source addr (look, routing cache
1932                            cannot know, that ttl is zero, so that packet
1933                            will not leave this host and route is valid).
1934                            Luckily, this hack is good workaround.
1935                          */
1936
1937                         fl4->flowi4_oif = dev_out->ifindex;
1938                         goto make_route;
1939                 }
1940
1941                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
1942                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1943                         if (!__ip_dev_find(net, fl4->saddr, false))
1944                                 goto out;
1945                 }
1946         }
1947
1948
1949         if (fl4->flowi4_oif) {
1950                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
1951                 rth = ERR_PTR(-ENODEV);
1952                 if (dev_out == NULL)
1953                         goto out;
1954
1955                 /* RACE: Check return value of inet_select_addr instead. */
1956                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
1957                         rth = ERR_PTR(-ENETUNREACH);
1958                         goto out;
1959                 }
1960                 if (ipv4_is_local_multicast(fl4->daddr) ||
1961                     ipv4_is_lbcast(fl4->daddr)) {
1962                         if (!fl4->saddr)
1963                                 fl4->saddr = inet_select_addr(dev_out, 0,
1964                                                               RT_SCOPE_LINK);
1965                         goto make_route;
1966                 }
1967                 if (fl4->saddr) {
1968                         if (ipv4_is_multicast(fl4->daddr))
1969                                 fl4->saddr = inet_select_addr(dev_out, 0,
1970                                                               fl4->flowi4_scope);
1971                         else if (!fl4->daddr)
1972                                 fl4->saddr = inet_select_addr(dev_out, 0,
1973                                                               RT_SCOPE_HOST);
1974                 }
1975         }
1976
1977         if (!fl4->daddr) {
1978                 fl4->daddr = fl4->saddr;
1979                 if (!fl4->daddr)
1980                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
1981                 dev_out = net->loopback_dev;
1982                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
1983                 res.type = RTN_LOCAL;
1984                 flags |= RTCF_LOCAL;
1985                 goto make_route;
1986         }
1987
1988         if (fib_lookup(net, fl4, &res)) {
1989                 res.fi = NULL;
1990                 res.table = NULL;
1991                 if (fl4->flowi4_oif) {
1992                         /* Apparently, routing tables are wrong. Assume,
1993                            that the destination is on link.
1994
1995                            WHY? DW.
1996                            Because we are allowed to send to iface
1997                            even if it has NO routes and NO assigned
1998                            addresses. When oif is specified, routing
1999                            tables are looked up with only one purpose:
2000                            to catch if destination is gatewayed, rather than
2001                            direct. Moreover, if MSG_DONTROUTE is set,
2002                            we send packet, ignoring both routing tables
2003                            and ifaddr state. --ANK
2004
2005
2006                            We could make it even if oif is unknown,
2007                            likely IPv6, but we do not.
2008                          */
2009
2010                         if (fl4->saddr == 0)
2011                                 fl4->saddr = inet_select_addr(dev_out, 0,
2012                                                               RT_SCOPE_LINK);
2013                         res.type = RTN_UNICAST;
2014                         goto make_route;
2015                 }
2016                 rth = ERR_PTR(-ENETUNREACH);
2017                 goto out;
2018         }
2019
2020         if (res.type == RTN_LOCAL) {
2021                 if (!fl4->saddr) {
2022                         if (res.fi->fib_prefsrc)
2023                                 fl4->saddr = res.fi->fib_prefsrc;
2024                         else
2025                                 fl4->saddr = fl4->daddr;
2026                 }
2027                 dev_out = net->loopback_dev;
2028                 fl4->flowi4_oif = dev_out->ifindex;
2029                 flags |= RTCF_LOCAL;
2030                 goto make_route;
2031         }
2032
2033 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2034         if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2035                 fib_select_multipath(&res);
2036         else
2037 #endif
2038         if (!res.prefixlen &&
2039             res.table->tb_num_default > 1 &&
2040             res.type == RTN_UNICAST && !fl4->flowi4_oif)
2041                 fib_select_default(&res);
2042
2043         if (!fl4->saddr)
2044                 fl4->saddr = FIB_RES_PREFSRC(net, res);
2045
2046         dev_out = FIB_RES_DEV(res);
2047         fl4->flowi4_oif = dev_out->ifindex;
2048
2049
2050 make_route:
2051         rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2052
2053 out:
2054         rcu_read_unlock();
2055         return rth;
2056 }
2057 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2058
2059 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2060 {
2061         return NULL;
2062 }
2063
2064 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2065 {
2066         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2067
2068         return mtu ? : dst->dev->mtu;
2069 }
2070
2071 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2072                                           struct sk_buff *skb, u32 mtu)
2073 {
2074 }
2075
2076 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2077                                        struct sk_buff *skb)
2078 {
2079 }
2080
2081 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2082                                           unsigned long old)
2083 {
2084         return NULL;
2085 }
2086
2087 static struct dst_ops ipv4_dst_blackhole_ops = {
2088         .family                 =       AF_INET,
2089         .protocol               =       cpu_to_be16(ETH_P_IP),
2090         .check                  =       ipv4_blackhole_dst_check,
2091         .mtu                    =       ipv4_blackhole_mtu,
2092         .default_advmss         =       ipv4_default_advmss,
2093         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2094         .redirect               =       ipv4_rt_blackhole_redirect,
2095         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
2096         .neigh_lookup           =       ipv4_neigh_lookup,
2097 };
2098
2099 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2100 {
2101         struct rtable *ort = (struct rtable *) dst_orig;
2102         struct rtable *rt;
2103
2104         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2105         if (rt) {
2106                 struct dst_entry *new = &rt->dst;
2107
2108                 new->__use = 1;
2109                 new->input = dst_discard;
2110                 new->output = dst_discard;
2111
2112                 new->dev = ort->dst.dev;
2113                 if (new->dev)
2114                         dev_hold(new->dev);
2115
2116                 rt->rt_is_input = ort->rt_is_input;
2117                 rt->rt_iif = ort->rt_iif;
2118                 rt->rt_pmtu = ort->rt_pmtu;
2119
2120                 rt->rt_genid = rt_genid(net);
2121                 rt->rt_flags = ort->rt_flags;
2122                 rt->rt_type = ort->rt_type;
2123                 rt->rt_gateway = ort->rt_gateway;
2124
2125                 INIT_LIST_HEAD(&rt->rt_uncached);
2126
2127                 dst_free(new);
2128         }
2129
2130         dst_release(dst_orig);
2131
2132         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2133 }
2134
2135 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2136                                     struct sock *sk)
2137 {
2138         struct rtable *rt = __ip_route_output_key(net, flp4);
2139
2140         if (IS_ERR(rt))
2141                 return rt;
2142
2143         if (flp4->flowi4_proto)
2144                 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2145                                                    flowi4_to_flowi(flp4),
2146                                                    sk, 0);
2147
2148         return rt;
2149 }
2150 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2151
2152 static int rt_fill_info(struct net *net,  __be32 dst, __be32 src,
2153                         struct flowi4 *fl4, struct sk_buff *skb, u32 pid,
2154                         u32 seq, int event, int nowait, unsigned int flags)
2155 {
2156         struct rtable *rt = skb_rtable(skb);
2157         struct rtmsg *r;
2158         struct nlmsghdr *nlh;
2159         unsigned long expires = 0;
2160         u32 error;
2161         u32 metrics[RTAX_MAX];
2162
2163         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2164         if (nlh == NULL)
2165                 return -EMSGSIZE;
2166
2167         r = nlmsg_data(nlh);
2168         r->rtm_family    = AF_INET;
2169         r->rtm_dst_len  = 32;
2170         r->rtm_src_len  = 0;
2171         r->rtm_tos      = fl4->flowi4_tos;
2172         r->rtm_table    = RT_TABLE_MAIN;
2173         if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2174                 goto nla_put_failure;
2175         r->rtm_type     = rt->rt_type;
2176         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2177         r->rtm_protocol = RTPROT_UNSPEC;
2178         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2179         if (rt->rt_flags & RTCF_NOTIFY)
2180                 r->rtm_flags |= RTM_F_NOTIFY;
2181
2182         if (nla_put_be32(skb, RTA_DST, dst))
2183                 goto nla_put_failure;
2184         if (src) {
2185                 r->rtm_src_len = 32;
2186                 if (nla_put_be32(skb, RTA_SRC, src))
2187                         goto nla_put_failure;
2188         }
2189         if (rt->dst.dev &&
2190             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2191                 goto nla_put_failure;
2192 #ifdef CONFIG_IP_ROUTE_CLASSID
2193         if (rt->dst.tclassid &&
2194             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2195                 goto nla_put_failure;
2196 #endif
2197         if (!rt_is_input_route(rt) &&
2198             fl4->saddr != src) {
2199                 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2200                         goto nla_put_failure;
2201         }
2202         if (rt->rt_gateway &&
2203             nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2204                 goto nla_put_failure;
2205
2206         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2207         if (rt->rt_pmtu)
2208                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2209         if (rtnetlink_put_metrics(skb, metrics) < 0)
2210                 goto nla_put_failure;
2211
2212         if (fl4->flowi4_mark &&
2213             nla_put_be32(skb, RTA_MARK, fl4->flowi4_mark))
2214                 goto nla_put_failure;
2215
2216         error = rt->dst.error;
2217         expires = rt->dst.expires;
2218         if (expires) {
2219                 if (time_before(jiffies, expires))
2220                         expires -= jiffies;
2221                 else
2222                         expires = 0;
2223         }
2224
2225         if (rt_is_input_route(rt)) {
2226                 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
2227                         goto nla_put_failure;
2228         }
2229
2230         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2231                 goto nla_put_failure;
2232
2233         return nlmsg_end(skb, nlh);
2234
2235 nla_put_failure:
2236         nlmsg_cancel(skb, nlh);
2237         return -EMSGSIZE;
2238 }
2239
2240 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
2241 {
2242         struct net *net = sock_net(in_skb->sk);
2243         struct rtmsg *rtm;
2244         struct nlattr *tb[RTA_MAX+1];
2245         struct rtable *rt = NULL;
2246         struct flowi4 fl4;
2247         __be32 dst = 0;
2248         __be32 src = 0;
2249         u32 iif;
2250         int err;
2251         int mark;
2252         struct sk_buff *skb;
2253
2254         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2255         if (err < 0)
2256                 goto errout;
2257
2258         rtm = nlmsg_data(nlh);
2259
2260         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2261         if (skb == NULL) {
2262                 err = -ENOBUFS;
2263                 goto errout;
2264         }
2265
2266         /* Reserve room for dummy headers, this skb can pass
2267            through good chunk of routing engine.
2268          */
2269         skb_reset_mac_header(skb);
2270         skb_reset_network_header(skb);
2271
2272         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2273         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2274         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2275
2276         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2277         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2278         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2279         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2280
2281         memset(&fl4, 0, sizeof(fl4));
2282         fl4.daddr = dst;
2283         fl4.saddr = src;
2284         fl4.flowi4_tos = rtm->rtm_tos;
2285         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2286         fl4.flowi4_mark = mark;
2287
2288         if (iif) {
2289                 struct net_device *dev;
2290
2291                 dev = __dev_get_by_index(net, iif);
2292                 if (dev == NULL) {
2293                         err = -ENODEV;
2294                         goto errout_free;
2295                 }
2296
2297                 skb->protocol   = htons(ETH_P_IP);
2298                 skb->dev        = dev;
2299                 skb->mark       = mark;
2300                 local_bh_disable();
2301                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2302                 local_bh_enable();
2303
2304                 rt = skb_rtable(skb);
2305                 if (err == 0 && rt->dst.error)
2306                         err = -rt->dst.error;
2307         } else {
2308                 rt = ip_route_output_key(net, &fl4);
2309
2310                 err = 0;
2311                 if (IS_ERR(rt))
2312                         err = PTR_ERR(rt);
2313         }
2314
2315         if (err)
2316                 goto errout_free;
2317
2318         skb_dst_set(skb, &rt->dst);
2319         if (rtm->rtm_flags & RTM_F_NOTIFY)
2320                 rt->rt_flags |= RTCF_NOTIFY;
2321
2322         err = rt_fill_info(net, dst, src, &fl4, skb,
2323                            NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2324                            RTM_NEWROUTE, 0, 0);
2325         if (err <= 0)
2326                 goto errout_free;
2327
2328         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2329 errout:
2330         return err;
2331
2332 errout_free:
2333         kfree_skb(skb);
2334         goto errout;
2335 }
2336
2337 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
2338 {
2339         return skb->len;
2340 }
2341
2342 void ip_rt_multicast_event(struct in_device *in_dev)
2343 {
2344         rt_cache_flush(dev_net(in_dev->dev));
2345 }
2346
2347 #ifdef CONFIG_SYSCTL
2348 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
2349                                         void __user *buffer,
2350                                         size_t *lenp, loff_t *ppos)
2351 {
2352         if (write) {
2353                 rt_cache_flush((struct net *)__ctl->extra1);
2354                 return 0;
2355         }
2356
2357         return -EINVAL;
2358 }
2359
2360 static ctl_table ipv4_route_table[] = {
2361         {
2362                 .procname       = "gc_thresh",
2363                 .data           = &ipv4_dst_ops.gc_thresh,
2364                 .maxlen         = sizeof(int),
2365                 .mode           = 0644,
2366                 .proc_handler   = proc_dointvec,
2367         },
2368         {
2369                 .procname       = "max_size",
2370                 .data           = &ip_rt_max_size,
2371                 .maxlen         = sizeof(int),
2372                 .mode           = 0644,
2373                 .proc_handler   = proc_dointvec,
2374         },
2375         {
2376                 /*  Deprecated. Use gc_min_interval_ms */
2377
2378                 .procname       = "gc_min_interval",
2379                 .data           = &ip_rt_gc_min_interval,
2380                 .maxlen         = sizeof(int),
2381                 .mode           = 0644,
2382                 .proc_handler   = proc_dointvec_jiffies,
2383         },
2384         {
2385                 .procname       = "gc_min_interval_ms",
2386                 .data           = &ip_rt_gc_min_interval,
2387                 .maxlen         = sizeof(int),
2388                 .mode           = 0644,
2389                 .proc_handler   = proc_dointvec_ms_jiffies,
2390         },
2391         {
2392                 .procname       = "gc_timeout",
2393                 .data           = &ip_rt_gc_timeout,
2394                 .maxlen         = sizeof(int),
2395                 .mode           = 0644,
2396                 .proc_handler   = proc_dointvec_jiffies,
2397         },
2398         {
2399                 .procname       = "gc_interval",
2400                 .data           = &ip_rt_gc_interval,
2401                 .maxlen         = sizeof(int),
2402                 .mode           = 0644,
2403                 .proc_handler   = proc_dointvec_jiffies,
2404         },
2405         {
2406                 .procname       = "redirect_load",
2407                 .data           = &ip_rt_redirect_load,
2408                 .maxlen         = sizeof(int),
2409                 .mode           = 0644,
2410                 .proc_handler   = proc_dointvec,
2411         },
2412         {
2413                 .procname       = "redirect_number",
2414                 .data           = &ip_rt_redirect_number,
2415                 .maxlen         = sizeof(int),
2416                 .mode           = 0644,
2417                 .proc_handler   = proc_dointvec,
2418         },
2419         {
2420                 .procname       = "redirect_silence",
2421                 .data           = &ip_rt_redirect_silence,
2422                 .maxlen         = sizeof(int),
2423                 .mode           = 0644,
2424                 .proc_handler   = proc_dointvec,
2425         },
2426         {
2427                 .procname       = "error_cost",
2428                 .data           = &ip_rt_error_cost,
2429                 .maxlen         = sizeof(int),
2430                 .mode           = 0644,
2431                 .proc_handler   = proc_dointvec,
2432         },
2433         {
2434                 .procname       = "error_burst",
2435                 .data           = &ip_rt_error_burst,
2436                 .maxlen         = sizeof(int),
2437                 .mode           = 0644,
2438                 .proc_handler   = proc_dointvec,
2439         },
2440         {
2441                 .procname       = "gc_elasticity",
2442                 .data           = &ip_rt_gc_elasticity,
2443                 .maxlen         = sizeof(int),
2444                 .mode           = 0644,
2445                 .proc_handler   = proc_dointvec,
2446         },
2447         {
2448                 .procname       = "mtu_expires",
2449                 .data           = &ip_rt_mtu_expires,
2450                 .maxlen         = sizeof(int),
2451                 .mode           = 0644,
2452                 .proc_handler   = proc_dointvec_jiffies,
2453         },
2454         {
2455                 .procname       = "min_pmtu",
2456                 .data           = &ip_rt_min_pmtu,
2457                 .maxlen         = sizeof(int),
2458                 .mode           = 0644,
2459                 .proc_handler   = proc_dointvec,
2460         },
2461         {
2462                 .procname       = "min_adv_mss",
2463                 .data           = &ip_rt_min_advmss,
2464                 .maxlen         = sizeof(int),
2465                 .mode           = 0644,
2466                 .proc_handler   = proc_dointvec,
2467         },
2468         { }
2469 };
2470
2471 static struct ctl_table ipv4_route_flush_table[] = {
2472         {
2473                 .procname       = "flush",
2474                 .maxlen         = sizeof(int),
2475                 .mode           = 0200,
2476                 .proc_handler   = ipv4_sysctl_rtcache_flush,
2477         },
2478         { },
2479 };
2480
2481 static __net_init int sysctl_route_net_init(struct net *net)
2482 {
2483         struct ctl_table *tbl;
2484
2485         tbl = ipv4_route_flush_table;
2486         if (!net_eq(net, &init_net)) {
2487                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2488                 if (tbl == NULL)
2489                         goto err_dup;
2490         }
2491         tbl[0].extra1 = net;
2492
2493         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2494         if (net->ipv4.route_hdr == NULL)
2495                 goto err_reg;
2496         return 0;
2497
2498 err_reg:
2499         if (tbl != ipv4_route_flush_table)
2500                 kfree(tbl);
2501 err_dup:
2502         return -ENOMEM;
2503 }
2504
2505 static __net_exit void sysctl_route_net_exit(struct net *net)
2506 {
2507         struct ctl_table *tbl;
2508
2509         tbl = net->ipv4.route_hdr->ctl_table_arg;
2510         unregister_net_sysctl_table(net->ipv4.route_hdr);
2511         BUG_ON(tbl == ipv4_route_flush_table);
2512         kfree(tbl);
2513 }
2514
2515 static __net_initdata struct pernet_operations sysctl_route_ops = {
2516         .init = sysctl_route_net_init,
2517         .exit = sysctl_route_net_exit,
2518 };
2519 #endif
2520
2521 static __net_init int rt_genid_init(struct net *net)
2522 {
2523         get_random_bytes(&net->ipv4.rt_genid,
2524                          sizeof(net->ipv4.rt_genid));
2525         get_random_bytes(&net->ipv4.dev_addr_genid,
2526                          sizeof(net->ipv4.dev_addr_genid));
2527         return 0;
2528 }
2529
2530 static __net_initdata struct pernet_operations rt_genid_ops = {
2531         .init = rt_genid_init,
2532 };
2533
2534 static int __net_init ipv4_inetpeer_init(struct net *net)
2535 {
2536         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2537
2538         if (!bp)
2539                 return -ENOMEM;
2540         inet_peer_base_init(bp);
2541         net->ipv4.peers = bp;
2542         return 0;
2543 }
2544
2545 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2546 {
2547         struct inet_peer_base *bp = net->ipv4.peers;
2548
2549         net->ipv4.peers = NULL;
2550         inetpeer_invalidate_tree(bp);
2551         kfree(bp);
2552 }
2553
2554 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2555         .init   =       ipv4_inetpeer_init,
2556         .exit   =       ipv4_inetpeer_exit,
2557 };
2558
2559 #ifdef CONFIG_IP_ROUTE_CLASSID
2560 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2561 #endif /* CONFIG_IP_ROUTE_CLASSID */
2562
2563 int __init ip_rt_init(void)
2564 {
2565         int rc = 0;
2566
2567 #ifdef CONFIG_IP_ROUTE_CLASSID
2568         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2569         if (!ip_rt_acct)
2570                 panic("IP: failed to allocate ip_rt_acct\n");
2571 #endif
2572
2573         ipv4_dst_ops.kmem_cachep =
2574                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2575                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2576
2577         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2578
2579         if (dst_entries_init(&ipv4_dst_ops) < 0)
2580                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2581
2582         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2583                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2584
2585         ipv4_dst_ops.gc_thresh = ~0;
2586         ip_rt_max_size = INT_MAX;
2587
2588         devinet_init();
2589         ip_fib_init();
2590
2591         if (ip_rt_proc_init())
2592                 pr_err("Unable to create route proc files\n");
2593 #ifdef CONFIG_XFRM
2594         xfrm_init();
2595         xfrm4_init(ip_rt_max_size);
2596 #endif
2597         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2598
2599 #ifdef CONFIG_SYSCTL
2600         register_pernet_subsys(&sysctl_route_ops);
2601 #endif
2602         register_pernet_subsys(&rt_genid_ops);
2603         register_pernet_subsys(&ipv4_inetpeer_ops);
2604         return rc;
2605 }
2606
2607 #ifdef CONFIG_SYSCTL
2608 /*
2609  * We really need to sanitize the damn ipv4 init order, then all
2610  * this nonsense will go away.
2611  */
2612 void __init ip_static_sysctl_init(void)
2613 {
2614         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
2615 }
2616 #endif