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 RT_GC_TIMEOUT (300*HZ)
 116
 117 static int ip_rt_max_size;
 118 static int ip_rt_redirect_number __read_mostly  = 9;
 119 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 120 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 121 static int ip_rt_error_cost __read_mostly       = HZ;
 122 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 123 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
 124 static int ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
 125 static int ip_rt_min_advmss __read_mostly       = 256;
 126
 127 /*
 128  *      Interface to generic destination cache.
 129  */
 130
 131 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 132 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 133 static unsigned int      ipv4_mtu(const struct dst_entry *dst);
 134 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 135 static void              ipv4_link_failure(struct sk_buff *skb);
 136 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 137                                            struct sk_buff *skb, u32 mtu);
 138 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 139                                         struct sk_buff *skb);
 140 static void             ipv4_dst_destroy(struct dst_entry *dst);
 141
 142 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 143                             int how)
 144 {
 145 }
 146
 147 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 148 {
 149         WARN_ON(1);
 150         return NULL;
 151 }
 152
 153 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 154                                            struct sk_buff *skb,
 155                                            const void *daddr);
 156
 157 static struct dst_ops ipv4_dst_ops = {
 158         .family =               AF_INET,
 159         .protocol =             cpu_to_be16(ETH_P_IP),
 160         .check =                ipv4_dst_check,
 161         .default_advmss =       ipv4_default_advmss,
 162         .mtu =                  ipv4_mtu,
 163         .cow_metrics =          ipv4_cow_metrics,
 164         .destroy =              ipv4_dst_destroy,
 165         .ifdown =               ipv4_dst_ifdown,
 166         .negative_advice =      ipv4_negative_advice,
 167         .link_failure =         ipv4_link_failure,
 168         .update_pmtu =          ip_rt_update_pmtu,
 169         .redirect =             ip_do_redirect,
 170         .local_out =            __ip_local_out,
 171         .neigh_lookup =         ipv4_neigh_lookup,
 172 };
 173
 174 #define ECN_OR_COST(class)      TC_PRIO_##class
 175
 176 const __u8 ip_tos2prio[16] = {
 177         TC_PRIO_BESTEFFORT,
 178         ECN_OR_COST(BESTEFFORT),
 179         TC_PRIO_BESTEFFORT,
 180         ECN_OR_COST(BESTEFFORT),
 181         TC_PRIO_BULK,
 182         ECN_OR_COST(BULK),
 183         TC_PRIO_BULK,
 184         ECN_OR_COST(BULK),
 185         TC_PRIO_INTERACTIVE,
 186         ECN_OR_COST(INTERACTIVE),
 187         TC_PRIO_INTERACTIVE,
 188         ECN_OR_COST(INTERACTIVE),
 189         TC_PRIO_INTERACTIVE_BULK,
 190         ECN_OR_COST(INTERACTIVE_BULK),
 191         TC_PRIO_INTERACTIVE_BULK,
 192         ECN_OR_COST(INTERACTIVE_BULK)
 193 };
 194 EXPORT_SYMBOL(ip_tos2prio);
 195
 196 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 197 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
 198
 199 #ifdef CONFIG_PROC_FS
 200 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 201 {
 202         if (*pos)
 203                 return NULL;
 204         return SEQ_START_TOKEN;
 205 }
 206
 207 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 208 {
 209         ++*pos;
 210         return NULL;
 211 }
 212
 213 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 214 {
 215 }
 216
 217 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 218 {
 219         if (v == SEQ_START_TOKEN)
 220                 seq_printf(seq, "%-127s\n",
 221                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 222                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 223                            "HHUptod\tSpecDst");
 224         return 0;
 225 }
 226
 227 static const struct seq_operations rt_cache_seq_ops = {
 228         .start  = rt_cache_seq_start,
 229         .next   = rt_cache_seq_next,
 230         .stop   = rt_cache_seq_stop,
 231         .show   = rt_cache_seq_show,
 232 };
 233
 234 static int rt_cache_seq_open(struct inode *inode, struct file *file)
 235 {
 236         return seq_open(file, &rt_cache_seq_ops);
 237 }
 238
 239 static const struct file_operations rt_cache_seq_fops = {
 240         .owner   = THIS_MODULE,
 241         .open    = rt_cache_seq_open,
 242         .read    = seq_read,
 243         .llseek  = seq_lseek,
 244         .release = seq_release,
 245 };
 246
 247
 248 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 249 {
 250         int cpu;
 251
 252         if (*pos == 0)
 253                 return SEQ_START_TOKEN;
 254
 255         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 256                 if (!cpu_possible(cpu))
 257                         continue;
 258                 *pos = cpu+1;
 259                 return &per_cpu(rt_cache_stat, cpu);
 260         }
 261         return NULL;
 262 }
 263
 264 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 265 {
 266         int cpu;
 267
 268         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 269                 if (!cpu_possible(cpu))
 270                         continue;
 271                 *pos = cpu+1;
 272                 return &per_cpu(rt_cache_stat, cpu);
 273         }
 274         return NULL;
 275
 276 }
 277
 278 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 279 {
 280
 281 }
 282
 283 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 284 {
 285         struct rt_cache_stat *st = v;
 286
 287         if (v == SEQ_START_TOKEN) {
 288                 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");
 289                 return 0;
 290         }
 291
 292         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 293                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 294                    dst_entries_get_slow(&ipv4_dst_ops),
 295                    0, /* st->in_hit */
 296                    st->in_slow_tot,
 297                    st->in_slow_mc,
 298                    st->in_no_route,
 299                    st->in_brd,
 300                    st->in_martian_dst,
 301                    st->in_martian_src,
 302
 303                    0, /* st->out_hit */
 304                    st->out_slow_tot,
 305                    st->out_slow_mc,
 306
 307                    0, /* st->gc_total */
 308                    0, /* st->gc_ignored */
 309                    0, /* st->gc_goal_miss */
 310                    0, /* st->gc_dst_overflow */
 311                    0, /* st->in_hlist_search */
 312                    0  /* st->out_hlist_search */
 313                 );
 314         return 0;
 315 }
 316
 317 static const struct seq_operations rt_cpu_seq_ops = {
 318         .start  = rt_cpu_seq_start,
 319         .next   = rt_cpu_seq_next,
 320         .stop   = rt_cpu_seq_stop,
 321         .show   = rt_cpu_seq_show,
 322 };
 323
 324
 325 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
 326 {
 327         return seq_open(file, &rt_cpu_seq_ops);
 328 }
 329
 330 static const struct file_operations rt_cpu_seq_fops = {
 331         .owner   = THIS_MODULE,
 332         .open    = rt_cpu_seq_open,
 333         .read    = seq_read,
 334         .llseek  = seq_lseek,
 335         .release = seq_release,
 336 };
 337
 338 #ifdef CONFIG_IP_ROUTE_CLASSID
 339 static int rt_acct_proc_show(struct seq_file *m, void *v)
 340 {
 341         struct ip_rt_acct *dst, *src;
 342         unsigned int i, j;
 343
 344         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 345         if (!dst)
 346                 return -ENOMEM;
 347
 348         for_each_possible_cpu(i) {
 349                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 350                 for (j = 0; j < 256; j++) {
 351                         dst[j].o_bytes   += src[j].o_bytes;
 352                         dst[j].o_packets += src[j].o_packets;
 353                         dst[j].i_bytes   += src[j].i_bytes;
 354                         dst[j].i_packets += src[j].i_packets;
 355                 }
 356         }
 357
 358         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 359         kfree(dst);
 360         return 0;
 361 }
 362
 363 static int rt_acct_proc_open(struct inode *inode, struct file *file)
 364 {
 365         return single_open(file, rt_acct_proc_show, NULL);
 366 }
 367
 368 static const struct file_operations rt_acct_proc_fops = {
 369         .owner          = THIS_MODULE,
 370         .open           = rt_acct_proc_open,
 371         .read           = seq_read,
 372         .llseek         = seq_lseek,
 373         .release        = single_release,
 374 };
 375 #endif
 376
 377 static int __net_init ip_rt_do_proc_init(struct net *net)
 378 {
 379         struct proc_dir_entry *pde;
 380
 381         pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
 382                           &rt_cache_seq_fops);
 383         if (!pde)
 384                 goto err1;
 385
 386         pde = proc_create("rt_cache", S_IRUGO,
 387                           net->proc_net_stat, &rt_cpu_seq_fops);
 388         if (!pde)
 389                 goto err2;
 390
 391 #ifdef CONFIG_IP_ROUTE_CLASSID
 392         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
 393         if (!pde)
 394                 goto err3;
 395 #endif
 396         return 0;
 397
 398 #ifdef CONFIG_IP_ROUTE_CLASSID
 399 err3:
 400         remove_proc_entry("rt_cache", net->proc_net_stat);
 401 #endif
 402 err2:
 403         remove_proc_entry("rt_cache", net->proc_net);
 404 err1:
 405         return -ENOMEM;
 406 }
 407
 408 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 409 {
 410         remove_proc_entry("rt_cache", net->proc_net_stat);
 411         remove_proc_entry("rt_cache", net->proc_net);
 412 #ifdef CONFIG_IP_ROUTE_CLASSID
 413         remove_proc_entry("rt_acct", net->proc_net);
 414 #endif
 415 }
 416
 417 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 418         .init = ip_rt_do_proc_init,
 419         .exit = ip_rt_do_proc_exit,
 420 };
 421
 422 static int __init ip_rt_proc_init(void)
 423 {
 424         return register_pernet_subsys(&ip_rt_proc_ops);
 425 }
 426
 427 #else
 428 static inline int ip_rt_proc_init(void)
 429 {
 430         return 0;
 431 }
 432 #endif /* CONFIG_PROC_FS */
 433
 434 static inline bool rt_is_expired(const struct rtable *rth)
 435 {
 436         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 437 }
 438
 439 void rt_cache_flush(struct net *net)
 440 {
 441         rt_genid_bump_ipv4(net);
 442 }
 443
 444 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 445                                            struct sk_buff *skb,
 446                                            const void *daddr)
 447 {
 448         struct net_device *dev = dst->dev;
 449         const __be32 *pkey = daddr;
 450         const struct rtable *rt;
 451         struct neighbour *n;
 452
 453         rt = (const struct rtable *) dst;
 454         if (rt->rt_gateway)
 455                 pkey = (const __be32 *) &rt->rt_gateway;
 456         else if (skb)
 457                 pkey = &ip_hdr(skb)->daddr;
 458
 459         n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
 460         if (n)
 461                 return n;
 462         return neigh_create(&arp_tbl, pkey, dev);
 463 }
 464
 465 /*
 466  * Peer allocation may fail only in serious out-of-memory conditions.  However
 467  * we still can generate some output.
 468  * Random ID selection looks a bit dangerous because we have no chances to
 469  * select ID being unique in a reasonable period of time.
 470  * But broken packet identifier may be better than no packet at all.
 471  */
 472 static void ip_select_fb_ident(struct iphdr *iph)
 473 {
 474         static DEFINE_SPINLOCK(ip_fb_id_lock);
 475         static u32 ip_fallback_id;
 476         u32 salt;
 477
 478         spin_lock_bh(&ip_fb_id_lock);
 479         salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
 480         iph->id = htons(salt & 0xFFFF);
 481         ip_fallback_id = salt;
 482         spin_unlock_bh(&ip_fb_id_lock);
 483 }
 484
 485 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
 486 {
 487         struct net *net = dev_net(dst->dev);
 488         struct inet_peer *peer;
 489
 490         peer = inet_getpeer_v4(net->ipv4.peers, iph->daddr, 1);
 491         if (peer) {
 492                 iph->id = htons(inet_getid(peer, more));
 493                 inet_putpeer(peer);
 494                 return;
 495         }
 496
 497         ip_select_fb_ident(iph);
 498 }
 499 EXPORT_SYMBOL(__ip_select_ident);
 500
 501 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 502                              const struct iphdr *iph,
 503                              int oif, u8 tos,
 504                              u8 prot, u32 mark, int flow_flags)
 505 {
 506         if (sk) {
 507                 const struct inet_sock *inet = inet_sk(sk);
 508
 509                 oif = sk->sk_bound_dev_if;
 510                 mark = sk->sk_mark;
 511                 tos = RT_CONN_FLAGS(sk);
 512                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 513         }
 514         flowi4_init_output(fl4, oif, mark, tos,
 515                            RT_SCOPE_UNIVERSE, prot,
 516                            flow_flags,
 517                            iph->daddr, iph->saddr, 0, 0);
 518 }
 519
 520 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 521                                const struct sock *sk)
 522 {
 523         const struct iphdr *iph = ip_hdr(skb);
 524         int oif = skb->dev->ifindex;
 525         u8 tos = RT_TOS(iph->tos);
 526         u8 prot = iph->protocol;
 527         u32 mark = skb->mark;
 528
 529         __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
 530 }
 531
 532 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 533 {
 534         const struct inet_sock *inet = inet_sk(sk);
 535         const struct ip_options_rcu *inet_opt;
 536         __be32 daddr = inet->inet_daddr;
 537
 538         rcu_read_lock();
 539         inet_opt = rcu_dereference(inet->inet_opt);
 540         if (inet_opt && inet_opt->opt.srr)
 541                 daddr = inet_opt->opt.faddr;
 542         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 543                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 544                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 545                            inet_sk_flowi_flags(sk),
 546                            daddr, inet->inet_saddr, 0, 0);
 547         rcu_read_unlock();
 548 }
 549
 550 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 551                                  const struct sk_buff *skb)
 552 {
 553         if (skb)
 554                 build_skb_flow_key(fl4, skb, sk);
 555         else
 556                 build_sk_flow_key(fl4, sk);
 557 }
 558
 559 static inline void rt_free(struct rtable *rt)
 560 {
 561         call_rcu(&rt->dst.rcu_head, dst_rcu_free);
 562 }
 563
 564 static DEFINE_SPINLOCK(fnhe_lock);
 565
 566 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 567 {
 568         struct rtable *rt;
 569
 570         rt = rcu_dereference(fnhe->fnhe_rth_input);
 571         if (rt) {
 572                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 573                 rt_free(rt);
 574         }
 575         rt = rcu_dereference(fnhe->fnhe_rth_output);
 576         if (rt) {
 577                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 578                 rt_free(rt);
 579         }
 580 }
 581
 582 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
 583 {
 584         struct fib_nh_exception *fnhe, *oldest;
 585
 586         oldest = rcu_dereference(hash->chain);
 587         for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
 588              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 589                 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
 590                         oldest = fnhe;
 591         }
 592         fnhe_flush_routes(oldest);
 593         return oldest;
 594 }
 595
 596 static inline u32 fnhe_hashfun(__be32 daddr)
 597 {
 598         u32 hval;
 599
 600         hval = (__force u32) daddr;
 601         hval ^= (hval >> 11) ^ (hval >> 22);
 602
 603         return hval & (FNHE_HASH_SIZE - 1);
 604 }
 605
 606 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 607 {
 608         rt->rt_pmtu = fnhe->fnhe_pmtu;
 609         rt->dst.expires = fnhe->fnhe_expires;
 610
 611         if (fnhe->fnhe_gw) {
 612                 rt->rt_flags |= RTCF_REDIRECTED;
 613                 rt->rt_gateway = fnhe->fnhe_gw;
 614                 rt->rt_uses_gateway = 1;
 615         }
 616 }
 617
 618 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
 619                                   u32 pmtu, unsigned long expires)
 620 {
 621         struct fnhe_hash_bucket *hash;
 622         struct fib_nh_exception *fnhe;
 623         struct rtable *rt;
 624         unsigned int i;
 625         int depth;
 626         u32 hval = fnhe_hashfun(daddr);
 627
 628         spin_lock_bh(&fnhe_lock);
 629
 630         hash = nh->nh_exceptions;
 631         if (!hash) {
 632                 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
 633                 if (!hash)
 634                         goto out_unlock;
 635                 nh->nh_exceptions = hash;
 636         }
 637
 638         hash += hval;
 639
 640         depth = 0;
 641         for (fnhe = rcu_dereference(hash->chain); fnhe;
 642              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 643                 if (fnhe->fnhe_daddr == daddr)
 644                         break;
 645                 depth++;
 646         }
 647
 648         if (fnhe) {
 649                 if (gw)
 650                         fnhe->fnhe_gw = gw;
 651                 if (pmtu) {
 652                         fnhe->fnhe_pmtu = pmtu;
 653                         fnhe->fnhe_expires = max(1UL, expires);
 654                 }
 655                 /* Update all cached dsts too */
 656                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 657                 if (rt)
 658                         fill_route_from_fnhe(rt, fnhe);
 659                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 660                 if (rt)
 661                         fill_route_from_fnhe(rt, fnhe);
 662         } else {
 663                 if (depth > FNHE_RECLAIM_DEPTH)
 664                         fnhe = fnhe_oldest(hash);
 665                 else {
 666                         fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 667                         if (!fnhe)
 668                                 goto out_unlock;
 669
 670                         fnhe->fnhe_next = hash->chain;
 671                         rcu_assign_pointer(hash->chain, fnhe);
 672                 }
 673                 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
 674                 fnhe->fnhe_daddr = daddr;
 675                 fnhe->fnhe_gw = gw;
 676                 fnhe->fnhe_pmtu = pmtu;
 677                 fnhe->fnhe_expires = expires;
 678
 679                 /* Exception created; mark the cached routes for the nexthop
 680                  * stale, so anyone caching it rechecks if this exception
 681                  * applies to them.
 682                  */
 683                 rt = rcu_dereference(nh->nh_rth_input);
 684                 if (rt)
 685                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 686
 687                 for_each_possible_cpu(i) {
 688                         struct rtable __rcu **prt;
 689                         prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
 690                         rt = rcu_dereference(*prt);
 691                         if (rt)
 692                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 693                 }
 694         }
 695
 696         fnhe->fnhe_stamp = jiffies;
 697
 698 out_unlock:
 699         spin_unlock_bh(&fnhe_lock);
 700         return;
 701 }
 702
 703 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 704                              bool kill_route)
 705 {
 706         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 707         __be32 old_gw = ip_hdr(skb)->saddr;
 708         struct net_device *dev = skb->dev;
 709         struct in_device *in_dev;
 710         struct fib_result res;
 711         struct neighbour *n;
 712         struct net *net;
 713
 714         switch (icmp_hdr(skb)->code & 7) {
 715         case ICMP_REDIR_NET:
 716         case ICMP_REDIR_NETTOS:
 717         case ICMP_REDIR_HOST:
 718         case ICMP_REDIR_HOSTTOS:
 719                 break;
 720
 721         default:
 722                 return;
 723         }
 724
 725         if (rt->rt_gateway != old_gw)
 726                 return;
 727
 728         in_dev = __in_dev_get_rcu(dev);
 729         if (!in_dev)
 730                 return;
 731
 732         net = dev_net(dev);
 733         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 734             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 735             ipv4_is_zeronet(new_gw))
 736                 goto reject_redirect;
 737
 738         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 739                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 740                         goto reject_redirect;
 741                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 742                         goto reject_redirect;
 743         } else {
 744                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 745                         goto reject_redirect;
 746         }
 747
 748         n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
 749         if (n) {
 750                 if (!(n->nud_state & NUD_VALID)) {
 751                         neigh_event_send(n, NULL);
 752                 } else {
 753                         if (fib_lookup(net, fl4, &res) == 0) {
 754                                 struct fib_nh *nh = &FIB_RES_NH(res);
 755
 756                                 update_or_create_fnhe(nh, fl4->daddr, new_gw,
 757                                                       0, 0);
 758                         }
 759                         if (kill_route)
 760                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 761                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 762                 }
 763                 neigh_release(n);
 764         }
 765         return;
 766
 767 reject_redirect:
 768 #ifdef CONFIG_IP_ROUTE_VERBOSE
 769         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 770                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 771                 __be32 daddr = iph->daddr;
 772                 __be32 saddr = iph->saddr;
 773
 774                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 775                                      "  Advised path = %pI4 -> %pI4\n",
 776                                      &old_gw, dev->name, &new_gw,
 777                                      &saddr, &daddr);
 778         }
 779 #endif
 780         ;
 781 }
 782
 783 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 784 {
 785         struct rtable *rt;
 786         struct flowi4 fl4;
 787         const struct iphdr *iph = (const struct iphdr *) skb->data;
 788         int oif = skb->dev->ifindex;
 789         u8 tos = RT_TOS(iph->tos);
 790         u8 prot = iph->protocol;
 791         u32 mark = skb->mark;
 792
 793         rt = (struct rtable *) dst;
 794
 795         __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
 796         __ip_do_redirect(rt, skb, &fl4, true);
 797 }
 798
 799 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 800 {
 801         struct rtable *rt = (struct rtable *)dst;
 802         struct dst_entry *ret = dst;
 803
 804         if (rt) {
 805                 if (dst->obsolete > 0) {
 806                         ip_rt_put(rt);
 807                         ret = NULL;
 808                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 809                            rt->dst.expires) {
 810                         ip_rt_put(rt);
 811                         ret = NULL;
 812                 }
 813         }
 814         return ret;
 815 }
 816
 817 /*
 818  * Algorithm:
 819  *      1. The first ip_rt_redirect_number redirects are sent
 820  *         with exponential backoff, then we stop sending them at all,
 821  *         assuming that the host ignores our redirects.
 822  *      2. If we did not see packets requiring redirects
 823  *         during ip_rt_redirect_silence, we assume that the host
 824  *         forgot redirected route and start to send redirects again.
 825  *
 826  * This algorithm is much cheaper and more intelligent than dumb load limiting
 827  * in icmp.c.
 828  *
 829  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 830  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 831  */
 832
 833 void ip_rt_send_redirect(struct sk_buff *skb)
 834 {
 835         struct rtable *rt = skb_rtable(skb);
 836         struct in_device *in_dev;
 837         struct inet_peer *peer;
 838         struct net *net;
 839         int log_martians;
 840
 841         rcu_read_lock();
 842         in_dev = __in_dev_get_rcu(rt->dst.dev);
 843         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 844                 rcu_read_unlock();
 845                 return;
 846         }
 847         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 848         rcu_read_unlock();
 849
 850         net = dev_net(rt->dst.dev);
 851         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
 852         if (!peer) {
 853                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 854                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 855                 return;
 856         }
 857
 858         /* No redirected packets during ip_rt_redirect_silence;
 859          * reset the algorithm.
 860          */
 861         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
 862                 peer->rate_tokens = 0;
 863
 864         /* Too many ignored redirects; do not send anything
 865          * set dst.rate_last to the last seen redirected packet.
 866          */
 867         if (peer->rate_tokens >= ip_rt_redirect_number) {
 868                 peer->rate_last = jiffies;
 869                 goto out_put_peer;
 870         }
 871
 872         /* Check for load limit; set rate_last to the latest sent
 873          * redirect.
 874          */
 875         if (peer->rate_tokens == 0 ||
 876             time_after(jiffies,
 877                        (peer->rate_last +
 878                         (ip_rt_redirect_load << peer->rate_tokens)))) {
 879                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 880
 881                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 882                 peer->rate_last = jiffies;
 883                 ++peer->rate_tokens;
 884 #ifdef CONFIG_IP_ROUTE_VERBOSE
 885                 if (log_martians &&
 886                     peer->rate_tokens == ip_rt_redirect_number)
 887                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 888                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 889                                              &ip_hdr(skb)->daddr, &gw);
 890 #endif
 891         }
 892 out_put_peer:
 893         inet_putpeer(peer);
 894 }
 895
 896 static int ip_error(struct sk_buff *skb)
 897 {
 898         struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
 899         struct rtable *rt = skb_rtable(skb);
 900         struct inet_peer *peer;
 901         unsigned long now;
 902         struct net *net;
 903         bool send;
 904         int code;
 905
 906         net = dev_net(rt->dst.dev);
 907         if (!IN_DEV_FORWARD(in_dev)) {
 908                 switch (rt->dst.error) {
 909                 case EHOSTUNREACH:
 910                         IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
 911                         break;
 912
 913                 case ENETUNREACH:
 914                         IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
 915                         break;
 916                 }
 917                 goto out;
 918         }
 919
 920         switch (rt->dst.error) {
 921         case EINVAL:
 922         default:
 923                 goto out;
 924         case EHOSTUNREACH:
 925                 code = ICMP_HOST_UNREACH;
 926                 break;
 927         case ENETUNREACH:
 928                 code = ICMP_NET_UNREACH;
 929                 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
 930                 break;
 931         case EACCES:
 932                 code = ICMP_PKT_FILTERED;
 933                 break;
 934         }
 935
 936         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
 937
 938         send = true;
 939         if (peer) {
 940                 now = jiffies;
 941                 peer->rate_tokens += now - peer->rate_last;
 942                 if (peer->rate_tokens > ip_rt_error_burst)
 943                         peer->rate_tokens = ip_rt_error_burst;
 944                 peer->rate_last = now;
 945                 if (peer->rate_tokens >= ip_rt_error_cost)
 946                         peer->rate_tokens -= ip_rt_error_cost;
 947                 else
 948                         send = false;
 949                 inet_putpeer(peer);
 950         }
 951         if (send)
 952                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
 953
 954 out:    kfree_skb(skb);
 955         return 0;
 956 }
 957
 958 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
 959 {
 960         struct dst_entry *dst = &rt->dst;
 961         struct fib_result res;
 962
 963         if (dst_metric_locked(dst, RTAX_MTU))
 964                 return;
 965
 966         if (dst->dev->mtu < mtu)
 967                 return;
 968
 969         if (mtu < ip_rt_min_pmtu)
 970                 mtu = ip_rt_min_pmtu;
 971
 972         if (rt->rt_pmtu == mtu &&
 973             time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
 974                 return;
 975
 976         rcu_read_lock();
 977         if (fib_lookup(dev_net(dst->dev), fl4, &res) == 0) {
 978                 struct fib_nh *nh = &FIB_RES_NH(res);
 979
 980                 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
 981                                       jiffies + ip_rt_mtu_expires);
 982         }
 983         rcu_read_unlock();
 984 }
 985
 986 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 987                               struct sk_buff *skb, u32 mtu)
 988 {
 989         struct rtable *rt = (struct rtable *) dst;
 990         struct flowi4 fl4;
 991
 992         ip_rt_build_flow_key(&fl4, sk, skb);
 993         __ip_rt_update_pmtu(rt, &fl4, mtu);
 994 }
 995
 996 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
 997                       int oif, u32 mark, u8 protocol, int flow_flags)
 998 {
 999         const struct iphdr *iph = (const struct iphdr *) skb->data;
1000         struct flowi4 fl4;
1001         struct rtable *rt;
1002
1003         __build_flow_key(&fl4, NULL, iph, oif,
1004                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1005         rt = __ip_route_output_key(net, &fl4);
1006         if (!IS_ERR(rt)) {
1007                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1008                 ip_rt_put(rt);
1009         }
1010 }
1011 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1012
1013 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1014 {
1015         const struct iphdr *iph = (const struct iphdr *) skb->data;
1016         struct flowi4 fl4;
1017         struct rtable *rt;
1018
1019         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1020         rt = __ip_route_output_key(sock_net(sk), &fl4);
1021         if (!IS_ERR(rt)) {
1022                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1023                 ip_rt_put(rt);
1024         }
1025 }
1026
1027 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1028 {
1029         const struct iphdr *iph = (const struct iphdr *) skb->data;
1030         struct flowi4 fl4;
1031         struct rtable *rt;
1032         struct dst_entry *dst;
1033         bool new = false;
1034
1035         bh_lock_sock(sk);
1036
1037         if (!ip_sk_accept_pmtu(sk))
1038                 goto out;
1039
1040         rt = (struct rtable *) __sk_dst_get(sk);
1041
1042         if (sock_owned_by_user(sk) || !rt) {
1043                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1044                 goto out;
1045         }
1046
1047         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1048
1049         if (!__sk_dst_check(sk, 0)) {
1050                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1051                 if (IS_ERR(rt))
1052                         goto out;
1053
1054                 new = true;
1055         }
1056
1057         __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1058
1059         dst = dst_check(&rt->dst, 0);
1060         if (!dst) {
1061                 if (new)
1062                         dst_release(&rt->dst);
1063
1064                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1065                 if (IS_ERR(rt))
1066                         goto out;
1067
1068                 new = true;
1069         }
1070
1071         if (new)
1072                 __sk_dst_set(sk, &rt->dst);
1073
1074 out:
1075         bh_unlock_sock(sk);
1076 }
1077 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1078
1079 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1080                    int oif, u32 mark, u8 protocol, int flow_flags)
1081 {
1082         const struct iphdr *iph = (const struct iphdr *) skb->data;
1083         struct flowi4 fl4;
1084         struct rtable *rt;
1085
1086         __build_flow_key(&fl4, NULL, iph, oif,
1087                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1088         rt = __ip_route_output_key(net, &fl4);
1089         if (!IS_ERR(rt)) {
1090                 __ip_do_redirect(rt, skb, &fl4, false);
1091                 ip_rt_put(rt);
1092         }
1093 }
1094 EXPORT_SYMBOL_GPL(ipv4_redirect);
1095
1096 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1097 {
1098         const struct iphdr *iph = (const struct iphdr *) skb->data;
1099         struct flowi4 fl4;
1100         struct rtable *rt;
1101
1102         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1103         rt = __ip_route_output_key(sock_net(sk), &fl4);
1104         if (!IS_ERR(rt)) {
1105                 __ip_do_redirect(rt, skb, &fl4, false);
1106                 ip_rt_put(rt);
1107         }
1108 }
1109 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1110
1111 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1112 {
1113         struct rtable *rt = (struct rtable *) dst;
1114
1115         /* All IPV4 dsts are created with ->obsolete set to the value
1116          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1117          * into this function always.
1118          *
1119          * When a PMTU/redirect information update invalidates a route,
1120          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1121          * DST_OBSOLETE_DEAD by dst_free().
1122          */
1123         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1124                 return NULL;
1125         return dst;
1126 }
1127
1128 static void ipv4_link_failure(struct sk_buff *skb)
1129 {
1130         struct rtable *rt;
1131
1132         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1133
1134         rt = skb_rtable(skb);
1135         if (rt)
1136                 dst_set_expires(&rt->dst, 0);
1137 }
1138
1139 static int ip_rt_bug(struct sk_buff *skb)
1140 {
1141         pr_debug("%s: %pI4 -> %pI4, %s\n",
1142                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1143                  skb->dev ? skb->dev->name : "?");
1144         kfree_skb(skb);
1145         WARN_ON(1);
1146         return 0;
1147 }
1148
1149 /*
1150    We do not cache source address of outgoing interface,
1151    because it is used only by IP RR, TS and SRR options,
1152    so that it out of fast path.
1153
1154    BTW remember: "addr" is allowed to be not aligned
1155    in IP options!
1156  */
1157
1158 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1159 {
1160         __be32 src;
1161
1162         if (rt_is_output_route(rt))
1163                 src = ip_hdr(skb)->saddr;
1164         else {
1165                 struct fib_result res;
1166                 struct flowi4 fl4;
1167                 struct iphdr *iph;
1168
1169                 iph = ip_hdr(skb);
1170
1171                 memset(&fl4, 0, sizeof(fl4));
1172                 fl4.daddr = iph->daddr;
1173                 fl4.saddr = iph->saddr;
1174                 fl4.flowi4_tos = RT_TOS(iph->tos);
1175                 fl4.flowi4_oif = rt->dst.dev->ifindex;
1176                 fl4.flowi4_iif = skb->dev->ifindex;
1177                 fl4.flowi4_mark = skb->mark;
1178
1179                 rcu_read_lock();
1180                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1181                         src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1182                 else
1183                         src = inet_select_addr(rt->dst.dev,
1184                                                rt_nexthop(rt, iph->daddr),
1185                                                RT_SCOPE_UNIVERSE);
1186                 rcu_read_unlock();
1187         }
1188         memcpy(addr, &src, 4);
1189 }
1190
1191 #ifdef CONFIG_IP_ROUTE_CLASSID
1192 static void set_class_tag(struct rtable *rt, u32 tag)
1193 {
1194         if (!(rt->dst.tclassid & 0xFFFF))
1195                 rt->dst.tclassid |= tag & 0xFFFF;
1196         if (!(rt->dst.tclassid & 0xFFFF0000))
1197                 rt->dst.tclassid |= tag & 0xFFFF0000;
1198 }
1199 #endif
1200
1201 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1202 {
1203         unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1204
1205         if (advmss == 0) {
1206                 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1207                                ip_rt_min_advmss);
1208                 if (advmss > 65535 - 40)
1209                         advmss = 65535 - 40;
1210         }
1211         return advmss;
1212 }
1213
1214 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1215 {
1216         const struct rtable *rt = (const struct rtable *) dst;
1217         unsigned int mtu = rt->rt_pmtu;
1218
1219         if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1220                 mtu = dst_metric_raw(dst, RTAX_MTU);
1221
1222         if (mtu)
1223                 return mtu;
1224
1225         mtu = dst->dev->mtu;
1226
1227         if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1228                 if (rt->rt_uses_gateway && mtu > 576)
1229                         mtu = 576;
1230         }
1231
1232         return min_t(unsigned int, mtu, IP_MAX_MTU);
1233 }
1234
1235 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1236 {
1237         struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1238         struct fib_nh_exception *fnhe;
1239         u32 hval;
1240
1241         if (!hash)
1242                 return NULL;
1243
1244         hval = fnhe_hashfun(daddr);
1245
1246         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1247              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1248                 if (fnhe->fnhe_daddr == daddr)
1249                         return fnhe;
1250         }
1251         return NULL;
1252 }
1253
1254 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1255                               __be32 daddr)
1256 {
1257         bool ret = false;
1258
1259         spin_lock_bh(&fnhe_lock);
1260
1261         if (daddr == fnhe->fnhe_daddr) {
1262                 struct rtable __rcu **porig;
1263                 struct rtable *orig;
1264                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1265
1266                 if (rt_is_input_route(rt))
1267                         porig = &fnhe->fnhe_rth_input;
1268                 else
1269                         porig = &fnhe->fnhe_rth_output;
1270                 orig = rcu_dereference(*porig);
1271
1272                 if (fnhe->fnhe_genid != genid) {
1273                         fnhe->fnhe_genid = genid;
1274                         fnhe->fnhe_gw = 0;
1275                         fnhe->fnhe_pmtu = 0;
1276                         fnhe->fnhe_expires = 0;
1277                         fnhe_flush_routes(fnhe);
1278                         orig = NULL;
1279                 }
1280                 fill_route_from_fnhe(rt, fnhe);
1281                 if (!rt->rt_gateway)
1282                         rt->rt_gateway = daddr;
1283
1284                 if (!(rt->dst.flags & DST_NOCACHE)) {
1285                         rcu_assign_pointer(*porig, rt);
1286                         if (orig)
1287                                 rt_free(orig);
1288                         ret = true;
1289                 }
1290
1291                 fnhe->fnhe_stamp = jiffies;
1292         }
1293         spin_unlock_bh(&fnhe_lock);
1294
1295         return ret;
1296 }
1297
1298 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1299 {
1300         struct rtable *orig, *prev, **p;
1301         bool ret = true;
1302
1303         if (rt_is_input_route(rt)) {
1304                 p = (struct rtable **)&nh->nh_rth_input;
1305         } else {
1306                 p = (struct rtable **)__this_cpu_ptr(nh->nh_pcpu_rth_output);
1307         }
1308         orig = *p;
1309
1310         prev = cmpxchg(p, orig, rt);
1311         if (prev == orig) {
1312                 if (orig)
1313                         rt_free(orig);
1314         } else
1315                 ret = false;
1316
1317         return ret;
1318 }
1319
1320 static DEFINE_SPINLOCK(rt_uncached_lock);
1321 static LIST_HEAD(rt_uncached_list);
1322
1323 static void rt_add_uncached_list(struct rtable *rt)
1324 {
1325         spin_lock_bh(&rt_uncached_lock);
1326         list_add_tail(&rt->rt_uncached, &rt_uncached_list);
1327         spin_unlock_bh(&rt_uncached_lock);
1328 }
1329
1330 static void ipv4_dst_destroy(struct dst_entry *dst)
1331 {
1332         struct rtable *rt = (struct rtable *) dst;
1333
1334         if (!list_empty(&rt->rt_uncached)) {
1335                 spin_lock_bh(&rt_uncached_lock);
1336                 list_del(&rt->rt_uncached);
1337                 spin_unlock_bh(&rt_uncached_lock);
1338         }
1339 }
1340
1341 void rt_flush_dev(struct net_device *dev)
1342 {
1343         if (!list_empty(&rt_uncached_list)) {
1344                 struct net *net = dev_net(dev);
1345                 struct rtable *rt;
1346
1347                 spin_lock_bh(&rt_uncached_lock);
1348                 list_for_each_entry(rt, &rt_uncached_list, rt_uncached) {
1349                         if (rt->dst.dev != dev)
1350                                 continue;
1351                         rt->dst.dev = net->loopback_dev;
1352                         dev_hold(rt->dst.dev);
1353                         dev_put(dev);
1354                 }
1355                 spin_unlock_bh(&rt_uncached_lock);
1356         }
1357 }
1358
1359 static bool rt_cache_valid(const struct rtable *rt)
1360 {
1361         return  rt &&
1362                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1363                 !rt_is_expired(rt);
1364 }
1365
1366 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1367                            const struct fib_result *res,
1368                            struct fib_nh_exception *fnhe,
1369                            struct fib_info *fi, u16 type, u32 itag)
1370 {
1371         bool cached = false;
1372
1373         if (fi) {
1374                 struct fib_nh *nh = &FIB_RES_NH(*res);
1375
1376                 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1377                         rt->rt_gateway = nh->nh_gw;
1378                         rt->rt_uses_gateway = 1;
1379                 }
1380                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1381 #ifdef CONFIG_IP_ROUTE_CLASSID
1382                 rt->dst.tclassid = nh->nh_tclassid;
1383 #endif
1384                 if (unlikely(fnhe))
1385                         cached = rt_bind_exception(rt, fnhe, daddr);
1386                 else if (!(rt->dst.flags & DST_NOCACHE))
1387                         cached = rt_cache_route(nh, rt);
1388                 if (unlikely(!cached)) {
1389                         /* Routes we intend to cache in nexthop exception or
1390                          * FIB nexthop have the DST_NOCACHE bit clear.
1391                          * However, if we are unsuccessful at storing this
1392                          * route into the cache we really need to set it.
1393                          */
1394                         rt->dst.flags |= DST_NOCACHE;
1395                         if (!rt->rt_gateway)
1396                                 rt->rt_gateway = daddr;
1397                         rt_add_uncached_list(rt);
1398                 }
1399         } else
1400                 rt_add_uncached_list(rt);
1401
1402 #ifdef CONFIG_IP_ROUTE_CLASSID
1403 #ifdef CONFIG_IP_MULTIPLE_TABLES
1404         set_class_tag(rt, res->tclassid);
1405 #endif
1406         set_class_tag(rt, itag);
1407 #endif
1408 }
1409
1410 static struct rtable *rt_dst_alloc(struct net_device *dev,
1411                                    bool nopolicy, bool noxfrm, bool will_cache)
1412 {
1413         return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1414                          (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1415                          (nopolicy ? DST_NOPOLICY : 0) |
1416                          (noxfrm ? DST_NOXFRM : 0));
1417 }
1418
1419 /* called in rcu_read_lock() section */
1420 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1421                                 u8 tos, struct net_device *dev, int our)
1422 {
1423         struct rtable *rth;
1424         struct in_device *in_dev = __in_dev_get_rcu(dev);
1425         u32 itag = 0;
1426         int err;
1427
1428         /* Primary sanity checks. */
1429
1430         if (in_dev == NULL)
1431                 return -EINVAL;
1432
1433         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1434             skb->protocol != htons(ETH_P_IP))
1435                 goto e_inval;
1436
1437         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1438                 if (ipv4_is_loopback(saddr))
1439                         goto e_inval;
1440
1441         if (ipv4_is_zeronet(saddr)) {
1442                 if (!ipv4_is_local_multicast(daddr))
1443                         goto e_inval;
1444         } else {
1445                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1446                                           in_dev, &itag);
1447                 if (err < 0)
1448                         goto e_err;
1449         }
1450         rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1451                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1452         if (!rth)
1453                 goto e_nobufs;
1454
1455 #ifdef CONFIG_IP_ROUTE_CLASSID
1456         rth->dst.tclassid = itag;
1457 #endif
1458         rth->dst.output = ip_rt_bug;
1459
1460         rth->rt_genid   = rt_genid_ipv4(dev_net(dev));
1461         rth->rt_flags   = RTCF_MULTICAST;
1462         rth->rt_type    = RTN_MULTICAST;
1463         rth->rt_is_input= 1;
1464         rth->rt_iif     = 0;
1465         rth->rt_pmtu    = 0;
1466         rth->rt_gateway = 0;
1467         rth->rt_uses_gateway = 0;
1468         INIT_LIST_HEAD(&rth->rt_uncached);
1469         if (our) {
1470                 rth->dst.input= ip_local_deliver;
1471                 rth->rt_flags |= RTCF_LOCAL;
1472         }
1473
1474 #ifdef CONFIG_IP_MROUTE
1475         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1476                 rth->dst.input = ip_mr_input;
1477 #endif
1478         RT_CACHE_STAT_INC(in_slow_mc);
1479
1480         skb_dst_set(skb, &rth->dst);
1481         return 0;
1482
1483 e_nobufs:
1484         return -ENOBUFS;
1485 e_inval:
1486         return -EINVAL;
1487 e_err:
1488         return err;
1489 }
1490
1491
1492 static void ip_handle_martian_source(struct net_device *dev,
1493                                      struct in_device *in_dev,
1494                                      struct sk_buff *skb,
1495                                      __be32 daddr,
1496                                      __be32 saddr)
1497 {
1498         RT_CACHE_STAT_INC(in_martian_src);
1499 #ifdef CONFIG_IP_ROUTE_VERBOSE
1500         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1501                 /*
1502                  *      RFC1812 recommendation, if source is martian,
1503                  *      the only hint is MAC header.
1504                  */
1505                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1506                         &daddr, &saddr, dev->name);
1507                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1508                         print_hex_dump(KERN_WARNING, "ll header: ",
1509                                        DUMP_PREFIX_OFFSET, 16, 1,
1510                                        skb_mac_header(skb),
1511                                        dev->hard_header_len, true);
1512                 }
1513         }
1514 #endif
1515 }
1516
1517 /* called in rcu_read_lock() section */
1518 static int __mkroute_input(struct sk_buff *skb,
1519                            const struct fib_result *res,
1520                            struct in_device *in_dev,
1521                            __be32 daddr, __be32 saddr, u32 tos)
1522 {
1523         struct fib_nh_exception *fnhe;
1524         struct rtable *rth;
1525         int err;
1526         struct in_device *out_dev;
1527         unsigned int flags = 0;
1528         bool do_cache;
1529         u32 itag;
1530
1531         /* get a working reference to the output device */
1532         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1533         if (out_dev == NULL) {
1534                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1535                 return -EINVAL;
1536         }
1537
1538         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1539                                   in_dev->dev, in_dev, &itag);
1540         if (err < 0) {
1541                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1542                                          saddr);
1543
1544                 goto cleanup;
1545         }
1546
1547         do_cache = res->fi && !itag;
1548         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1549             (IN_DEV_SHARED_MEDIA(out_dev) ||
1550              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) {
1551                 flags |= RTCF_DOREDIRECT;
1552                 do_cache = false;
1553         }
1554
1555         if (skb->protocol != htons(ETH_P_IP)) {
1556                 /* Not IP (i.e. ARP). Do not create route, if it is
1557                  * invalid for proxy arp. DNAT routes are always valid.
1558                  *
1559                  * Proxy arp feature have been extended to allow, ARP
1560                  * replies back to the same interface, to support
1561                  * Private VLAN switch technologies. See arp.c.
1562                  */
1563                 if (out_dev == in_dev &&
1564                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1565                         err = -EINVAL;
1566                         goto cleanup;
1567                 }
1568         }
1569
1570         fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1571         if (do_cache) {
1572                 if (fnhe != NULL)
1573                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1574                 else
1575                         rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1576
1577                 if (rt_cache_valid(rth)) {
1578                         skb_dst_set_noref(skb, &rth->dst);
1579                         goto out;
1580                 }
1581         }
1582
1583         rth = rt_dst_alloc(out_dev->dev,
1584                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1585                            IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1586         if (!rth) {
1587                 err = -ENOBUFS;
1588                 goto cleanup;
1589         }
1590
1591         rth->rt_genid = rt_genid_ipv4(dev_net(rth->dst.dev));
1592         rth->rt_flags = flags;
1593         rth->rt_type = res->type;
1594         rth->rt_is_input = 1;
1595         rth->rt_iif     = 0;
1596         rth->rt_pmtu    = 0;
1597         rth->rt_gateway = 0;
1598         rth->rt_uses_gateway = 0;
1599         INIT_LIST_HEAD(&rth->rt_uncached);
1600
1601         rth->dst.input = ip_forward;
1602         rth->dst.output = ip_output;
1603
1604         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1605         skb_dst_set(skb, &rth->dst);
1606 out:
1607         err = 0;
1608  cleanup:
1609         return err;
1610 }
1611
1612 static int ip_mkroute_input(struct sk_buff *skb,
1613                             struct fib_result *res,
1614                             const struct flowi4 *fl4,
1615                             struct in_device *in_dev,
1616                             __be32 daddr, __be32 saddr, u32 tos)
1617 {
1618 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1619         if (res->fi && res->fi->fib_nhs > 1)
1620                 fib_select_multipath(res);
1621 #endif
1622
1623         /* create a routing cache entry */
1624         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1625 }
1626
1627 /*
1628  *      NOTE. We drop all the packets that has local source
1629  *      addresses, because every properly looped back packet
1630  *      must have correct destination already attached by output routine.
1631  *
1632  *      Such approach solves two big problems:
1633  *      1. Not simplex devices are handled properly.
1634  *      2. IP spoofing attempts are filtered with 100% of guarantee.
1635  *      called with rcu_read_lock()
1636  */
1637
1638 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1639                                u8 tos, struct net_device *dev)
1640 {
1641         struct fib_result res;
1642         struct in_device *in_dev = __in_dev_get_rcu(dev);
1643         struct flowi4   fl4;
1644         unsigned int    flags = 0;
1645         u32             itag = 0;
1646         struct rtable   *rth;
1647         int             err = -EINVAL;
1648         struct net    *net = dev_net(dev);
1649         bool do_cache;
1650
1651         /* IP on this device is disabled. */
1652
1653         if (!in_dev)
1654                 goto out;
1655
1656         /* Check for the most weird martians, which can be not detected
1657            by fib_lookup.
1658          */
1659
1660         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1661                 goto martian_source;
1662
1663         res.fi = NULL;
1664         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1665                 goto brd_input;
1666
1667         /* Accept zero addresses only to limited broadcast;
1668          * I even do not know to fix it or not. Waiting for complains :-)
1669          */
1670         if (ipv4_is_zeronet(saddr))
1671                 goto martian_source;
1672
1673         if (ipv4_is_zeronet(daddr))
1674                 goto martian_destination;
1675
1676         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1677          * and call it once if daddr or/and saddr are loopback addresses
1678          */
1679         if (ipv4_is_loopback(daddr)) {
1680                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1681                         goto martian_destination;
1682         } else if (ipv4_is_loopback(saddr)) {
1683                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1684                         goto martian_source;
1685         }
1686
1687         /*
1688          *      Now we are ready to route packet.
1689          */
1690         fl4.flowi4_oif = 0;
1691         fl4.flowi4_iif = dev->ifindex;
1692         fl4.flowi4_mark = skb->mark;
1693         fl4.flowi4_tos = tos;
1694         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1695         fl4.daddr = daddr;
1696         fl4.saddr = saddr;
1697         err = fib_lookup(net, &fl4, &res);
1698         if (err != 0)
1699                 goto no_route;
1700
1701         RT_CACHE_STAT_INC(in_slow_tot);
1702
1703         if (res.type == RTN_BROADCAST)
1704                 goto brd_input;
1705
1706         if (res.type == RTN_LOCAL) {
1707                 err = fib_validate_source(skb, saddr, daddr, tos,
1708                                           LOOPBACK_IFINDEX,
1709                                           dev, in_dev, &itag);
1710                 if (err < 0)
1711                         goto martian_source_keep_err;
1712                 goto local_input;
1713         }
1714
1715         if (!IN_DEV_FORWARD(in_dev))
1716                 goto no_route;
1717         if (res.type != RTN_UNICAST)
1718                 goto martian_destination;
1719
1720         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1721 out:    return err;
1722
1723 brd_input:
1724         if (skb->protocol != htons(ETH_P_IP))
1725                 goto e_inval;
1726
1727         if (!ipv4_is_zeronet(saddr)) {
1728                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1729                                           in_dev, &itag);
1730                 if (err < 0)
1731                         goto martian_source_keep_err;
1732         }
1733         flags |= RTCF_BROADCAST;
1734         res.type = RTN_BROADCAST;
1735         RT_CACHE_STAT_INC(in_brd);
1736
1737 local_input:
1738         do_cache = false;
1739         if (res.fi) {
1740                 if (!itag) {
1741                         rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1742                         if (rt_cache_valid(rth)) {
1743                                 skb_dst_set_noref(skb, &rth->dst);
1744                                 err = 0;
1745                                 goto out;
1746                         }
1747                         do_cache = true;
1748                 }
1749         }
1750
1751         rth = rt_dst_alloc(net->loopback_dev,
1752                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1753         if (!rth)
1754                 goto e_nobufs;
1755
1756         rth->dst.input= ip_local_deliver;
1757         rth->dst.output= ip_rt_bug;
1758 #ifdef CONFIG_IP_ROUTE_CLASSID
1759         rth->dst.tclassid = itag;
1760 #endif
1761
1762         rth->rt_genid = rt_genid_ipv4(net);
1763         rth->rt_flags   = flags|RTCF_LOCAL;
1764         rth->rt_type    = res.type;
1765         rth->rt_is_input = 1;
1766         rth->rt_iif     = 0;
1767         rth->rt_pmtu    = 0;
1768         rth->rt_gateway = 0;
1769         rth->rt_uses_gateway = 0;
1770         INIT_LIST_HEAD(&rth->rt_uncached);
1771         if (res.type == RTN_UNREACHABLE) {
1772                 rth->dst.input= ip_error;
1773                 rth->dst.error= -err;
1774                 rth->rt_flags   &= ~RTCF_LOCAL;
1775         }
1776         if (do_cache) {
1777                 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1778                         rth->dst.flags |= DST_NOCACHE;
1779                         rt_add_uncached_list(rth);
1780                 }
1781         }
1782         skb_dst_set(skb, &rth->dst);
1783         err = 0;
1784         goto out;
1785
1786 no_route:
1787         RT_CACHE_STAT_INC(in_no_route);
1788         res.type = RTN_UNREACHABLE;
1789         if (err == -ESRCH)
1790                 err = -ENETUNREACH;
1791         goto local_input;
1792
1793         /*
1794          *      Do not cache martian addresses: they should be logged (RFC1812)
1795          */
1796 martian_destination:
1797         RT_CACHE_STAT_INC(in_martian_dst);
1798 #ifdef CONFIG_IP_ROUTE_VERBOSE
1799         if (IN_DEV_LOG_MARTIANS(in_dev))
1800                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1801                                      &daddr, &saddr, dev->name);
1802 #endif
1803
1804 e_inval:
1805         err = -EINVAL;
1806         goto out;
1807
1808 e_nobufs:
1809         err = -ENOBUFS;
1810         goto out;
1811
1812 martian_source:
1813         err = -EINVAL;
1814 martian_source_keep_err:
1815         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1816         goto out;
1817 }
1818
1819 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1820                          u8 tos, struct net_device *dev)
1821 {
1822         int res;
1823
1824         rcu_read_lock();
1825
1826         /* Multicast recognition logic is moved from route cache to here.
1827            The problem was that too many Ethernet cards have broken/missing
1828            hardware multicast filters :-( As result the host on multicasting
1829            network acquires a lot of useless route cache entries, sort of
1830            SDR messages from all the world. Now we try to get rid of them.
1831            Really, provided software IP multicast filter is organized
1832            reasonably (at least, hashed), it does not result in a slowdown
1833            comparing with route cache reject entries.
1834            Note, that multicast routers are not affected, because
1835            route cache entry is created eventually.
1836          */
1837         if (ipv4_is_multicast(daddr)) {
1838                 struct in_device *in_dev = __in_dev_get_rcu(dev);
1839
1840                 if (in_dev) {
1841                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1842                                                   ip_hdr(skb)->protocol);
1843                         if (our
1844 #ifdef CONFIG_IP_MROUTE
1845                                 ||
1846                             (!ipv4_is_local_multicast(daddr) &&
1847                              IN_DEV_MFORWARD(in_dev))
1848 #endif
1849                            ) {
1850                                 int res = ip_route_input_mc(skb, daddr, saddr,
1851                                                             tos, dev, our);
1852                                 rcu_read_unlock();
1853                                 return res;
1854                         }
1855                 }
1856                 rcu_read_unlock();
1857                 return -EINVAL;
1858         }
1859         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1860         rcu_read_unlock();
1861         return res;
1862 }
1863 EXPORT_SYMBOL(ip_route_input_noref);
1864
1865 /* called with rcu_read_lock() */
1866 static struct rtable *__mkroute_output(const struct fib_result *res,
1867                                        const struct flowi4 *fl4, int orig_oif,
1868                                        struct net_device *dev_out,
1869                                        unsigned int flags)
1870 {
1871         struct fib_info *fi = res->fi;
1872         struct fib_nh_exception *fnhe;
1873         struct in_device *in_dev;
1874         u16 type = res->type;
1875         struct rtable *rth;
1876         bool do_cache;
1877
1878         in_dev = __in_dev_get_rcu(dev_out);
1879         if (!in_dev)
1880                 return ERR_PTR(-EINVAL);
1881
1882         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1883                 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1884                         return ERR_PTR(-EINVAL);
1885
1886         if (ipv4_is_lbcast(fl4->daddr))
1887                 type = RTN_BROADCAST;
1888         else if (ipv4_is_multicast(fl4->daddr))
1889                 type = RTN_MULTICAST;
1890         else if (ipv4_is_zeronet(fl4->daddr))
1891                 return ERR_PTR(-EINVAL);
1892
1893         if (dev_out->flags & IFF_LOOPBACK)
1894                 flags |= RTCF_LOCAL;
1895
1896         do_cache = true;
1897         if (type == RTN_BROADCAST) {
1898                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1899                 fi = NULL;
1900         } else if (type == RTN_MULTICAST) {
1901                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1902                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1903                                      fl4->flowi4_proto))
1904                         flags &= ~RTCF_LOCAL;
1905                 else
1906                         do_cache = false;
1907                 /* If multicast route do not exist use
1908                  * default one, but do not gateway in this case.
1909                  * Yes, it is hack.
1910                  */
1911                 if (fi && res->prefixlen < 4)
1912                         fi = NULL;
1913         }
1914
1915         fnhe = NULL;
1916         do_cache &= fi != NULL;
1917         if (do_cache) {
1918                 struct rtable __rcu **prth;
1919                 struct fib_nh *nh = &FIB_RES_NH(*res);
1920
1921                 fnhe = find_exception(nh, fl4->daddr);
1922                 if (fnhe)
1923                         prth = &fnhe->fnhe_rth_output;
1924                 else {
1925                         if (unlikely(fl4->flowi4_flags &
1926                                      FLOWI_FLAG_KNOWN_NH &&
1927                                      !(nh->nh_gw &&
1928                                        nh->nh_scope == RT_SCOPE_LINK))) {
1929                                 do_cache = false;
1930                                 goto add;
1931                         }
1932                         prth = __this_cpu_ptr(nh->nh_pcpu_rth_output);
1933                 }
1934                 rth = rcu_dereference(*prth);
1935                 if (rt_cache_valid(rth)) {
1936                         dst_hold(&rth->dst);
1937                         return rth;
1938                 }
1939         }
1940
1941 add:
1942         rth = rt_dst_alloc(dev_out,
1943                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1944                            IN_DEV_CONF_GET(in_dev, NOXFRM),
1945                            do_cache);
1946         if (!rth)
1947                 return ERR_PTR(-ENOBUFS);
1948
1949         rth->dst.output = ip_output;
1950
1951         rth->rt_genid = rt_genid_ipv4(dev_net(dev_out));
1952         rth->rt_flags   = flags;
1953         rth->rt_type    = type;
1954         rth->rt_is_input = 0;
1955         rth->rt_iif     = orig_oif ? : 0;
1956         rth->rt_pmtu    = 0;
1957         rth->rt_gateway = 0;
1958         rth->rt_uses_gateway = 0;
1959         INIT_LIST_HEAD(&rth->rt_uncached);
1960
1961         RT_CACHE_STAT_INC(out_slow_tot);
1962
1963         if (flags & RTCF_LOCAL)
1964                 rth->dst.input = ip_local_deliver;
1965         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1966                 if (flags & RTCF_LOCAL &&
1967                     !(dev_out->flags & IFF_LOOPBACK)) {
1968                         rth->dst.output = ip_mc_output;
1969                         RT_CACHE_STAT_INC(out_slow_mc);
1970                 }
1971 #ifdef CONFIG_IP_MROUTE
1972                 if (type == RTN_MULTICAST) {
1973                         if (IN_DEV_MFORWARD(in_dev) &&
1974                             !ipv4_is_local_multicast(fl4->daddr)) {
1975                                 rth->dst.input = ip_mr_input;
1976                                 rth->dst.output = ip_mc_output;
1977                         }
1978                 }
1979 #endif
1980         }
1981
1982         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
1983
1984         return rth;
1985 }
1986
1987 /*
1988  * Major route resolver routine.
1989  */
1990
1991 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
1992 {
1993         struct net_device *dev_out = NULL;
1994         __u8 tos = RT_FL_TOS(fl4);
1995         unsigned int flags = 0;
1996         struct fib_result res;
1997         struct rtable *rth;
1998         int orig_oif;
1999
2000         res.tclassid    = 0;
2001         res.fi          = NULL;
2002         res.table       = NULL;
2003
2004         orig_oif = fl4->flowi4_oif;
2005
2006         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2007         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2008         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2009                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2010
2011         rcu_read_lock();
2012         if (fl4->saddr) {
2013                 rth = ERR_PTR(-EINVAL);
2014                 if (ipv4_is_multicast(fl4->saddr) ||
2015                     ipv4_is_lbcast(fl4->saddr) ||
2016                     ipv4_is_zeronet(fl4->saddr))
2017                         goto out;
2018
2019                 /* I removed check for oif == dev_out->oif here.
2020                    It was wrong for two reasons:
2021                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2022                       is assigned to multiple interfaces.
2023                    2. Moreover, we are allowed to send packets with saddr
2024                       of another iface. --ANK
2025                  */
2026
2027                 if (fl4->flowi4_oif == 0 &&
2028                     (ipv4_is_multicast(fl4->daddr) ||
2029                      ipv4_is_lbcast(fl4->daddr))) {
2030                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2031                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2032                         if (dev_out == NULL)
2033                                 goto out;
2034
2035                         /* Special hack: user can direct multicasts
2036                            and limited broadcast via necessary interface
2037                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2038                            This hack is not just for fun, it allows
2039                            vic,vat and friends to work.
2040                            They bind socket to loopback, set ttl to zero
2041                            and expect that it will work.
2042                            From the viewpoint of routing cache they are broken,
2043                            because we are not allowed to build multicast path
2044                            with loopback source addr (look, routing cache
2045                            cannot know, that ttl is zero, so that packet
2046                            will not leave this host and route is valid).
2047                            Luckily, this hack is good workaround.
2048                          */
2049
2050                         fl4->flowi4_oif = dev_out->ifindex;
2051                         goto make_route;
2052                 }
2053
2054                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2055                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2056                         if (!__ip_dev_find(net, fl4->saddr, false))
2057                                 goto out;
2058                 }
2059         }
2060
2061
2062         if (fl4->flowi4_oif) {
2063                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2064                 rth = ERR_PTR(-ENODEV);
2065                 if (dev_out == NULL)
2066                         goto out;
2067
2068                 /* RACE: Check return value of inet_select_addr instead. */
2069                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2070                         rth = ERR_PTR(-ENETUNREACH);
2071                         goto out;
2072                 }
2073                 if (ipv4_is_local_multicast(fl4->daddr) ||
2074                     ipv4_is_lbcast(fl4->daddr)) {
2075                         if (!fl4->saddr)
2076                                 fl4->saddr = inet_select_addr(dev_out, 0,
2077                                                               RT_SCOPE_LINK);
2078                         goto make_route;
2079                 }
2080                 if (!fl4->saddr) {
2081                         if (ipv4_is_multicast(fl4->daddr))
2082                                 fl4->saddr = inet_select_addr(dev_out, 0,
2083                                                               fl4->flowi4_scope);
2084                         else if (!fl4->daddr)
2085                                 fl4->saddr = inet_select_addr(dev_out, 0,
2086                                                               RT_SCOPE_HOST);
2087                 }
2088         }
2089
2090         if (!fl4->daddr) {
2091                 fl4->daddr = fl4->saddr;
2092                 if (!fl4->daddr)
2093                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2094                 dev_out = net->loopback_dev;
2095                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2096                 res.type = RTN_LOCAL;
2097                 flags |= RTCF_LOCAL;
2098                 goto make_route;
2099         }
2100
2101         if (fib_lookup(net, fl4, &res)) {
2102                 res.fi = NULL;
2103                 res.table = NULL;
2104                 if (fl4->flowi4_oif) {
2105                         /* Apparently, routing tables are wrong. Assume,
2106                            that the destination is on link.
2107
2108                            WHY? DW.
2109                            Because we are allowed to send to iface
2110                            even if it has NO routes and NO assigned
2111                            addresses. When oif is specified, routing
2112                            tables are looked up with only one purpose:
2113                            to catch if destination is gatewayed, rather than
2114                            direct. Moreover, if MSG_DONTROUTE is set,
2115                            we send packet, ignoring both routing tables
2116                            and ifaddr state. --ANK
2117
2118
2119                            We could make it even if oif is unknown,
2120                            likely IPv6, but we do not.
2121                          */
2122
2123                         if (fl4->saddr == 0)
2124                                 fl4->saddr = inet_select_addr(dev_out, 0,
2125                                                               RT_SCOPE_LINK);
2126                         res.type = RTN_UNICAST;
2127                         goto make_route;
2128                 }
2129                 rth = ERR_PTR(-ENETUNREACH);
2130                 goto out;
2131         }
2132
2133         if (res.type == RTN_LOCAL) {
2134                 if (!fl4->saddr) {
2135                         if (res.fi->fib_prefsrc)
2136                                 fl4->saddr = res.fi->fib_prefsrc;
2137                         else
2138                                 fl4->saddr = fl4->daddr;
2139                 }
2140                 dev_out = net->loopback_dev;
2141                 fl4->flowi4_oif = dev_out->ifindex;
2142                 flags |= RTCF_LOCAL;
2143                 goto make_route;
2144         }
2145
2146 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2147         if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2148                 fib_select_multipath(&res);
2149         else
2150 #endif
2151         if (!res.prefixlen &&
2152             res.table->tb_num_default > 1 &&
2153             res.type == RTN_UNICAST && !fl4->flowi4_oif)
2154                 fib_select_default(&res);
2155
2156         if (!fl4->saddr)
2157                 fl4->saddr = FIB_RES_PREFSRC(net, res);
2158
2159         dev_out = FIB_RES_DEV(res);
2160         fl4->flowi4_oif = dev_out->ifindex;
2161
2162
2163 make_route:
2164         rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2165
2166 out:
2167         rcu_read_unlock();
2168         return rth;
2169 }
2170 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2171
2172 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2173 {
2174         return NULL;
2175 }
2176
2177 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2178 {
2179         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2180
2181         return mtu ? : dst->dev->mtu;
2182 }
2183
2184 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2185                                           struct sk_buff *skb, u32 mtu)
2186 {
2187 }
2188
2189 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2190                                        struct sk_buff *skb)
2191 {
2192 }
2193
2194 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2195                                           unsigned long old)
2196 {
2197         return NULL;
2198 }
2199
2200 static struct dst_ops ipv4_dst_blackhole_ops = {
2201         .family                 =       AF_INET,
2202         .protocol               =       cpu_to_be16(ETH_P_IP),
2203         .check                  =       ipv4_blackhole_dst_check,
2204         .mtu                    =       ipv4_blackhole_mtu,
2205         .default_advmss         =       ipv4_default_advmss,
2206         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2207         .redirect               =       ipv4_rt_blackhole_redirect,
2208         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
2209         .neigh_lookup           =       ipv4_neigh_lookup,
2210 };
2211
2212 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2213 {
2214         struct rtable *ort = (struct rtable *) dst_orig;
2215         struct rtable *rt;
2216
2217         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2218         if (rt) {
2219                 struct dst_entry *new = &rt->dst;
2220
2221                 new->__use = 1;
2222                 new->input = dst_discard;
2223                 new->output = dst_discard;
2224
2225                 new->dev = ort->dst.dev;
2226                 if (new->dev)
2227                         dev_hold(new->dev);
2228
2229                 rt->rt_is_input = ort->rt_is_input;
2230                 rt->rt_iif = ort->rt_iif;
2231                 rt->rt_pmtu = ort->rt_pmtu;
2232
2233                 rt->rt_genid = rt_genid_ipv4(net);
2234                 rt->rt_flags = ort->rt_flags;
2235                 rt->rt_type = ort->rt_type;
2236                 rt->rt_gateway = ort->rt_gateway;
2237                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2238
2239                 INIT_LIST_HEAD(&rt->rt_uncached);
2240
2241                 dst_free(new);
2242         }
2243
2244         dst_release(dst_orig);
2245
2246         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2247 }
2248
2249 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2250                                     struct sock *sk)
2251 {
2252         struct rtable *rt = __ip_route_output_key(net, flp4);
2253
2254         if (IS_ERR(rt))
2255                 return rt;
2256
2257         if (flp4->flowi4_proto)
2258                 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2259                                                    flowi4_to_flowi(flp4),
2260                                                    sk, 0);
2261
2262         return rt;
2263 }
2264 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2265
2266 static int rt_fill_info(struct net *net,  __be32 dst, __be32 src,
2267                         struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2268                         u32 seq, int event, int nowait, unsigned int flags)
2269 {
2270         struct rtable *rt = skb_rtable(skb);
2271         struct rtmsg *r;
2272         struct nlmsghdr *nlh;
2273         unsigned long expires = 0;
2274         u32 error;
2275         u32 metrics[RTAX_MAX];
2276
2277         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2278         if (nlh == NULL)
2279                 return -EMSGSIZE;
2280
2281         r = nlmsg_data(nlh);
2282         r->rtm_family    = AF_INET;
2283         r->rtm_dst_len  = 32;
2284         r->rtm_src_len  = 0;
2285         r->rtm_tos      = fl4->flowi4_tos;
2286         r->rtm_table    = RT_TABLE_MAIN;
2287         if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2288                 goto nla_put_failure;
2289         r->rtm_type     = rt->rt_type;
2290         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2291         r->rtm_protocol = RTPROT_UNSPEC;
2292         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2293         if (rt->rt_flags & RTCF_NOTIFY)
2294                 r->rtm_flags |= RTM_F_NOTIFY;
2295
2296         if (nla_put_be32(skb, RTA_DST, dst))
2297                 goto nla_put_failure;
2298         if (src) {
2299                 r->rtm_src_len = 32;
2300                 if (nla_put_be32(skb, RTA_SRC, src))
2301                         goto nla_put_failure;
2302         }
2303         if (rt->dst.dev &&
2304             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2305                 goto nla_put_failure;
2306 #ifdef CONFIG_IP_ROUTE_CLASSID
2307         if (rt->dst.tclassid &&
2308             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2309                 goto nla_put_failure;
2310 #endif
2311         if (!rt_is_input_route(rt) &&
2312             fl4->saddr != src) {
2313                 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2314                         goto nla_put_failure;
2315         }
2316         if (rt->rt_uses_gateway &&
2317             nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2318                 goto nla_put_failure;
2319
2320         expires = rt->dst.expires;
2321         if (expires) {
2322                 unsigned long now = jiffies;
2323
2324                 if (time_before(now, expires))
2325                         expires -= now;
2326                 else
2327                         expires = 0;
2328         }
2329
2330         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2331         if (rt->rt_pmtu && expires)
2332                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2333         if (rtnetlink_put_metrics(skb, metrics) < 0)
2334                 goto nla_put_failure;
2335
2336         if (fl4->flowi4_mark &&
2337             nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2338                 goto nla_put_failure;
2339
2340         error = rt->dst.error;
2341
2342         if (rt_is_input_route(rt)) {
2343 #ifdef CONFIG_IP_MROUTE
2344                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2345                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2346                         int err = ipmr_get_route(net, skb,
2347                                                  fl4->saddr, fl4->daddr,
2348                                                  r, nowait);
2349                         if (err <= 0) {
2350                                 if (!nowait) {
2351                                         if (err == 0)
2352                                                 return 0;
2353                                         goto nla_put_failure;
2354                                 } else {
2355                                         if (err == -EMSGSIZE)
2356                                                 goto nla_put_failure;
2357                                         error = err;
2358                                 }
2359                         }
2360                 } else
2361 #endif
2362                         if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
2363                                 goto nla_put_failure;
2364         }
2365
2366         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2367                 goto nla_put_failure;
2368
2369         return nlmsg_end(skb, nlh);
2370
2371 nla_put_failure:
2372         nlmsg_cancel(skb, nlh);
2373         return -EMSGSIZE;
2374 }
2375
2376 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2377 {
2378         struct net *net = sock_net(in_skb->sk);
2379         struct rtmsg *rtm;
2380         struct nlattr *tb[RTA_MAX+1];
2381         struct rtable *rt = NULL;
2382         struct flowi4 fl4;
2383         __be32 dst = 0;
2384         __be32 src = 0;
2385         u32 iif;
2386         int err;
2387         int mark;
2388         struct sk_buff *skb;
2389
2390         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2391         if (err < 0)
2392                 goto errout;
2393
2394         rtm = nlmsg_data(nlh);
2395
2396         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2397         if (skb == NULL) {
2398                 err = -ENOBUFS;
2399                 goto errout;
2400         }
2401
2402         /* Reserve room for dummy headers, this skb can pass
2403            through good chunk of routing engine.
2404          */
2405         skb_reset_mac_header(skb);
2406         skb_reset_network_header(skb);
2407
2408         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2409         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2410         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2411
2412         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2413         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2414         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2415         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2416
2417         memset(&fl4, 0, sizeof(fl4));
2418         fl4.daddr = dst;
2419         fl4.saddr = src;
2420         fl4.flowi4_tos = rtm->rtm_tos;
2421         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2422         fl4.flowi4_mark = mark;
2423
2424         if (iif) {
2425                 struct net_device *dev;
2426
2427                 dev = __dev_get_by_index(net, iif);
2428                 if (dev == NULL) {
2429                         err = -ENODEV;
2430                         goto errout_free;
2431                 }
2432
2433                 skb->protocol   = htons(ETH_P_IP);
2434                 skb->dev        = dev;
2435                 skb->mark       = mark;
2436                 local_bh_disable();
2437                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2438                 local_bh_enable();
2439
2440                 rt = skb_rtable(skb);
2441                 if (err == 0 && rt->dst.error)
2442                         err = -rt->dst.error;
2443         } else {
2444                 rt = ip_route_output_key(net, &fl4);
2445
2446                 err = 0;
2447                 if (IS_ERR(rt))
2448                         err = PTR_ERR(rt);
2449         }
2450
2451         if (err)
2452                 goto errout_free;
2453
2454         skb_dst_set(skb, &rt->dst);
2455         if (rtm->rtm_flags & RTM_F_NOTIFY)
2456                 rt->rt_flags |= RTCF_NOTIFY;
2457
2458         err = rt_fill_info(net, dst, src, &fl4, skb,
2459                            NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2460                            RTM_NEWROUTE, 0, 0);
2461         if (err <= 0)
2462                 goto errout_free;
2463
2464         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2465 errout:
2466         return err;
2467
2468 errout_free:
2469         kfree_skb(skb);
2470         goto errout;
2471 }
2472
2473 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
2474 {
2475         return skb->len;
2476 }
2477
2478 void ip_rt_multicast_event(struct in_device *in_dev)
2479 {
2480         rt_cache_flush(dev_net(in_dev->dev));
2481 }
2482
2483 #ifdef CONFIG_SYSCTL
2484 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
2485 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
2486 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
2487 static int ip_rt_gc_elasticity __read_mostly    = 8;
2488
2489 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2490                                         void __user *buffer,
2491                                         size_t *lenp, loff_t *ppos)
2492 {
2493         struct net *net = (struct net *)__ctl->extra1;
2494
2495         if (write) {
2496                 rt_cache_flush(net);
2497                 fnhe_genid_bump(net);
2498                 return 0;
2499         }
2500
2501         return -EINVAL;
2502 }
2503
2504 static struct ctl_table ipv4_route_table[] = {
2505         {
2506                 .procname       = "gc_thresh",
2507                 .data           = &ipv4_dst_ops.gc_thresh,
2508                 .maxlen         = sizeof(int),
2509                 .mode           = 0644,
2510                 .proc_handler   = proc_dointvec,
2511         },
2512         {
2513                 .procname       = "max_size",
2514                 .data           = &ip_rt_max_size,
2515                 .maxlen         = sizeof(int),
2516                 .mode           = 0644,
2517                 .proc_handler   = proc_dointvec,
2518         },
2519         {
2520                 /*  Deprecated. Use gc_min_interval_ms */
2521
2522                 .procname       = "gc_min_interval",
2523                 .data           = &ip_rt_gc_min_interval,
2524                 .maxlen         = sizeof(int),
2525                 .mode           = 0644,
2526                 .proc_handler   = proc_dointvec_jiffies,
2527         },
2528         {
2529                 .procname       = "gc_min_interval_ms",
2530                 .data           = &ip_rt_gc_min_interval,
2531                 .maxlen         = sizeof(int),
2532                 .mode           = 0644,
2533                 .proc_handler   = proc_dointvec_ms_jiffies,
2534         },
2535         {
2536                 .procname       = "gc_timeout",
2537                 .data           = &ip_rt_gc_timeout,
2538                 .maxlen         = sizeof(int),
2539                 .mode           = 0644,
2540                 .proc_handler   = proc_dointvec_jiffies,
2541         },
2542         {
2543                 .procname       = "gc_interval",
2544                 .data           = &ip_rt_gc_interval,
2545                 .maxlen         = sizeof(int),
2546                 .mode           = 0644,
2547                 .proc_handler   = proc_dointvec_jiffies,
2548         },
2549         {
2550                 .procname       = "redirect_load",
2551                 .data           = &ip_rt_redirect_load,
2552                 .maxlen         = sizeof(int),
2553                 .mode           = 0644,
2554                 .proc_handler   = proc_dointvec,
2555         },
2556         {
2557                 .procname       = "redirect_number",
2558                 .data           = &ip_rt_redirect_number,
2559                 .maxlen         = sizeof(int),
2560                 .mode           = 0644,
2561                 .proc_handler   = proc_dointvec,
2562         },
2563         {
2564                 .procname       = "redirect_silence",
2565                 .data           = &ip_rt_redirect_silence,
2566                 .maxlen         = sizeof(int),
2567                 .mode           = 0644,
2568                 .proc_handler   = proc_dointvec,
2569         },
2570         {
2571                 .procname       = "error_cost",
2572                 .data           = &ip_rt_error_cost,
2573                 .maxlen         = sizeof(int),
2574                 .mode           = 0644,
2575                 .proc_handler   = proc_dointvec,
2576         },
2577         {
2578                 .procname       = "error_burst",
2579                 .data           = &ip_rt_error_burst,
2580                 .maxlen         = sizeof(int),
2581                 .mode           = 0644,
2582                 .proc_handler   = proc_dointvec,
2583         },
2584         {
2585                 .procname       = "gc_elasticity",
2586                 .data           = &ip_rt_gc_elasticity,
2587                 .maxlen         = sizeof(int),
2588                 .mode           = 0644,
2589                 .proc_handler   = proc_dointvec,
2590         },
2591         {
2592                 .procname       = "mtu_expires",
2593                 .data           = &ip_rt_mtu_expires,
2594                 .maxlen         = sizeof(int),
2595                 .mode           = 0644,
2596                 .proc_handler   = proc_dointvec_jiffies,
2597         },
2598         {
2599                 .procname       = "min_pmtu",
2600                 .data           = &ip_rt_min_pmtu,
2601                 .maxlen         = sizeof(int),
2602                 .mode           = 0644,
2603                 .proc_handler   = proc_dointvec,
2604         },
2605         {
2606                 .procname       = "min_adv_mss",
2607                 .data           = &ip_rt_min_advmss,
2608                 .maxlen         = sizeof(int),
2609                 .mode           = 0644,
2610                 .proc_handler   = proc_dointvec,
2611         },
2612         { }
2613 };
2614
2615 static struct ctl_table ipv4_route_flush_table[] = {
2616         {
2617                 .procname       = "flush",
2618                 .maxlen         = sizeof(int),
2619                 .mode           = 0200,
2620                 .proc_handler   = ipv4_sysctl_rtcache_flush,
2621         },
2622         { },
2623 };
2624
2625 static __net_init int sysctl_route_net_init(struct net *net)
2626 {
2627         struct ctl_table *tbl;
2628
2629         tbl = ipv4_route_flush_table;
2630         if (!net_eq(net, &init_net)) {
2631                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2632                 if (tbl == NULL)
2633                         goto err_dup;
2634
2635                 /* Don't export sysctls to unprivileged users */
2636                 if (net->user_ns != &init_user_ns)
2637                         tbl[0].procname = NULL;
2638         }
2639         tbl[0].extra1 = net;
2640
2641         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2642         if (net->ipv4.route_hdr == NULL)
2643                 goto err_reg;
2644         return 0;
2645
2646 err_reg:
2647         if (tbl != ipv4_route_flush_table)
2648                 kfree(tbl);
2649 err_dup:
2650         return -ENOMEM;
2651 }
2652
2653 static __net_exit void sysctl_route_net_exit(struct net *net)
2654 {
2655         struct ctl_table *tbl;
2656
2657         tbl = net->ipv4.route_hdr->ctl_table_arg;
2658         unregister_net_sysctl_table(net->ipv4.route_hdr);
2659         BUG_ON(tbl == ipv4_route_flush_table);
2660         kfree(tbl);
2661 }
2662
2663 static __net_initdata struct pernet_operations sysctl_route_ops = {
2664         .init = sysctl_route_net_init,
2665         .exit = sysctl_route_net_exit,
2666 };
2667 #endif
2668
2669 static __net_init int rt_genid_init(struct net *net)
2670 {
2671         atomic_set(&net->ipv4.rt_genid, 0);
2672         atomic_set(&net->fnhe_genid, 0);
2673         get_random_bytes(&net->ipv4.dev_addr_genid,
2674                          sizeof(net->ipv4.dev_addr_genid));
2675         return 0;
2676 }
2677
2678 static __net_initdata struct pernet_operations rt_genid_ops = {
2679         .init = rt_genid_init,
2680 };
2681
2682 static int __net_init ipv4_inetpeer_init(struct net *net)
2683 {
2684         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2685
2686         if (!bp)
2687                 return -ENOMEM;
2688         inet_peer_base_init(bp);
2689         net->ipv4.peers = bp;
2690         return 0;
2691 }
2692
2693 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2694 {
2695         struct inet_peer_base *bp = net->ipv4.peers;
2696
2697         net->ipv4.peers = NULL;
2698         inetpeer_invalidate_tree(bp);
2699         kfree(bp);
2700 }
2701
2702 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2703         .init   =       ipv4_inetpeer_init,
2704         .exit   =       ipv4_inetpeer_exit,
2705 };
2706
2707 #ifdef CONFIG_IP_ROUTE_CLASSID
2708 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2709 #endif /* CONFIG_IP_ROUTE_CLASSID */
2710
2711 int __init ip_rt_init(void)
2712 {
2713         int rc = 0;
2714
2715 #ifdef CONFIG_IP_ROUTE_CLASSID
2716         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2717         if (!ip_rt_acct)
2718                 panic("IP: failed to allocate ip_rt_acct\n");
2719 #endif
2720
2721         ipv4_dst_ops.kmem_cachep =
2722                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2723                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2724
2725         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2726
2727         if (dst_entries_init(&ipv4_dst_ops) < 0)
2728                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2729
2730         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2731                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2732
2733         ipv4_dst_ops.gc_thresh = ~0;
2734         ip_rt_max_size = INT_MAX;
2735
2736         devinet_init();
2737         ip_fib_init();
2738
2739         if (ip_rt_proc_init())
2740                 pr_err("Unable to create route proc files\n");
2741 #ifdef CONFIG_XFRM
2742         xfrm_init();
2743         xfrm4_init();
2744 #endif
2745         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2746
2747 #ifdef CONFIG_SYSCTL
2748         register_pernet_subsys(&sysctl_route_ops);
2749 #endif
2750         register_pernet_subsys(&rt_genid_ops);
2751         register_pernet_subsys(&ipv4_inetpeer_ops);
2752         return rc;
2753 }
2754
2755 #ifdef CONFIG_SYSCTL
2756 /*
2757  * We really need to sanitize the damn ipv4 init order, then all
2758  * this nonsense will go away.
2759  */
2760 void __init ip_static_sysctl_init(void)
2761 {
2762         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
2763 }
2764 #endif