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