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, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *              Linus Torvalds, <Linus.Torvalds@helsinki.fi>
  12  *              Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  13  *
  14  * Fixes:
  15  *              Alan Cox        :       Verify area fixes.
  16  *              Alan Cox        :       cli() protects routing changes
  17  *              Rui Oliveira    :       ICMP routing table updates
  18  *              (rco@di.uminho.pt)      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_redirect_number __read_mostly  = 9;
 121 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 122 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 123 static int ip_rt_error_cost __read_mostly       = HZ;
 124 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 125 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
 126 static int ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
 127 static int ip_rt_min_advmss __read_mostly       = 256;
 128
 129 /*
 130  *      Interface to generic destination cache.
 131  */
 132
 133 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 134 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 135 static unsigned int      ipv4_mtu(const struct dst_entry *dst);
 136 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 137 static void              ipv4_link_failure(struct sk_buff *skb);
 138 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 139                                            struct sk_buff *skb, u32 mtu);
 140 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 141                                         struct sk_buff *skb);
 142 static void             ipv4_dst_destroy(struct dst_entry *dst);
 143
 144 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 145                             int how)
 146 {
 147 }
 148
 149 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 150 {
 151         WARN_ON(1);
 152         return NULL;
 153 }
 154
 155 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 156                                            struct sk_buff *skb,
 157                                            const void *daddr);
 158
 159 static struct dst_ops ipv4_dst_ops = {
 160         .family =               AF_INET,
 161         .protocol =             cpu_to_be16(ETH_P_IP),
 162         .check =                ipv4_dst_check,
 163         .default_advmss =       ipv4_default_advmss,
 164         .mtu =                  ipv4_mtu,
 165         .cow_metrics =          ipv4_cow_metrics,
 166         .destroy =              ipv4_dst_destroy,
 167         .ifdown =               ipv4_dst_ifdown,
 168         .negative_advice =      ipv4_negative_advice,
 169         .link_failure =         ipv4_link_failure,
 170         .update_pmtu =          ip_rt_update_pmtu,
 171         .redirect =             ip_do_redirect,
 172         .local_out =            __ip_local_out,
 173         .neigh_lookup =         ipv4_neigh_lookup,
 174 };
 175
 176 #define ECN_OR_COST(class)      TC_PRIO_##class
 177
 178 const __u8 ip_tos2prio[16] = {
 179         TC_PRIO_BESTEFFORT,
 180         ECN_OR_COST(BESTEFFORT),
 181         TC_PRIO_BESTEFFORT,
 182         ECN_OR_COST(BESTEFFORT),
 183         TC_PRIO_BULK,
 184         ECN_OR_COST(BULK),
 185         TC_PRIO_BULK,
 186         ECN_OR_COST(BULK),
 187         TC_PRIO_INTERACTIVE,
 188         ECN_OR_COST(INTERACTIVE),
 189         TC_PRIO_INTERACTIVE,
 190         ECN_OR_COST(INTERACTIVE),
 191         TC_PRIO_INTERACTIVE_BULK,
 192         ECN_OR_COST(INTERACTIVE_BULK),
 193         TC_PRIO_INTERACTIVE_BULK,
 194         ECN_OR_COST(INTERACTIVE_BULK)
 195 };
 196 EXPORT_SYMBOL(ip_tos2prio);
 197
 198 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 199 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
 200
 201 #ifdef CONFIG_PROC_FS
 202 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 203 {
 204         if (*pos)
 205                 return NULL;
 206         return SEQ_START_TOKEN;
 207 }
 208
 209 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 210 {
 211         ++*pos;
 212         return NULL;
 213 }
 214
 215 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 216 {
 217 }
 218
 219 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 220 {
 221         if (v == SEQ_START_TOKEN)
 222                 seq_printf(seq, "%-127s\n",
 223                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 224                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 225                            "HHUptod\tSpecDst");
 226         return 0;
 227 }
 228
 229 static const struct seq_operations rt_cache_seq_ops = {
 230         .start  = rt_cache_seq_start,
 231         .next   = rt_cache_seq_next,
 232         .stop   = rt_cache_seq_stop,
 233         .show   = rt_cache_seq_show,
 234 };
 235
 236 static int rt_cache_seq_open(struct inode *inode, struct file *file)
 237 {
 238         return seq_open(file, &rt_cache_seq_ops);
 239 }
 240
 241 static const struct file_operations rt_cache_seq_fops = {
 242         .owner   = THIS_MODULE,
 243         .open    = rt_cache_seq_open,
 244         .read    = seq_read,
 245         .llseek  = seq_lseek,
 246         .release = seq_release,
 247 };
 248
 249
 250 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 251 {
 252         int cpu;
 253
 254         if (*pos == 0)
 255                 return SEQ_START_TOKEN;
 256
 257         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 258                 if (!cpu_possible(cpu))
 259                         continue;
 260                 *pos = cpu+1;
 261                 return &per_cpu(rt_cache_stat, cpu);
 262         }
 263         return NULL;
 264 }
 265
 266 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 267 {
 268         int cpu;
 269
 270         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 271                 if (!cpu_possible(cpu))
 272                         continue;
 273                 *pos = cpu+1;
 274                 return &per_cpu(rt_cache_stat, cpu);
 275         }
 276         return NULL;
 277
 278 }
 279
 280 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 281 {
 282
 283 }
 284
 285 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 286 {
 287         struct rt_cache_stat *st = v;
 288
 289         if (v == SEQ_START_TOKEN) {
 290                 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");
 291                 return 0;
 292         }
 293
 294         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 295                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 296                    dst_entries_get_slow(&ipv4_dst_ops),
 297                    st->in_hit,
 298                    st->in_slow_tot,
 299                    st->in_slow_mc,
 300                    st->in_no_route,
 301                    st->in_brd,
 302                    st->in_martian_dst,
 303                    st->in_martian_src,
 304
 305                    st->out_hit,
 306                    st->out_slow_tot,
 307                    st->out_slow_mc,
 308
 309                    st->gc_total,
 310                    st->gc_ignored,
 311                    st->gc_goal_miss,
 312                    st->gc_dst_overflow,
 313                    st->in_hlist_search,
 314                    st->out_hlist_search
 315                 );
 316         return 0;
 317 }
 318
 319 static const struct seq_operations rt_cpu_seq_ops = {
 320         .start  = rt_cpu_seq_start,
 321         .next   = rt_cpu_seq_next,
 322         .stop   = rt_cpu_seq_stop,
 323         .show   = rt_cpu_seq_show,
 324 };
 325
 326
 327 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
 328 {
 329         return seq_open(file, &rt_cpu_seq_ops);
 330 }
 331
 332 static const struct file_operations rt_cpu_seq_fops = {
 333         .owner   = THIS_MODULE,
 334         .open    = rt_cpu_seq_open,
 335         .read    = seq_read,
 336         .llseek  = seq_lseek,
 337         .release = seq_release,
 338 };
 339
 340 #ifdef CONFIG_IP_ROUTE_CLASSID
 341 static int rt_acct_proc_show(struct seq_file *m, void *v)
 342 {
 343         struct ip_rt_acct *dst, *src;
 344         unsigned int i, j;
 345
 346         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 347         if (!dst)
 348                 return -ENOMEM;
 349
 350         for_each_possible_cpu(i) {
 351                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 352                 for (j = 0; j < 256; j++) {
 353                         dst[j].o_bytes   += src[j].o_bytes;
 354                         dst[j].o_packets += src[j].o_packets;
 355                         dst[j].i_bytes   += src[j].i_bytes;
 356                         dst[j].i_packets += src[j].i_packets;
 357                 }
 358         }
 359
 360         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 361         kfree(dst);
 362         return 0;
 363 }
 364
 365 static int rt_acct_proc_open(struct inode *inode, struct file *file)
 366 {
 367         return single_open(file, rt_acct_proc_show, NULL);
 368 }
 369
 370 static const struct file_operations rt_acct_proc_fops = {
 371         .owner          = THIS_MODULE,
 372         .open           = rt_acct_proc_open,
 373         .read           = seq_read,
 374         .llseek         = seq_lseek,
 375         .release        = single_release,
 376 };
 377 #endif
 378
 379 static int __net_init ip_rt_do_proc_init(struct net *net)
 380 {
 381         struct proc_dir_entry *pde;
 382
 383         pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
 384                           &rt_cache_seq_fops);
 385         if (!pde)
 386                 goto err1;
 387
 388         pde = proc_create("rt_cache", S_IRUGO,
 389                           net->proc_net_stat, &rt_cpu_seq_fops);
 390         if (!pde)
 391                 goto err2;
 392
 393 #ifdef CONFIG_IP_ROUTE_CLASSID
 394         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
 395         if (!pde)
 396                 goto err3;
 397 #endif
 398         return 0;
 399
 400 #ifdef CONFIG_IP_ROUTE_CLASSID
 401 err3:
 402         remove_proc_entry("rt_cache", net->proc_net_stat);
 403 #endif
 404 err2:
 405         remove_proc_entry("rt_cache", net->proc_net);
 406 err1:
 407         return -ENOMEM;
 408 }
 409
 410 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 411 {
 412         remove_proc_entry("rt_cache", net->proc_net_stat);
 413         remove_proc_entry("rt_cache", net->proc_net);
 414 #ifdef CONFIG_IP_ROUTE_CLASSID
 415         remove_proc_entry("rt_acct", net->proc_net);
 416 #endif
 417 }
 418
 419 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 420         .init = ip_rt_do_proc_init,
 421         .exit = ip_rt_do_proc_exit,
 422 };
 423
 424 static int __init ip_rt_proc_init(void)
 425 {
 426         return register_pernet_subsys(&ip_rt_proc_ops);
 427 }
 428
 429 #else
 430 static inline int ip_rt_proc_init(void)
 431 {
 432         return 0;
 433 }
 434 #endif /* CONFIG_PROC_FS */
 435
 436 static inline bool rt_is_expired(const struct rtable *rth)
 437 {
 438         return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
 439 }
 440
 441 void rt_cache_flush(struct net *net)
 442 {
 443         rt_genid_bump(net);
 444 }
 445
 446 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 447                                            struct sk_buff *skb,
 448                                            const void *daddr)
 449 {
 450         struct net_device *dev = dst->dev;
 451         const __be32 *pkey = daddr;
 452         const struct rtable *rt;
 453         struct neighbour *n;
 454
 455         rt = (const struct rtable *) dst;
 456         if (rt->rt_gateway)
 457                 pkey = (const __be32 *) &rt->rt_gateway;
 458         else if (skb)
 459                 pkey = &ip_hdr(skb)->daddr;
 460
 461         n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
 462         if (n)
 463                 return n;
 464         return neigh_create(&arp_tbl, pkey, dev);
 465 }
 466
 467 /*
 468  * Peer allocation may fail only in serious out-of-memory conditions.  However
 469  * we still can generate some output.
 470  * Random ID selection looks a bit dangerous because we have no chances to
 471  * select ID being unique in a reasonable period of time.
 472  * But broken packet identifier may be better than no packet at all.
 473  */
 474 static void ip_select_fb_ident(struct iphdr *iph)
 475 {
 476         static DEFINE_SPINLOCK(ip_fb_id_lock);
 477         static u32 ip_fallback_id;
 478         u32 salt;
 479
 480         spin_lock_bh(&ip_fb_id_lock);
 481         salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
 482         iph->id = htons(salt & 0xFFFF);
 483         ip_fallback_id = salt;
 484         spin_unlock_bh(&ip_fb_id_lock);
 485 }
 486
 487 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
 488 {
 489         struct net *net = dev_net(dst->dev);
 490         struct inet_peer *peer;
 491
 492         peer = inet_getpeer_v4(net->ipv4.peers, iph->daddr, 1);
 493         if (peer) {
 494                 iph->id = htons(inet_getid(peer, more));
 495                 inet_putpeer(peer);
 496                 return;
 497         }
 498
 499         ip_select_fb_ident(iph);
 500 }
 501 EXPORT_SYMBOL(__ip_select_ident);
 502
 503 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 504                              const struct iphdr *iph,
 505                              int oif, u8 tos,
 506                              u8 prot, u32 mark, int flow_flags)
 507 {
 508         if (sk) {
 509                 const struct inet_sock *inet = inet_sk(sk);
 510
 511                 oif = sk->sk_bound_dev_if;
 512                 mark = sk->sk_mark;
 513                 tos = RT_CONN_FLAGS(sk);
 514                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 515         }
 516         flowi4_init_output(fl4, oif, mark, tos,
 517                            RT_SCOPE_UNIVERSE, prot,
 518                            flow_flags,
 519                            iph->daddr, iph->saddr, 0, 0);
 520 }
 521
 522 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 523                                const struct sock *sk)
 524 {
 525         const struct iphdr *iph = ip_hdr(skb);
 526         int oif = skb->dev->ifindex;
 527         u8 tos = RT_TOS(iph->tos);
 528         u8 prot = iph->protocol;
 529         u32 mark = skb->mark;
 530
 531         __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
 532 }
 533
 534 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 535 {
 536         const struct inet_sock *inet = inet_sk(sk);
 537         const struct ip_options_rcu *inet_opt;
 538         __be32 daddr = inet->inet_daddr;
 539
 540         rcu_read_lock();
 541         inet_opt = rcu_dereference(inet->inet_opt);
 542         if (inet_opt && inet_opt->opt.srr)
 543                 daddr = inet_opt->opt.faddr;
 544         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 545                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 546                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 547                            inet_sk_flowi_flags(sk),
 548                            daddr, inet->inet_saddr, 0, 0);
 549         rcu_read_unlock();
 550 }
 551
 552 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 553                                  const struct sk_buff *skb)
 554 {
 555         if (skb)
 556                 build_skb_flow_key(fl4, skb, sk);
 557         else
 558                 build_sk_flow_key(fl4, sk);
 559 }
 560
 561 static inline void rt_free(struct rtable *rt)
 562 {
 563         call_rcu(&rt->dst.rcu_head, dst_rcu_free);
 564 }
 565
 566 static DEFINE_SPINLOCK(fnhe_lock);
 567
 568 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
 569 {
 570         struct fib_nh_exception *fnhe, *oldest;
 571         struct rtable *orig;
 572
 573         oldest = rcu_dereference(hash->chain);
 574         for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
 575              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 576                 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
 577                         oldest = fnhe;
 578         }
 579         orig = rcu_dereference(oldest->fnhe_rth);
 580         if (orig) {
 581                 RCU_INIT_POINTER(oldest->fnhe_rth, NULL);
 582                 rt_free(orig);
 583         }
 584         return oldest;
 585 }
 586
 587 static inline u32 fnhe_hashfun(__be32 daddr)
 588 {
 589         u32 hval;
 590
 591         hval = (__force u32) daddr;
 592         hval ^= (hval >> 11) ^ (hval >> 22);
 593
 594         return hval & (FNHE_HASH_SIZE - 1);
 595 }
 596
 597 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
 598                                   u32 pmtu, unsigned long expires)
 599 {
 600         struct fnhe_hash_bucket *hash;
 601         struct fib_nh_exception *fnhe;
 602         int depth;
 603         u32 hval = fnhe_hashfun(daddr);
 604
 605         spin_lock_bh(&fnhe_lock);
 606
 607         hash = nh->nh_exceptions;
 608         if (!hash) {
 609                 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
 610                 if (!hash)
 611                         goto out_unlock;
 612                 nh->nh_exceptions = hash;
 613         }
 614
 615         hash += hval;
 616
 617         depth = 0;
 618         for (fnhe = rcu_dereference(hash->chain); fnhe;
 619              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 620                 if (fnhe->fnhe_daddr == daddr)
 621                         break;
 622                 depth++;
 623         }
 624
 625         if (fnhe) {
 626                 if (gw)
 627                         fnhe->fnhe_gw = gw;
 628                 if (pmtu) {
 629                         fnhe->fnhe_pmtu = pmtu;
 630                         fnhe->fnhe_expires = expires;
 631                 }
 632         } else {
 633                 if (depth > FNHE_RECLAIM_DEPTH)
 634                         fnhe = fnhe_oldest(hash);
 635                 else {
 636                         fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 637                         if (!fnhe)
 638                                 goto out_unlock;
 639
 640                         fnhe->fnhe_next = hash->chain;
 641                         rcu_assign_pointer(hash->chain, fnhe);
 642                 }
 643                 fnhe->fnhe_daddr = daddr;
 644                 fnhe->fnhe_gw = gw;
 645                 fnhe->fnhe_pmtu = pmtu;
 646                 fnhe->fnhe_expires = expires;
 647         }
 648
 649         fnhe->fnhe_stamp = jiffies;
 650
 651 out_unlock:
 652         spin_unlock_bh(&fnhe_lock);
 653         return;
 654 }
 655
 656 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 657                              bool kill_route)
 658 {
 659         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 660         __be32 old_gw = ip_hdr(skb)->saddr;
 661         struct net_device *dev = skb->dev;
 662         struct in_device *in_dev;
 663         struct fib_result res;
 664         struct neighbour *n;
 665         struct net *net;
 666
 667         switch (icmp_hdr(skb)->code & 7) {
 668         case ICMP_REDIR_NET:
 669         case ICMP_REDIR_NETTOS:
 670         case ICMP_REDIR_HOST:
 671         case ICMP_REDIR_HOSTTOS:
 672                 break;
 673
 674         default:
 675                 return;
 676         }
 677
 678         if (rt->rt_gateway != old_gw)
 679                 return;
 680
 681         in_dev = __in_dev_get_rcu(dev);
 682         if (!in_dev)
 683                 return;
 684
 685         net = dev_net(dev);
 686         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 687             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 688             ipv4_is_zeronet(new_gw))
 689                 goto reject_redirect;
 690
 691         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 692                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 693                         goto reject_redirect;
 694                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 695                         goto reject_redirect;
 696         } else {
 697                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 698                         goto reject_redirect;
 699         }
 700
 701         n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
 702         if (n) {
 703                 if (!(n->nud_state & NUD_VALID)) {
 704                         neigh_event_send(n, NULL);
 705                 } else {
 706                         if (fib_lookup(net, fl4, &res) == 0) {
 707                                 struct fib_nh *nh = &FIB_RES_NH(res);
 708
 709                                 update_or_create_fnhe(nh, fl4->daddr, new_gw,
 710                                                       0, 0);
 711                         }
 712                         if (kill_route)
 713                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 714                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 715                 }
 716                 neigh_release(n);
 717         }
 718         return;
 719
 720 reject_redirect:
 721 #ifdef CONFIG_IP_ROUTE_VERBOSE
 722         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 723                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 724                 __be32 daddr = iph->daddr;
 725                 __be32 saddr = iph->saddr;
 726
 727                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 728                                      "  Advised path = %pI4 -> %pI4\n",
 729                                      &old_gw, dev->name, &new_gw,
 730                                      &saddr, &daddr);
 731         }
 732 #endif
 733         ;
 734 }
 735
 736 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 737 {
 738         struct rtable *rt;
 739         struct flowi4 fl4;
 740         const struct iphdr *iph = (const struct iphdr *) skb->data;
 741         int oif = skb->dev->ifindex;
 742         u8 tos = RT_TOS(iph->tos);
 743         u8 prot = iph->protocol;
 744         u32 mark = skb->mark;
 745
 746         rt = (struct rtable *) dst;
 747
 748         __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
 749         __ip_do_redirect(rt, skb, &fl4, true);
 750 }
 751
 752 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 753 {
 754         struct rtable *rt = (struct rtable *)dst;
 755         struct dst_entry *ret = dst;
 756
 757         if (rt) {
 758                 if (dst->obsolete > 0) {
 759                         ip_rt_put(rt);
 760                         ret = NULL;
 761                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 762                            rt->dst.expires) {
 763                         ip_rt_put(rt);
 764                         ret = NULL;
 765                 }
 766         }
 767         return ret;
 768 }
 769
 770 /*
 771  * Algorithm:
 772  *      1. The first ip_rt_redirect_number redirects are sent
 773  *         with exponential backoff, then we stop sending them at all,
 774  *         assuming that the host ignores our redirects.
 775  *      2. If we did not see packets requiring redirects
 776  *         during ip_rt_redirect_silence, we assume that the host
 777  *         forgot redirected route and start to send redirects again.
 778  *
 779  * This algorithm is much cheaper and more intelligent than dumb load limiting
 780  * in icmp.c.
 781  *
 782  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 783  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 784  */
 785
 786 void ip_rt_send_redirect(struct sk_buff *skb)
 787 {
 788         struct rtable *rt = skb_rtable(skb);
 789         struct in_device *in_dev;
 790         struct inet_peer *peer;
 791         struct net *net;
 792         int log_martians;
 793
 794         rcu_read_lock();
 795         in_dev = __in_dev_get_rcu(rt->dst.dev);
 796         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 797                 rcu_read_unlock();
 798                 return;
 799         }
 800         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 801         rcu_read_unlock();
 802
 803         net = dev_net(rt->dst.dev);
 804         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
 805         if (!peer) {
 806                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 807                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 808                 return;
 809         }
 810
 811         /* No redirected packets during ip_rt_redirect_silence;
 812          * reset the algorithm.
 813          */
 814         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
 815                 peer->rate_tokens = 0;
 816
 817         /* Too many ignored redirects; do not send anything
 818          * set dst.rate_last to the last seen redirected packet.
 819          */
 820         if (peer->rate_tokens >= ip_rt_redirect_number) {
 821                 peer->rate_last = jiffies;
 822                 goto out_put_peer;
 823         }
 824
 825         /* Check for load limit; set rate_last to the latest sent
 826          * redirect.
 827          */
 828         if (peer->rate_tokens == 0 ||
 829             time_after(jiffies,
 830                        (peer->rate_last +
 831                         (ip_rt_redirect_load << peer->rate_tokens)))) {
 832                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 833
 834                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 835                 peer->rate_last = jiffies;
 836                 ++peer->rate_tokens;
 837 #ifdef CONFIG_IP_ROUTE_VERBOSE
 838                 if (log_martians &&
 839                     peer->rate_tokens == ip_rt_redirect_number)
 840                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 841                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 842                                              &ip_hdr(skb)->daddr, &gw);
 843 #endif
 844         }
 845 out_put_peer:
 846         inet_putpeer(peer);
 847 }
 848
 849 static int ip_error(struct sk_buff *skb)
 850 {
 851         struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
 852         struct rtable *rt = skb_rtable(skb);
 853         struct inet_peer *peer;
 854         unsigned long now;
 855         struct net *net;
 856         bool send;
 857         int code;
 858
 859         net = dev_net(rt->dst.dev);
 860         if (!IN_DEV_FORWARD(in_dev)) {
 861                 switch (rt->dst.error) {
 862                 case EHOSTUNREACH:
 863                         IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
 864                         break;
 865
 866                 case ENETUNREACH:
 867                         IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
 868                         break;
 869                 }
 870                 goto out;
 871         }
 872
 873         switch (rt->dst.error) {
 874         case EINVAL:
 875         default:
 876                 goto out;
 877         case EHOSTUNREACH:
 878                 code = ICMP_HOST_UNREACH;
 879                 break;
 880         case ENETUNREACH:
 881                 code = ICMP_NET_UNREACH;
 882                 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
 883                 break;
 884         case EACCES:
 885                 code = ICMP_PKT_FILTERED;
 886                 break;
 887         }
 888
 889         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
 890
 891         send = true;
 892         if (peer) {
 893                 now = jiffies;
 894                 peer->rate_tokens += now - peer->rate_last;
 895                 if (peer->rate_tokens > ip_rt_error_burst)
 896                         peer->rate_tokens = ip_rt_error_burst;
 897                 peer->rate_last = now;
 898                 if (peer->rate_tokens >= ip_rt_error_cost)
 899                         peer->rate_tokens -= ip_rt_error_cost;
 900                 else
 901                         send = false;
 902                 inet_putpeer(peer);
 903         }
 904         if (send)
 905                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
 906
 907 out:    kfree_skb(skb);
 908         return 0;
 909 }
 910
 911 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
 912 {
 913         struct dst_entry *dst = &rt->dst;
 914         struct fib_result res;
 915
 916         if (dst_metric_locked(dst, RTAX_MTU))
 917                 return;
 918
 919         if (dst->dev->mtu < mtu)
 920                 return;
 921
 922         if (mtu < ip_rt_min_pmtu)
 923                 mtu = ip_rt_min_pmtu;
 924
 925         if (!rt->rt_pmtu) {
 926                 dst->obsolete = DST_OBSOLETE_KILL;
 927         } else {
 928                 rt->rt_pmtu = mtu;
 929                 dst->expires = max(1UL, jiffies + ip_rt_mtu_expires);
 930         }
 931
 932         rcu_read_lock();
 933         if (fib_lookup(dev_net(dst->dev), fl4, &res) == 0) {
 934                 struct fib_nh *nh = &FIB_RES_NH(res);
 935
 936                 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
 937                                       jiffies + ip_rt_mtu_expires);
 938         }
 939         rcu_read_unlock();
 940 }
 941
 942 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 943                               struct sk_buff *skb, u32 mtu)
 944 {
 945         struct rtable *rt = (struct rtable *) dst;
 946         struct flowi4 fl4;
 947
 948         ip_rt_build_flow_key(&fl4, sk, skb);
 949         __ip_rt_update_pmtu(rt, &fl4, mtu);
 950 }
 951
 952 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
 953                       int oif, u32 mark, u8 protocol, int flow_flags)
 954 {
 955         const struct iphdr *iph = (const struct iphdr *) skb->data;
 956         struct flowi4 fl4;
 957         struct rtable *rt;
 958
 959         if (!mark)
 960                 mark = IP4_REPLY_MARK(net, skb->mark);
 961
 962         __build_flow_key(&fl4, NULL, iph, oif,
 963                          RT_TOS(iph->tos), protocol, mark, flow_flags);
 964         rt = __ip_route_output_key(net, &fl4);
 965         if (!IS_ERR(rt)) {
 966                 __ip_rt_update_pmtu(rt, &fl4, mtu);
 967                 ip_rt_put(rt);
 968         }
 969 }
 970 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
 971
 972 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
 973 {
 974         const struct iphdr *iph = (const struct iphdr *) skb->data;
 975         struct flowi4 fl4;
 976         struct rtable *rt;
 977
 978         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
 979
 980         if (!fl4.flowi4_mark)
 981                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
 982
 983         rt = __ip_route_output_key(sock_net(sk), &fl4);
 984         if (!IS_ERR(rt)) {
 985                 __ip_rt_update_pmtu(rt, &fl4, mtu);
 986                 ip_rt_put(rt);
 987         }
 988 }
 989
 990 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
 991 {
 992         const struct iphdr *iph = (const struct iphdr *) skb->data;
 993         struct flowi4 fl4;
 994         struct rtable *rt;
 995         struct dst_entry *dst;
 996         bool new = false;
 997
 998         bh_lock_sock(sk);
 999         rt = (struct rtable *) __sk_dst_get(sk);
1000
1001         if (sock_owned_by_user(sk) || !rt) {
1002                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1003                 goto out;
1004         }
1005
1006         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1007
1008         if (!__sk_dst_check(sk, 0)) {
1009                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1010                 if (IS_ERR(rt))
1011                         goto out;
1012
1013                 new = true;
1014         }
1015
1016         __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1017
1018         dst = dst_check(&rt->dst, 0);
1019         if (!dst) {
1020                 if (new)
1021                         dst_release(&rt->dst);
1022
1023                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1024                 if (IS_ERR(rt))
1025                         goto out;
1026
1027                 new = true;
1028         }
1029
1030         if (new)
1031                 __sk_dst_set(sk, &rt->dst);
1032
1033 out:
1034         bh_unlock_sock(sk);
1035 }
1036 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1037
1038 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1039                    int oif, u32 mark, u8 protocol, int flow_flags)
1040 {
1041         const struct iphdr *iph = (const struct iphdr *) skb->data;
1042         struct flowi4 fl4;
1043         struct rtable *rt;
1044
1045         __build_flow_key(&fl4, NULL, iph, oif,
1046                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1047         rt = __ip_route_output_key(net, &fl4);
1048         if (!IS_ERR(rt)) {
1049                 __ip_do_redirect(rt, skb, &fl4, false);
1050                 ip_rt_put(rt);
1051         }
1052 }
1053 EXPORT_SYMBOL_GPL(ipv4_redirect);
1054
1055 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1056 {
1057         const struct iphdr *iph = (const struct iphdr *) skb->data;
1058         struct flowi4 fl4;
1059         struct rtable *rt;
1060
1061         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1062         rt = __ip_route_output_key(sock_net(sk), &fl4);
1063         if (!IS_ERR(rt)) {
1064                 __ip_do_redirect(rt, skb, &fl4, false);
1065                 ip_rt_put(rt);
1066         }
1067 }
1068 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1069
1070 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1071 {
1072         struct rtable *rt = (struct rtable *) dst;
1073
1074         /* All IPV4 dsts are created with ->obsolete set to the value
1075          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1076          * into this function always.
1077          *
1078          * When a PMTU/redirect information update invalidates a
1079          * route, this is indicated by setting obsolete to
1080          * DST_OBSOLETE_KILL.
1081          */
1082         if (dst->obsolete == DST_OBSOLETE_KILL || rt_is_expired(rt))
1083                 return NULL;
1084         return dst;
1085 }
1086
1087 static void ipv4_link_failure(struct sk_buff *skb)
1088 {
1089         struct rtable *rt;
1090
1091         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1092
1093         rt = skb_rtable(skb);
1094         if (rt)
1095                 dst_set_expires(&rt->dst, 0);
1096 }
1097
1098 static int ip_rt_bug(struct sk_buff *skb)
1099 {
1100         pr_debug("%s: %pI4 -> %pI4, %s\n",
1101                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1102                  skb->dev ? skb->dev->name : "?");
1103         kfree_skb(skb);
1104         WARN_ON(1);
1105         return 0;
1106 }
1107
1108 /*
1109    We do not cache source address of outgoing interface,
1110    because it is used only by IP RR, TS and SRR options,
1111    so that it out of fast path.
1112
1113    BTW remember: "addr" is allowed to be not aligned
1114    in IP options!
1115  */
1116
1117 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1118 {
1119         __be32 src;
1120
1121         if (rt_is_output_route(rt))
1122                 src = ip_hdr(skb)->saddr;
1123         else {
1124                 struct fib_result res;
1125                 struct flowi4 fl4;
1126                 struct iphdr *iph;
1127
1128                 iph = ip_hdr(skb);
1129
1130                 memset(&fl4, 0, sizeof(fl4));
1131                 fl4.daddr = iph->daddr;
1132                 fl4.saddr = iph->saddr;
1133                 fl4.flowi4_tos = RT_TOS(iph->tos);
1134                 fl4.flowi4_oif = rt->dst.dev->ifindex;
1135                 fl4.flowi4_iif = skb->dev->ifindex;
1136                 fl4.flowi4_mark = skb->mark;
1137
1138                 rcu_read_lock();
1139                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1140                         src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1141                 else
1142                         src = inet_select_addr(rt->dst.dev,
1143                                                rt_nexthop(rt, iph->daddr),
1144                                                RT_SCOPE_UNIVERSE);
1145                 rcu_read_unlock();
1146         }
1147         memcpy(addr, &src, 4);
1148 }
1149
1150 #ifdef CONFIG_IP_ROUTE_CLASSID
1151 static void set_class_tag(struct rtable *rt, u32 tag)
1152 {
1153         if (!(rt->dst.tclassid & 0xFFFF))
1154                 rt->dst.tclassid |= tag & 0xFFFF;
1155         if (!(rt->dst.tclassid & 0xFFFF0000))
1156                 rt->dst.tclassid |= tag & 0xFFFF0000;
1157 }
1158 #endif
1159
1160 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1161 {
1162         unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1163
1164         if (advmss == 0) {
1165                 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1166                                ip_rt_min_advmss);
1167                 if (advmss > 65535 - 40)
1168                         advmss = 65535 - 40;
1169         }
1170         return advmss;
1171 }
1172
1173 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1174 {
1175         const struct rtable *rt = (const struct rtable *) dst;
1176         unsigned int mtu = rt->rt_pmtu;
1177
1178         if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1179                 mtu = dst_metric_raw(dst, RTAX_MTU);
1180
1181         if (mtu)
1182                 return mtu;
1183
1184         mtu = dst->dev->mtu;
1185
1186         if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1187                 if (rt->rt_uses_gateway && mtu > 576)
1188                         mtu = 576;
1189         }
1190
1191         if (mtu > IP_MAX_MTU)
1192                 mtu = IP_MAX_MTU;
1193
1194         return mtu;
1195 }
1196
1197 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1198 {
1199         struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1200         struct fib_nh_exception *fnhe;
1201         u32 hval;
1202
1203         if (!hash)
1204                 return NULL;
1205
1206         hval = fnhe_hashfun(daddr);
1207
1208         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1209              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1210                 if (fnhe->fnhe_daddr == daddr)
1211                         return fnhe;
1212         }
1213         return NULL;
1214 }
1215
1216 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1217                               __be32 daddr)
1218 {
1219         bool ret = false;
1220
1221         spin_lock_bh(&fnhe_lock);
1222
1223         if (daddr == fnhe->fnhe_daddr) {
1224                 struct rtable *orig = rcu_dereference(fnhe->fnhe_rth);
1225                 if (orig && rt_is_expired(orig)) {
1226                         fnhe->fnhe_gw = 0;
1227                         fnhe->fnhe_pmtu = 0;
1228                         fnhe->fnhe_expires = 0;
1229                 }
1230                 if (fnhe->fnhe_pmtu) {
1231                         unsigned long expires = fnhe->fnhe_expires;
1232                         unsigned long diff = expires - jiffies;
1233
1234                         if (time_before(jiffies, expires)) {
1235                                 rt->rt_pmtu = fnhe->fnhe_pmtu;
1236                                 dst_set_expires(&rt->dst, diff);
1237                         }
1238                 }
1239                 if (fnhe->fnhe_gw) {
1240                         rt->rt_flags |= RTCF_REDIRECTED;
1241                         rt->rt_gateway = fnhe->fnhe_gw;
1242                         rt->rt_uses_gateway = 1;
1243                 } else if (!rt->rt_gateway)
1244                         rt->rt_gateway = daddr;
1245
1246                 rcu_assign_pointer(fnhe->fnhe_rth, rt);
1247                 if (orig)
1248                         rt_free(orig);
1249
1250                 fnhe->fnhe_stamp = jiffies;
1251                 ret = true;
1252         }
1253         spin_unlock_bh(&fnhe_lock);
1254
1255         return ret;
1256 }
1257
1258 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1259 {
1260         struct rtable *orig, *prev, **p;
1261         bool ret = true;
1262
1263         if (rt_is_input_route(rt)) {
1264                 p = (struct rtable **)&nh->nh_rth_input;
1265         } else {
1266                 p = (struct rtable **)__this_cpu_ptr(nh->nh_pcpu_rth_output);
1267         }
1268         orig = *p;
1269
1270         prev = cmpxchg(p, orig, rt);
1271         if (prev == orig) {
1272                 if (orig)
1273                         rt_free(orig);
1274         } else
1275                 ret = false;
1276
1277         return ret;
1278 }
1279
1280 static DEFINE_SPINLOCK(rt_uncached_lock);
1281 static LIST_HEAD(rt_uncached_list);
1282
1283 static void rt_add_uncached_list(struct rtable *rt)
1284 {
1285         spin_lock_bh(&rt_uncached_lock);
1286         list_add_tail(&rt->rt_uncached, &rt_uncached_list);
1287         spin_unlock_bh(&rt_uncached_lock);
1288 }
1289
1290 static void ipv4_dst_destroy(struct dst_entry *dst)
1291 {
1292         struct rtable *rt = (struct rtable *) dst;
1293
1294         if (!list_empty(&rt->rt_uncached)) {
1295                 spin_lock_bh(&rt_uncached_lock);
1296                 list_del(&rt->rt_uncached);
1297                 spin_unlock_bh(&rt_uncached_lock);
1298         }
1299 }
1300
1301 void rt_flush_dev(struct net_device *dev)
1302 {
1303         if (!list_empty(&rt_uncached_list)) {
1304                 struct net *net = dev_net(dev);
1305                 struct rtable *rt;
1306
1307                 spin_lock_bh(&rt_uncached_lock);
1308                 list_for_each_entry(rt, &rt_uncached_list, rt_uncached) {
1309                         if (rt->dst.dev != dev)
1310                                 continue;
1311                         rt->dst.dev = net->loopback_dev;
1312                         dev_hold(rt->dst.dev);
1313                         dev_put(dev);
1314                 }
1315                 spin_unlock_bh(&rt_uncached_lock);
1316         }
1317 }
1318
1319 static bool rt_cache_valid(const struct rtable *rt)
1320 {
1321         return  rt &&
1322                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1323                 !rt_is_expired(rt);
1324 }
1325
1326 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1327                            const struct fib_result *res,
1328                            struct fib_nh_exception *fnhe,
1329                            struct fib_info *fi, u16 type, u32 itag)
1330 {
1331         bool cached = false;
1332
1333         if (fi) {
1334                 struct fib_nh *nh = &FIB_RES_NH(*res);
1335
1336                 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1337                         rt->rt_gateway = nh->nh_gw;
1338                         rt->rt_uses_gateway = 1;
1339                 }
1340                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1341 #ifdef CONFIG_IP_ROUTE_CLASSID
1342                 rt->dst.tclassid = nh->nh_tclassid;
1343 #endif
1344                 if (unlikely(fnhe))
1345                         cached = rt_bind_exception(rt, fnhe, daddr);
1346                 else if (!(rt->dst.flags & DST_NOCACHE))
1347                         cached = rt_cache_route(nh, rt);
1348                 if (unlikely(!cached)) {
1349                         /* Routes we intend to cache in nexthop exception or
1350                          * FIB nexthop have the DST_NOCACHE bit clear.
1351                          * However, if we are unsuccessful at storing this
1352                          * route into the cache we really need to set it.
1353                          */
1354                         rt->dst.flags |= DST_NOCACHE;
1355                         if (!rt->rt_gateway)
1356                                 rt->rt_gateway = daddr;
1357                         rt_add_uncached_list(rt);
1358                 }
1359         } else
1360                 rt_add_uncached_list(rt);
1361
1362 #ifdef CONFIG_IP_ROUTE_CLASSID
1363 #ifdef CONFIG_IP_MULTIPLE_TABLES
1364         set_class_tag(rt, res->tclassid);
1365 #endif
1366         set_class_tag(rt, itag);
1367 #endif
1368 }
1369
1370 static struct rtable *rt_dst_alloc(struct net_device *dev,
1371                                    bool nopolicy, bool noxfrm, bool will_cache)
1372 {
1373         return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1374                          (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1375                          (nopolicy ? DST_NOPOLICY : 0) |
1376                          (noxfrm ? DST_NOXFRM : 0));
1377 }
1378
1379 /* called in rcu_read_lock() section */
1380 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1381                                 u8 tos, struct net_device *dev, int our)
1382 {
1383         struct rtable *rth;
1384         struct in_device *in_dev = __in_dev_get_rcu(dev);
1385         u32 itag = 0;
1386         int err;
1387
1388         /* Primary sanity checks. */
1389
1390         if (in_dev == NULL)
1391                 return -EINVAL;
1392
1393         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1394             skb->protocol != htons(ETH_P_IP))
1395                 goto e_inval;
1396
1397         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1398                 if (ipv4_is_loopback(saddr))
1399                         goto e_inval;
1400
1401         if (ipv4_is_zeronet(saddr)) {
1402                 if (!ipv4_is_local_multicast(daddr))
1403                         goto e_inval;
1404         } else {
1405                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1406                                           in_dev, &itag);
1407                 if (err < 0)
1408                         goto e_err;
1409         }
1410         rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1411                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1412         if (!rth)
1413                 goto e_nobufs;
1414
1415 #ifdef CONFIG_IP_ROUTE_CLASSID
1416         rth->dst.tclassid = itag;
1417 #endif
1418         rth->dst.output = ip_rt_bug;
1419
1420         rth->rt_genid   = rt_genid(dev_net(dev));
1421         rth->rt_flags   = RTCF_MULTICAST;
1422         rth->rt_type    = RTN_MULTICAST;
1423         rth->rt_is_input= 1;
1424         rth->rt_iif     = 0;
1425         rth->rt_pmtu    = 0;
1426         rth->rt_gateway = 0;
1427         rth->rt_uses_gateway = 0;
1428         INIT_LIST_HEAD(&rth->rt_uncached);
1429         if (our) {
1430                 rth->dst.input= ip_local_deliver;
1431                 rth->rt_flags |= RTCF_LOCAL;
1432         }
1433
1434 #ifdef CONFIG_IP_MROUTE
1435         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1436                 rth->dst.input = ip_mr_input;
1437 #endif
1438         RT_CACHE_STAT_INC(in_slow_mc);
1439
1440         skb_dst_set(skb, &rth->dst);
1441         return 0;
1442
1443 e_nobufs:
1444         return -ENOBUFS;
1445 e_inval:
1446         return -EINVAL;
1447 e_err:
1448         return err;
1449 }
1450
1451
1452 static void ip_handle_martian_source(struct net_device *dev,
1453                                      struct in_device *in_dev,
1454                                      struct sk_buff *skb,
1455                                      __be32 daddr,
1456                                      __be32 saddr)
1457 {
1458         RT_CACHE_STAT_INC(in_martian_src);
1459 #ifdef CONFIG_IP_ROUTE_VERBOSE
1460         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1461                 /*
1462                  *      RFC1812 recommendation, if source is martian,
1463                  *      the only hint is MAC header.
1464                  */
1465                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1466                         &daddr, &saddr, dev->name);
1467                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1468                         print_hex_dump(KERN_WARNING, "ll header: ",
1469                                        DUMP_PREFIX_OFFSET, 16, 1,
1470                                        skb_mac_header(skb),
1471                                        dev->hard_header_len, true);
1472                 }
1473         }
1474 #endif
1475 }
1476
1477 /* called in rcu_read_lock() section */
1478 static int __mkroute_input(struct sk_buff *skb,
1479                            const struct fib_result *res,
1480                            struct in_device *in_dev,
1481                            __be32 daddr, __be32 saddr, u32 tos)
1482 {
1483         struct rtable *rth;
1484         int err;
1485         struct in_device *out_dev;
1486         unsigned int flags = 0;
1487         bool do_cache;
1488         u32 itag = 0;
1489
1490         /* get a working reference to the output device */
1491         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1492         if (out_dev == NULL) {
1493                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1494                 return -EINVAL;
1495         }
1496
1497         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1498                                   in_dev->dev, in_dev, &itag);
1499         if (err < 0) {
1500                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1501                                          saddr);
1502
1503                 goto cleanup;
1504         }
1505
1506         do_cache = res->fi && !itag;
1507         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1508             (IN_DEV_SHARED_MEDIA(out_dev) ||
1509              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) {
1510                 flags |= RTCF_DOREDIRECT;
1511                 do_cache = false;
1512         }
1513
1514         if (skb->protocol != htons(ETH_P_IP)) {
1515                 /* Not IP (i.e. ARP). Do not create route, if it is
1516                  * invalid for proxy arp. DNAT routes are always valid.
1517                  *
1518                  * Proxy arp feature have been extended to allow, ARP
1519                  * replies back to the same interface, to support
1520                  * Private VLAN switch technologies. See arp.c.
1521                  */
1522                 if (out_dev == in_dev &&
1523                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1524                         err = -EINVAL;
1525                         goto cleanup;
1526                 }
1527         }
1528
1529         if (do_cache) {
1530                 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1531                 if (rt_cache_valid(rth)) {
1532                         skb_dst_set_noref(skb, &rth->dst);
1533                         goto out;
1534                 }
1535         }
1536
1537         rth = rt_dst_alloc(out_dev->dev,
1538                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1539                            IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1540         if (!rth) {
1541                 err = -ENOBUFS;
1542                 goto cleanup;
1543         }
1544
1545         rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
1546         rth->rt_flags = flags;
1547         rth->rt_type = res->type;
1548         rth->rt_is_input = 1;
1549         rth->rt_iif     = 0;
1550         rth->rt_pmtu    = 0;
1551         rth->rt_gateway = 0;
1552         rth->rt_uses_gateway = 0;
1553         INIT_LIST_HEAD(&rth->rt_uncached);
1554         RT_CACHE_STAT_INC(in_slow_tot);
1555
1556         rth->dst.input = ip_forward;
1557         rth->dst.output = ip_output;
1558
1559         rt_set_nexthop(rth, daddr, res, NULL, res->fi, res->type, itag);
1560         skb_dst_set(skb, &rth->dst);
1561 out:
1562         err = 0;
1563  cleanup:
1564         return err;
1565 }
1566
1567 static int ip_mkroute_input(struct sk_buff *skb,
1568                             struct fib_result *res,
1569                             const struct flowi4 *fl4,
1570                             struct in_device *in_dev,
1571                             __be32 daddr, __be32 saddr, u32 tos)
1572 {
1573 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1574         if (res->fi && res->fi->fib_nhs > 1)
1575                 fib_select_multipath(res);
1576 #endif
1577
1578         /* create a routing cache entry */
1579         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1580 }
1581
1582 /*
1583  *      NOTE. We drop all the packets that has local source
1584  *      addresses, because every properly looped back packet
1585  *      must have correct destination already attached by output routine.
1586  *
1587  *      Such approach solves two big problems:
1588  *      1. Not simplex devices are handled properly.
1589  *      2. IP spoofing attempts are filtered with 100% of guarantee.
1590  *      called with rcu_read_lock()
1591  */
1592
1593 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1594                                u8 tos, struct net_device *dev)
1595 {
1596         struct fib_result res;
1597         struct in_device *in_dev = __in_dev_get_rcu(dev);
1598         struct flowi4   fl4;
1599         unsigned int    flags = 0;
1600         u32             itag = 0;
1601         struct rtable   *rth;
1602         int             err = -EINVAL;
1603         struct net    *net = dev_net(dev);
1604         bool do_cache;
1605
1606         /* IP on this device is disabled. */
1607
1608         if (!in_dev)
1609                 goto out;
1610
1611         /* Check for the most weird martians, which can be not detected
1612            by fib_lookup.
1613          */
1614
1615         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1616                 goto martian_source;
1617
1618         res.fi = NULL;
1619         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1620                 goto brd_input;
1621
1622         /* Accept zero addresses only to limited broadcast;
1623          * I even do not know to fix it or not. Waiting for complains :-)
1624          */
1625         if (ipv4_is_zeronet(saddr))
1626                 goto martian_source;
1627
1628         if (ipv4_is_zeronet(daddr))
1629                 goto martian_destination;
1630
1631         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1632          * and call it once if daddr or/and saddr are loopback addresses
1633          */
1634         if (ipv4_is_loopback(daddr)) {
1635                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1636                         goto martian_destination;
1637         } else if (ipv4_is_loopback(saddr)) {
1638                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1639                         goto martian_source;
1640         }
1641
1642         /*
1643          *      Now we are ready to route packet.
1644          */
1645         fl4.flowi4_oif = 0;
1646         fl4.flowi4_iif = dev->ifindex;
1647         fl4.flowi4_mark = skb->mark;
1648         fl4.flowi4_tos = tos;
1649         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1650         fl4.daddr = daddr;
1651         fl4.saddr = saddr;
1652         err = fib_lookup(net, &fl4, &res);
1653         if (err != 0)
1654                 goto no_route;
1655
1656         if (res.type == RTN_BROADCAST)
1657                 goto brd_input;
1658
1659         if (res.type == RTN_LOCAL) {
1660                 err = fib_validate_source(skb, saddr, daddr, tos,
1661                                           LOOPBACK_IFINDEX,
1662                                           dev, in_dev, &itag);
1663                 if (err < 0)
1664                         goto martian_source_keep_err;
1665                 goto local_input;
1666         }
1667
1668         if (!IN_DEV_FORWARD(in_dev))
1669                 goto no_route;
1670         if (res.type != RTN_UNICAST)
1671                 goto martian_destination;
1672
1673         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1674 out:    return err;
1675
1676 brd_input:
1677         if (skb->protocol != htons(ETH_P_IP))
1678                 goto e_inval;
1679
1680         if (!ipv4_is_zeronet(saddr)) {
1681                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1682                                           in_dev, &itag);
1683                 if (err < 0)
1684                         goto martian_source_keep_err;
1685         }
1686         flags |= RTCF_BROADCAST;
1687         res.type = RTN_BROADCAST;
1688         RT_CACHE_STAT_INC(in_brd);
1689
1690 local_input:
1691         do_cache = false;
1692         if (res.fi) {
1693                 if (!itag) {
1694                         rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1695                         if (rt_cache_valid(rth)) {
1696                                 skb_dst_set_noref(skb, &rth->dst);
1697                                 err = 0;
1698                                 goto out;
1699                         }
1700                         do_cache = true;
1701                 }
1702         }
1703
1704         rth = rt_dst_alloc(net->loopback_dev,
1705                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1706         if (!rth)
1707                 goto e_nobufs;
1708
1709         rth->dst.input= ip_local_deliver;
1710         rth->dst.output= ip_rt_bug;
1711 #ifdef CONFIG_IP_ROUTE_CLASSID
1712         rth->dst.tclassid = itag;
1713 #endif
1714
1715         rth->rt_genid = rt_genid(net);
1716         rth->rt_flags   = flags|RTCF_LOCAL;
1717         rth->rt_type    = res.type;
1718         rth->rt_is_input = 1;
1719         rth->rt_iif     = 0;
1720         rth->rt_pmtu    = 0;
1721         rth->rt_gateway = 0;
1722         rth->rt_uses_gateway = 0;
1723         INIT_LIST_HEAD(&rth->rt_uncached);
1724         RT_CACHE_STAT_INC(in_slow_tot);
1725         if (res.type == RTN_UNREACHABLE) {
1726                 rth->dst.input= ip_error;
1727                 rth->dst.error= -err;
1728                 rth->rt_flags   &= ~RTCF_LOCAL;
1729         }
1730         if (do_cache) {
1731                 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1732                         rth->dst.flags |= DST_NOCACHE;
1733                         rt_add_uncached_list(rth);
1734                 }
1735         }
1736         skb_dst_set(skb, &rth->dst);
1737         err = 0;
1738         goto out;
1739
1740 no_route:
1741         RT_CACHE_STAT_INC(in_no_route);
1742         res.type = RTN_UNREACHABLE;
1743         if (err == -ESRCH)
1744                 err = -ENETUNREACH;
1745         goto local_input;
1746
1747         /*
1748          *      Do not cache martian addresses: they should be logged (RFC1812)
1749          */
1750 martian_destination:
1751         RT_CACHE_STAT_INC(in_martian_dst);
1752 #ifdef CONFIG_IP_ROUTE_VERBOSE
1753         if (IN_DEV_LOG_MARTIANS(in_dev))
1754                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1755                                      &daddr, &saddr, dev->name);
1756 #endif
1757
1758 e_inval:
1759         err = -EINVAL;
1760         goto out;
1761
1762 e_nobufs:
1763         err = -ENOBUFS;
1764         goto out;
1765
1766 martian_source:
1767         err = -EINVAL;
1768 martian_source_keep_err:
1769         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1770         goto out;
1771 }
1772
1773 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1774                          u8 tos, struct net_device *dev)
1775 {
1776         int res;
1777
1778         rcu_read_lock();
1779
1780         /* Multicast recognition logic is moved from route cache to here.
1781            The problem was that too many Ethernet cards have broken/missing
1782            hardware multicast filters :-( As result the host on multicasting
1783            network acquires a lot of useless route cache entries, sort of
1784            SDR messages from all the world. Now we try to get rid of them.
1785            Really, provided software IP multicast filter is organized
1786            reasonably (at least, hashed), it does not result in a slowdown
1787            comparing with route cache reject entries.
1788            Note, that multicast routers are not affected, because
1789            route cache entry is created eventually.
1790          */
1791         if (ipv4_is_multicast(daddr)) {
1792                 struct in_device *in_dev = __in_dev_get_rcu(dev);
1793
1794                 if (in_dev) {
1795                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1796                                                   ip_hdr(skb)->protocol);
1797                         if (our
1798 #ifdef CONFIG_IP_MROUTE
1799                                 ||
1800                             (!ipv4_is_local_multicast(daddr) &&
1801                              IN_DEV_MFORWARD(in_dev))
1802 #endif
1803                            ) {
1804                                 int res = ip_route_input_mc(skb, daddr, saddr,
1805                                                             tos, dev, our);
1806                                 rcu_read_unlock();
1807                                 return res;
1808                         }
1809                 }
1810                 rcu_read_unlock();
1811                 return -EINVAL;
1812         }
1813         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1814         rcu_read_unlock();
1815         return res;
1816 }
1817 EXPORT_SYMBOL(ip_route_input_noref);
1818
1819 /* called with rcu_read_lock() */
1820 static struct rtable *__mkroute_output(const struct fib_result *res,
1821                                        const struct flowi4 *fl4, int orig_oif,
1822                                        struct net_device *dev_out,
1823                                        unsigned int flags)
1824 {
1825         struct fib_info *fi = res->fi;
1826         struct fib_nh_exception *fnhe;
1827         struct in_device *in_dev;
1828         u16 type = res->type;
1829         struct rtable *rth;
1830         bool do_cache;
1831
1832         in_dev = __in_dev_get_rcu(dev_out);
1833         if (!in_dev)
1834                 return ERR_PTR(-EINVAL);
1835
1836         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1837                 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1838                         return ERR_PTR(-EINVAL);
1839
1840         if (ipv4_is_lbcast(fl4->daddr))
1841                 type = RTN_BROADCAST;
1842         else if (ipv4_is_multicast(fl4->daddr))
1843                 type = RTN_MULTICAST;
1844         else if (ipv4_is_zeronet(fl4->daddr))
1845                 return ERR_PTR(-EINVAL);
1846
1847         if (dev_out->flags & IFF_LOOPBACK)
1848                 flags |= RTCF_LOCAL;
1849
1850         do_cache = true;
1851         if (type == RTN_BROADCAST) {
1852                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1853                 fi = NULL;
1854         } else if (type == RTN_MULTICAST) {
1855                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1856                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1857                                      fl4->flowi4_proto))
1858                         flags &= ~RTCF_LOCAL;
1859                 else
1860                         do_cache = false;
1861                 /* If multicast route do not exist use
1862                  * default one, but do not gateway in this case.
1863                  * Yes, it is hack.
1864                  */
1865                 if (fi && res->prefixlen < 4)
1866                         fi = NULL;
1867         }
1868
1869         fnhe = NULL;
1870         do_cache &= fi != NULL;
1871         if (do_cache) {
1872                 struct rtable __rcu **prth;
1873                 struct fib_nh *nh = &FIB_RES_NH(*res);
1874
1875                 fnhe = find_exception(nh, fl4->daddr);
1876                 if (fnhe)
1877                         prth = &fnhe->fnhe_rth;
1878                 else {
1879                         if (unlikely(fl4->flowi4_flags &
1880                                      FLOWI_FLAG_KNOWN_NH &&
1881                                      !(nh->nh_gw &&
1882                                        nh->nh_scope == RT_SCOPE_LINK))) {
1883                                 do_cache = false;
1884                                 goto add;
1885                         }
1886                         prth = __this_cpu_ptr(nh->nh_pcpu_rth_output);
1887                 }
1888                 rth = rcu_dereference(*prth);
1889                 if (rt_cache_valid(rth)) {
1890                         dst_hold(&rth->dst);
1891                         return rth;
1892                 }
1893         }
1894
1895 add:
1896         rth = rt_dst_alloc(dev_out,
1897                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1898                            IN_DEV_CONF_GET(in_dev, NOXFRM),
1899                            do_cache);
1900         if (!rth)
1901                 return ERR_PTR(-ENOBUFS);
1902
1903         rth->dst.output = ip_output;
1904
1905         rth->rt_genid = rt_genid(dev_net(dev_out));
1906         rth->rt_flags   = flags;
1907         rth->rt_type    = type;
1908         rth->rt_is_input = 0;
1909         rth->rt_iif     = orig_oif ? : 0;
1910         rth->rt_pmtu    = 0;
1911         rth->rt_gateway = 0;
1912         rth->rt_uses_gateway = 0;
1913         INIT_LIST_HEAD(&rth->rt_uncached);
1914
1915         RT_CACHE_STAT_INC(out_slow_tot);
1916
1917         if (flags & RTCF_LOCAL)
1918                 rth->dst.input = ip_local_deliver;
1919         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1920                 if (flags & RTCF_LOCAL &&
1921                     !(dev_out->flags & IFF_LOOPBACK)) {
1922                         rth->dst.output = ip_mc_output;
1923                         RT_CACHE_STAT_INC(out_slow_mc);
1924                 }
1925 #ifdef CONFIG_IP_MROUTE
1926                 if (type == RTN_MULTICAST) {
1927                         if (IN_DEV_MFORWARD(in_dev) &&
1928                             !ipv4_is_local_multicast(fl4->daddr)) {
1929                                 rth->dst.input = ip_mr_input;
1930                                 rth->dst.output = ip_mc_output;
1931                         }
1932                 }
1933 #endif
1934         }
1935
1936         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
1937
1938         return rth;
1939 }
1940
1941 /*
1942  * Major route resolver routine.
1943  */
1944
1945 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
1946 {
1947         struct net_device *dev_out = NULL;
1948         __u8 tos = RT_FL_TOS(fl4);
1949         unsigned int flags = 0;
1950         struct fib_result res;
1951         struct rtable *rth;
1952         int orig_oif;
1953
1954         res.tclassid    = 0;
1955         res.fi          = NULL;
1956         res.table       = NULL;
1957
1958         orig_oif = fl4->flowi4_oif;
1959
1960         fl4->flowi4_iif = LOOPBACK_IFINDEX;
1961         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
1962         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
1963                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
1964
1965         rcu_read_lock();
1966         if (fl4->saddr) {
1967                 rth = ERR_PTR(-EINVAL);
1968                 if (ipv4_is_multicast(fl4->saddr) ||
1969                     ipv4_is_lbcast(fl4->saddr) ||
1970                     ipv4_is_zeronet(fl4->saddr))
1971                         goto out;
1972
1973                 /* I removed check for oif == dev_out->oif here.
1974                    It was wrong for two reasons:
1975                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
1976                       is assigned to multiple interfaces.
1977                    2. Moreover, we are allowed to send packets with saddr
1978                       of another iface. --ANK
1979                  */
1980
1981                 if (fl4->flowi4_oif == 0 &&
1982                     (ipv4_is_multicast(fl4->daddr) ||
1983                      ipv4_is_lbcast(fl4->daddr))) {
1984                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1985                         dev_out = __ip_dev_find(net, fl4->saddr, false);
1986                         if (dev_out == NULL)
1987                                 goto out;
1988
1989                         /* Special hack: user can direct multicasts
1990                            and limited broadcast via necessary interface
1991                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
1992                            This hack is not just for fun, it allows
1993                            vic,vat and friends to work.
1994                            They bind socket to loopback, set ttl to zero
1995                            and expect that it will work.
1996                            From the viewpoint of routing cache they are broken,
1997                            because we are not allowed to build multicast path
1998                            with loopback source addr (look, routing cache
1999                            cannot know, that ttl is zero, so that packet
2000                            will not leave this host and route is valid).
2001                            Luckily, this hack is good workaround.
2002                          */
2003
2004                         fl4->flowi4_oif = dev_out->ifindex;
2005                         goto make_route;
2006                 }
2007
2008                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2009                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2010                         if (!__ip_dev_find(net, fl4->saddr, false))
2011                                 goto out;
2012                 }
2013         }
2014
2015
2016         if (fl4->flowi4_oif) {
2017                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2018                 rth = ERR_PTR(-ENODEV);
2019                 if (dev_out == NULL)
2020                         goto out;
2021
2022                 /* RACE: Check return value of inet_select_addr instead. */
2023                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2024                         rth = ERR_PTR(-ENETUNREACH);
2025                         goto out;
2026                 }
2027                 if (ipv4_is_local_multicast(fl4->daddr) ||
2028                     ipv4_is_lbcast(fl4->daddr)) {
2029                         if (!fl4->saddr)
2030                                 fl4->saddr = inet_select_addr(dev_out, 0,
2031                                                               RT_SCOPE_LINK);
2032                         goto make_route;
2033                 }
2034                 if (!fl4->saddr) {
2035                         if (ipv4_is_multicast(fl4->daddr))
2036                                 fl4->saddr = inet_select_addr(dev_out, 0,
2037                                                               fl4->flowi4_scope);
2038                         else if (!fl4->daddr)
2039                                 fl4->saddr = inet_select_addr(dev_out, 0,
2040                                                               RT_SCOPE_HOST);
2041                 }
2042         }
2043
2044         if (!fl4->daddr) {
2045                 fl4->daddr = fl4->saddr;
2046                 if (!fl4->daddr)
2047                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2048                 dev_out = net->loopback_dev;
2049                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2050                 res.type = RTN_LOCAL;
2051                 flags |= RTCF_LOCAL;
2052                 goto make_route;
2053         }
2054
2055         if (fib_lookup(net, fl4, &res)) {
2056                 res.fi = NULL;
2057                 res.table = NULL;
2058                 if (fl4->flowi4_oif) {
2059                         /* Apparently, routing tables are wrong. Assume,
2060                            that the destination is on link.
2061
2062                            WHY? DW.
2063                            Because we are allowed to send to iface
2064                            even if it has NO routes and NO assigned
2065                            addresses. When oif is specified, routing
2066                            tables are looked up with only one purpose:
2067                            to catch if destination is gatewayed, rather than
2068                            direct. Moreover, if MSG_DONTROUTE is set,
2069                            we send packet, ignoring both routing tables
2070                            and ifaddr state. --ANK
2071
2072
2073                            We could make it even if oif is unknown,
2074                            likely IPv6, but we do not.
2075                          */
2076
2077                         if (fl4->saddr == 0)
2078                                 fl4->saddr = inet_select_addr(dev_out, 0,
2079                                                               RT_SCOPE_LINK);
2080                         res.type = RTN_UNICAST;
2081                         goto make_route;
2082                 }
2083                 rth = ERR_PTR(-ENETUNREACH);
2084                 goto out;
2085         }
2086
2087         if (res.type == RTN_LOCAL) {
2088                 if (!fl4->saddr) {
2089                         if (res.fi->fib_prefsrc)
2090                                 fl4->saddr = res.fi->fib_prefsrc;
2091                         else
2092                                 fl4->saddr = fl4->daddr;
2093                 }
2094                 dev_out = net->loopback_dev;
2095                 fl4->flowi4_oif = dev_out->ifindex;
2096                 flags |= RTCF_LOCAL;
2097                 goto make_route;
2098         }
2099
2100 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2101         if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2102                 fib_select_multipath(&res);
2103         else
2104 #endif
2105         if (!res.prefixlen &&
2106             res.table->tb_num_default > 1 &&
2107             res.type == RTN_UNICAST && !fl4->flowi4_oif)
2108                 fib_select_default(&res);
2109
2110         if (!fl4->saddr)
2111                 fl4->saddr = FIB_RES_PREFSRC(net, res);
2112
2113         dev_out = FIB_RES_DEV(res);
2114         fl4->flowi4_oif = dev_out->ifindex;
2115
2116
2117 make_route:
2118         rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2119
2120 out:
2121         rcu_read_unlock();
2122         return rth;
2123 }
2124 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2125
2126 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2127 {
2128         return NULL;
2129 }
2130
2131 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2132 {
2133         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2134
2135         return mtu ? : dst->dev->mtu;
2136 }
2137
2138 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2139                                           struct sk_buff *skb, u32 mtu)
2140 {
2141 }
2142
2143 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2144                                        struct sk_buff *skb)
2145 {
2146 }
2147
2148 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2149                                           unsigned long old)
2150 {
2151         return NULL;
2152 }
2153
2154 static struct dst_ops ipv4_dst_blackhole_ops = {
2155         .family                 =       AF_INET,
2156         .protocol               =       cpu_to_be16(ETH_P_IP),
2157         .check                  =       ipv4_blackhole_dst_check,
2158         .mtu                    =       ipv4_blackhole_mtu,
2159         .default_advmss         =       ipv4_default_advmss,
2160         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2161         .redirect               =       ipv4_rt_blackhole_redirect,
2162         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
2163         .neigh_lookup           =       ipv4_neigh_lookup,
2164 };
2165
2166 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2167 {
2168         struct rtable *ort = (struct rtable *) dst_orig;
2169         struct rtable *rt;
2170
2171         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2172         if (rt) {
2173                 struct dst_entry *new = &rt->dst;
2174
2175                 new->__use = 1;
2176                 new->input = dst_discard;
2177                 new->output = dst_discard;
2178
2179                 new->dev = ort->dst.dev;
2180                 if (new->dev)
2181                         dev_hold(new->dev);
2182
2183                 rt->rt_is_input = ort->rt_is_input;
2184                 rt->rt_iif = ort->rt_iif;
2185                 rt->rt_pmtu = ort->rt_pmtu;
2186
2187                 rt->rt_genid = rt_genid(net);
2188                 rt->rt_flags = ort->rt_flags;
2189                 rt->rt_type = ort->rt_type;
2190                 rt->rt_gateway = ort->rt_gateway;
2191                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2192
2193                 INIT_LIST_HEAD(&rt->rt_uncached);
2194
2195                 dst_free(new);
2196         }
2197
2198         dst_release(dst_orig);
2199
2200         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2201 }
2202
2203 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2204                                     struct sock *sk)
2205 {
2206         struct rtable *rt = __ip_route_output_key(net, flp4);
2207
2208         if (IS_ERR(rt))
2209                 return rt;
2210
2211         if (flp4->flowi4_proto)
2212                 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2213                                                    flowi4_to_flowi(flp4),
2214                                                    sk, 0);
2215
2216         return rt;
2217 }
2218 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2219
2220 static int rt_fill_info(struct net *net,  __be32 dst, __be32 src,
2221                         struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2222                         u32 seq, int event, int nowait, unsigned int flags)
2223 {
2224         struct rtable *rt = skb_rtable(skb);
2225         struct rtmsg *r;
2226         struct nlmsghdr *nlh;
2227         unsigned long expires = 0;
2228         u32 error;
2229         u32 metrics[RTAX_MAX];
2230
2231         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2232         if (nlh == NULL)
2233                 return -EMSGSIZE;
2234
2235         r = nlmsg_data(nlh);
2236         r->rtm_family    = AF_INET;
2237         r->rtm_dst_len  = 32;
2238         r->rtm_src_len  = 0;
2239         r->rtm_tos      = fl4->flowi4_tos;
2240         r->rtm_table    = RT_TABLE_MAIN;
2241         if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2242                 goto nla_put_failure;
2243         r->rtm_type     = rt->rt_type;
2244         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2245         r->rtm_protocol = RTPROT_UNSPEC;
2246         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2247         if (rt->rt_flags & RTCF_NOTIFY)
2248                 r->rtm_flags |= RTM_F_NOTIFY;
2249
2250         if (nla_put_be32(skb, RTA_DST, dst))
2251                 goto nla_put_failure;
2252         if (src) {
2253                 r->rtm_src_len = 32;
2254                 if (nla_put_be32(skb, RTA_SRC, src))
2255                         goto nla_put_failure;
2256         }
2257         if (rt->dst.dev &&
2258             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2259                 goto nla_put_failure;
2260 #ifdef CONFIG_IP_ROUTE_CLASSID
2261         if (rt->dst.tclassid &&
2262             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2263                 goto nla_put_failure;
2264 #endif
2265         if (!rt_is_input_route(rt) &&
2266             fl4->saddr != src) {
2267                 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2268                         goto nla_put_failure;
2269         }
2270         if (rt->rt_uses_gateway &&
2271             nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2272                 goto nla_put_failure;
2273
2274         expires = rt->dst.expires;
2275         if (expires) {
2276                 unsigned long now = jiffies;
2277
2278                 if (time_before(now, expires))
2279                         expires -= now;
2280                 else
2281                         expires = 0;
2282         }
2283
2284         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2285         if (rt->rt_pmtu && expires)
2286                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2287         if (rtnetlink_put_metrics(skb, metrics) < 0)
2288                 goto nla_put_failure;
2289
2290         if (fl4->flowi4_mark &&
2291             nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2292                 goto nla_put_failure;
2293
2294         error = rt->dst.error;
2295
2296         if (rt_is_input_route(rt)) {
2297 #ifdef CONFIG_IP_MROUTE
2298                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2299                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2300                         int err = ipmr_get_route(net, skb,
2301                                                  fl4->saddr, fl4->daddr,
2302                                                  r, nowait);
2303                         if (err <= 0) {
2304                                 if (!nowait) {
2305                                         if (err == 0)
2306                                                 return 0;
2307                                         goto nla_put_failure;
2308                                 } else {
2309                                         if (err == -EMSGSIZE)
2310                                                 goto nla_put_failure;
2311                                         error = err;
2312                                 }
2313                         }
2314                 } else
2315 #endif
2316                         if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2317                                 goto nla_put_failure;
2318         }
2319
2320         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2321                 goto nla_put_failure;
2322
2323         return nlmsg_end(skb, nlh);
2324
2325 nla_put_failure:
2326         nlmsg_cancel(skb, nlh);
2327         return -EMSGSIZE;
2328 }
2329
2330 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2331 {
2332         struct net *net = sock_net(in_skb->sk);
2333         struct rtmsg *rtm;
2334         struct nlattr *tb[RTA_MAX+1];
2335         struct rtable *rt = NULL;
2336         struct flowi4 fl4;
2337         __be32 dst = 0;
2338         __be32 src = 0;
2339         u32 iif;
2340         int err;
2341         int mark;
2342         struct sk_buff *skb;
2343
2344         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2345         if (err < 0)
2346                 goto errout;
2347
2348         rtm = nlmsg_data(nlh);
2349
2350         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2351         if (skb == NULL) {
2352                 err = -ENOBUFS;
2353                 goto errout;
2354         }
2355
2356         /* Reserve room for dummy headers, this skb can pass
2357            through good chunk of routing engine.
2358          */
2359         skb_reset_mac_header(skb);
2360         skb_reset_network_header(skb);
2361
2362         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2363         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2364         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2365
2366         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2367         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2368         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2369         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2370
2371         memset(&fl4, 0, sizeof(fl4));
2372         fl4.daddr = dst;
2373         fl4.saddr = src;
2374         fl4.flowi4_tos = rtm->rtm_tos;
2375         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2376         fl4.flowi4_mark = mark;
2377
2378         if (iif) {
2379                 struct net_device *dev;
2380
2381                 dev = __dev_get_by_index(net, iif);
2382                 if (dev == NULL) {
2383                         err = -ENODEV;
2384                         goto errout_free;
2385                 }
2386
2387                 skb->protocol   = htons(ETH_P_IP);
2388                 skb->dev        = dev;
2389                 skb->mark       = mark;
2390                 local_bh_disable();
2391                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2392                 local_bh_enable();
2393
2394                 rt = skb_rtable(skb);
2395                 if (err == 0 && rt->dst.error)
2396                         err = -rt->dst.error;
2397         } else {
2398                 rt = ip_route_output_key(net, &fl4);
2399
2400                 err = 0;
2401                 if (IS_ERR(rt))
2402                         err = PTR_ERR(rt);
2403         }
2404
2405         if (err)
2406                 goto errout_free;
2407
2408         skb_dst_set(skb, &rt->dst);
2409         if (rtm->rtm_flags & RTM_F_NOTIFY)
2410                 rt->rt_flags |= RTCF_NOTIFY;
2411
2412         err = rt_fill_info(net, dst, src, &fl4, skb,
2413                            NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2414                            RTM_NEWROUTE, 0, 0);
2415         if (err <= 0)
2416                 goto errout_free;
2417
2418         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2419 errout:
2420         return err;
2421
2422 errout_free:
2423         kfree_skb(skb);
2424         goto errout;
2425 }
2426
2427 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
2428 {
2429         return skb->len;
2430 }
2431
2432 void ip_rt_multicast_event(struct in_device *in_dev)
2433 {
2434         rt_cache_flush(dev_net(in_dev->dev));
2435 }
2436
2437 #ifdef CONFIG_SYSCTL
2438 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
2439 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
2440 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
2441 static int ip_rt_gc_elasticity __read_mostly    = 8;
2442
2443 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
2444                                         void __user *buffer,
2445                                         size_t *lenp, loff_t *ppos)
2446 {
2447         if (write) {
2448                 rt_cache_flush((struct net *)__ctl->extra1);
2449                 return 0;
2450         }
2451
2452         return -EINVAL;
2453 }
2454
2455 static ctl_table ipv4_route_table[] = {
2456         {
2457                 .procname       = "gc_thresh",
2458                 .data           = &ipv4_dst_ops.gc_thresh,
2459                 .maxlen         = sizeof(int),
2460                 .mode           = 0644,
2461                 .proc_handler   = proc_dointvec,
2462         },
2463         {
2464                 .procname       = "max_size",
2465                 .data           = &ip_rt_max_size,
2466                 .maxlen         = sizeof(int),
2467                 .mode           = 0644,
2468                 .proc_handler   = proc_dointvec,
2469         },
2470         {
2471                 /*  Deprecated. Use gc_min_interval_ms */
2472
2473                 .procname       = "gc_min_interval",
2474                 .data           = &ip_rt_gc_min_interval,
2475                 .maxlen         = sizeof(int),
2476                 .mode           = 0644,
2477                 .proc_handler   = proc_dointvec_jiffies,
2478         },
2479         {
2480                 .procname       = "gc_min_interval_ms",
2481                 .data           = &ip_rt_gc_min_interval,
2482                 .maxlen         = sizeof(int),
2483                 .mode           = 0644,
2484                 .proc_handler   = proc_dointvec_ms_jiffies,
2485         },
2486         {
2487                 .procname       = "gc_timeout",
2488                 .data           = &ip_rt_gc_timeout,
2489                 .maxlen         = sizeof(int),
2490                 .mode           = 0644,
2491                 .proc_handler   = proc_dointvec_jiffies,
2492         },
2493         {
2494                 .procname       = "gc_interval",
2495                 .data           = &ip_rt_gc_interval,
2496                 .maxlen         = sizeof(int),
2497                 .mode           = 0644,
2498                 .proc_handler   = proc_dointvec_jiffies,
2499         },
2500         {
2501                 .procname       = "redirect_load",
2502                 .data           = &ip_rt_redirect_load,
2503                 .maxlen         = sizeof(int),
2504                 .mode           = 0644,
2505                 .proc_handler   = proc_dointvec,
2506         },
2507         {
2508                 .procname       = "redirect_number",
2509                 .data           = &ip_rt_redirect_number,
2510                 .maxlen         = sizeof(int),
2511                 .mode           = 0644,
2512                 .proc_handler   = proc_dointvec,
2513         },
2514         {
2515                 .procname       = "redirect_silence",
2516                 .data           = &ip_rt_redirect_silence,
2517                 .maxlen         = sizeof(int),
2518                 .mode           = 0644,
2519                 .proc_handler   = proc_dointvec,
2520         },
2521         {
2522                 .procname       = "error_cost",
2523                 .data           = &ip_rt_error_cost,
2524                 .maxlen         = sizeof(int),
2525                 .mode           = 0644,
2526                 .proc_handler   = proc_dointvec,
2527         },
2528         {
2529                 .procname       = "error_burst",
2530                 .data           = &ip_rt_error_burst,
2531                 .maxlen         = sizeof(int),
2532                 .mode           = 0644,
2533                 .proc_handler   = proc_dointvec,
2534         },
2535         {
2536                 .procname       = "gc_elasticity",
2537                 .data           = &ip_rt_gc_elasticity,
2538                 .maxlen         = sizeof(int),
2539                 .mode           = 0644,
2540                 .proc_handler   = proc_dointvec,
2541         },
2542         {
2543                 .procname       = "mtu_expires",
2544                 .data           = &ip_rt_mtu_expires,
2545                 .maxlen         = sizeof(int),
2546                 .mode           = 0644,
2547                 .proc_handler   = proc_dointvec_jiffies,
2548         },
2549         {
2550                 .procname       = "min_pmtu",
2551                 .data           = &ip_rt_min_pmtu,
2552                 .maxlen         = sizeof(int),
2553                 .mode           = 0644,
2554                 .proc_handler   = proc_dointvec,
2555         },
2556         {
2557                 .procname       = "min_adv_mss",
2558                 .data           = &ip_rt_min_advmss,
2559                 .maxlen         = sizeof(int),
2560                 .mode           = 0644,
2561                 .proc_handler   = proc_dointvec,
2562         },
2563         { }
2564 };
2565
2566 static struct ctl_table ipv4_route_flush_table[] = {
2567         {
2568                 .procname       = "flush",
2569                 .maxlen         = sizeof(int),
2570                 .mode           = 0200,
2571                 .proc_handler   = ipv4_sysctl_rtcache_flush,
2572         },
2573         { },
2574 };
2575
2576 static __net_init int sysctl_route_net_init(struct net *net)
2577 {
2578         struct ctl_table *tbl;
2579
2580         tbl = ipv4_route_flush_table;
2581         if (!net_eq(net, &init_net)) {
2582                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2583                 if (tbl == NULL)
2584                         goto err_dup;
2585
2586                 /* Don't export sysctls to unprivileged users */
2587                 if (net->user_ns != &init_user_ns)
2588                         tbl[0].procname = NULL;
2589         }
2590         tbl[0].extra1 = net;
2591
2592         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2593         if (net->ipv4.route_hdr == NULL)
2594                 goto err_reg;
2595         return 0;
2596
2597 err_reg:
2598         if (tbl != ipv4_route_flush_table)
2599                 kfree(tbl);
2600 err_dup:
2601         return -ENOMEM;
2602 }
2603
2604 static __net_exit void sysctl_route_net_exit(struct net *net)
2605 {
2606         struct ctl_table *tbl;
2607
2608         tbl = net->ipv4.route_hdr->ctl_table_arg;
2609         unregister_net_sysctl_table(net->ipv4.route_hdr);
2610         BUG_ON(tbl == ipv4_route_flush_table);
2611         kfree(tbl);
2612 }
2613
2614 static __net_initdata struct pernet_operations sysctl_route_ops = {
2615         .init = sysctl_route_net_init,
2616         .exit = sysctl_route_net_exit,
2617 };
2618 #endif
2619
2620 static __net_init int rt_genid_init(struct net *net)
2621 {
2622         atomic_set(&net->rt_genid, 0);
2623         get_random_bytes(&net->ipv4.dev_addr_genid,
2624                          sizeof(net->ipv4.dev_addr_genid));
2625         return 0;
2626 }
2627
2628 static __net_initdata struct pernet_operations rt_genid_ops = {
2629         .init = rt_genid_init,
2630 };
2631
2632 static int __net_init ipv4_inetpeer_init(struct net *net)
2633 {
2634         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2635
2636         if (!bp)
2637                 return -ENOMEM;
2638         inet_peer_base_init(bp);
2639         net->ipv4.peers = bp;
2640         return 0;
2641 }
2642
2643 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2644 {
2645         struct inet_peer_base *bp = net->ipv4.peers;
2646
2647         net->ipv4.peers = NULL;
2648         inetpeer_invalidate_tree(bp);
2649         kfree(bp);
2650 }
2651
2652 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2653         .init   =       ipv4_inetpeer_init,
2654         .exit   =       ipv4_inetpeer_exit,
2655 };
2656
2657 #ifdef CONFIG_IP_ROUTE_CLASSID
2658 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2659 #endif /* CONFIG_IP_ROUTE_CLASSID */
2660
2661 int __init ip_rt_init(void)
2662 {
2663         int rc = 0;
2664
2665 #ifdef CONFIG_IP_ROUTE_CLASSID
2666         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2667         if (!ip_rt_acct)
2668                 panic("IP: failed to allocate ip_rt_acct\n");
2669 #endif
2670
2671         ipv4_dst_ops.kmem_cachep =
2672                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2673                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2674
2675         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2676
2677         if (dst_entries_init(&ipv4_dst_ops) < 0)
2678                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2679
2680         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2681                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2682
2683         ipv4_dst_ops.gc_thresh = ~0;
2684         ip_rt_max_size = INT_MAX;
2685
2686         devinet_init();
2687         ip_fib_init();
2688
2689         if (ip_rt_proc_init())
2690                 pr_err("Unable to create route proc files\n");
2691 #ifdef CONFIG_XFRM
2692         xfrm_init();
2693         xfrm4_init();
2694 #endif
2695         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2696
2697 #ifdef CONFIG_SYSCTL
2698         register_pernet_subsys(&sysctl_route_ops);
2699 #endif
2700         register_pernet_subsys(&rt_genid_ops);
2701         register_pernet_subsys(&ipv4_inetpeer_ops);
2702         return rc;
2703 }
2704
2705 #ifdef CONFIG_SYSCTL
2706 /*
2707  * We really need to sanitize the damn ipv4 init order, then all
2708  * this nonsense will go away.
2709  */
2710 void __init ip_static_sysctl_init(void)
2711 {
2712         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
2713 }
2714 #endif