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 <linux/jhash.h>
  93 #include <net/dst.h>
  94 #include <net/dst_metadata.h>
  95 #include <net/net_namespace.h>
  96 #include <net/protocol.h>
  97 #include <net/ip.h>
  98 #include <net/route.h>
  99 #include <net/inetpeer.h>
 100 #include <net/sock.h>
 101 #include <net/ip_fib.h>
 102 #include <net/arp.h>
 103 #include <net/tcp.h>
 104 #include <net/icmp.h>
 105 #include <net/xfrm.h>
 106 #include <net/lwtunnel.h>
 107 #include <net/netevent.h>
 108 #include <net/rtnetlink.h>
 109 #ifdef CONFIG_SYSCTL
 110 #include <linux/sysctl.h>
 111 #include <linux/kmemleak.h>
 112 #endif
 113 #include <net/secure_seq.h>
 114 #include <net/ip_tunnels.h>
 115 #include <net/l3mdev.h>
 116
 117 #define RT_FL_TOS(oldflp4) \
 118         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
 119
 120 #define RT_GC_TIMEOUT (300*HZ)
 121
 122 static int ip_rt_max_size;
 123 static int ip_rt_redirect_number __read_mostly  = 9;
 124 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 125 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 126 static int ip_rt_error_cost __read_mostly       = HZ;
 127 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 128 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
 129 static int ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
 130 static int ip_rt_min_advmss __read_mostly       = 256;
 131
 132 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
 133 /*
 134  *      Interface to generic destination cache.
 135  */
 136
 137 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 138 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 139 static unsigned int      ipv4_mtu(const struct dst_entry *dst);
 140 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 141 static void              ipv4_link_failure(struct sk_buff *skb);
 142 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 143                                            struct sk_buff *skb, u32 mtu);
 144 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 145                                         struct sk_buff *skb);
 146 static void             ipv4_dst_destroy(struct dst_entry *dst);
 147
 148 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 149 {
 150         WARN_ON(1);
 151         return NULL;
 152 }
 153
 154 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 155                                            struct sk_buff *skb,
 156                                            const void *daddr);
 157
 158 static struct dst_ops ipv4_dst_ops = {
 159         .family =               AF_INET,
 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         .negative_advice =      ipv4_negative_advice,
 166         .link_failure =         ipv4_link_failure,
 167         .update_pmtu =          ip_rt_update_pmtu,
 168         .redirect =             ip_do_redirect,
 169         .local_out =            __ip_local_out,
 170         .neigh_lookup =         ipv4_neigh_lookup,
 171 };
 172
 173 #define ECN_OR_COST(class)      TC_PRIO_##class
 174
 175 const __u8 ip_tos2prio[16] = {
 176         TC_PRIO_BESTEFFORT,
 177         ECN_OR_COST(BESTEFFORT),
 178         TC_PRIO_BESTEFFORT,
 179         ECN_OR_COST(BESTEFFORT),
 180         TC_PRIO_BULK,
 181         ECN_OR_COST(BULK),
 182         TC_PRIO_BULK,
 183         ECN_OR_COST(BULK),
 184         TC_PRIO_INTERACTIVE,
 185         ECN_OR_COST(INTERACTIVE),
 186         TC_PRIO_INTERACTIVE,
 187         ECN_OR_COST(INTERACTIVE),
 188         TC_PRIO_INTERACTIVE_BULK,
 189         ECN_OR_COST(INTERACTIVE_BULK),
 190         TC_PRIO_INTERACTIVE_BULK,
 191         ECN_OR_COST(INTERACTIVE_BULK)
 192 };
 193 EXPORT_SYMBOL(ip_tos2prio);
 194
 195 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 196 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
 197
 198 #ifdef CONFIG_PROC_FS
 199 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 200 {
 201         if (*pos)
 202                 return NULL;
 203         return SEQ_START_TOKEN;
 204 }
 205
 206 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 207 {
 208         ++*pos;
 209         return NULL;
 210 }
 211
 212 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 213 {
 214 }
 215
 216 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 217 {
 218         if (v == SEQ_START_TOKEN)
 219                 seq_printf(seq, "%-127s\n",
 220                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 221                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 222                            "HHUptod\tSpecDst");
 223         return 0;
 224 }
 225
 226 static const struct seq_operations rt_cache_seq_ops = {
 227         .start  = rt_cache_seq_start,
 228         .next   = rt_cache_seq_next,
 229         .stop   = rt_cache_seq_stop,
 230         .show   = rt_cache_seq_show,
 231 };
 232
 233 static int rt_cache_seq_open(struct inode *inode, struct file *file)
 234 {
 235         return seq_open(file, &rt_cache_seq_ops);
 236 }
 237
 238 static const struct file_operations rt_cache_seq_fops = {
 239         .owner   = THIS_MODULE,
 240         .open    = rt_cache_seq_open,
 241         .read    = seq_read,
 242         .llseek  = seq_lseek,
 243         .release = seq_release,
 244 };
 245
 246
 247 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 248 {
 249         int cpu;
 250
 251         if (*pos == 0)
 252                 return SEQ_START_TOKEN;
 253
 254         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 255                 if (!cpu_possible(cpu))
 256                         continue;
 257                 *pos = cpu+1;
 258                 return &per_cpu(rt_cache_stat, cpu);
 259         }
 260         return NULL;
 261 }
 262
 263 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 264 {
 265         int cpu;
 266
 267         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 268                 if (!cpu_possible(cpu))
 269                         continue;
 270                 *pos = cpu+1;
 271                 return &per_cpu(rt_cache_stat, cpu);
 272         }
 273         return NULL;
 274
 275 }
 276
 277 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 278 {
 279
 280 }
 281
 282 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 283 {
 284         struct rt_cache_stat *st = v;
 285
 286         if (v == SEQ_START_TOKEN) {
 287                 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");
 288                 return 0;
 289         }
 290
 291         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 292                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 293                    dst_entries_get_slow(&ipv4_dst_ops),
 294                    0, /* st->in_hit */
 295                    st->in_slow_tot,
 296                    st->in_slow_mc,
 297                    st->in_no_route,
 298                    st->in_brd,
 299                    st->in_martian_dst,
 300                    st->in_martian_src,
 301
 302                    0, /* st->out_hit */
 303                    st->out_slow_tot,
 304                    st->out_slow_mc,
 305
 306                    0, /* st->gc_total */
 307                    0, /* st->gc_ignored */
 308                    0, /* st->gc_goal_miss */
 309                    0, /* st->gc_dst_overflow */
 310                    0, /* st->in_hlist_search */
 311                    0  /* st->out_hlist_search */
 312                 );
 313         return 0;
 314 }
 315
 316 static const struct seq_operations rt_cpu_seq_ops = {
 317         .start  = rt_cpu_seq_start,
 318         .next   = rt_cpu_seq_next,
 319         .stop   = rt_cpu_seq_stop,
 320         .show   = rt_cpu_seq_show,
 321 };
 322
 323
 324 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
 325 {
 326         return seq_open(file, &rt_cpu_seq_ops);
 327 }
 328
 329 static const struct file_operations rt_cpu_seq_fops = {
 330         .owner   = THIS_MODULE,
 331         .open    = rt_cpu_seq_open,
 332         .read    = seq_read,
 333         .llseek  = seq_lseek,
 334         .release = seq_release,
 335 };
 336
 337 #ifdef CONFIG_IP_ROUTE_CLASSID
 338 static int rt_acct_proc_show(struct seq_file *m, void *v)
 339 {
 340         struct ip_rt_acct *dst, *src;
 341         unsigned int i, j;
 342
 343         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 344         if (!dst)
 345                 return -ENOMEM;
 346
 347         for_each_possible_cpu(i) {
 348                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 349                 for (j = 0; j < 256; j++) {
 350                         dst[j].o_bytes   += src[j].o_bytes;
 351                         dst[j].o_packets += src[j].o_packets;
 352                         dst[j].i_bytes   += src[j].i_bytes;
 353                         dst[j].i_packets += src[j].i_packets;
 354                 }
 355         }
 356
 357         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 358         kfree(dst);
 359         return 0;
 360 }
 361
 362 static int rt_acct_proc_open(struct inode *inode, struct file *file)
 363 {
 364         return single_open(file, rt_acct_proc_show, NULL);
 365 }
 366
 367 static const struct file_operations rt_acct_proc_fops = {
 368         .owner          = THIS_MODULE,
 369         .open           = rt_acct_proc_open,
 370         .read           = seq_read,
 371         .llseek         = seq_lseek,
 372         .release        = single_release,
 373 };
 374 #endif
 375
 376 static int __net_init ip_rt_do_proc_init(struct net *net)
 377 {
 378         struct proc_dir_entry *pde;
 379
 380         pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
 381                           &rt_cache_seq_fops);
 382         if (!pde)
 383                 goto err1;
 384
 385         pde = proc_create("rt_cache", S_IRUGO,
 386                           net->proc_net_stat, &rt_cpu_seq_fops);
 387         if (!pde)
 388                 goto err2;
 389
 390 #ifdef CONFIG_IP_ROUTE_CLASSID
 391         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
 392         if (!pde)
 393                 goto err3;
 394 #endif
 395         return 0;
 396
 397 #ifdef CONFIG_IP_ROUTE_CLASSID
 398 err3:
 399         remove_proc_entry("rt_cache", net->proc_net_stat);
 400 #endif
 401 err2:
 402         remove_proc_entry("rt_cache", net->proc_net);
 403 err1:
 404         return -ENOMEM;
 405 }
 406
 407 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 408 {
 409         remove_proc_entry("rt_cache", net->proc_net_stat);
 410         remove_proc_entry("rt_cache", net->proc_net);
 411 #ifdef CONFIG_IP_ROUTE_CLASSID
 412         remove_proc_entry("rt_acct", net->proc_net);
 413 #endif
 414 }
 415
 416 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 417         .init = ip_rt_do_proc_init,
 418         .exit = ip_rt_do_proc_exit,
 419 };
 420
 421 static int __init ip_rt_proc_init(void)
 422 {
 423         return register_pernet_subsys(&ip_rt_proc_ops);
 424 }
 425
 426 #else
 427 static inline int ip_rt_proc_init(void)
 428 {
 429         return 0;
 430 }
 431 #endif /* CONFIG_PROC_FS */
 432
 433 static inline bool rt_is_expired(const struct rtable *rth)
 434 {
 435         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 436 }
 437
 438 void rt_cache_flush(struct net *net)
 439 {
 440         rt_genid_bump_ipv4(net);
 441 }
 442
 443 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 444                                            struct sk_buff *skb,
 445                                            const void *daddr)
 446 {
 447         struct net_device *dev = dst->dev;
 448         const __be32 *pkey = daddr;
 449         const struct rtable *rt;
 450         struct neighbour *n;
 451
 452         rt = (const struct rtable *) dst;
 453         if (rt->rt_gateway)
 454                 pkey = (const __be32 *) &rt->rt_gateway;
 455         else if (skb)
 456                 pkey = &ip_hdr(skb)->daddr;
 457
 458         n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
 459         if (n)
 460                 return n;
 461         return neigh_create(&arp_tbl, pkey, dev);
 462 }
 463
 464 #define IP_IDENTS_SZ 2048u
 465
 466 static atomic_t *ip_idents __read_mostly;
 467 static u32 *ip_tstamps __read_mostly;
 468
 469 /* In order to protect privacy, we add a perturbation to identifiers
 470  * if one generator is seldom used. This makes hard for an attacker
 471  * to infer how many packets were sent between two points in time.
 472  */
 473 u32 ip_idents_reserve(u32 hash, int segs)
 474 {
 475         u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
 476         atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
 477         u32 old = ACCESS_ONCE(*p_tstamp);
 478         u32 now = (u32)jiffies;
 479         u32 new, delta = 0;
 480
 481         if (old != now && cmpxchg(p_tstamp, old, now) == old)
 482                 delta = prandom_u32_max(now - old);
 483
 484         /* Do not use atomic_add_return() as it makes UBSAN unhappy */
 485         do {
 486                 old = (u32)atomic_read(p_id);
 487                 new = old + delta + segs;
 488         } while (atomic_cmpxchg(p_id, old, new) != old);
 489
 490         return new - segs;
 491 }
 492 EXPORT_SYMBOL(ip_idents_reserve);
 493
 494 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
 495 {
 496         static u32 ip_idents_hashrnd __read_mostly;
 497         u32 hash, id;
 498
 499         net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
 500
 501         hash = jhash_3words((__force u32)iph->daddr,
 502                             (__force u32)iph->saddr,
 503                             iph->protocol ^ net_hash_mix(net),
 504                             ip_idents_hashrnd);
 505         id = ip_idents_reserve(hash, segs);
 506         iph->id = htons(id);
 507 }
 508 EXPORT_SYMBOL(__ip_select_ident);
 509
 510 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 511                              const struct iphdr *iph,
 512                              int oif, u8 tos,
 513                              u8 prot, u32 mark, int flow_flags)
 514 {
 515         if (sk) {
 516                 const struct inet_sock *inet = inet_sk(sk);
 517
 518                 oif = sk->sk_bound_dev_if;
 519                 mark = sk->sk_mark;
 520                 tos = RT_CONN_FLAGS(sk);
 521                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 522         }
 523         flowi4_init_output(fl4, oif, mark, tos,
 524                            RT_SCOPE_UNIVERSE, prot,
 525                            flow_flags,
 526                            iph->daddr, iph->saddr, 0, 0);
 527 }
 528
 529 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 530                                const struct sock *sk)
 531 {
 532         const struct iphdr *iph = ip_hdr(skb);
 533         int oif = skb->dev->ifindex;
 534         u8 tos = RT_TOS(iph->tos);
 535         u8 prot = iph->protocol;
 536         u32 mark = skb->mark;
 537
 538         __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
 539 }
 540
 541 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 542 {
 543         const struct inet_sock *inet = inet_sk(sk);
 544         const struct ip_options_rcu *inet_opt;
 545         __be32 daddr = inet->inet_daddr;
 546
 547         rcu_read_lock();
 548         inet_opt = rcu_dereference(inet->inet_opt);
 549         if (inet_opt && inet_opt->opt.srr)
 550                 daddr = inet_opt->opt.faddr;
 551         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 552                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 553                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 554                            inet_sk_flowi_flags(sk),
 555                            daddr, inet->inet_saddr, 0, 0);
 556         rcu_read_unlock();
 557 }
 558
 559 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 560                                  const struct sk_buff *skb)
 561 {
 562         if (skb)
 563                 build_skb_flow_key(fl4, skb, sk);
 564         else
 565                 build_sk_flow_key(fl4, sk);
 566 }
 567
 568 static inline void rt_free(struct rtable *rt)
 569 {
 570         call_rcu(&rt->dst.rcu_head, dst_rcu_free);
 571 }
 572
 573 static DEFINE_SPINLOCK(fnhe_lock);
 574
 575 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 576 {
 577         struct rtable *rt;
 578
 579         rt = rcu_dereference(fnhe->fnhe_rth_input);
 580         if (rt) {
 581                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 582                 rt_free(rt);
 583         }
 584         rt = rcu_dereference(fnhe->fnhe_rth_output);
 585         if (rt) {
 586                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 587                 rt_free(rt);
 588         }
 589 }
 590
 591 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
 592 {
 593         struct fib_nh_exception *fnhe, *oldest;
 594
 595         oldest = rcu_dereference(hash->chain);
 596         for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
 597              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 598                 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
 599                         oldest = fnhe;
 600         }
 601         fnhe_flush_routes(oldest);
 602         return oldest;
 603 }
 604
 605 static inline u32 fnhe_hashfun(__be32 daddr)
 606 {
 607         static u32 fnhe_hashrnd __read_mostly;
 608         u32 hval;
 609
 610         net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
 611         hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
 612         return hash_32(hval, FNHE_HASH_SHIFT);
 613 }
 614
 615 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 616 {
 617         rt->rt_pmtu = fnhe->fnhe_pmtu;
 618         rt->dst.expires = fnhe->fnhe_expires;
 619
 620         if (fnhe->fnhe_gw) {
 621                 rt->rt_flags |= RTCF_REDIRECTED;
 622                 rt->rt_gateway = fnhe->fnhe_gw;
 623                 rt->rt_uses_gateway = 1;
 624         }
 625 }
 626
 627 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
 628                                   u32 pmtu, unsigned long expires)
 629 {
 630         struct fnhe_hash_bucket *hash;
 631         struct fib_nh_exception *fnhe;
 632         struct rtable *rt;
 633         unsigned int i;
 634         int depth;
 635         u32 hval = fnhe_hashfun(daddr);
 636
 637         spin_lock_bh(&fnhe_lock);
 638
 639         hash = rcu_dereference(nh->nh_exceptions);
 640         if (!hash) {
 641                 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
 642                 if (!hash)
 643                         goto out_unlock;
 644                 rcu_assign_pointer(nh->nh_exceptions, hash);
 645         }
 646
 647         hash += hval;
 648
 649         depth = 0;
 650         for (fnhe = rcu_dereference(hash->chain); fnhe;
 651              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 652                 if (fnhe->fnhe_daddr == daddr)
 653                         break;
 654                 depth++;
 655         }
 656
 657         if (fnhe) {
 658                 if (gw)
 659                         fnhe->fnhe_gw = gw;
 660                 if (pmtu) {
 661                         fnhe->fnhe_pmtu = pmtu;
 662                         fnhe->fnhe_expires = max(1UL, expires);
 663                 }
 664                 /* Update all cached dsts too */
 665                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 666                 if (rt)
 667                         fill_route_from_fnhe(rt, fnhe);
 668                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 669                 if (rt)
 670                         fill_route_from_fnhe(rt, fnhe);
 671         } else {
 672                 if (depth > FNHE_RECLAIM_DEPTH)
 673                         fnhe = fnhe_oldest(hash);
 674                 else {
 675                         fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 676                         if (!fnhe)
 677                                 goto out_unlock;
 678
 679                         fnhe->fnhe_next = hash->chain;
 680                         rcu_assign_pointer(hash->chain, fnhe);
 681                 }
 682                 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
 683                 fnhe->fnhe_daddr = daddr;
 684                 fnhe->fnhe_gw = gw;
 685                 fnhe->fnhe_pmtu = pmtu;
 686                 fnhe->fnhe_expires = expires;
 687
 688                 /* Exception created; mark the cached routes for the nexthop
 689                  * stale, so anyone caching it rechecks if this exception
 690                  * applies to them.
 691                  */
 692                 rt = rcu_dereference(nh->nh_rth_input);
 693                 if (rt)
 694                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 695
 696                 for_each_possible_cpu(i) {
 697                         struct rtable __rcu **prt;
 698                         prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
 699                         rt = rcu_dereference(*prt);
 700                         if (rt)
 701                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 702                 }
 703         }
 704
 705         fnhe->fnhe_stamp = jiffies;
 706
 707 out_unlock:
 708         spin_unlock_bh(&fnhe_lock);
 709 }
 710
 711 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 712                              bool kill_route)
 713 {
 714         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 715         __be32 old_gw = ip_hdr(skb)->saddr;
 716         struct net_device *dev = skb->dev;
 717         struct in_device *in_dev;
 718         struct fib_result res;
 719         struct neighbour *n;
 720         struct net *net;
 721
 722         switch (icmp_hdr(skb)->code & 7) {
 723         case ICMP_REDIR_NET:
 724         case ICMP_REDIR_NETTOS:
 725         case ICMP_REDIR_HOST:
 726         case ICMP_REDIR_HOSTTOS:
 727                 break;
 728
 729         default:
 730                 return;
 731         }
 732
 733         if (rt->rt_gateway != old_gw)
 734                 return;
 735
 736         in_dev = __in_dev_get_rcu(dev);
 737         if (!in_dev)
 738                 return;
 739
 740         net = dev_net(dev);
 741         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 742             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 743             ipv4_is_zeronet(new_gw))
 744                 goto reject_redirect;
 745
 746         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 747                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 748                         goto reject_redirect;
 749                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 750                         goto reject_redirect;
 751         } else {
 752                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 753                         goto reject_redirect;
 754         }
 755
 756         n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
 757         if (!n)
 758                 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
 759         if (!IS_ERR(n)) {
 760                 if (!(n->nud_state & NUD_VALID)) {
 761                         neigh_event_send(n, NULL);
 762                 } else {
 763                         if (fib_lookup(net, fl4, &res, 0) == 0) {
 764                                 struct fib_nh *nh = &FIB_RES_NH(res);
 765
 766                                 update_or_create_fnhe(nh, fl4->daddr, new_gw,
 767                                                 0, jiffies + ip_rt_gc_timeout);
 768                         }
 769                         if (kill_route)
 770                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 771                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 772                 }
 773                 neigh_release(n);
 774         }
 775         return;
 776
 777 reject_redirect:
 778 #ifdef CONFIG_IP_ROUTE_VERBOSE
 779         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 780                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 781                 __be32 daddr = iph->daddr;
 782                 __be32 saddr = iph->saddr;
 783
 784                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 785                                      "  Advised path = %pI4 -> %pI4\n",
 786                                      &old_gw, dev->name, &new_gw,
 787                                      &saddr, &daddr);
 788         }
 789 #endif
 790         ;
 791 }
 792
 793 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 794 {
 795         struct rtable *rt;
 796         struct flowi4 fl4;
 797         const struct iphdr *iph = (const struct iphdr *) skb->data;
 798         int oif = skb->dev->ifindex;
 799         u8 tos = RT_TOS(iph->tos);
 800         u8 prot = iph->protocol;
 801         u32 mark = skb->mark;
 802
 803         rt = (struct rtable *) dst;
 804
 805         __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
 806         __ip_do_redirect(rt, skb, &fl4, true);
 807 }
 808
 809 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 810 {
 811         struct rtable *rt = (struct rtable *)dst;
 812         struct dst_entry *ret = dst;
 813
 814         if (rt) {
 815                 if (dst->obsolete > 0) {
 816                         ip_rt_put(rt);
 817                         ret = NULL;
 818                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 819                            rt->dst.expires) {
 820                         ip_rt_put(rt);
 821                         ret = NULL;
 822                 }
 823         }
 824         return ret;
 825 }
 826
 827 /*
 828  * Algorithm:
 829  *      1. The first ip_rt_redirect_number redirects are sent
 830  *         with exponential backoff, then we stop sending them at all,
 831  *         assuming that the host ignores our redirects.
 832  *      2. If we did not see packets requiring redirects
 833  *         during ip_rt_redirect_silence, we assume that the host
 834  *         forgot redirected route and start to send redirects again.
 835  *
 836  * This algorithm is much cheaper and more intelligent than dumb load limiting
 837  * in icmp.c.
 838  *
 839  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 840  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 841  */
 842
 843 void ip_rt_send_redirect(struct sk_buff *skb)
 844 {
 845         struct rtable *rt = skb_rtable(skb);
 846         struct in_device *in_dev;
 847         struct inet_peer *peer;
 848         struct net *net;
 849         int log_martians;
 850         int vif;
 851
 852         rcu_read_lock();
 853         in_dev = __in_dev_get_rcu(rt->dst.dev);
 854         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 855                 rcu_read_unlock();
 856                 return;
 857         }
 858         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 859         vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
 860         rcu_read_unlock();
 861
 862         net = dev_net(rt->dst.dev);
 863         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
 864         if (!peer) {
 865                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 866                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 867                 return;
 868         }
 869
 870         /* No redirected packets during ip_rt_redirect_silence;
 871          * reset the algorithm.
 872          */
 873         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
 874                 peer->rate_tokens = 0;
 875
 876         /* Too many ignored redirects; do not send anything
 877          * set dst.rate_last to the last seen redirected packet.
 878          */
 879         if (peer->rate_tokens >= ip_rt_redirect_number) {
 880                 peer->rate_last = jiffies;
 881                 goto out_put_peer;
 882         }
 883
 884         /* Check for load limit; set rate_last to the latest sent
 885          * redirect.
 886          */
 887         if (peer->rate_tokens == 0 ||
 888             time_after(jiffies,
 889                        (peer->rate_last +
 890                         (ip_rt_redirect_load << peer->rate_tokens)))) {
 891                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 892
 893                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 894                 peer->rate_last = jiffies;
 895                 ++peer->rate_tokens;
 896 #ifdef CONFIG_IP_ROUTE_VERBOSE
 897                 if (log_martians &&
 898                     peer->rate_tokens == ip_rt_redirect_number)
 899                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 900                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 901                                              &ip_hdr(skb)->daddr, &gw);
 902 #endif
 903         }
 904 out_put_peer:
 905         inet_putpeer(peer);
 906 }
 907
 908 static int ip_error(struct sk_buff *skb)
 909 {
 910         struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
 911         struct rtable *rt = skb_rtable(skb);
 912         struct inet_peer *peer;
 913         unsigned long now;
 914         struct net *net;
 915         bool send;
 916         int code;
 917
 918         /* IP on this device is disabled. */
 919         if (!in_dev)
 920                 goto out;
 921
 922         net = dev_net(rt->dst.dev);
 923         if (!IN_DEV_FORWARD(in_dev)) {
 924                 switch (rt->dst.error) {
 925                 case EHOSTUNREACH:
 926                         __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
 927                         break;
 928
 929                 case ENETUNREACH:
 930                         __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 931                         break;
 932                 }
 933                 goto out;
 934         }
 935
 936         switch (rt->dst.error) {
 937         case EINVAL:
 938         default:
 939                 goto out;
 940         case EHOSTUNREACH:
 941                 code = ICMP_HOST_UNREACH;
 942                 break;
 943         case ENETUNREACH:
 944                 code = ICMP_NET_UNREACH;
 945                 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 946                 break;
 947         case EACCES:
 948                 code = ICMP_PKT_FILTERED;
 949                 break;
 950         }
 951
 952         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
 953                                l3mdev_master_ifindex(skb->dev), 1);
 954
 955         send = true;
 956         if (peer) {
 957                 now = jiffies;
 958                 peer->rate_tokens += now - peer->rate_last;
 959                 if (peer->rate_tokens > ip_rt_error_burst)
 960                         peer->rate_tokens = ip_rt_error_burst;
 961                 peer->rate_last = now;
 962                 if (peer->rate_tokens >= ip_rt_error_cost)
 963                         peer->rate_tokens -= ip_rt_error_cost;
 964                 else
 965                         send = false;
 966                 inet_putpeer(peer);
 967         }
 968         if (send)
 969                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
 970
 971 out:    kfree_skb(skb);
 972         return 0;
 973 }
 974
 975 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
 976 {
 977         struct dst_entry *dst = &rt->dst;
 978         struct fib_result res;
 979
 980         if (dst_metric_locked(dst, RTAX_MTU))
 981                 return;
 982
 983         if (ipv4_mtu(dst) < mtu)
 984                 return;
 985
 986         if (mtu < ip_rt_min_pmtu)
 987                 mtu = ip_rt_min_pmtu;
 988
 989         if (rt->rt_pmtu == mtu &&
 990             time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
 991                 return;
 992
 993         rcu_read_lock();
 994         if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
 995                 struct fib_nh *nh = &FIB_RES_NH(res);
 996
 997                 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
 998                                       jiffies + ip_rt_mtu_expires);
 999         }
1000         rcu_read_unlock();
1001 }
1002
1003 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1004                               struct sk_buff *skb, u32 mtu)
1005 {
1006         struct rtable *rt = (struct rtable *) dst;
1007         struct flowi4 fl4;
1008
1009         ip_rt_build_flow_key(&fl4, sk, skb);
1010         __ip_rt_update_pmtu(rt, &fl4, mtu);
1011 }
1012
1013 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1014                       int oif, u32 mark, u8 protocol, int flow_flags)
1015 {
1016         const struct iphdr *iph = (const struct iphdr *) skb->data;
1017         struct flowi4 fl4;
1018         struct rtable *rt;
1019
1020         if (!mark)
1021                 mark = IP4_REPLY_MARK(net, skb->mark);
1022
1023         __build_flow_key(&fl4, NULL, iph, oif,
1024                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1025         rt = __ip_route_output_key(net, &fl4);
1026         if (!IS_ERR(rt)) {
1027                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1028                 ip_rt_put(rt);
1029         }
1030 }
1031 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1032
1033 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1034 {
1035         const struct iphdr *iph = (const struct iphdr *) skb->data;
1036         struct flowi4 fl4;
1037         struct rtable *rt;
1038
1039         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1040
1041         if (!fl4.flowi4_mark)
1042                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1043
1044         rt = __ip_route_output_key(sock_net(sk), &fl4);
1045         if (!IS_ERR(rt)) {
1046                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1047                 ip_rt_put(rt);
1048         }
1049 }
1050
1051 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1052 {
1053         const struct iphdr *iph = (const struct iphdr *) skb->data;
1054         struct flowi4 fl4;
1055         struct rtable *rt;
1056         struct dst_entry *odst = NULL;
1057         bool new = false;
1058
1059         bh_lock_sock(sk);
1060
1061         if (!ip_sk_accept_pmtu(sk))
1062                 goto out;
1063
1064         odst = sk_dst_get(sk);
1065
1066         if (sock_owned_by_user(sk) || !odst) {
1067                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1068                 goto out;
1069         }
1070
1071         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1072
1073         rt = (struct rtable *)odst;
1074         if (odst->obsolete && !odst->ops->check(odst, 0)) {
1075                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1076                 if (IS_ERR(rt))
1077                         goto out;
1078
1079                 new = true;
1080         }
1081
1082         __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1083
1084         if (!dst_check(&rt->dst, 0)) {
1085                 if (new)
1086                         dst_release(&rt->dst);
1087
1088                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1089                 if (IS_ERR(rt))
1090                         goto out;
1091
1092                 new = true;
1093         }
1094
1095         if (new)
1096                 sk_dst_set(sk, &rt->dst);
1097
1098 out:
1099         bh_unlock_sock(sk);
1100         dst_release(odst);
1101 }
1102 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1103
1104 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1105                    int oif, u32 mark, u8 protocol, int flow_flags)
1106 {
1107         const struct iphdr *iph = (const struct iphdr *) skb->data;
1108         struct flowi4 fl4;
1109         struct rtable *rt;
1110
1111         __build_flow_key(&fl4, NULL, iph, oif,
1112                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1113         rt = __ip_route_output_key(net, &fl4);
1114         if (!IS_ERR(rt)) {
1115                 __ip_do_redirect(rt, skb, &fl4, false);
1116                 ip_rt_put(rt);
1117         }
1118 }
1119 EXPORT_SYMBOL_GPL(ipv4_redirect);
1120
1121 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1122 {
1123         const struct iphdr *iph = (const struct iphdr *) skb->data;
1124         struct flowi4 fl4;
1125         struct rtable *rt;
1126
1127         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1128         rt = __ip_route_output_key(sock_net(sk), &fl4);
1129         if (!IS_ERR(rt)) {
1130                 __ip_do_redirect(rt, skb, &fl4, false);
1131                 ip_rt_put(rt);
1132         }
1133 }
1134 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1135
1136 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1137 {
1138         struct rtable *rt = (struct rtable *) dst;
1139
1140         /* All IPV4 dsts are created with ->obsolete set to the value
1141          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1142          * into this function always.
1143          *
1144          * When a PMTU/redirect information update invalidates a route,
1145          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1146          * DST_OBSOLETE_DEAD by dst_free().
1147          */
1148         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1149                 return NULL;
1150         return dst;
1151 }
1152
1153 static void ipv4_link_failure(struct sk_buff *skb)
1154 {
1155         struct rtable *rt;
1156
1157         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1158
1159         rt = skb_rtable(skb);
1160         if (rt)
1161                 dst_set_expires(&rt->dst, 0);
1162 }
1163
1164 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1165 {
1166         pr_debug("%s: %pI4 -> %pI4, %s\n",
1167                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1168                  skb->dev ? skb->dev->name : "?");
1169         kfree_skb(skb);
1170         WARN_ON(1);
1171         return 0;
1172 }
1173
1174 /*
1175    We do not cache source address of outgoing interface,
1176    because it is used only by IP RR, TS and SRR options,
1177    so that it out of fast path.
1178
1179    BTW remember: "addr" is allowed to be not aligned
1180    in IP options!
1181  */
1182
1183 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1184 {
1185         __be32 src;
1186
1187         if (rt_is_output_route(rt))
1188                 src = ip_hdr(skb)->saddr;
1189         else {
1190                 struct fib_result res;
1191                 struct flowi4 fl4;
1192                 struct iphdr *iph;
1193
1194                 iph = ip_hdr(skb);
1195
1196                 memset(&fl4, 0, sizeof(fl4));
1197                 fl4.daddr = iph->daddr;
1198                 fl4.saddr = iph->saddr;
1199                 fl4.flowi4_tos = RT_TOS(iph->tos);
1200                 fl4.flowi4_oif = rt->dst.dev->ifindex;
1201                 fl4.flowi4_iif = skb->dev->ifindex;
1202                 fl4.flowi4_mark = skb->mark;
1203
1204                 rcu_read_lock();
1205                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1206                         src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1207                 else
1208                         src = inet_select_addr(rt->dst.dev,
1209                                                rt_nexthop(rt, iph->daddr),
1210                                                RT_SCOPE_UNIVERSE);
1211                 rcu_read_unlock();
1212         }
1213         memcpy(addr, &src, 4);
1214 }
1215
1216 #ifdef CONFIG_IP_ROUTE_CLASSID
1217 static void set_class_tag(struct rtable *rt, u32 tag)
1218 {
1219         if (!(rt->dst.tclassid & 0xFFFF))
1220                 rt->dst.tclassid |= tag & 0xFFFF;
1221         if (!(rt->dst.tclassid & 0xFFFF0000))
1222                 rt->dst.tclassid |= tag & 0xFFFF0000;
1223 }
1224 #endif
1225
1226 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1227 {
1228         unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1229
1230         if (advmss == 0) {
1231                 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1232                                ip_rt_min_advmss);
1233                 if (advmss > 65535 - 40)
1234                         advmss = 65535 - 40;
1235         }
1236         return advmss;
1237 }
1238
1239 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1240 {
1241         const struct rtable *rt = (const struct rtable *) dst;
1242         unsigned int mtu = rt->rt_pmtu;
1243
1244         if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1245                 mtu = dst_metric_raw(dst, RTAX_MTU);
1246
1247         if (mtu)
1248                 return mtu;
1249
1250         mtu = dst->dev->mtu;
1251
1252         if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1253                 if (rt->rt_uses_gateway && mtu > 576)
1254                         mtu = 576;
1255         }
1256
1257         mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1258
1259         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1260 }
1261
1262 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1263 {
1264         struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1265         struct fib_nh_exception *fnhe;
1266         u32 hval;
1267
1268         if (!hash)
1269                 return NULL;
1270
1271         hval = fnhe_hashfun(daddr);
1272
1273         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1274              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1275                 if (fnhe->fnhe_daddr == daddr)
1276                         return fnhe;
1277         }
1278         return NULL;
1279 }
1280
1281 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1282                               __be32 daddr)
1283 {
1284         bool ret = false;
1285
1286         spin_lock_bh(&fnhe_lock);
1287
1288         if (daddr == fnhe->fnhe_daddr) {
1289                 struct rtable __rcu **porig;
1290                 struct rtable *orig;
1291                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1292
1293                 if (rt_is_input_route(rt))
1294                         porig = &fnhe->fnhe_rth_input;
1295                 else
1296                         porig = &fnhe->fnhe_rth_output;
1297                 orig = rcu_dereference(*porig);
1298
1299                 if (fnhe->fnhe_genid != genid) {
1300                         fnhe->fnhe_genid = genid;
1301                         fnhe->fnhe_gw = 0;
1302                         fnhe->fnhe_pmtu = 0;
1303                         fnhe->fnhe_expires = 0;
1304                         fnhe_flush_routes(fnhe);
1305                         orig = NULL;
1306                 }
1307                 fill_route_from_fnhe(rt, fnhe);
1308                 if (!rt->rt_gateway)
1309                         rt->rt_gateway = daddr;
1310
1311                 if (!(rt->dst.flags & DST_NOCACHE)) {
1312                         rcu_assign_pointer(*porig, rt);
1313                         if (orig)
1314                                 rt_free(orig);
1315                         ret = true;
1316                 }
1317
1318                 fnhe->fnhe_stamp = jiffies;
1319         }
1320         spin_unlock_bh(&fnhe_lock);
1321
1322         return ret;
1323 }
1324
1325 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1326 {
1327         struct rtable *orig, *prev, **p;
1328         bool ret = true;
1329
1330         if (rt_is_input_route(rt)) {
1331                 p = (struct rtable **)&nh->nh_rth_input;
1332         } else {
1333                 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1334         }
1335         orig = *p;
1336
1337         prev = cmpxchg(p, orig, rt);
1338         if (prev == orig) {
1339                 if (orig)
1340                         rt_free(orig);
1341         } else
1342                 ret = false;
1343
1344         return ret;
1345 }
1346
1347 struct uncached_list {
1348         spinlock_t              lock;
1349         struct list_head        head;
1350 };
1351
1352 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1353
1354 static void rt_add_uncached_list(struct rtable *rt)
1355 {
1356         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1357
1358         rt->rt_uncached_list = ul;
1359
1360         spin_lock_bh(&ul->lock);
1361         list_add_tail(&rt->rt_uncached, &ul->head);
1362         spin_unlock_bh(&ul->lock);
1363 }
1364
1365 static void ipv4_dst_destroy(struct dst_entry *dst)
1366 {
1367         struct rtable *rt = (struct rtable *) dst;
1368
1369         if (!list_empty(&rt->rt_uncached)) {
1370                 struct uncached_list *ul = rt->rt_uncached_list;
1371
1372                 spin_lock_bh(&ul->lock);
1373                 list_del(&rt->rt_uncached);
1374                 spin_unlock_bh(&ul->lock);
1375         }
1376 }
1377
1378 void rt_flush_dev(struct net_device *dev)
1379 {
1380         struct net *net = dev_net(dev);
1381         struct rtable *rt;
1382         int cpu;
1383
1384         for_each_possible_cpu(cpu) {
1385                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1386
1387                 spin_lock_bh(&ul->lock);
1388                 list_for_each_entry(rt, &ul->head, rt_uncached) {
1389                         if (rt->dst.dev != dev)
1390                                 continue;
1391                         rt->dst.dev = net->loopback_dev;
1392                         dev_hold(rt->dst.dev);
1393                         dev_put(dev);
1394                 }
1395                 spin_unlock_bh(&ul->lock);
1396         }
1397 }
1398
1399 static bool rt_cache_valid(const struct rtable *rt)
1400 {
1401         return  rt &&
1402                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1403                 !rt_is_expired(rt);
1404 }
1405
1406 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1407                            const struct fib_result *res,
1408                            struct fib_nh_exception *fnhe,
1409                            struct fib_info *fi, u16 type, u32 itag)
1410 {
1411         bool cached = false;
1412
1413         if (fi) {
1414                 struct fib_nh *nh = &FIB_RES_NH(*res);
1415
1416                 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1417                         rt->rt_gateway = nh->nh_gw;
1418                         rt->rt_uses_gateway = 1;
1419                 }
1420                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1421 #ifdef CONFIG_IP_ROUTE_CLASSID
1422                 rt->dst.tclassid = nh->nh_tclassid;
1423 #endif
1424                 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1425                 if (unlikely(fnhe))
1426                         cached = rt_bind_exception(rt, fnhe, daddr);
1427                 else if (!(rt->dst.flags & DST_NOCACHE))
1428                         cached = rt_cache_route(nh, rt);
1429                 if (unlikely(!cached)) {
1430                         /* Routes we intend to cache in nexthop exception or
1431                          * FIB nexthop have the DST_NOCACHE bit clear.
1432                          * However, if we are unsuccessful at storing this
1433                          * route into the cache we really need to set it.
1434                          */
1435                         rt->dst.flags |= DST_NOCACHE;
1436                         if (!rt->rt_gateway)
1437                                 rt->rt_gateway = daddr;
1438                         rt_add_uncached_list(rt);
1439                 }
1440         } else
1441                 rt_add_uncached_list(rt);
1442
1443 #ifdef CONFIG_IP_ROUTE_CLASSID
1444 #ifdef CONFIG_IP_MULTIPLE_TABLES
1445         set_class_tag(rt, res->tclassid);
1446 #endif
1447         set_class_tag(rt, itag);
1448 #endif
1449 }
1450
1451 struct rtable *rt_dst_alloc(struct net_device *dev,
1452                             unsigned int flags, u16 type,
1453                             bool nopolicy, bool noxfrm, bool will_cache)
1454 {
1455         struct rtable *rt;
1456
1457         rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1458                        (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1459                        (nopolicy ? DST_NOPOLICY : 0) |
1460                        (noxfrm ? DST_NOXFRM : 0));
1461
1462         if (rt) {
1463                 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1464                 rt->rt_flags = flags;
1465                 rt->rt_type = type;
1466                 rt->rt_is_input = 0;
1467                 rt->rt_iif = 0;
1468                 rt->rt_pmtu = 0;
1469                 rt->rt_gateway = 0;
1470                 rt->rt_uses_gateway = 0;
1471                 rt->rt_table_id = 0;
1472                 INIT_LIST_HEAD(&rt->rt_uncached);
1473
1474                 rt->dst.output = ip_output;
1475                 if (flags & RTCF_LOCAL)
1476                         rt->dst.input = ip_local_deliver;
1477         }
1478
1479         return rt;
1480 }
1481 EXPORT_SYMBOL(rt_dst_alloc);
1482
1483 /* called in rcu_read_lock() section */
1484 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1485                                 u8 tos, struct net_device *dev, int our)
1486 {
1487         struct rtable *rth;
1488         struct in_device *in_dev = __in_dev_get_rcu(dev);
1489         unsigned int flags = RTCF_MULTICAST;
1490         u32 itag = 0;
1491         int err;
1492
1493         /* Primary sanity checks. */
1494
1495         if (!in_dev)
1496                 return -EINVAL;
1497
1498         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1499             skb->protocol != htons(ETH_P_IP))
1500                 goto e_inval;
1501
1502         if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1503                 goto e_inval;
1504
1505         if (ipv4_is_zeronet(saddr)) {
1506                 if (!ipv4_is_local_multicast(daddr))
1507                         goto e_inval;
1508         } else {
1509                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1510                                           in_dev, &itag);
1511                 if (err < 0)
1512                         goto e_err;
1513         }
1514         if (our)
1515                 flags |= RTCF_LOCAL;
1516
1517         rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1518                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1519         if (!rth)
1520                 goto e_nobufs;
1521
1522 #ifdef CONFIG_IP_ROUTE_CLASSID
1523         rth->dst.tclassid = itag;
1524 #endif
1525         rth->dst.output = ip_rt_bug;
1526         rth->rt_is_input= 1;
1527
1528 #ifdef CONFIG_IP_MROUTE
1529         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1530                 rth->dst.input = ip_mr_input;
1531 #endif
1532         RT_CACHE_STAT_INC(in_slow_mc);
1533
1534         skb_dst_set(skb, &rth->dst);
1535         return 0;
1536
1537 e_nobufs:
1538         return -ENOBUFS;
1539 e_inval:
1540         return -EINVAL;
1541 e_err:
1542         return err;
1543 }
1544
1545
1546 static void ip_handle_martian_source(struct net_device *dev,
1547                                      struct in_device *in_dev,
1548                                      struct sk_buff *skb,
1549                                      __be32 daddr,
1550                                      __be32 saddr)
1551 {
1552         RT_CACHE_STAT_INC(in_martian_src);
1553 #ifdef CONFIG_IP_ROUTE_VERBOSE
1554         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1555                 /*
1556                  *      RFC1812 recommendation, if source is martian,
1557                  *      the only hint is MAC header.
1558                  */
1559                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1560                         &daddr, &saddr, dev->name);
1561                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1562                         print_hex_dump(KERN_WARNING, "ll header: ",
1563                                        DUMP_PREFIX_OFFSET, 16, 1,
1564                                        skb_mac_header(skb),
1565                                        dev->hard_header_len, true);
1566                 }
1567         }
1568 #endif
1569 }
1570
1571 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1572 {
1573         struct fnhe_hash_bucket *hash;
1574         struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1575         u32 hval = fnhe_hashfun(daddr);
1576
1577         spin_lock_bh(&fnhe_lock);
1578
1579         hash = rcu_dereference_protected(nh->nh_exceptions,
1580                                          lockdep_is_held(&fnhe_lock));
1581         hash += hval;
1582
1583         fnhe_p = &hash->chain;
1584         fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1585         while (fnhe) {
1586                 if (fnhe->fnhe_daddr == daddr) {
1587                         rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1588                                 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1589                         fnhe_flush_routes(fnhe);
1590                         kfree_rcu(fnhe, rcu);
1591                         break;
1592                 }
1593                 fnhe_p = &fnhe->fnhe_next;
1594                 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1595                                                  lockdep_is_held(&fnhe_lock));
1596         }
1597
1598         spin_unlock_bh(&fnhe_lock);
1599 }
1600
1601 /* called in rcu_read_lock() section */
1602 static int __mkroute_input(struct sk_buff *skb,
1603                            const struct fib_result *res,
1604                            struct in_device *in_dev,
1605                            __be32 daddr, __be32 saddr, u32 tos)
1606 {
1607         struct fib_nh_exception *fnhe;
1608         struct rtable *rth;
1609         int err;
1610         struct in_device *out_dev;
1611         bool do_cache;
1612         u32 itag = 0;
1613
1614         /* get a working reference to the output device */
1615         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1616         if (!out_dev) {
1617                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1618                 return -EINVAL;
1619         }
1620
1621         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1622                                   in_dev->dev, in_dev, &itag);
1623         if (err < 0) {
1624                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1625                                          saddr);
1626
1627                 goto cleanup;
1628         }
1629
1630         do_cache = res->fi && !itag;
1631         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1632             skb->protocol == htons(ETH_P_IP) &&
1633             (IN_DEV_SHARED_MEDIA(out_dev) ||
1634              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1635                 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1636
1637         if (skb->protocol != htons(ETH_P_IP)) {
1638                 /* Not IP (i.e. ARP). Do not create route, if it is
1639                  * invalid for proxy arp. DNAT routes are always valid.
1640                  *
1641                  * Proxy arp feature have been extended to allow, ARP
1642                  * replies back to the same interface, to support
1643                  * Private VLAN switch technologies. See arp.c.
1644                  */
1645                 if (out_dev == in_dev &&
1646                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1647                         err = -EINVAL;
1648                         goto cleanup;
1649                 }
1650         }
1651
1652         fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1653         if (do_cache) {
1654                 if (fnhe) {
1655                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1656                         if (rth && rth->dst.expires &&
1657                             time_after(jiffies, rth->dst.expires)) {
1658                                 ip_del_fnhe(&FIB_RES_NH(*res), daddr);
1659                                 fnhe = NULL;
1660                         } else {
1661                                 goto rt_cache;
1662                         }
1663                 }
1664
1665                 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1666
1667 rt_cache:
1668                 if (rt_cache_valid(rth)) {
1669                         skb_dst_set_noref(skb, &rth->dst);
1670                         goto out;
1671                 }
1672         }
1673
1674         rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1675                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1676                            IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1677         if (!rth) {
1678                 err = -ENOBUFS;
1679                 goto cleanup;
1680         }
1681
1682         rth->rt_is_input = 1;
1683         if (res->table)
1684                 rth->rt_table_id = res->table->tb_id;
1685         RT_CACHE_STAT_INC(in_slow_tot);
1686
1687         rth->dst.input = ip_forward;
1688
1689         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1690         if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1691                 rth->dst.lwtstate->orig_output = rth->dst.output;
1692                 rth->dst.output = lwtunnel_output;
1693         }
1694         if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1695                 rth->dst.lwtstate->orig_input = rth->dst.input;
1696                 rth->dst.input = lwtunnel_input;
1697         }
1698         skb_dst_set(skb, &rth->dst);
1699 out:
1700         err = 0;
1701  cleanup:
1702         return err;
1703 }
1704
1705 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1706
1707 /* To make ICMP packets follow the right flow, the multipath hash is
1708  * calculated from the inner IP addresses in reverse order.
1709  */
1710 static int ip_multipath_icmp_hash(struct sk_buff *skb)
1711 {
1712         const struct iphdr *outer_iph = ip_hdr(skb);
1713         struct icmphdr _icmph;
1714         const struct icmphdr *icmph;
1715         struct iphdr _inner_iph;
1716         const struct iphdr *inner_iph;
1717
1718         if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1719                 goto standard_hash;
1720
1721         icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1722                                    &_icmph);
1723         if (!icmph)
1724                 goto standard_hash;
1725
1726         if (icmph->type != ICMP_DEST_UNREACH &&
1727             icmph->type != ICMP_REDIRECT &&
1728             icmph->type != ICMP_TIME_EXCEEDED &&
1729             icmph->type != ICMP_PARAMETERPROB) {
1730                 goto standard_hash;
1731         }
1732
1733         inner_iph = skb_header_pointer(skb,
1734                                        outer_iph->ihl * 4 + sizeof(_icmph),
1735                                        sizeof(_inner_iph), &_inner_iph);
1736         if (!inner_iph)
1737                 goto standard_hash;
1738
1739         return fib_multipath_hash(inner_iph->daddr, inner_iph->saddr);
1740
1741 standard_hash:
1742         return fib_multipath_hash(outer_iph->saddr, outer_iph->daddr);
1743 }
1744
1745 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1746
1747 static int ip_mkroute_input(struct sk_buff *skb,
1748                             struct fib_result *res,
1749                             const struct flowi4 *fl4,
1750                             struct in_device *in_dev,
1751                             __be32 daddr, __be32 saddr, u32 tos)
1752 {
1753 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1754         if (res->fi && res->fi->fib_nhs > 1) {
1755                 int h;
1756
1757                 if (unlikely(ip_hdr(skb)->protocol == IPPROTO_ICMP))
1758                         h = ip_multipath_icmp_hash(skb);
1759                 else
1760                         h = fib_multipath_hash(saddr, daddr);
1761                 fib_select_multipath(res, h);
1762         }
1763 #endif
1764
1765         /* create a routing cache entry */
1766         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1767 }
1768
1769 /*
1770  *      NOTE. We drop all the packets that has local source
1771  *      addresses, because every properly looped back packet
1772  *      must have correct destination already attached by output routine.
1773  *
1774  *      Such approach solves two big problems:
1775  *      1. Not simplex devices are handled properly.
1776  *      2. IP spoofing attempts are filtered with 100% of guarantee.
1777  *      called with rcu_read_lock()
1778  */
1779
1780 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1781                                u8 tos, struct net_device *dev)
1782 {
1783         struct fib_result res;
1784         struct in_device *in_dev = __in_dev_get_rcu(dev);
1785         struct ip_tunnel_info *tun_info;
1786         struct flowi4   fl4;
1787         unsigned int    flags = 0;
1788         u32             itag = 0;
1789         struct rtable   *rth;
1790         int             err = -EINVAL;
1791         struct net    *net = dev_net(dev);
1792         bool do_cache;
1793
1794         /* IP on this device is disabled. */
1795
1796         if (!in_dev)
1797                 goto out;
1798
1799         /* Check for the most weird martians, which can be not detected
1800            by fib_lookup.
1801          */
1802
1803         tun_info = skb_tunnel_info(skb);
1804         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1805                 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1806         else
1807                 fl4.flowi4_tun_key.tun_id = 0;
1808         skb_dst_drop(skb);
1809
1810         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1811                 goto martian_source;
1812
1813         res.fi = NULL;
1814         res.table = NULL;
1815         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1816                 goto brd_input;
1817
1818         /* Accept zero addresses only to limited broadcast;
1819          * I even do not know to fix it or not. Waiting for complains :-)
1820          */
1821         if (ipv4_is_zeronet(saddr))
1822                 goto martian_source;
1823
1824         if (ipv4_is_zeronet(daddr))
1825                 goto martian_destination;
1826
1827         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1828          * and call it once if daddr or/and saddr are loopback addresses
1829          */
1830         if (ipv4_is_loopback(daddr)) {
1831                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1832                         goto martian_destination;
1833         } else if (ipv4_is_loopback(saddr)) {
1834                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1835                         goto martian_source;
1836         }
1837
1838         /*
1839          *      Now we are ready to route packet.
1840          */
1841         fl4.flowi4_oif = 0;
1842         fl4.flowi4_iif = dev->ifindex;
1843         fl4.flowi4_mark = skb->mark;
1844         fl4.flowi4_tos = tos;
1845         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1846         fl4.flowi4_flags = 0;
1847         fl4.daddr = daddr;
1848         fl4.saddr = saddr;
1849         err = fib_lookup(net, &fl4, &res, 0);
1850         if (err != 0) {
1851                 if (!IN_DEV_FORWARD(in_dev))
1852                         err = -EHOSTUNREACH;
1853                 goto no_route;
1854         }
1855
1856         if (res.type == RTN_BROADCAST)
1857                 goto brd_input;
1858
1859         if (res.type == RTN_LOCAL) {
1860                 err = fib_validate_source(skb, saddr, daddr, tos,
1861                                           0, dev, in_dev, &itag);
1862                 if (err < 0)
1863                         goto martian_source;
1864                 goto local_input;
1865         }
1866
1867         if (!IN_DEV_FORWARD(in_dev)) {
1868                 err = -EHOSTUNREACH;
1869                 goto no_route;
1870         }
1871         if (res.type != RTN_UNICAST)
1872                 goto martian_destination;
1873
1874         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1875 out:    return err;
1876
1877 brd_input:
1878         if (skb->protocol != htons(ETH_P_IP))
1879                 goto e_inval;
1880
1881         if (!ipv4_is_zeronet(saddr)) {
1882                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1883                                           in_dev, &itag);
1884                 if (err < 0)
1885                         goto martian_source;
1886         }
1887         flags |= RTCF_BROADCAST;
1888         res.type = RTN_BROADCAST;
1889         RT_CACHE_STAT_INC(in_brd);
1890
1891 local_input:
1892         do_cache = false;
1893         if (res.fi) {
1894                 if (!itag) {
1895                         rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1896                         if (rt_cache_valid(rth)) {
1897                                 skb_dst_set_noref(skb, &rth->dst);
1898                                 err = 0;
1899                                 goto out;
1900                         }
1901                         do_cache = true;
1902                 }
1903         }
1904
1905         rth = rt_dst_alloc(net->loopback_dev, flags | RTCF_LOCAL, res.type,
1906                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1907         if (!rth)
1908                 goto e_nobufs;
1909
1910         rth->dst.output= ip_rt_bug;
1911 #ifdef CONFIG_IP_ROUTE_CLASSID
1912         rth->dst.tclassid = itag;
1913 #endif
1914         rth->rt_is_input = 1;
1915         if (res.table)
1916                 rth->rt_table_id = res.table->tb_id;
1917
1918         RT_CACHE_STAT_INC(in_slow_tot);
1919         if (res.type == RTN_UNREACHABLE) {
1920                 rth->dst.input= ip_error;
1921                 rth->dst.error= -err;
1922                 rth->rt_flags   &= ~RTCF_LOCAL;
1923         }
1924         if (do_cache) {
1925                 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1926                         rth->dst.flags |= DST_NOCACHE;
1927                         rt_add_uncached_list(rth);
1928                 }
1929         }
1930         skb_dst_set(skb, &rth->dst);
1931         err = 0;
1932         goto out;
1933
1934 no_route:
1935         RT_CACHE_STAT_INC(in_no_route);
1936         res.type = RTN_UNREACHABLE;
1937         res.fi = NULL;
1938         res.table = NULL;
1939         goto local_input;
1940
1941         /*
1942          *      Do not cache martian addresses: they should be logged (RFC1812)
1943          */
1944 martian_destination:
1945         RT_CACHE_STAT_INC(in_martian_dst);
1946 #ifdef CONFIG_IP_ROUTE_VERBOSE
1947         if (IN_DEV_LOG_MARTIANS(in_dev))
1948                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1949                                      &daddr, &saddr, dev->name);
1950 #endif
1951
1952 e_inval:
1953         err = -EINVAL;
1954         goto out;
1955
1956 e_nobufs:
1957         err = -ENOBUFS;
1958         goto out;
1959
1960 martian_source:
1961         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1962         goto out;
1963 }
1964
1965 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1966                          u8 tos, struct net_device *dev)
1967 {
1968         int res;
1969
1970         rcu_read_lock();
1971
1972         /* Multicast recognition logic is moved from route cache to here.
1973            The problem was that too many Ethernet cards have broken/missing
1974            hardware multicast filters :-( As result the host on multicasting
1975            network acquires a lot of useless route cache entries, sort of
1976            SDR messages from all the world. Now we try to get rid of them.
1977            Really, provided software IP multicast filter is organized
1978            reasonably (at least, hashed), it does not result in a slowdown
1979            comparing with route cache reject entries.
1980            Note, that multicast routers are not affected, because
1981            route cache entry is created eventually.
1982          */
1983         if (ipv4_is_multicast(daddr)) {
1984                 struct in_device *in_dev = __in_dev_get_rcu(dev);
1985
1986                 if (in_dev) {
1987                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1988                                                   ip_hdr(skb)->protocol);
1989                         if (our
1990 #ifdef CONFIG_IP_MROUTE
1991                                 ||
1992                             (!ipv4_is_local_multicast(daddr) &&
1993                              IN_DEV_MFORWARD(in_dev))
1994 #endif
1995                            ) {
1996                                 int res = ip_route_input_mc(skb, daddr, saddr,
1997                                                             tos, dev, our);
1998                                 rcu_read_unlock();
1999                                 return res;
2000                         }
2001                 }
2002                 rcu_read_unlock();
2003                 return -EINVAL;
2004         }
2005         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2006         rcu_read_unlock();
2007         return res;
2008 }
2009 EXPORT_SYMBOL(ip_route_input_noref);
2010
2011 /* called with rcu_read_lock() */
2012 static struct rtable *__mkroute_output(const struct fib_result *res,
2013                                        const struct flowi4 *fl4, int orig_oif,
2014                                        struct net_device *dev_out,
2015                                        unsigned int flags)
2016 {
2017         struct fib_info *fi = res->fi;
2018         struct fib_nh_exception *fnhe;
2019         struct in_device *in_dev;
2020         u16 type = res->type;
2021         struct rtable *rth;
2022         bool do_cache;
2023
2024         in_dev = __in_dev_get_rcu(dev_out);
2025         if (!in_dev)
2026                 return ERR_PTR(-EINVAL);
2027
2028         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2029                 if (ipv4_is_loopback(fl4->saddr) &&
2030                     !(dev_out->flags & IFF_LOOPBACK) &&
2031                     !netif_is_l3_master(dev_out))
2032                         return ERR_PTR(-EINVAL);
2033
2034         if (ipv4_is_lbcast(fl4->daddr))
2035                 type = RTN_BROADCAST;
2036         else if (ipv4_is_multicast(fl4->daddr))
2037                 type = RTN_MULTICAST;
2038         else if (ipv4_is_zeronet(fl4->daddr))
2039                 return ERR_PTR(-EINVAL);
2040
2041         if (dev_out->flags & IFF_LOOPBACK)
2042                 flags |= RTCF_LOCAL;
2043
2044         do_cache = true;
2045         if (type == RTN_BROADCAST) {
2046                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2047                 fi = NULL;
2048         } else if (type == RTN_MULTICAST) {
2049                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2050                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2051                                      fl4->flowi4_proto))
2052                         flags &= ~RTCF_LOCAL;
2053                 else
2054                         do_cache = false;
2055                 /* If multicast route do not exist use
2056                  * default one, but do not gateway in this case.
2057                  * Yes, it is hack.
2058                  */
2059                 if (fi && res->prefixlen < 4)
2060                         fi = NULL;
2061         } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2062                    (orig_oif != dev_out->ifindex)) {
2063                 /* For local routes that require a particular output interface
2064                  * we do not want to cache the result.  Caching the result
2065                  * causes incorrect behaviour when there are multiple source
2066                  * addresses on the interface, the end result being that if the
2067                  * intended recipient is waiting on that interface for the
2068                  * packet he won't receive it because it will be delivered on
2069                  * the loopback interface and the IP_PKTINFO ipi_ifindex will
2070                  * be set to the loopback interface as well.
2071                  */
2072                 fi = NULL;
2073         }
2074
2075         fnhe = NULL;
2076         do_cache &= fi != NULL;
2077         if (do_cache) {
2078                 struct rtable __rcu **prth;
2079                 struct fib_nh *nh = &FIB_RES_NH(*res);
2080
2081                 fnhe = find_exception(nh, fl4->daddr);
2082                 if (fnhe) {
2083                         prth = &fnhe->fnhe_rth_output;
2084                         rth = rcu_dereference(*prth);
2085                         if (rth && rth->dst.expires &&
2086                             time_after(jiffies, rth->dst.expires)) {
2087                                 ip_del_fnhe(nh, fl4->daddr);
2088                                 fnhe = NULL;
2089                         } else {
2090                                 goto rt_cache;
2091                         }
2092                 }
2093
2094                 if (unlikely(fl4->flowi4_flags &
2095                              FLOWI_FLAG_KNOWN_NH &&
2096                              !(nh->nh_gw &&
2097                                nh->nh_scope == RT_SCOPE_LINK))) {
2098                         do_cache = false;
2099                         goto add;
2100                 }
2101                 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2102                 rth = rcu_dereference(*prth);
2103
2104 rt_cache:
2105                 if (rt_cache_valid(rth)) {
2106                         dst_hold(&rth->dst);
2107                         return rth;
2108                 }
2109         }
2110
2111 add:
2112         rth = rt_dst_alloc(dev_out, flags, type,
2113                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
2114                            IN_DEV_CONF_GET(in_dev, NOXFRM),
2115                            do_cache);
2116         if (!rth)
2117                 return ERR_PTR(-ENOBUFS);
2118
2119         rth->rt_iif     = orig_oif ? : 0;
2120         if (res->table)
2121                 rth->rt_table_id = res->table->tb_id;
2122
2123         RT_CACHE_STAT_INC(out_slow_tot);
2124
2125         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2126                 if (flags & RTCF_LOCAL &&
2127                     !(dev_out->flags & IFF_LOOPBACK)) {
2128                         rth->dst.output = ip_mc_output;
2129                         RT_CACHE_STAT_INC(out_slow_mc);
2130                 }
2131 #ifdef CONFIG_IP_MROUTE
2132                 if (type == RTN_MULTICAST) {
2133                         if (IN_DEV_MFORWARD(in_dev) &&
2134                             !ipv4_is_local_multicast(fl4->daddr)) {
2135                                 rth->dst.input = ip_mr_input;
2136                                 rth->dst.output = ip_mc_output;
2137                         }
2138                 }
2139 #endif
2140         }
2141
2142         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2143         if (lwtunnel_output_redirect(rth->dst.lwtstate))
2144                 rth->dst.output = lwtunnel_output;
2145
2146         return rth;
2147 }
2148
2149 /*
2150  * Major route resolver routine.
2151  */
2152
2153 struct rtable *__ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2154                                           int mp_hash)
2155 {
2156         struct net_device *dev_out = NULL;
2157         __u8 tos = RT_FL_TOS(fl4);
2158         unsigned int flags = 0;
2159         struct fib_result res;
2160         struct rtable *rth;
2161         int orig_oif;
2162         int err = -ENETUNREACH;
2163
2164         res.tclassid    = 0;
2165         res.fi          = NULL;
2166         res.table       = NULL;
2167
2168         orig_oif = fl4->flowi4_oif;
2169
2170         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2171         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2172         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2173                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2174
2175         rcu_read_lock();
2176         if (fl4->saddr) {
2177                 rth = ERR_PTR(-EINVAL);
2178                 if (ipv4_is_multicast(fl4->saddr) ||
2179                     ipv4_is_lbcast(fl4->saddr) ||
2180                     ipv4_is_zeronet(fl4->saddr))
2181                         goto out;
2182
2183                 /* I removed check for oif == dev_out->oif here.
2184                    It was wrong for two reasons:
2185                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2186                       is assigned to multiple interfaces.
2187                    2. Moreover, we are allowed to send packets with saddr
2188                       of another iface. --ANK
2189                  */
2190
2191                 if (fl4->flowi4_oif == 0 &&
2192                     (ipv4_is_multicast(fl4->daddr) ||
2193                      ipv4_is_lbcast(fl4->daddr))) {
2194                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2195                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2196                         if (!dev_out)
2197                                 goto out;
2198
2199                         /* Special hack: user can direct multicasts
2200                            and limited broadcast via necessary interface
2201                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2202                            This hack is not just for fun, it allows
2203                            vic,vat and friends to work.
2204                            They bind socket to loopback, set ttl to zero
2205                            and expect that it will work.
2206                            From the viewpoint of routing cache they are broken,
2207                            because we are not allowed to build multicast path
2208                            with loopback source addr (look, routing cache
2209                            cannot know, that ttl is zero, so that packet
2210                            will not leave this host and route is valid).
2211                            Luckily, this hack is good workaround.
2212                          */
2213
2214                         fl4->flowi4_oif = dev_out->ifindex;
2215                         goto make_route;
2216                 }
2217
2218                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2219                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2220                         if (!__ip_dev_find(net, fl4->saddr, false))
2221                                 goto out;
2222                 }
2223         }
2224
2225
2226         if (fl4->flowi4_oif) {
2227                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2228                 rth = ERR_PTR(-ENODEV);
2229                 if (!dev_out)
2230                         goto out;
2231
2232                 /* RACE: Check return value of inet_select_addr instead. */
2233                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2234                         rth = ERR_PTR(-ENETUNREACH);
2235                         goto out;
2236                 }
2237                 if (ipv4_is_local_multicast(fl4->daddr) ||
2238                     ipv4_is_lbcast(fl4->daddr) ||
2239                     fl4->flowi4_proto == IPPROTO_IGMP) {
2240                         if (!fl4->saddr)
2241                                 fl4->saddr = inet_select_addr(dev_out, 0,
2242                                                               RT_SCOPE_LINK);
2243                         goto make_route;
2244                 }
2245                 if (!fl4->saddr) {
2246                         if (ipv4_is_multicast(fl4->daddr))
2247                                 fl4->saddr = inet_select_addr(dev_out, 0,
2248                                                               fl4->flowi4_scope);
2249                         else if (!fl4->daddr)
2250                                 fl4->saddr = inet_select_addr(dev_out, 0,
2251                                                               RT_SCOPE_HOST);
2252                 }
2253         }
2254
2255         if (!fl4->daddr) {
2256                 fl4->daddr = fl4->saddr;
2257                 if (!fl4->daddr)
2258                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2259                 dev_out = net->loopback_dev;
2260                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2261                 res.type = RTN_LOCAL;
2262                 flags |= RTCF_LOCAL;
2263                 goto make_route;
2264         }
2265
2266         err = fib_lookup(net, fl4, &res, 0);
2267         if (err) {
2268                 res.fi = NULL;
2269                 res.table = NULL;
2270                 if (fl4->flowi4_oif &&
2271                     !netif_index_is_l3_master(net, fl4->flowi4_oif)) {
2272                         /* Apparently, routing tables are wrong. Assume,
2273                            that the destination is on link.
2274
2275                            WHY? DW.
2276                            Because we are allowed to send to iface
2277                            even if it has NO routes and NO assigned
2278                            addresses. When oif is specified, routing
2279                            tables are looked up with only one purpose:
2280                            to catch if destination is gatewayed, rather than
2281                            direct. Moreover, if MSG_DONTROUTE is set,
2282                            we send packet, ignoring both routing tables
2283                            and ifaddr state. --ANK
2284
2285
2286                            We could make it even if oif is unknown,
2287                            likely IPv6, but we do not.
2288                          */
2289
2290                         if (fl4->saddr == 0)
2291                                 fl4->saddr = inet_select_addr(dev_out, 0,
2292                                                               RT_SCOPE_LINK);
2293                         res.type = RTN_UNICAST;
2294                         goto make_route;
2295                 }
2296                 rth = ERR_PTR(err);
2297                 goto out;
2298         }
2299
2300         if (res.type == RTN_LOCAL) {
2301                 if (!fl4->saddr) {
2302                         if (res.fi->fib_prefsrc)
2303                                 fl4->saddr = res.fi->fib_prefsrc;
2304                         else
2305                                 fl4->saddr = fl4->daddr;
2306                 }
2307
2308                 /* L3 master device is the loopback for that domain */
2309                 dev_out = l3mdev_master_dev_rcu(dev_out) ? : net->loopback_dev;
2310                 fl4->flowi4_oif = dev_out->ifindex;
2311                 flags |= RTCF_LOCAL;
2312                 goto make_route;
2313         }
2314
2315         fib_select_path(net, &res, fl4, mp_hash);
2316
2317         dev_out = FIB_RES_DEV(res);
2318         fl4->flowi4_oif = dev_out->ifindex;
2319
2320
2321 make_route:
2322         rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2323
2324 out:
2325         rcu_read_unlock();
2326         return rth;
2327 }
2328 EXPORT_SYMBOL_GPL(__ip_route_output_key_hash);
2329
2330 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2331 {
2332         return NULL;
2333 }
2334
2335 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2336 {
2337         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2338
2339         return mtu ? : dst->dev->mtu;
2340 }
2341
2342 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2343                                           struct sk_buff *skb, u32 mtu)
2344 {
2345 }
2346
2347 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2348                                        struct sk_buff *skb)
2349 {
2350 }
2351
2352 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2353                                           unsigned long old)
2354 {
2355         return NULL;
2356 }
2357
2358 static struct dst_ops ipv4_dst_blackhole_ops = {
2359         .family                 =       AF_INET,
2360         .check                  =       ipv4_blackhole_dst_check,
2361         .mtu                    =       ipv4_blackhole_mtu,
2362         .default_advmss         =       ipv4_default_advmss,
2363         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2364         .redirect               =       ipv4_rt_blackhole_redirect,
2365         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
2366         .neigh_lookup           =       ipv4_neigh_lookup,
2367 };
2368
2369 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2370 {
2371         struct rtable *ort = (struct rtable *) dst_orig;
2372         struct rtable *rt;
2373
2374         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2375         if (rt) {
2376                 struct dst_entry *new = &rt->dst;
2377
2378                 new->__use = 1;
2379                 new->input = dst_discard;
2380                 new->output = dst_discard_out;
2381
2382                 new->dev = ort->dst.dev;
2383                 if (new->dev)
2384                         dev_hold(new->dev);
2385
2386                 rt->rt_is_input = ort->rt_is_input;
2387                 rt->rt_iif = ort->rt_iif;
2388                 rt->rt_pmtu = ort->rt_pmtu;
2389
2390                 rt->rt_genid = rt_genid_ipv4(net);
2391                 rt->rt_flags = ort->rt_flags;
2392                 rt->rt_type = ort->rt_type;
2393                 rt->rt_gateway = ort->rt_gateway;
2394                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2395
2396                 INIT_LIST_HEAD(&rt->rt_uncached);
2397                 dst_free(new);
2398         }
2399
2400         dst_release(dst_orig);
2401
2402         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2403 }
2404
2405 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2406                                     const struct sock *sk)
2407 {
2408         struct rtable *rt = __ip_route_output_key(net, flp4);
2409
2410         if (IS_ERR(rt))
2411                 return rt;
2412
2413         if (flp4->flowi4_proto)
2414                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2415                                                         flowi4_to_flowi(flp4),
2416                                                         sk, 0);
2417
2418         return rt;
2419 }
2420 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2421
2422 static int rt_fill_info(struct net *net,  __be32 dst, __be32 src, u32 table_id,
2423                         struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2424                         u32 seq, int event, int nowait, unsigned int flags)
2425 {
2426         struct rtable *rt = skb_rtable(skb);
2427         struct rtmsg *r;
2428         struct nlmsghdr *nlh;
2429         unsigned long expires = 0;
2430         u32 error;
2431         u32 metrics[RTAX_MAX];
2432
2433         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2434         if (!nlh)
2435                 return -EMSGSIZE;
2436
2437         r = nlmsg_data(nlh);
2438         r->rtm_family    = AF_INET;
2439         r->rtm_dst_len  = 32;
2440         r->rtm_src_len  = 0;
2441         r->rtm_tos      = fl4->flowi4_tos;
2442         r->rtm_table    = table_id;
2443         if (nla_put_u32(skb, RTA_TABLE, table_id))
2444                 goto nla_put_failure;
2445         r->rtm_type     = rt->rt_type;
2446         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2447         r->rtm_protocol = RTPROT_UNSPEC;
2448         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2449         if (rt->rt_flags & RTCF_NOTIFY)
2450                 r->rtm_flags |= RTM_F_NOTIFY;
2451         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2452                 r->rtm_flags |= RTCF_DOREDIRECT;
2453
2454         if (nla_put_in_addr(skb, RTA_DST, dst))
2455                 goto nla_put_failure;
2456         if (src) {
2457                 r->rtm_src_len = 32;
2458                 if (nla_put_in_addr(skb, RTA_SRC, src))
2459                         goto nla_put_failure;
2460         }
2461         if (rt->dst.dev &&
2462             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2463                 goto nla_put_failure;
2464 #ifdef CONFIG_IP_ROUTE_CLASSID
2465         if (rt->dst.tclassid &&
2466             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2467                 goto nla_put_failure;
2468 #endif
2469         if (!rt_is_input_route(rt) &&
2470             fl4->saddr != src) {
2471                 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2472                         goto nla_put_failure;
2473         }
2474         if (rt->rt_uses_gateway &&
2475             nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2476                 goto nla_put_failure;
2477
2478         expires = rt->dst.expires;
2479         if (expires) {
2480                 unsigned long now = jiffies;
2481
2482                 if (time_before(now, expires))
2483                         expires -= now;
2484                 else
2485                         expires = 0;
2486         }
2487
2488         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2489         if (rt->rt_pmtu && expires)
2490                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2491         if (rtnetlink_put_metrics(skb, metrics) < 0)
2492                 goto nla_put_failure;
2493
2494         if (fl4->flowi4_mark &&
2495             nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2496                 goto nla_put_failure;
2497
2498         error = rt->dst.error;
2499
2500         if (rt_is_input_route(rt)) {
2501 #ifdef CONFIG_IP_MROUTE
2502                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2503                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2504                         int err = ipmr_get_route(net, skb,
2505                                                  fl4->saddr, fl4->daddr,
2506                                                  r, nowait, portid);
2507
2508                         if (err <= 0) {
2509                                 if (!nowait) {
2510                                         if (err == 0)
2511                                                 return 0;
2512                                         goto nla_put_failure;
2513                                 } else {
2514                                         if (err == -EMSGSIZE)
2515                                                 goto nla_put_failure;
2516                                         error = err;
2517                                 }
2518                         }
2519                 } else
2520 #endif
2521                         if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2522                                 goto nla_put_failure;
2523         }
2524
2525         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2526                 goto nla_put_failure;
2527
2528         nlmsg_end(skb, nlh);
2529         return 0;
2530
2531 nla_put_failure:
2532         nlmsg_cancel(skb, nlh);
2533         return -EMSGSIZE;
2534 }
2535
2536 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2537 {
2538         struct net *net = sock_net(in_skb->sk);
2539         struct rtmsg *rtm;
2540         struct nlattr *tb[RTA_MAX+1];
2541         struct rtable *rt = NULL;
2542         struct flowi4 fl4;
2543         __be32 dst = 0;
2544         __be32 src = 0;
2545         u32 iif;
2546         int err;
2547         int mark;
2548         struct sk_buff *skb;
2549         u32 table_id = RT_TABLE_MAIN;
2550
2551         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2552         if (err < 0)
2553                 goto errout;
2554
2555         rtm = nlmsg_data(nlh);
2556
2557         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2558         if (!skb) {
2559                 err = -ENOBUFS;
2560                 goto errout;
2561         }
2562
2563         /* Reserve room for dummy headers, this skb can pass
2564            through good chunk of routing engine.
2565          */
2566         skb_reset_mac_header(skb);
2567         skb_reset_network_header(skb);
2568
2569         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2570         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2571         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2572
2573         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2574         dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2575         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2576         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2577
2578         memset(&fl4, 0, sizeof(fl4));
2579         fl4.daddr = dst;
2580         fl4.saddr = src;
2581         fl4.flowi4_tos = rtm->rtm_tos;
2582         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2583         fl4.flowi4_mark = mark;
2584
2585         if (iif) {
2586                 struct net_device *dev;
2587
2588                 dev = __dev_get_by_index(net, iif);
2589                 if (!dev) {
2590                         err = -ENODEV;
2591                         goto errout_free;
2592                 }
2593
2594                 skb->protocol   = htons(ETH_P_IP);
2595                 skb->dev        = dev;
2596                 skb->mark       = mark;
2597                 local_bh_disable();
2598                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2599                 local_bh_enable();
2600
2601                 rt = skb_rtable(skb);
2602                 if (err == 0 && rt->dst.error)
2603                         err = -rt->dst.error;
2604         } else {
2605                 rt = ip_route_output_key(net, &fl4);
2606
2607                 err = 0;
2608                 if (IS_ERR(rt))
2609                         err = PTR_ERR(rt);
2610         }
2611
2612         if (err)
2613                 goto errout_free;
2614
2615         skb_dst_set(skb, &rt->dst);
2616         if (rtm->rtm_flags & RTM_F_NOTIFY)
2617                 rt->rt_flags |= RTCF_NOTIFY;
2618
2619         if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2620                 table_id = rt->rt_table_id;
2621
2622         err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2623                            NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2624                            RTM_NEWROUTE, 0, 0);
2625         if (err < 0)
2626                 goto errout_free;
2627
2628         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2629 errout:
2630         return err;
2631
2632 errout_free:
2633         kfree_skb(skb);
2634         goto errout;
2635 }
2636
2637 void ip_rt_multicast_event(struct in_device *in_dev)
2638 {
2639         rt_cache_flush(dev_net(in_dev->dev));
2640 }
2641
2642 #ifdef CONFIG_SYSCTL
2643 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
2644 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
2645 static int ip_rt_gc_elasticity __read_mostly    = 8;
2646
2647 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2648                                         void __user *buffer,
2649                                         size_t *lenp, loff_t *ppos)
2650 {
2651         struct net *net = (struct net *)__ctl->extra1;
2652
2653         if (write) {
2654                 rt_cache_flush(net);
2655                 fnhe_genid_bump(net);
2656                 return 0;
2657         }
2658
2659         return -EINVAL;
2660 }
2661
2662 static struct ctl_table ipv4_route_table[] = {
2663         {
2664                 .procname       = "gc_thresh",
2665                 .data           = &ipv4_dst_ops.gc_thresh,
2666                 .maxlen         = sizeof(int),
2667                 .mode           = 0644,
2668                 .proc_handler   = proc_dointvec,
2669         },
2670         {
2671                 .procname       = "max_size",
2672                 .data           = &ip_rt_max_size,
2673                 .maxlen         = sizeof(int),
2674                 .mode           = 0644,
2675                 .proc_handler   = proc_dointvec,
2676         },
2677         {
2678                 /*  Deprecated. Use gc_min_interval_ms */
2679
2680                 .procname       = "gc_min_interval",
2681                 .data           = &ip_rt_gc_min_interval,
2682                 .maxlen         = sizeof(int),
2683                 .mode           = 0644,
2684                 .proc_handler   = proc_dointvec_jiffies,
2685         },
2686         {
2687                 .procname       = "gc_min_interval_ms",
2688                 .data           = &ip_rt_gc_min_interval,
2689                 .maxlen         = sizeof(int),
2690                 .mode           = 0644,
2691                 .proc_handler   = proc_dointvec_ms_jiffies,
2692         },
2693         {
2694                 .procname       = "gc_timeout",
2695                 .data           = &ip_rt_gc_timeout,
2696                 .maxlen         = sizeof(int),
2697                 .mode           = 0644,
2698                 .proc_handler   = proc_dointvec_jiffies,
2699         },
2700         {
2701                 .procname       = "gc_interval",
2702                 .data           = &ip_rt_gc_interval,
2703                 .maxlen         = sizeof(int),
2704                 .mode           = 0644,
2705                 .proc_handler   = proc_dointvec_jiffies,
2706         },
2707         {
2708                 .procname       = "redirect_load",
2709                 .data           = &ip_rt_redirect_load,
2710                 .maxlen         = sizeof(int),
2711                 .mode           = 0644,
2712                 .proc_handler   = proc_dointvec,
2713         },
2714         {
2715                 .procname       = "redirect_number",
2716                 .data           = &ip_rt_redirect_number,
2717                 .maxlen         = sizeof(int),
2718                 .mode           = 0644,
2719                 .proc_handler   = proc_dointvec,
2720         },
2721         {
2722                 .procname       = "redirect_silence",
2723                 .data           = &ip_rt_redirect_silence,
2724                 .maxlen         = sizeof(int),
2725                 .mode           = 0644,
2726                 .proc_handler   = proc_dointvec,
2727         },
2728         {
2729                 .procname       = "error_cost",
2730                 .data           = &ip_rt_error_cost,
2731                 .maxlen         = sizeof(int),
2732                 .mode           = 0644,
2733                 .proc_handler   = proc_dointvec,
2734         },
2735         {
2736                 .procname       = "error_burst",
2737                 .data           = &ip_rt_error_burst,
2738                 .maxlen         = sizeof(int),
2739                 .mode           = 0644,
2740                 .proc_handler   = proc_dointvec,
2741         },
2742         {
2743                 .procname       = "gc_elasticity",
2744                 .data           = &ip_rt_gc_elasticity,
2745                 .maxlen         = sizeof(int),
2746                 .mode           = 0644,
2747                 .proc_handler   = proc_dointvec,
2748         },
2749         {
2750                 .procname       = "mtu_expires",
2751                 .data           = &ip_rt_mtu_expires,
2752                 .maxlen         = sizeof(int),
2753                 .mode           = 0644,
2754                 .proc_handler   = proc_dointvec_jiffies,
2755         },
2756         {
2757                 .procname       = "min_pmtu",
2758                 .data           = &ip_rt_min_pmtu,
2759                 .maxlen         = sizeof(int),
2760                 .mode           = 0644,
2761                 .proc_handler   = proc_dointvec,
2762         },
2763         {
2764                 .procname       = "min_adv_mss",
2765                 .data           = &ip_rt_min_advmss,
2766                 .maxlen         = sizeof(int),
2767                 .mode           = 0644,
2768                 .proc_handler   = proc_dointvec,
2769         },
2770         { }
2771 };
2772
2773 static struct ctl_table ipv4_route_flush_table[] = {
2774         {
2775                 .procname       = "flush",
2776                 .maxlen         = sizeof(int),
2777                 .mode           = 0200,
2778                 .proc_handler   = ipv4_sysctl_rtcache_flush,
2779         },
2780         { },
2781 };
2782
2783 static __net_init int sysctl_route_net_init(struct net *net)
2784 {
2785         struct ctl_table *tbl;
2786
2787         tbl = ipv4_route_flush_table;
2788         if (!net_eq(net, &init_net)) {
2789                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2790                 if (!tbl)
2791                         goto err_dup;
2792
2793                 /* Don't export sysctls to unprivileged users */
2794                 if (net->user_ns != &init_user_ns)
2795                         tbl[0].procname = NULL;
2796         }
2797         tbl[0].extra1 = net;
2798
2799         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2800         if (!net->ipv4.route_hdr)
2801                 goto err_reg;
2802         return 0;
2803
2804 err_reg:
2805         if (tbl != ipv4_route_flush_table)
2806                 kfree(tbl);
2807 err_dup:
2808         return -ENOMEM;
2809 }
2810
2811 static __net_exit void sysctl_route_net_exit(struct net *net)
2812 {
2813         struct ctl_table *tbl;
2814
2815         tbl = net->ipv4.route_hdr->ctl_table_arg;
2816         unregister_net_sysctl_table(net->ipv4.route_hdr);
2817         BUG_ON(tbl == ipv4_route_flush_table);
2818         kfree(tbl);
2819 }
2820
2821 static __net_initdata struct pernet_operations sysctl_route_ops = {
2822         .init = sysctl_route_net_init,
2823         .exit = sysctl_route_net_exit,
2824 };
2825 #endif
2826
2827 static __net_init int rt_genid_init(struct net *net)
2828 {
2829         atomic_set(&net->ipv4.rt_genid, 0);
2830         atomic_set(&net->fnhe_genid, 0);
2831         get_random_bytes(&net->ipv4.dev_addr_genid,
2832                          sizeof(net->ipv4.dev_addr_genid));
2833         return 0;
2834 }
2835
2836 static __net_initdata struct pernet_operations rt_genid_ops = {
2837         .init = rt_genid_init,
2838 };
2839
2840 static int __net_init ipv4_inetpeer_init(struct net *net)
2841 {
2842         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2843
2844         if (!bp)
2845                 return -ENOMEM;
2846         inet_peer_base_init(bp);
2847         net->ipv4.peers = bp;
2848         return 0;
2849 }
2850
2851 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2852 {
2853         struct inet_peer_base *bp = net->ipv4.peers;
2854
2855         net->ipv4.peers = NULL;
2856         inetpeer_invalidate_tree(bp);
2857         kfree(bp);
2858 }
2859
2860 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2861         .init   =       ipv4_inetpeer_init,
2862         .exit   =       ipv4_inetpeer_exit,
2863 };
2864
2865 #ifdef CONFIG_IP_ROUTE_CLASSID
2866 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2867 #endif /* CONFIG_IP_ROUTE_CLASSID */
2868
2869 int __init ip_rt_init(void)
2870 {
2871         int rc = 0;
2872         int cpu;
2873
2874         ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2875         if (!ip_idents)
2876                 panic("IP: failed to allocate ip_idents\n");
2877
2878         prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2879
2880         ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
2881         if (!ip_tstamps)
2882                 panic("IP: failed to allocate ip_tstamps\n");
2883
2884         for_each_possible_cpu(cpu) {
2885                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
2886
2887                 INIT_LIST_HEAD(&ul->head);
2888                 spin_lock_init(&ul->lock);
2889         }
2890 #ifdef CONFIG_IP_ROUTE_CLASSID
2891         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2892         if (!ip_rt_acct)
2893                 panic("IP: failed to allocate ip_rt_acct\n");
2894 #endif
2895
2896         ipv4_dst_ops.kmem_cachep =
2897                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2898                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2899
2900         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2901
2902         if (dst_entries_init(&ipv4_dst_ops) < 0)
2903                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2904
2905         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2906                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2907
2908         ipv4_dst_ops.gc_thresh = ~0;
2909         ip_rt_max_size = INT_MAX;
2910
2911         devinet_init();
2912         ip_fib_init();
2913
2914         if (ip_rt_proc_init())
2915                 pr_err("Unable to create route proc files\n");
2916 #ifdef CONFIG_XFRM
2917         xfrm_init();
2918         xfrm4_init();
2919 #endif
2920         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2921
2922 #ifdef CONFIG_SYSCTL
2923         register_pernet_subsys(&sysctl_route_ops);
2924 #endif
2925         register_pernet_subsys(&rt_genid_ops);
2926         register_pernet_subsys(&ipv4_inetpeer_ops);
2927         return rc;
2928 }
2929
2930 #ifdef CONFIG_SYSCTL
2931 /*
2932  * We really need to sanitize the damn ipv4 init order, then all
2933  * this nonsense will go away.
2934  */
2935 void __init ip_static_sysctl_init(void)
2936 {
2937         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
2938 }
2939 #endif