net/ipv4/route.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              ROUTE - implementation of the IP router.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *              Linus Torvalds, <Linus.Torvalds@helsinki.fi>
  12  *              Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  13  *
  14  * Fixes:
  15  *              Alan Cox        :       Verify area fixes.
  16  *              Alan Cox        :       cli() protects routing changes
  17  *              Rui Oliveira    :       ICMP routing table updates
  18  *              (rco@di.uminho.pt)      Routing table insertion and update
  19  *              Linus Torvalds  :       Rewrote bits to be sensible
  20  *              Alan Cox        :       Added BSD route gw semantics
  21  *              Alan Cox        :       Super /proc >4K
  22  *              Alan Cox        :       MTU in route table
  23  *              Alan Cox        :       MSS actually. Also added the window
  24  *                                      clamper.
  25  *              Sam Lantinga    :       Fixed route matching in rt_del()
  26  *              Alan Cox        :       Routing cache support.
  27  *              Alan Cox        :       Removed compatibility cruft.
  28  *              Alan Cox        :       RTF_REJECT support.
  29  *              Alan Cox        :       TCP irtt support.
  30  *              Jonathan Naylor :       Added Metric support.
  31  *      Miquel van Smoorenburg  :       BSD API fixes.
  32  *      Miquel van Smoorenburg  :       Metrics.
  33  *              Alan Cox        :       Use __u32 properly
  34  *              Alan Cox        :       Aligned routing errors more closely with BSD
  35  *                                      our system is still very different.
  36  *              Alan Cox        :       Faster /proc handling
  37  *      Alexey Kuznetsov        :       Massive rework to support tree based routing,
  38  *                                      routing caches and better behaviour.
  39  *
  40  *              Olaf Erb        :       irtt wasn't being copied right.
  41  *              Bjorn Ekwall    :       Kerneld route support.
  42  *              Alan Cox        :       Multicast fixed (I hope)
  43  *              Pavel Krauz     :       Limited broadcast fixed
  44  *              Mike McLagan    :       Routing by source
  45  *      Alexey Kuznetsov        :       End of old history. Split to fib.c and
  46  *                                      route.c and rewritten from scratch.
  47  *              Andi Kleen      :       Load-limit warning messages.
  48  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  49  *      Vitaly E. Lavrov        :       Race condition in ip_route_input_slow.
  50  *      Tobias Ringstrom        :       Uninitialized res.type in ip_route_output_slow.
  51  *      Vladimir V. Ivanov      :       IP rule info (flowid) is really useful.
  52  *              Marc Boucher    :       routing by fwmark
  53  *      Robert Olsson           :       Added rt_cache statistics
  54  *      Arnaldo C. Melo         :       Convert proc stuff to seq_file
  55  *      Eric Dumazet            :       hashed spinlocks and rt_check_expire() fixes.
  56  *      Ilia Sotnikov           :       Ignore TOS on PMTUD and Redirect
  57  *      Ilia Sotnikov           :       Removed TOS from hash calculations
  58  *
  59  *              This program is free software; you can redistribute it and/or
  60  *              modify it under the terms of the GNU General Public License
  61  *              as published by the Free Software Foundation; either version
  62  *              2 of the License, or (at your option) any later version.
  63  */
  64
  65 #define pr_fmt(fmt) "IPv4: " fmt
  66
  67 #include <linux/module.h>
  68 #include <asm/uaccess.h>
  69 #include <linux/bitops.h>
  70 #include <linux/types.h>
  71 #include <linux/kernel.h>
  72 #include <linux/mm.h>
  73 #include <linux/string.h>
  74 #include <linux/socket.h>
  75 #include <linux/sockios.h>
  76 #include <linux/errno.h>
  77 #include <linux/in.h>
  78 #include <linux/inet.h>
  79 #include <linux/netdevice.h>
  80 #include <linux/proc_fs.h>
  81 #include <linux/init.h>
  82 #include <linux/skbuff.h>
  83 #include <linux/inetdevice.h>
  84 #include <linux/igmp.h>
  85 #include <linux/pkt_sched.h>
  86 #include <linux/mroute.h>
  87 #include <linux/netfilter_ipv4.h>
  88 #include <linux/random.h>
  89 #include <linux/rcupdate.h>
  90 #include <linux/times.h>
  91 #include <linux/slab.h>
  92 #include <linux/jhash.h>
  93 #include <net/dst.h>
  94 #include <net/dst_metadata.h>
  95 #include <net/net_namespace.h>
  96 #include <net/protocol.h>
  97 #include <net/ip.h>
  98 #include <net/route.h>
  99 #include <net/inetpeer.h>
 100 #include <net/sock.h>
 101 #include <net/ip_fib.h>
 102 #include <net/arp.h>
 103 #include <net/tcp.h>
 104 #include <net/icmp.h>
 105 #include <net/xfrm.h>
 106 #include <net/lwtunnel.h>
 107 #include <net/netevent.h>
 108 #include <net/rtnetlink.h>
 109 #ifdef CONFIG_SYSCTL
 110 #include <linux/sysctl.h>
 111 #include <linux/kmemleak.h>
 112 #endif
 113 #include <net/secure_seq.h>
 114 #include <net/ip_tunnels.h>
 115 #include <net/l3mdev.h>
 116
 117 #define RT_FL_TOS(oldflp4) \
 118         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
 119
 120 #define RT_GC_TIMEOUT (300*HZ)
 121
 122 static int ip_rt_max_size;
 123 static int ip_rt_redirect_number __read_mostly  = 9;
 124 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 125 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 126 static int ip_rt_error_cost __read_mostly       = HZ;
 127 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 128 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
 129 static int ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
 130 static int ip_rt_min_advmss __read_mostly       = 256;
 131
 132 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
 133 /*
 134  *      Interface to generic destination cache.
 135  */
 136
 137 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 138 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 139 static unsigned int      ipv4_mtu(const struct dst_entry *dst);
 140 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 141 static void              ipv4_link_failure(struct sk_buff *skb);
 142 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 143                                            struct sk_buff *skb, u32 mtu);
 144 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 145                                         struct sk_buff *skb);
 146 static void             ipv4_dst_destroy(struct dst_entry *dst);
 147
 148 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 149 {
 150         WARN_ON(1);
 151         return NULL;
 152 }
 153
 154 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 155                                            struct sk_buff *skb,
 156                                            const void *daddr);
 157
 158 static struct dst_ops ipv4_dst_ops = {
 159         .family =               AF_INET,
 160         .check =                ipv4_dst_check,
 161         .default_advmss =       ipv4_default_advmss,
 162         .mtu =                  ipv4_mtu,
 163         .cow_metrics =          ipv4_cow_metrics,
 164         .destroy =              ipv4_dst_destroy,
 165         .negative_advice =      ipv4_negative_advice,
 166         .link_failure =         ipv4_link_failure,
 167         .update_pmtu =          ip_rt_update_pmtu,
 168         .redirect =             ip_do_redirect,
 169         .local_out =            __ip_local_out,
 170         .neigh_lookup =         ipv4_neigh_lookup,
 171 };
 172
 173 #define ECN_OR_COST(class)      TC_PRIO_##class
 174
 175 const __u8 ip_tos2prio[16] = {
 176         TC_PRIO_BESTEFFORT,
 177         ECN_OR_COST(BESTEFFORT),
 178         TC_PRIO_BESTEFFORT,
 179         ECN_OR_COST(BESTEFFORT),
 180         TC_PRIO_BULK,
 181         ECN_OR_COST(BULK),
 182         TC_PRIO_BULK,
 183         ECN_OR_COST(BULK),
 184         TC_PRIO_INTERACTIVE,
 185         ECN_OR_COST(INTERACTIVE),
 186         TC_PRIO_INTERACTIVE,
 187         ECN_OR_COST(INTERACTIVE),
 188         TC_PRIO_INTERACTIVE_BULK,
 189         ECN_OR_COST(INTERACTIVE_BULK),
 190         TC_PRIO_INTERACTIVE_BULK,
 191         ECN_OR_COST(INTERACTIVE_BULK)
 192 };
 193 EXPORT_SYMBOL(ip_tos2prio);
 194
 195 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 196 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
 197
 198 #ifdef CONFIG_PROC_FS
 199 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 200 {
 201         if (*pos)
 202                 return NULL;
 203         return SEQ_START_TOKEN;
 204 }
 205
 206 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 207 {
 208         ++*pos;
 209         return NULL;
 210 }
 211
 212 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 213 {
 214 }
 215
 216 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 217 {
 218         if (v == SEQ_START_TOKEN)
 219                 seq_printf(seq, "%-127s\n",
 220                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 221                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 222                            "HHUptod\tSpecDst");
 223         return 0;
 224 }
 225
 226 static const struct seq_operations rt_cache_seq_ops = {
 227         .start  = rt_cache_seq_start,
 228         .next   = rt_cache_seq_next,
 229         .stop   = rt_cache_seq_stop,
 230         .show   = rt_cache_seq_show,
 231 };
 232
 233 static int rt_cache_seq_open(struct inode *inode, struct file *file)
 234 {
 235         return seq_open(file, &rt_cache_seq_ops);
 236 }
 237
 238 static const struct file_operations rt_cache_seq_fops = {
 239         .owner   = THIS_MODULE,
 240         .open    = rt_cache_seq_open,
 241         .read    = seq_read,
 242         .llseek  = seq_lseek,
 243         .release = seq_release,
 244 };
 245
 246
 247 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 248 {
 249         int cpu;
 250
 251         if (*pos == 0)
 252                 return SEQ_START_TOKEN;
 253
 254         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 255                 if (!cpu_possible(cpu))
 256                         continue;
 257                 *pos = cpu+1;
 258                 return &per_cpu(rt_cache_stat, cpu);
 259         }
 260         return NULL;
 261 }
 262
 263 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 264 {
 265         int cpu;
 266
 267         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 268                 if (!cpu_possible(cpu))
 269                         continue;
 270                 *pos = cpu+1;
 271                 return &per_cpu(rt_cache_stat, cpu);
 272         }
 273         return NULL;
 274
 275 }
 276
 277 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 278 {
 279
 280 }
 281
 282 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 283 {
 284         struct rt_cache_stat *st = v;
 285
 286         if (v == SEQ_START_TOKEN) {
 287                 seq_printf(seq, "entries  in_hit in_slow_tot in_slow_mc in_no_route in_brd in_martian_dst in_martian_src  out_hit out_slow_tot out_slow_mc  gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
 288                 return 0;
 289         }
 290
 291         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 292                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 293                    dst_entries_get_slow(&ipv4_dst_ops),
 294                    0, /* st->in_hit */
 295                    st->in_slow_tot,
 296                    st->in_slow_mc,
 297                    st->in_no_route,
 298                    st->in_brd,
 299                    st->in_martian_dst,
 300                    st->in_martian_src,
 301
 302                    0, /* st->out_hit */
 303                    st->out_slow_tot,
 304                    st->out_slow_mc,
 305
 306                    0, /* st->gc_total */
 307                    0, /* st->gc_ignored */
 308                    0, /* st->gc_goal_miss */
 309                    0, /* st->gc_dst_overflow */
 310                    0, /* st->in_hlist_search */
 311                    0  /* st->out_hlist_search */
 312                 );
 313         return 0;
 314 }
 315
 316 static const struct seq_operations rt_cpu_seq_ops = {
 317         .start  = rt_cpu_seq_start,
 318         .next   = rt_cpu_seq_next,
 319         .stop   = rt_cpu_seq_stop,
 320         .show   = rt_cpu_seq_show,
 321 };
 322
 323
 324 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
 325 {
 326         return seq_open(file, &rt_cpu_seq_ops);
 327 }
 328
 329 static const struct file_operations rt_cpu_seq_fops = {
 330         .owner   = THIS_MODULE,
 331         .open    = rt_cpu_seq_open,
 332         .read    = seq_read,
 333         .llseek  = seq_lseek,
 334         .release = seq_release,
 335 };
 336
 337 #ifdef CONFIG_IP_ROUTE_CLASSID
 338 static int rt_acct_proc_show(struct seq_file *m, void *v)
 339 {
 340         struct ip_rt_acct *dst, *src;
 341         unsigned int i, j;
 342
 343         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 344         if (!dst)
 345                 return -ENOMEM;
 346
 347         for_each_possible_cpu(i) {
 348                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 349                 for (j = 0; j < 256; j++) {
 350                         dst[j].o_bytes   += src[j].o_bytes;
 351                         dst[j].o_packets += src[j].o_packets;
 352                         dst[j].i_bytes   += src[j].i_bytes;
 353                         dst[j].i_packets += src[j].i_packets;
 354                 }
 355         }
 356
 357         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 358         kfree(dst);
 359         return 0;
 360 }
 361
 362 static int rt_acct_proc_open(struct inode *inode, struct file *file)
 363 {
 364         return single_open(file, rt_acct_proc_show, NULL);
 365 }
 366
 367 static const struct file_operations rt_acct_proc_fops = {
 368         .owner          = THIS_MODULE,
 369         .open           = rt_acct_proc_open,
 370         .read           = seq_read,
 371         .llseek         = seq_lseek,
 372         .release        = single_release,
 373 };
 374 #endif
 375
 376 static int __net_init ip_rt_do_proc_init(struct net *net)
 377 {
 378         struct proc_dir_entry *pde;
 379
 380         pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
 381                           &rt_cache_seq_fops);
 382         if (!pde)
 383                 goto err1;
 384
 385         pde = proc_create("rt_cache", S_IRUGO,
 386                           net->proc_net_stat, &rt_cpu_seq_fops);
 387         if (!pde)
 388                 goto err2;
 389
 390 #ifdef CONFIG_IP_ROUTE_CLASSID
 391         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
 392         if (!pde)
 393                 goto err3;
 394 #endif
 395         return 0;
 396
 397 #ifdef CONFIG_IP_ROUTE_CLASSID
 398 err3:
 399         remove_proc_entry("rt_cache", net->proc_net_stat);
 400 #endif
 401 err2:
 402         remove_proc_entry("rt_cache", net->proc_net);
 403 err1:
 404         return -ENOMEM;
 405 }
 406
 407 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 408 {
 409         remove_proc_entry("rt_cache", net->proc_net_stat);
 410         remove_proc_entry("rt_cache", net->proc_net);
 411 #ifdef CONFIG_IP_ROUTE_CLASSID
 412         remove_proc_entry("rt_acct", net->proc_net);
 413 #endif
 414 }
 415
 416 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 417         .init = ip_rt_do_proc_init,
 418         .exit = ip_rt_do_proc_exit,
 419 };
 420
 421 static int __init ip_rt_proc_init(void)
 422 {
 423         return register_pernet_subsys(&ip_rt_proc_ops);
 424 }
 425
 426 #else
 427 static inline int ip_rt_proc_init(void)
 428 {
 429         return 0;
 430 }
 431 #endif /* CONFIG_PROC_FS */
 432
 433 static inline bool rt_is_expired(const struct rtable *rth)
 434 {
 435         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 436 }
 437
 438 void rt_cache_flush(struct net *net)
 439 {
 440         rt_genid_bump_ipv4(net);
 441 }
 442
 443 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 444                                            struct sk_buff *skb,
 445                                            const void *daddr)
 446 {
 447         struct net_device *dev = dst->dev;
 448         const __be32 *pkey = daddr;
 449         const struct rtable *rt;
 450         struct neighbour *n;
 451
 452         rt = (const struct rtable *) dst;
 453         if (rt->rt_gateway)
 454                 pkey = (const __be32 *) &rt->rt_gateway;
 455         else if (skb)
 456                 pkey = &ip_hdr(skb)->daddr;
 457
 458         n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
 459         if (n)
 460                 return n;
 461         return neigh_create(&arp_tbl, pkey, dev);
 462 }
 463
 464 #define IP_IDENTS_SZ 2048u
 465
 466 static atomic_t *ip_idents __read_mostly;
 467 static u32 *ip_tstamps __read_mostly;
 468
 469 /* In order to protect privacy, we add a perturbation to identifiers
 470  * if one generator is seldom used. This makes hard for an attacker
 471  * to infer how many packets were sent between two points in time.
 472  */
 473 u32 ip_idents_reserve(u32 hash, int segs)
 474 {
 475         u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
 476         atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
 477         u32 old = ACCESS_ONCE(*p_tstamp);
 478         u32 now = (u32)jiffies;
 479         u32 delta = 0;
 480
 481         if (old != now && cmpxchg(p_tstamp, old, now) == old)
 482                 delta = prandom_u32_max(now - old);
 483
 484         return atomic_add_return(segs + delta, p_id) - segs;
 485 }
 486 EXPORT_SYMBOL(ip_idents_reserve);
 487
 488 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
 489 {
 490         static u32 ip_idents_hashrnd __read_mostly;
 491         u32 hash, id;
 492
 493         net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
 494
 495         hash = jhash_3words((__force u32)iph->daddr,
 496                             (__force u32)iph->saddr,
 497                             iph->protocol ^ net_hash_mix(net),
 498                             ip_idents_hashrnd);
 499         id = ip_idents_reserve(hash, segs);
 500         iph->id = htons(id);
 501 }
 502 EXPORT_SYMBOL(__ip_select_ident);
 503
 504 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 505                              const struct iphdr *iph,
 506                              int oif, u8 tos,
 507                              u8 prot, u32 mark, int flow_flags)
 508 {
 509         if (sk) {
 510                 const struct inet_sock *inet = inet_sk(sk);
 511
 512                 oif = sk->sk_bound_dev_if;
 513                 mark = sk->sk_mark;
 514                 tos = RT_CONN_FLAGS(sk);
 515                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 516         }
 517         flowi4_init_output(fl4, oif, mark, tos,
 518                            RT_SCOPE_UNIVERSE, prot,
 519                            flow_flags,
 520                            iph->daddr, iph->saddr, 0, 0);
 521 }
 522
 523 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 524                                const struct sock *sk)
 525 {
 526         const struct iphdr *iph = ip_hdr(skb);
 527         int oif = skb->dev->ifindex;
 528         u8 tos = RT_TOS(iph->tos);
 529         u8 prot = iph->protocol;
 530         u32 mark = skb->mark;
 531
 532         __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
 533 }
 534
 535 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 536 {
 537         const struct inet_sock *inet = inet_sk(sk);
 538         const struct ip_options_rcu *inet_opt;
 539         __be32 daddr = inet->inet_daddr;
 540
 541         rcu_read_lock();
 542         inet_opt = rcu_dereference(inet->inet_opt);
 543         if (inet_opt && inet_opt->opt.srr)
 544                 daddr = inet_opt->opt.faddr;
 545         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 546                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 547                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 548                            inet_sk_flowi_flags(sk),
 549                            daddr, inet->inet_saddr, 0, 0);
 550         rcu_read_unlock();
 551 }
 552
 553 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 554                                  const struct sk_buff *skb)
 555 {
 556         if (skb)
 557                 build_skb_flow_key(fl4, skb, sk);
 558         else
 559                 build_sk_flow_key(fl4, sk);
 560 }
 561
 562 static inline void rt_free(struct rtable *rt)
 563 {
 564         call_rcu(&rt->dst.rcu_head, dst_rcu_free);
 565 }
 566
 567 static DEFINE_SPINLOCK(fnhe_lock);
 568
 569 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 570 {
 571         struct rtable *rt;
 572
 573         rt = rcu_dereference(fnhe->fnhe_rth_input);
 574         if (rt) {
 575                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 576                 rt_free(rt);
 577         }
 578         rt = rcu_dereference(fnhe->fnhe_rth_output);
 579         if (rt) {
 580                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 581                 rt_free(rt);
 582         }
 583 }
 584
 585 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
 586 {
 587         struct fib_nh_exception *fnhe, *oldest;
 588
 589         oldest = rcu_dereference(hash->chain);
 590         for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
 591              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 592                 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
 593                         oldest = fnhe;
 594         }
 595         fnhe_flush_routes(oldest);
 596         return oldest;
 597 }
 598
 599 static inline u32 fnhe_hashfun(__be32 daddr)
 600 {
 601         static u32 fnhe_hashrnd __read_mostly;
 602         u32 hval;
 603
 604         net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
 605         hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
 606         return hash_32(hval, FNHE_HASH_SHIFT);
 607 }
 608
 609 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 610 {
 611         rt->rt_pmtu = fnhe->fnhe_pmtu;
 612         rt->dst.expires = fnhe->fnhe_expires;
 613
 614         if (fnhe->fnhe_gw) {
 615                 rt->rt_flags |= RTCF_REDIRECTED;
 616                 rt->rt_gateway = fnhe->fnhe_gw;
 617                 rt->rt_uses_gateway = 1;
 618         }
 619 }
 620
 621 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
 622                                   u32 pmtu, unsigned long expires)
 623 {
 624         struct fnhe_hash_bucket *hash;
 625         struct fib_nh_exception *fnhe;
 626         struct rtable *rt;
 627         unsigned int i;
 628         int depth;
 629         u32 hval = fnhe_hashfun(daddr);
 630
 631         spin_lock_bh(&fnhe_lock);
 632
 633         hash = rcu_dereference(nh->nh_exceptions);
 634         if (!hash) {
 635                 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
 636                 if (!hash)
 637                         goto out_unlock;
 638                 rcu_assign_pointer(nh->nh_exceptions, hash);
 639         }
 640
 641         hash += hval;
 642
 643         depth = 0;
 644         for (fnhe = rcu_dereference(hash->chain); fnhe;
 645              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 646                 if (fnhe->fnhe_daddr == daddr)
 647                         break;
 648                 depth++;
 649         }
 650
 651         if (fnhe) {
 652                 if (gw)
 653                         fnhe->fnhe_gw = gw;
 654                 if (pmtu) {
 655                         fnhe->fnhe_pmtu = pmtu;
 656                         fnhe->fnhe_expires = max(1UL, expires);
 657                 }
 658                 /* Update all cached dsts too */
 659                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 660                 if (rt)
 661                         fill_route_from_fnhe(rt, fnhe);
 662                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 663                 if (rt)
 664                         fill_route_from_fnhe(rt, fnhe);
 665         } else {
 666                 if (depth > FNHE_RECLAIM_DEPTH)
 667                         fnhe = fnhe_oldest(hash);
 668                 else {
 669                         fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 670                         if (!fnhe)
 671                                 goto out_unlock;
 672
 673                         fnhe->fnhe_next = hash->chain;
 674                         rcu_assign_pointer(hash->chain, fnhe);
 675                 }
 676                 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
 677                 fnhe->fnhe_daddr = daddr;
 678                 fnhe->fnhe_gw = gw;
 679                 fnhe->fnhe_pmtu = pmtu;
 680                 fnhe->fnhe_expires = expires;
 681
 682                 /* Exception created; mark the cached routes for the nexthop
 683                  * stale, so anyone caching it rechecks if this exception
 684                  * applies to them.
 685                  */
 686                 rt = rcu_dereference(nh->nh_rth_input);
 687                 if (rt)
 688                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 689
 690                 for_each_possible_cpu(i) {
 691                         struct rtable __rcu **prt;
 692                         prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
 693                         rt = rcu_dereference(*prt);
 694                         if (rt)
 695                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 696                 }
 697         }
 698
 699         fnhe->fnhe_stamp = jiffies;
 700
 701 out_unlock:
 702         spin_unlock_bh(&fnhe_lock);
 703 }
 704
 705 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 706                              bool kill_route)
 707 {
 708         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 709         __be32 old_gw = ip_hdr(skb)->saddr;
 710         struct net_device *dev = skb->dev;
 711         struct in_device *in_dev;
 712         struct fib_result res;
 713         struct neighbour *n;
 714         struct net *net;
 715
 716         switch (icmp_hdr(skb)->code & 7) {
 717         case ICMP_REDIR_NET:
 718         case ICMP_REDIR_NETTOS:
 719         case ICMP_REDIR_HOST:
 720         case ICMP_REDIR_HOSTTOS:
 721                 break;
 722
 723         default:
 724                 return;
 725         }
 726
 727         if (rt->rt_gateway != old_gw)
 728                 return;
 729
 730         in_dev = __in_dev_get_rcu(dev);
 731         if (!in_dev)
 732                 return;
 733
 734         net = dev_net(dev);
 735         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 736             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 737             ipv4_is_zeronet(new_gw))
 738                 goto reject_redirect;
 739
 740         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 741                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 742                         goto reject_redirect;
 743                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 744                         goto reject_redirect;
 745         } else {
 746                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 747                         goto reject_redirect;
 748         }
 749
 750         n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
 751         if (!IS_ERR(n)) {
 752                 if (!(n->nud_state & NUD_VALID)) {
 753                         neigh_event_send(n, NULL);
 754                 } else {
 755                         if (fib_lookup(net, fl4, &res, 0) == 0) {
 756                                 struct fib_nh *nh = &FIB_RES_NH(res);
 757
 758                                 update_or_create_fnhe(nh, fl4->daddr, new_gw,
 759                                                 0, jiffies + ip_rt_gc_timeout);
 760                         }
 761                         if (kill_route)
 762                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 763                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 764                 }
 765                 neigh_release(n);
 766         }
 767         return;
 768
 769 reject_redirect:
 770 #ifdef CONFIG_IP_ROUTE_VERBOSE
 771         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 772                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 773                 __be32 daddr = iph->daddr;
 774                 __be32 saddr = iph->saddr;
 775
 776                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 777                                      "  Advised path = %pI4 -> %pI4\n",
 778                                      &old_gw, dev->name, &new_gw,
 779                                      &saddr, &daddr);
 780         }
 781 #endif
 782         ;
 783 }
 784
 785 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 786 {
 787         struct rtable *rt;
 788         struct flowi4 fl4;
 789         const struct iphdr *iph = (const struct iphdr *) skb->data;
 790         int oif = skb->dev->ifindex;
 791         u8 tos = RT_TOS(iph->tos);
 792         u8 prot = iph->protocol;
 793         u32 mark = skb->mark;
 794
 795         rt = (struct rtable *) dst;
 796
 797         __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
 798         __ip_do_redirect(rt, skb, &fl4, true);
 799 }
 800
 801 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 802 {
 803         struct rtable *rt = (struct rtable *)dst;
 804         struct dst_entry *ret = dst;
 805
 806         if (rt) {
 807                 if (dst->obsolete > 0) {
 808                         ip_rt_put(rt);
 809                         ret = NULL;
 810                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 811                            rt->dst.expires) {
 812                         ip_rt_put(rt);
 813                         ret = NULL;
 814                 }
 815         }
 816         return ret;
 817 }
 818
 819 /*
 820  * Algorithm:
 821  *      1. The first ip_rt_redirect_number redirects are sent
 822  *         with exponential backoff, then we stop sending them at all,
 823  *         assuming that the host ignores our redirects.
 824  *      2. If we did not see packets requiring redirects
 825  *         during ip_rt_redirect_silence, we assume that the host
 826  *         forgot redirected route and start to send redirects again.
 827  *
 828  * This algorithm is much cheaper and more intelligent than dumb load limiting
 829  * in icmp.c.
 830  *
 831  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 832  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 833  */
 834
 835 void ip_rt_send_redirect(struct sk_buff *skb)
 836 {
 837         struct rtable *rt = skb_rtable(skb);
 838         struct in_device *in_dev;
 839         struct inet_peer *peer;
 840         struct net *net;
 841         int log_martians;
 842         int vif;
 843
 844         rcu_read_lock();
 845         in_dev = __in_dev_get_rcu(rt->dst.dev);
 846         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 847                 rcu_read_unlock();
 848                 return;
 849         }
 850         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 851         vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
 852         rcu_read_unlock();
 853
 854         net = dev_net(rt->dst.dev);
 855         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
 856         if (!peer) {
 857                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 858                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 859                 return;
 860         }
 861
 862         /* No redirected packets during ip_rt_redirect_silence;
 863          * reset the algorithm.
 864          */
 865         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
 866                 peer->rate_tokens = 0;
 867
 868         /* Too many ignored redirects; do not send anything
 869          * set dst.rate_last to the last seen redirected packet.
 870          */
 871         if (peer->rate_tokens >= ip_rt_redirect_number) {
 872                 peer->rate_last = jiffies;
 873                 goto out_put_peer;
 874         }
 875
 876         /* Check for load limit; set rate_last to the latest sent
 877          * redirect.
 878          */
 879         if (peer->rate_tokens == 0 ||
 880             time_after(jiffies,
 881                        (peer->rate_last +
 882                         (ip_rt_redirect_load << peer->rate_tokens)))) {
 883                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 884
 885                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 886                 peer->rate_last = jiffies;
 887                 ++peer->rate_tokens;
 888 #ifdef CONFIG_IP_ROUTE_VERBOSE
 889                 if (log_martians &&
 890                     peer->rate_tokens == ip_rt_redirect_number)
 891                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 892                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 893                                              &ip_hdr(skb)->daddr, &gw);
 894 #endif
 895         }
 896 out_put_peer:
 897         inet_putpeer(peer);
 898 }
 899
 900 static int ip_error(struct sk_buff *skb)
 901 {
 902         struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
 903         struct rtable *rt = skb_rtable(skb);
 904         struct inet_peer *peer;
 905         unsigned long now;
 906         struct net *net;
 907         bool send;
 908         int code;
 909
 910         /* IP on this device is disabled. */
 911         if (!in_dev)
 912                 goto out;
 913
 914         net = dev_net(rt->dst.dev);
 915         if (!IN_DEV_FORWARD(in_dev)) {
 916                 switch (rt->dst.error) {
 917                 case EHOSTUNREACH:
 918                         __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
 919                         break;
 920
 921                 case ENETUNREACH:
 922                         __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 923                         break;
 924                 }
 925                 goto out;
 926         }
 927
 928         switch (rt->dst.error) {
 929         case EINVAL:
 930         default:
 931                 goto out;
 932         case EHOSTUNREACH:
 933                 code = ICMP_HOST_UNREACH;
 934                 break;
 935         case ENETUNREACH:
 936                 code = ICMP_NET_UNREACH;
 937                 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 938                 break;
 939         case EACCES:
 940                 code = ICMP_PKT_FILTERED;
 941                 break;
 942         }
 943
 944         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
 945                                l3mdev_master_ifindex(skb->dev), 1);
 946
 947         send = true;
 948         if (peer) {
 949                 now = jiffies;
 950                 peer->rate_tokens += now - peer->rate_last;
 951                 if (peer->rate_tokens > ip_rt_error_burst)
 952                         peer->rate_tokens = ip_rt_error_burst;
 953                 peer->rate_last = now;
 954                 if (peer->rate_tokens >= ip_rt_error_cost)
 955                         peer->rate_tokens -= ip_rt_error_cost;
 956                 else
 957                         send = false;
 958                 inet_putpeer(peer);
 959         }
 960         if (send)
 961                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
 962
 963 out:    kfree_skb(skb);
 964         return 0;
 965 }
 966
 967 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
 968 {
 969         struct dst_entry *dst = &rt->dst;
 970         struct fib_result res;
 971
 972         if (dst_metric_locked(dst, RTAX_MTU))
 973                 return;
 974
 975         if (ipv4_mtu(dst) < mtu)
 976                 return;
 977
 978         if (mtu < ip_rt_min_pmtu)
 979                 mtu = ip_rt_min_pmtu;
 980
 981         if (rt->rt_pmtu == mtu &&
 982             time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
 983                 return;
 984
 985         rcu_read_lock();
 986         if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
 987                 struct fib_nh *nh = &FIB_RES_NH(res);
 988
 989                 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
 990                                       jiffies + ip_rt_mtu_expires);
 991         }
 992         rcu_read_unlock();
 993 }
 994
 995 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 996                               struct sk_buff *skb, u32 mtu)
 997 {
 998         struct rtable *rt = (struct rtable *) dst;
 999         struct flowi4 fl4;
1000
1001         ip_rt_build_flow_key(&fl4, sk, skb);
1002         __ip_rt_update_pmtu(rt, &fl4, mtu);
1003 }
1004
1005 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1006                       int oif, u32 mark, u8 protocol, int flow_flags)
1007 {
1008         const struct iphdr *iph = (const struct iphdr *) skb->data;
1009         struct flowi4 fl4;
1010         struct rtable *rt;
1011
1012         if (!mark)
1013                 mark = IP4_REPLY_MARK(net, skb->mark);
1014
1015         __build_flow_key(&fl4, NULL, iph, oif,
1016                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1017         rt = __ip_route_output_key(net, &fl4);
1018         if (!IS_ERR(rt)) {
1019                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1020                 ip_rt_put(rt);
1021         }
1022 }
1023 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1024
1025 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1026 {
1027         const struct iphdr *iph = (const struct iphdr *) skb->data;
1028         struct flowi4 fl4;
1029         struct rtable *rt;
1030
1031         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1032
1033         if (!fl4.flowi4_mark)
1034                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1035
1036         rt = __ip_route_output_key(sock_net(sk), &fl4);
1037         if (!IS_ERR(rt)) {
1038                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1039                 ip_rt_put(rt);
1040         }
1041 }
1042
1043 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1044 {
1045         const struct iphdr *iph = (const struct iphdr *) skb->data;
1046         struct flowi4 fl4;
1047         struct rtable *rt;
1048         struct dst_entry *odst = NULL;
1049         bool new = false;
1050
1051         bh_lock_sock(sk);
1052
1053         if (!ip_sk_accept_pmtu(sk))
1054                 goto out;
1055
1056         odst = sk_dst_get(sk);
1057
1058         if (sock_owned_by_user(sk) || !odst) {
1059                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1060                 goto out;
1061         }
1062
1063         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1064
1065         rt = (struct rtable *)odst;
1066         if (odst->obsolete && !odst->ops->check(odst, 0)) {
1067                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1068                 if (IS_ERR(rt))
1069                         goto out;
1070
1071                 new = true;
1072         }
1073
1074         __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1075
1076         if (!dst_check(&rt->dst, 0)) {
1077                 if (new)
1078                         dst_release(&rt->dst);
1079
1080                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1081                 if (IS_ERR(rt))
1082                         goto out;
1083
1084                 new = true;
1085         }
1086
1087         if (new)
1088                 sk_dst_set(sk, &rt->dst);
1089
1090 out:
1091         bh_unlock_sock(sk);
1092         dst_release(odst);
1093 }
1094 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1095
1096 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1097                    int oif, u32 mark, u8 protocol, int flow_flags)
1098 {
1099         const struct iphdr *iph = (const struct iphdr *) skb->data;
1100         struct flowi4 fl4;
1101         struct rtable *rt;
1102
1103         __build_flow_key(&fl4, NULL, iph, oif,
1104                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1105         rt = __ip_route_output_key(net, &fl4);
1106         if (!IS_ERR(rt)) {
1107                 __ip_do_redirect(rt, skb, &fl4, false);
1108                 ip_rt_put(rt);
1109         }
1110 }
1111 EXPORT_SYMBOL_GPL(ipv4_redirect);
1112
1113 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1114 {
1115         const struct iphdr *iph = (const struct iphdr *) skb->data;
1116         struct flowi4 fl4;
1117         struct rtable *rt;
1118
1119         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1120         rt = __ip_route_output_key(sock_net(sk), &fl4);
1121         if (!IS_ERR(rt)) {
1122                 __ip_do_redirect(rt, skb, &fl4, false);
1123                 ip_rt_put(rt);
1124         }
1125 }
1126 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1127
1128 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1129 {
1130         struct rtable *rt = (struct rtable *) dst;
1131
1132         /* All IPV4 dsts are created with ->obsolete set to the value
1133          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1134          * into this function always.
1135          *
1136          * When a PMTU/redirect information update invalidates a route,
1137          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1138          * DST_OBSOLETE_DEAD by dst_free().
1139          */
1140         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1141                 return NULL;
1142         return dst;
1143 }
1144
1145 static void ipv4_link_failure(struct sk_buff *skb)
1146 {
1147         struct rtable *rt;
1148
1149         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1150
1151         rt = skb_rtable(skb);
1152         if (rt)
1153                 dst_set_expires(&rt->dst, 0);
1154 }
1155
1156 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1157 {
1158         pr_debug("%s: %pI4 -> %pI4, %s\n",
1159                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1160                  skb->dev ? skb->dev->name : "?");
1161         kfree_skb(skb);
1162         WARN_ON(1);
1163         return 0;
1164 }
1165
1166 /*
1167    We do not cache source address of outgoing interface,
1168    because it is used only by IP RR, TS and SRR options,
1169    so that it out of fast path.
1170
1171    BTW remember: "addr" is allowed to be not aligned
1172    in IP options!
1173  */
1174
1175 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1176 {
1177         __be32 src;
1178
1179         if (rt_is_output_route(rt))
1180                 src = ip_hdr(skb)->saddr;
1181         else {
1182                 struct fib_result res;
1183                 struct flowi4 fl4;
1184                 struct iphdr *iph;
1185
1186                 iph = ip_hdr(skb);
1187
1188                 memset(&fl4, 0, sizeof(fl4));
1189                 fl4.daddr = iph->daddr;
1190                 fl4.saddr = iph->saddr;
1191                 fl4.flowi4_tos = RT_TOS(iph->tos);
1192                 fl4.flowi4_oif = rt->dst.dev->ifindex;
1193                 fl4.flowi4_iif = skb->dev->ifindex;
1194                 fl4.flowi4_mark = skb->mark;
1195
1196                 rcu_read_lock();
1197                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1198                         src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1199                 else
1200                         src = inet_select_addr(rt->dst.dev,
1201                                                rt_nexthop(rt, iph->daddr),
1202                                                RT_SCOPE_UNIVERSE);
1203                 rcu_read_unlock();
1204         }
1205         memcpy(addr, &src, 4);
1206 }
1207
1208 #ifdef CONFIG_IP_ROUTE_CLASSID
1209 static void set_class_tag(struct rtable *rt, u32 tag)
1210 {
1211         if (!(rt->dst.tclassid & 0xFFFF))
1212                 rt->dst.tclassid |= tag & 0xFFFF;
1213         if (!(rt->dst.tclassid & 0xFFFF0000))
1214                 rt->dst.tclassid |= tag & 0xFFFF0000;
1215 }
1216 #endif
1217
1218 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1219 {
1220         unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1221
1222         if (advmss == 0) {
1223                 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1224                                ip_rt_min_advmss);
1225                 if (advmss > 65535 - 40)
1226                         advmss = 65535 - 40;
1227         }
1228         return advmss;
1229 }
1230
1231 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1232 {
1233         const struct rtable *rt = (const struct rtable *) dst;
1234         unsigned int mtu = rt->rt_pmtu;
1235
1236         if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1237                 mtu = dst_metric_raw(dst, RTAX_MTU);
1238
1239         if (mtu)
1240                 return mtu;
1241
1242         mtu = dst->dev->mtu;
1243
1244         if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1245                 if (rt->rt_uses_gateway && mtu > 576)
1246                         mtu = 576;
1247         }
1248
1249         return min_t(unsigned int, mtu, IP_MAX_MTU);
1250 }
1251
1252 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1253 {
1254         struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1255         struct fib_nh_exception *fnhe;
1256         u32 hval;
1257
1258         if (!hash)
1259                 return NULL;
1260
1261         hval = fnhe_hashfun(daddr);
1262
1263         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1264              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1265                 if (fnhe->fnhe_daddr == daddr)
1266                         return fnhe;
1267         }
1268         return NULL;
1269 }
1270
1271 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1272                               __be32 daddr)
1273 {
1274         bool ret = false;
1275
1276         spin_lock_bh(&fnhe_lock);
1277
1278         if (daddr == fnhe->fnhe_daddr) {
1279                 struct rtable __rcu **porig;
1280                 struct rtable *orig;
1281                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1282
1283                 if (rt_is_input_route(rt))
1284                         porig = &fnhe->fnhe_rth_input;
1285                 else
1286                         porig = &fnhe->fnhe_rth_output;
1287                 orig = rcu_dereference(*porig);
1288
1289                 if (fnhe->fnhe_genid != genid) {
1290                         fnhe->fnhe_genid = genid;
1291                         fnhe->fnhe_gw = 0;
1292                         fnhe->fnhe_pmtu = 0;
1293                         fnhe->fnhe_expires = 0;
1294                         fnhe_flush_routes(fnhe);
1295                         orig = NULL;
1296                 }
1297                 fill_route_from_fnhe(rt, fnhe);
1298                 if (!rt->rt_gateway)
1299                         rt->rt_gateway = daddr;
1300
1301                 if (!(rt->dst.flags & DST_NOCACHE)) {
1302                         rcu_assign_pointer(*porig, rt);
1303                         if (orig)
1304                                 rt_free(orig);
1305                         ret = true;
1306                 }
1307
1308                 fnhe->fnhe_stamp = jiffies;
1309         }
1310         spin_unlock_bh(&fnhe_lock);
1311
1312         return ret;
1313 }
1314
1315 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1316 {
1317         struct rtable *orig, *prev, **p;
1318         bool ret = true;
1319
1320         if (rt_is_input_route(rt)) {
1321                 p = (struct rtable **)&nh->nh_rth_input;
1322         } else {
1323                 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1324         }
1325         orig = *p;
1326
1327         prev = cmpxchg(p, orig, rt);
1328         if (prev == orig) {
1329                 if (orig)
1330                         rt_free(orig);
1331         } else
1332                 ret = false;
1333
1334         return ret;
1335 }
1336
1337 struct uncached_list {
1338         spinlock_t              lock;
1339         struct list_head        head;
1340 };
1341
1342 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1343
1344 static void rt_add_uncached_list(struct rtable *rt)
1345 {
1346         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1347
1348         rt->rt_uncached_list = ul;
1349
1350         spin_lock_bh(&ul->lock);
1351         list_add_tail(&rt->rt_uncached, &ul->head);
1352         spin_unlock_bh(&ul->lock);
1353 }
1354
1355 static void ipv4_dst_destroy(struct dst_entry *dst)
1356 {
1357         struct rtable *rt = (struct rtable *) dst;
1358
1359         if (!list_empty(&rt->rt_uncached)) {
1360                 struct uncached_list *ul = rt->rt_uncached_list;
1361
1362                 spin_lock_bh(&ul->lock);
1363                 list_del(&rt->rt_uncached);
1364                 spin_unlock_bh(&ul->lock);
1365         }
1366 }
1367
1368 void rt_flush_dev(struct net_device *dev)
1369 {
1370         struct net *net = dev_net(dev);
1371         struct rtable *rt;
1372         int cpu;
1373
1374         for_each_possible_cpu(cpu) {
1375                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1376
1377                 spin_lock_bh(&ul->lock);
1378                 list_for_each_entry(rt, &ul->head, rt_uncached) {
1379                         if (rt->dst.dev != dev)
1380                                 continue;
1381                         rt->dst.dev = net->loopback_dev;
1382                         dev_hold(rt->dst.dev);
1383                         dev_put(dev);
1384                 }
1385                 spin_unlock_bh(&ul->lock);
1386         }
1387 }
1388
1389 static bool rt_cache_valid(const struct rtable *rt)
1390 {
1391         return  rt &&
1392                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1393                 !rt_is_expired(rt);
1394 }
1395
1396 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1397                            const struct fib_result *res,
1398                            struct fib_nh_exception *fnhe,
1399                            struct fib_info *fi, u16 type, u32 itag)
1400 {
1401         bool cached = false;
1402
1403         if (fi) {
1404                 struct fib_nh *nh = &FIB_RES_NH(*res);
1405
1406                 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1407                         rt->rt_gateway = nh->nh_gw;
1408                         rt->rt_uses_gateway = 1;
1409                 }
1410                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1411 #ifdef CONFIG_IP_ROUTE_CLASSID
1412                 rt->dst.tclassid = nh->nh_tclassid;
1413 #endif
1414                 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1415                 if (unlikely(fnhe))
1416                         cached = rt_bind_exception(rt, fnhe, daddr);
1417                 else if (!(rt->dst.flags & DST_NOCACHE))
1418                         cached = rt_cache_route(nh, rt);
1419                 if (unlikely(!cached)) {
1420                         /* Routes we intend to cache in nexthop exception or
1421                          * FIB nexthop have the DST_NOCACHE bit clear.
1422                          * However, if we are unsuccessful at storing this
1423                          * route into the cache we really need to set it.
1424                          */
1425                         rt->dst.flags |= DST_NOCACHE;
1426                         if (!rt->rt_gateway)
1427                                 rt->rt_gateway = daddr;
1428                         rt_add_uncached_list(rt);
1429                 }
1430         } else
1431                 rt_add_uncached_list(rt);
1432
1433 #ifdef CONFIG_IP_ROUTE_CLASSID
1434 #ifdef CONFIG_IP_MULTIPLE_TABLES
1435         set_class_tag(rt, res->tclassid);
1436 #endif
1437         set_class_tag(rt, itag);
1438 #endif
1439 }
1440
1441 struct rtable *rt_dst_alloc(struct net_device *dev,
1442                             unsigned int flags, u16 type,
1443                             bool nopolicy, bool noxfrm, bool will_cache)
1444 {
1445         struct rtable *rt;
1446
1447         rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1448                        (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1449                        (nopolicy ? DST_NOPOLICY : 0) |
1450                        (noxfrm ? DST_NOXFRM : 0));
1451
1452         if (rt) {
1453                 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1454                 rt->rt_flags = flags;
1455                 rt->rt_type = type;
1456                 rt->rt_is_input = 0;
1457                 rt->rt_iif = 0;
1458                 rt->rt_pmtu = 0;
1459                 rt->rt_gateway = 0;
1460                 rt->rt_uses_gateway = 0;
1461                 rt->rt_table_id = 0;
1462                 INIT_LIST_HEAD(&rt->rt_uncached);
1463
1464                 rt->dst.output = ip_output;
1465                 if (flags & RTCF_LOCAL)
1466                         rt->dst.input = ip_local_deliver;
1467         }
1468
1469         return rt;
1470 }
1471 EXPORT_SYMBOL(rt_dst_alloc);
1472
1473 /* called in rcu_read_lock() section */
1474 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1475                                 u8 tos, struct net_device *dev, int our)
1476 {
1477         struct rtable *rth;
1478         struct in_device *in_dev = __in_dev_get_rcu(dev);
1479         unsigned int flags = RTCF_MULTICAST;
1480         u32 itag = 0;
1481         int err;
1482
1483         /* Primary sanity checks. */
1484
1485         if (!in_dev)
1486                 return -EINVAL;
1487
1488         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1489             skb->protocol != htons(ETH_P_IP))
1490                 goto e_inval;
1491
1492         if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1493                 goto e_inval;
1494
1495         if (ipv4_is_zeronet(saddr)) {
1496                 if (!ipv4_is_local_multicast(daddr))
1497                         goto e_inval;
1498         } else {
1499                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1500                                           in_dev, &itag);
1501                 if (err < 0)
1502                         goto e_err;
1503         }
1504         if (our)
1505                 flags |= RTCF_LOCAL;
1506
1507         rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1508                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1509         if (!rth)
1510                 goto e_nobufs;
1511
1512 #ifdef CONFIG_IP_ROUTE_CLASSID
1513         rth->dst.tclassid = itag;
1514 #endif
1515         rth->dst.output = ip_rt_bug;
1516         rth->rt_is_input= 1;
1517
1518 #ifdef CONFIG_IP_MROUTE
1519         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1520                 rth->dst.input = ip_mr_input;
1521 #endif
1522         RT_CACHE_STAT_INC(in_slow_mc);
1523
1524         skb_dst_set(skb, &rth->dst);
1525         return 0;
1526
1527 e_nobufs:
1528         return -ENOBUFS;
1529 e_inval:
1530         return -EINVAL;
1531 e_err:
1532         return err;
1533 }
1534
1535
1536 static void ip_handle_martian_source(struct net_device *dev,
1537                                      struct in_device *in_dev,
1538                                      struct sk_buff *skb,
1539                                      __be32 daddr,
1540                                      __be32 saddr)
1541 {
1542         RT_CACHE_STAT_INC(in_martian_src);
1543 #ifdef CONFIG_IP_ROUTE_VERBOSE
1544         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1545                 /*
1546                  *      RFC1812 recommendation, if source is martian,
1547                  *      the only hint is MAC header.
1548                  */
1549                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1550                         &daddr, &saddr, dev->name);
1551                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1552                         print_hex_dump(KERN_WARNING, "ll header: ",
1553                                        DUMP_PREFIX_OFFSET, 16, 1,
1554                                        skb_mac_header(skb),
1555                                        dev->hard_header_len, true);
1556                 }
1557         }
1558 #endif
1559 }
1560
1561 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1562 {
1563         struct fnhe_hash_bucket *hash;
1564         struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1565         u32 hval = fnhe_hashfun(daddr);
1566
1567         spin_lock_bh(&fnhe_lock);
1568
1569         hash = rcu_dereference_protected(nh->nh_exceptions,
1570                                          lockdep_is_held(&fnhe_lock));
1571         hash += hval;
1572
1573         fnhe_p = &hash->chain;
1574         fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1575         while (fnhe) {
1576                 if (fnhe->fnhe_daddr == daddr) {
1577                         rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1578                                 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1579                         fnhe_flush_routes(fnhe);
1580                         kfree_rcu(fnhe, rcu);
1581                         break;
1582                 }
1583                 fnhe_p = &fnhe->fnhe_next;
1584                 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1585                                                  lockdep_is_held(&fnhe_lock));
1586         }
1587
1588         spin_unlock_bh(&fnhe_lock);
1589 }
1590
1591 /* called in rcu_read_lock() section */
1592 static int __mkroute_input(struct sk_buff *skb,
1593                            const struct fib_result *res,
1594                            struct in_device *in_dev,
1595                            __be32 daddr, __be32 saddr, u32 tos)
1596 {
1597         struct fib_nh_exception *fnhe;
1598         struct rtable *rth;
1599         int err;
1600         struct in_device *out_dev;
1601         bool do_cache;
1602         u32 itag = 0;
1603
1604         /* get a working reference to the output device */
1605         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1606         if (!out_dev) {
1607                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1608                 return -EINVAL;
1609         }
1610
1611         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1612                                   in_dev->dev, in_dev, &itag);
1613         if (err < 0) {
1614                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1615                                          saddr);
1616
1617                 goto cleanup;
1618         }
1619
1620         do_cache = res->fi && !itag;
1621         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1622             skb->protocol == htons(ETH_P_IP) &&
1623             (IN_DEV_SHARED_MEDIA(out_dev) ||
1624              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1625                 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1626
1627         if (skb->protocol != htons(ETH_P_IP)) {
1628                 /* Not IP (i.e. ARP). Do not create route, if it is
1629                  * invalid for proxy arp. DNAT routes are always valid.
1630                  *
1631                  * Proxy arp feature have been extended to allow, ARP
1632                  * replies back to the same interface, to support
1633                  * Private VLAN switch technologies. See arp.c.
1634                  */
1635                 if (out_dev == in_dev &&
1636                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1637                         err = -EINVAL;
1638                         goto cleanup;
1639                 }
1640         }
1641
1642         fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1643         if (do_cache) {
1644                 if (fnhe) {
1645                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1646                         if (rth && rth->dst.expires &&
1647                             time_after(jiffies, rth->dst.expires)) {
1648                                 ip_del_fnhe(&FIB_RES_NH(*res), daddr);
1649                                 fnhe = NULL;
1650                         } else {
1651                                 goto rt_cache;
1652                         }
1653                 }
1654
1655                 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1656
1657 rt_cache:
1658                 if (rt_cache_valid(rth)) {
1659                         skb_dst_set_noref(skb, &rth->dst);
1660                         goto out;
1661                 }
1662         }
1663
1664         rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1665                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1666                            IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1667         if (!rth) {
1668                 err = -ENOBUFS;
1669                 goto cleanup;
1670         }
1671
1672         rth->rt_is_input = 1;
1673         if (res->table)
1674                 rth->rt_table_id = res->table->tb_id;
1675         RT_CACHE_STAT_INC(in_slow_tot);
1676
1677         rth->dst.input = ip_forward;
1678
1679         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1680         if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1681                 rth->dst.lwtstate->orig_output = rth->dst.output;
1682                 rth->dst.output = lwtunnel_output;
1683         }
1684         if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1685                 rth->dst.lwtstate->orig_input = rth->dst.input;
1686                 rth->dst.input = lwtunnel_input;
1687         }
1688         skb_dst_set(skb, &rth->dst);
1689 out:
1690         err = 0;
1691  cleanup:
1692         return err;
1693 }
1694
1695 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1696
1697 /* To make ICMP packets follow the right flow, the multipath hash is
1698  * calculated from the inner IP addresses in reverse order.
1699  */
1700 static int ip_multipath_icmp_hash(struct sk_buff *skb)
1701 {
1702         const struct iphdr *outer_iph = ip_hdr(skb);
1703         struct icmphdr _icmph;
1704         const struct icmphdr *icmph;
1705         struct iphdr _inner_iph;
1706         const struct iphdr *inner_iph;
1707
1708         if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1709                 goto standard_hash;
1710
1711         icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1712                                    &_icmph);
1713         if (!icmph)
1714                 goto standard_hash;
1715
1716         if (icmph->type != ICMP_DEST_UNREACH &&
1717             icmph->type != ICMP_REDIRECT &&
1718             icmph->type != ICMP_TIME_EXCEEDED &&
1719             icmph->type != ICMP_PARAMETERPROB) {
1720                 goto standard_hash;
1721         }
1722
1723         inner_iph = skb_header_pointer(skb,
1724                                        outer_iph->ihl * 4 + sizeof(_icmph),
1725                                        sizeof(_inner_iph), &_inner_iph);
1726         if (!inner_iph)
1727                 goto standard_hash;
1728
1729         return fib_multipath_hash(inner_iph->daddr, inner_iph->saddr);
1730
1731 standard_hash:
1732         return fib_multipath_hash(outer_iph->saddr, outer_iph->daddr);
1733 }
1734
1735 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1736
1737 static int ip_mkroute_input(struct sk_buff *skb,
1738                             struct fib_result *res,
1739                             const struct flowi4 *fl4,
1740                             struct in_device *in_dev,
1741                             __be32 daddr, __be32 saddr, u32 tos)
1742 {
1743 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1744         if (res->fi && res->fi->fib_nhs > 1) {
1745                 int h;
1746
1747                 if (unlikely(ip_hdr(skb)->protocol == IPPROTO_ICMP))
1748                         h = ip_multipath_icmp_hash(skb);
1749                 else
1750                         h = fib_multipath_hash(saddr, daddr);
1751                 fib_select_multipath(res, h);
1752         }
1753 #endif
1754
1755         /* create a routing cache entry */
1756         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1757 }
1758
1759 /*
1760  *      NOTE. We drop all the packets that has local source
1761  *      addresses, because every properly looped back packet
1762  *      must have correct destination already attached by output routine.
1763  *
1764  *      Such approach solves two big problems:
1765  *      1. Not simplex devices are handled properly.
1766  *      2. IP spoofing attempts are filtered with 100% of guarantee.
1767  *      called with rcu_read_lock()
1768  */
1769
1770 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1771                                u8 tos, struct net_device *dev)
1772 {
1773         struct fib_result res;
1774         struct in_device *in_dev = __in_dev_get_rcu(dev);
1775         struct ip_tunnel_info *tun_info;
1776         struct flowi4   fl4;
1777         unsigned int    flags = 0;
1778         u32             itag = 0;
1779         struct rtable   *rth;
1780         int             err = -EINVAL;
1781         struct net    *net = dev_net(dev);
1782         bool do_cache;
1783
1784         /* IP on this device is disabled. */
1785
1786         if (!in_dev)
1787                 goto out;
1788
1789         /* Check for the most weird martians, which can be not detected
1790            by fib_lookup.
1791          */
1792
1793         tun_info = skb_tunnel_info(skb);
1794         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1795                 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1796         else
1797                 fl4.flowi4_tun_key.tun_id = 0;
1798         skb_dst_drop(skb);
1799
1800         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1801                 goto martian_source;
1802
1803         res.fi = NULL;
1804         res.table = NULL;
1805         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1806                 goto brd_input;
1807
1808         /* Accept zero addresses only to limited broadcast;
1809          * I even do not know to fix it or not. Waiting for complains :-)
1810          */
1811         if (ipv4_is_zeronet(saddr))
1812                 goto martian_source;
1813
1814         if (ipv4_is_zeronet(daddr))
1815                 goto martian_destination;
1816
1817         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1818          * and call it once if daddr or/and saddr are loopback addresses
1819          */
1820         if (ipv4_is_loopback(daddr)) {
1821                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1822                         goto martian_destination;
1823         } else if (ipv4_is_loopback(saddr)) {
1824                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1825                         goto martian_source;
1826         }
1827
1828         /*
1829          *      Now we are ready to route packet.
1830          */
1831         fl4.flowi4_oif = 0;
1832         fl4.flowi4_iif = l3mdev_fib_oif_rcu(dev);
1833         fl4.flowi4_mark = skb->mark;
1834         fl4.flowi4_tos = tos;
1835         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1836         fl4.flowi4_flags = 0;
1837         fl4.daddr = daddr;
1838         fl4.saddr = saddr;
1839         err = fib_lookup(net, &fl4, &res, 0);
1840         if (err != 0) {
1841                 if (!IN_DEV_FORWARD(in_dev))
1842                         err = -EHOSTUNREACH;
1843                 goto no_route;
1844         }
1845
1846         if (res.type == RTN_BROADCAST)
1847                 goto brd_input;
1848
1849         if (res.type == RTN_LOCAL) {
1850                 err = fib_validate_source(skb, saddr, daddr, tos,
1851                                           0, dev, in_dev, &itag);
1852                 if (err < 0)
1853                         goto martian_source;
1854                 goto local_input;
1855         }
1856
1857         if (!IN_DEV_FORWARD(in_dev)) {
1858                 err = -EHOSTUNREACH;
1859                 goto no_route;
1860         }
1861         if (res.type != RTN_UNICAST)
1862                 goto martian_destination;
1863
1864         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1865 out:    return err;
1866
1867 brd_input:
1868         if (skb->protocol != htons(ETH_P_IP))
1869                 goto e_inval;
1870
1871         if (!ipv4_is_zeronet(saddr)) {
1872                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1873                                           in_dev, &itag);
1874                 if (err < 0)
1875                         goto martian_source;
1876         }
1877         flags |= RTCF_BROADCAST;
1878         res.type = RTN_BROADCAST;
1879         RT_CACHE_STAT_INC(in_brd);
1880
1881 local_input:
1882         do_cache = false;
1883         if (res.fi) {
1884                 if (!itag) {
1885                         rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1886                         if (rt_cache_valid(rth)) {
1887                                 skb_dst_set_noref(skb, &rth->dst);
1888                                 err = 0;
1889                                 goto out;
1890                         }
1891                         do_cache = true;
1892                 }
1893         }
1894
1895         rth = rt_dst_alloc(net->loopback_dev, flags | RTCF_LOCAL, res.type,
1896                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1897         if (!rth)
1898                 goto e_nobufs;
1899
1900         rth->dst.output= ip_rt_bug;
1901 #ifdef CONFIG_IP_ROUTE_CLASSID
1902         rth->dst.tclassid = itag;
1903 #endif
1904         rth->rt_is_input = 1;
1905         if (res.table)
1906                 rth->rt_table_id = res.table->tb_id;
1907
1908         RT_CACHE_STAT_INC(in_slow_tot);
1909         if (res.type == RTN_UNREACHABLE) {
1910                 rth->dst.input= ip_error;
1911                 rth->dst.error= -err;
1912                 rth->rt_flags   &= ~RTCF_LOCAL;
1913         }
1914         if (do_cache) {
1915                 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1916                         rth->dst.flags |= DST_NOCACHE;
1917                         rt_add_uncached_list(rth);
1918                 }
1919         }
1920         skb_dst_set(skb, &rth->dst);
1921         err = 0;
1922         goto out;
1923
1924 no_route:
1925         RT_CACHE_STAT_INC(in_no_route);
1926         res.type = RTN_UNREACHABLE;
1927         res.fi = NULL;
1928         res.table = NULL;
1929         goto local_input;
1930
1931         /*
1932          *      Do not cache martian addresses: they should be logged (RFC1812)
1933          */
1934 martian_destination:
1935         RT_CACHE_STAT_INC(in_martian_dst);
1936 #ifdef CONFIG_IP_ROUTE_VERBOSE
1937         if (IN_DEV_LOG_MARTIANS(in_dev))
1938                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1939                                      &daddr, &saddr, dev->name);
1940 #endif
1941
1942 e_inval:
1943         err = -EINVAL;
1944         goto out;
1945
1946 e_nobufs:
1947         err = -ENOBUFS;
1948         goto out;
1949
1950 martian_source:
1951         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1952         goto out;
1953 }
1954
1955 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1956                          u8 tos, struct net_device *dev)
1957 {
1958         int res;
1959
1960         rcu_read_lock();
1961
1962         /* Multicast recognition logic is moved from route cache to here.
1963            The problem was that too many Ethernet cards have broken/missing
1964            hardware multicast filters :-( As result the host on multicasting
1965            network acquires a lot of useless route cache entries, sort of
1966            SDR messages from all the world. Now we try to get rid of them.
1967            Really, provided software IP multicast filter is organized
1968            reasonably (at least, hashed), it does not result in a slowdown
1969            comparing with route cache reject entries.
1970            Note, that multicast routers are not affected, because
1971            route cache entry is created eventually.
1972          */
1973         if (ipv4_is_multicast(daddr)) {
1974                 struct in_device *in_dev = __in_dev_get_rcu(dev);
1975
1976                 if (in_dev) {
1977                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1978                                                   ip_hdr(skb)->protocol);
1979                         if (our
1980 #ifdef CONFIG_IP_MROUTE
1981                                 ||
1982                             (!ipv4_is_local_multicast(daddr) &&
1983                              IN_DEV_MFORWARD(in_dev))
1984 #endif
1985                            ) {
1986                                 int res = ip_route_input_mc(skb, daddr, saddr,
1987                                                             tos, dev, our);
1988                                 rcu_read_unlock();
1989                                 return res;
1990                         }
1991                 }
1992                 rcu_read_unlock();
1993                 return -EINVAL;
1994         }
1995         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1996         rcu_read_unlock();
1997         return res;
1998 }
1999 EXPORT_SYMBOL(ip_route_input_noref);
2000
2001 /* called with rcu_read_lock() */
2002 static struct rtable *__mkroute_output(const struct fib_result *res,
2003                                        const struct flowi4 *fl4, int orig_oif,
2004                                        struct net_device *dev_out,
2005                                        unsigned int flags)
2006 {
2007         struct fib_info *fi = res->fi;
2008         struct fib_nh_exception *fnhe;
2009         struct in_device *in_dev;
2010         u16 type = res->type;
2011         struct rtable *rth;
2012         bool do_cache;
2013
2014         in_dev = __in_dev_get_rcu(dev_out);
2015         if (!in_dev)
2016                 return ERR_PTR(-EINVAL);
2017
2018         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2019                 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2020                         return ERR_PTR(-EINVAL);
2021
2022         if (ipv4_is_lbcast(fl4->daddr))
2023                 type = RTN_BROADCAST;
2024         else if (ipv4_is_multicast(fl4->daddr))
2025                 type = RTN_MULTICAST;
2026         else if (ipv4_is_zeronet(fl4->daddr))
2027                 return ERR_PTR(-EINVAL);
2028
2029         if (dev_out->flags & IFF_LOOPBACK)
2030                 flags |= RTCF_LOCAL;
2031
2032         do_cache = true;
2033         if (type == RTN_BROADCAST) {
2034                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2035                 fi = NULL;
2036         } else if (type == RTN_MULTICAST) {
2037                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2038                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2039                                      fl4->flowi4_proto))
2040                         flags &= ~RTCF_LOCAL;
2041                 else
2042                         do_cache = false;
2043                 /* If multicast route do not exist use
2044                  * default one, but do not gateway in this case.
2045                  * Yes, it is hack.
2046                  */
2047                 if (fi && res->prefixlen < 4)
2048                         fi = NULL;
2049         } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2050                    (orig_oif != dev_out->ifindex)) {
2051                 /* For local routes that require a particular output interface
2052                  * we do not want to cache the result.  Caching the result
2053                  * causes incorrect behaviour when there are multiple source
2054                  * addresses on the interface, the end result being that if the
2055                  * intended recipient is waiting on that interface for the
2056                  * packet he won't receive it because it will be delivered on
2057                  * the loopback interface and the IP_PKTINFO ipi_ifindex will
2058                  * be set to the loopback interface as well.
2059                  */
2060                 fi = NULL;
2061         }
2062
2063         fnhe = NULL;
2064         do_cache &= fi != NULL;
2065         if (do_cache) {
2066                 struct rtable __rcu **prth;
2067                 struct fib_nh *nh = &FIB_RES_NH(*res);
2068
2069                 fnhe = find_exception(nh, fl4->daddr);
2070                 if (fnhe) {
2071                         prth = &fnhe->fnhe_rth_output;
2072                         rth = rcu_dereference(*prth);
2073                         if (rth && rth->dst.expires &&
2074                             time_after(jiffies, rth->dst.expires)) {
2075                                 ip_del_fnhe(nh, fl4->daddr);
2076                                 fnhe = NULL;
2077                         } else {
2078                                 goto rt_cache;
2079                         }
2080                 }
2081
2082                 if (unlikely(fl4->flowi4_flags &
2083                              FLOWI_FLAG_KNOWN_NH &&
2084                              !(nh->nh_gw &&
2085                                nh->nh_scope == RT_SCOPE_LINK))) {
2086                         do_cache = false;
2087                         goto add;
2088                 }
2089                 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2090                 rth = rcu_dereference(*prth);
2091
2092 rt_cache:
2093                 if (rt_cache_valid(rth)) {
2094                         dst_hold(&rth->dst);
2095                         return rth;
2096                 }
2097         }
2098
2099 add:
2100         rth = rt_dst_alloc(dev_out, flags, type,
2101                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
2102                            IN_DEV_CONF_GET(in_dev, NOXFRM),
2103                            do_cache);
2104         if (!rth)
2105                 return ERR_PTR(-ENOBUFS);
2106
2107         rth->rt_iif     = orig_oif ? : 0;
2108         if (res->table)
2109                 rth->rt_table_id = res->table->tb_id;
2110
2111         RT_CACHE_STAT_INC(out_slow_tot);
2112
2113         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2114                 if (flags & RTCF_LOCAL &&
2115                     !(dev_out->flags & IFF_LOOPBACK)) {
2116                         rth->dst.output = ip_mc_output;
2117                         RT_CACHE_STAT_INC(out_slow_mc);
2118                 }
2119 #ifdef CONFIG_IP_MROUTE
2120                 if (type == RTN_MULTICAST) {
2121                         if (IN_DEV_MFORWARD(in_dev) &&
2122                             !ipv4_is_local_multicast(fl4->daddr)) {
2123                                 rth->dst.input = ip_mr_input;
2124                                 rth->dst.output = ip_mc_output;
2125                         }
2126                 }
2127 #endif
2128         }
2129
2130         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2131         if (lwtunnel_output_redirect(rth->dst.lwtstate))
2132                 rth->dst.output = lwtunnel_output;
2133
2134         return rth;
2135 }
2136
2137 /*
2138  * Major route resolver routine.
2139  */
2140
2141 struct rtable *__ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2142                                           int mp_hash)
2143 {
2144         struct net_device *dev_out = NULL;
2145         __u8 tos = RT_FL_TOS(fl4);
2146         unsigned int flags = 0;
2147         struct fib_result res;
2148         struct rtable *rth;
2149         int master_idx;
2150         int orig_oif;
2151         int err = -ENETUNREACH;
2152
2153         res.tclassid    = 0;
2154         res.fi          = NULL;
2155         res.table       = NULL;
2156
2157         orig_oif = fl4->flowi4_oif;
2158
2159         master_idx = l3mdev_master_ifindex_by_index(net, fl4->flowi4_oif);
2160         if (master_idx)
2161                 fl4->flowi4_oif = master_idx;
2162         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2163         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2164         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2165                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2166
2167         rcu_read_lock();
2168         if (fl4->saddr) {
2169                 rth = ERR_PTR(-EINVAL);
2170                 if (ipv4_is_multicast(fl4->saddr) ||
2171                     ipv4_is_lbcast(fl4->saddr) ||
2172                     ipv4_is_zeronet(fl4->saddr))
2173                         goto out;
2174
2175                 /* I removed check for oif == dev_out->oif here.
2176                    It was wrong for two reasons:
2177                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2178                       is assigned to multiple interfaces.
2179                    2. Moreover, we are allowed to send packets with saddr
2180                       of another iface. --ANK
2181                  */
2182
2183                 if (fl4->flowi4_oif == 0 &&
2184                     (ipv4_is_multicast(fl4->daddr) ||
2185                      ipv4_is_lbcast(fl4->daddr))) {
2186                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2187                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2188                         if (!dev_out)
2189                                 goto out;
2190
2191                         /* Special hack: user can direct multicasts
2192                            and limited broadcast via necessary interface
2193                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2194                            This hack is not just for fun, it allows
2195                            vic,vat and friends to work.
2196                            They bind socket to loopback, set ttl to zero
2197                            and expect that it will work.
2198                            From the viewpoint of routing cache they are broken,
2199                            because we are not allowed to build multicast path
2200                            with loopback source addr (look, routing cache
2201                            cannot know, that ttl is zero, so that packet
2202                            will not leave this host and route is valid).
2203                            Luckily, this hack is good workaround.
2204                          */
2205
2206                         fl4->flowi4_oif = dev_out->ifindex;
2207                         goto make_route;
2208                 }
2209
2210                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2211                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2212                         if (!__ip_dev_find(net, fl4->saddr, false))
2213                                 goto out;
2214                 }
2215         }
2216
2217
2218         if (fl4->flowi4_oif) {
2219                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2220                 rth = ERR_PTR(-ENODEV);
2221                 if (!dev_out)
2222                         goto out;
2223
2224                 /* RACE: Check return value of inet_select_addr instead. */
2225                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2226                         rth = ERR_PTR(-ENETUNREACH);
2227                         goto out;
2228                 }
2229                 if (ipv4_is_local_multicast(fl4->daddr) ||
2230                     ipv4_is_lbcast(fl4->daddr) ||
2231                     fl4->flowi4_proto == IPPROTO_IGMP) {
2232                         if (!fl4->saddr)
2233                                 fl4->saddr = inet_select_addr(dev_out, 0,
2234                                                               RT_SCOPE_LINK);
2235                         goto make_route;
2236                 }
2237                 if (!fl4->saddr) {
2238                         if (ipv4_is_multicast(fl4->daddr))
2239                                 fl4->saddr = inet_select_addr(dev_out, 0,
2240                                                               fl4->flowi4_scope);
2241                         else if (!fl4->daddr)
2242                                 fl4->saddr = inet_select_addr(dev_out, 0,
2243                                                               RT_SCOPE_HOST);
2244                 }
2245
2246                 rth = l3mdev_get_rtable(dev_out, fl4);
2247                 if (rth)
2248                         goto out;
2249         }
2250
2251         if (!fl4->daddr) {
2252                 fl4->daddr = fl4->saddr;
2253                 if (!fl4->daddr)
2254                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2255                 dev_out = net->loopback_dev;
2256                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2257                 res.type = RTN_LOCAL;
2258                 flags |= RTCF_LOCAL;
2259                 goto make_route;
2260         }
2261
2262         err = fib_lookup(net, fl4, &res, 0);
2263         if (err) {
2264                 res.fi = NULL;
2265                 res.table = NULL;
2266                 if (fl4->flowi4_oif &&
2267                     !netif_index_is_l3_master(net, fl4->flowi4_oif)) {
2268                         /* Apparently, routing tables are wrong. Assume,
2269                            that the destination is on link.
2270
2271                            WHY? DW.
2272                            Because we are allowed to send to iface
2273                            even if it has NO routes and NO assigned
2274                            addresses. When oif is specified, routing
2275                            tables are looked up with only one purpose:
2276                            to catch if destination is gatewayed, rather than
2277                            direct. Moreover, if MSG_DONTROUTE is set,
2278                            we send packet, ignoring both routing tables
2279                            and ifaddr state. --ANK
2280
2281
2282                            We could make it even if oif is unknown,
2283                            likely IPv6, but we do not.
2284                          */
2285
2286                         if (fl4->saddr == 0)
2287                                 fl4->saddr = inet_select_addr(dev_out, 0,
2288                                                               RT_SCOPE_LINK);
2289                         res.type = RTN_UNICAST;
2290                         goto make_route;
2291                 }
2292                 rth = ERR_PTR(err);
2293                 goto out;
2294         }
2295
2296         if (res.type == RTN_LOCAL) {
2297                 if (!fl4->saddr) {
2298                         if (res.fi->fib_prefsrc)
2299                                 fl4->saddr = res.fi->fib_prefsrc;
2300                         else
2301                                 fl4->saddr = fl4->daddr;
2302                 }
2303                 dev_out = net->loopback_dev;
2304                 fl4->flowi4_oif = dev_out->ifindex;
2305                 flags |= RTCF_LOCAL;
2306                 goto make_route;
2307         }
2308
2309         fib_select_path(net, &res, fl4, mp_hash);
2310
2311         dev_out = FIB_RES_DEV(res);
2312         fl4->flowi4_oif = dev_out->ifindex;
2313
2314
2315 make_route:
2316         rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2317
2318 out:
2319         rcu_read_unlock();
2320         return rth;
2321 }
2322 EXPORT_SYMBOL_GPL(__ip_route_output_key_hash);
2323
2324 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2325 {
2326         return NULL;
2327 }
2328
2329 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2330 {
2331         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2332
2333         return mtu ? : dst->dev->mtu;
2334 }
2335
2336 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2337                                           struct sk_buff *skb, u32 mtu)
2338 {
2339 }
2340
2341 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2342                                        struct sk_buff *skb)
2343 {
2344 }
2345
2346 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2347                                           unsigned long old)
2348 {
2349         return NULL;
2350 }
2351
2352 static struct dst_ops ipv4_dst_blackhole_ops = {
2353         .family                 =       AF_INET,
2354         .check                  =       ipv4_blackhole_dst_check,
2355         .mtu                    =       ipv4_blackhole_mtu,
2356         .default_advmss         =       ipv4_default_advmss,
2357         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2358         .redirect               =       ipv4_rt_blackhole_redirect,
2359         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
2360         .neigh_lookup           =       ipv4_neigh_lookup,
2361 };
2362
2363 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2364 {
2365         struct rtable *ort = (struct rtable *) dst_orig;
2366         struct rtable *rt;
2367
2368         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2369         if (rt) {
2370                 struct dst_entry *new = &rt->dst;
2371
2372                 new->__use = 1;
2373                 new->input = dst_discard;
2374                 new->output = dst_discard_out;
2375
2376                 new->dev = ort->dst.dev;
2377                 if (new->dev)
2378                         dev_hold(new->dev);
2379
2380                 rt->rt_is_input = ort->rt_is_input;
2381                 rt->rt_iif = ort->rt_iif;
2382                 rt->rt_pmtu = ort->rt_pmtu;
2383
2384                 rt->rt_genid = rt_genid_ipv4(net);
2385                 rt->rt_flags = ort->rt_flags;
2386                 rt->rt_type = ort->rt_type;
2387                 rt->rt_gateway = ort->rt_gateway;
2388                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2389
2390                 INIT_LIST_HEAD(&rt->rt_uncached);
2391                 dst_free(new);
2392         }
2393
2394         dst_release(dst_orig);
2395
2396         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2397 }
2398
2399 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2400                                     const struct sock *sk)
2401 {
2402         struct rtable *rt = __ip_route_output_key(net, flp4);
2403
2404         if (IS_ERR(rt))
2405                 return rt;
2406
2407         if (flp4->flowi4_proto)
2408                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2409                                                         flowi4_to_flowi(flp4),
2410                                                         sk, 0);
2411
2412         return rt;
2413 }
2414 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2415
2416 static int rt_fill_info(struct net *net,  __be32 dst, __be32 src, u32 table_id,
2417                         struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2418                         u32 seq, int event, int nowait, unsigned int flags)
2419 {
2420         struct rtable *rt = skb_rtable(skb);
2421         struct rtmsg *r;
2422         struct nlmsghdr *nlh;
2423         unsigned long expires = 0;
2424         u32 error;
2425         u32 metrics[RTAX_MAX];
2426
2427         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2428         if (!nlh)
2429                 return -EMSGSIZE;
2430
2431         r = nlmsg_data(nlh);
2432         r->rtm_family    = AF_INET;
2433         r->rtm_dst_len  = 32;
2434         r->rtm_src_len  = 0;
2435         r->rtm_tos      = fl4->flowi4_tos;
2436         r->rtm_table    = table_id;
2437         if (nla_put_u32(skb, RTA_TABLE, table_id))
2438                 goto nla_put_failure;
2439         r->rtm_type     = rt->rt_type;
2440         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2441         r->rtm_protocol = RTPROT_UNSPEC;
2442         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2443         if (rt->rt_flags & RTCF_NOTIFY)
2444                 r->rtm_flags |= RTM_F_NOTIFY;
2445         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2446                 r->rtm_flags |= RTCF_DOREDIRECT;
2447
2448         if (nla_put_in_addr(skb, RTA_DST, dst))
2449                 goto nla_put_failure;
2450         if (src) {
2451                 r->rtm_src_len = 32;
2452                 if (nla_put_in_addr(skb, RTA_SRC, src))
2453                         goto nla_put_failure;
2454         }
2455         if (rt->dst.dev &&
2456             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2457                 goto nla_put_failure;
2458 #ifdef CONFIG_IP_ROUTE_CLASSID
2459         if (rt->dst.tclassid &&
2460             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2461                 goto nla_put_failure;
2462 #endif
2463         if (!rt_is_input_route(rt) &&
2464             fl4->saddr != src) {
2465                 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2466                         goto nla_put_failure;
2467         }
2468         if (rt->rt_uses_gateway &&
2469             nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2470                 goto nla_put_failure;
2471
2472         expires = rt->dst.expires;
2473         if (expires) {
2474                 unsigned long now = jiffies;
2475
2476                 if (time_before(now, expires))
2477                         expires -= now;
2478                 else
2479                         expires = 0;
2480         }
2481
2482         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2483         if (rt->rt_pmtu && expires)
2484                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2485         if (rtnetlink_put_metrics(skb, metrics) < 0)
2486                 goto nla_put_failure;
2487
2488         if (fl4->flowi4_mark &&
2489             nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2490                 goto nla_put_failure;
2491
2492         error = rt->dst.error;
2493
2494         if (rt_is_input_route(rt)) {
2495 #ifdef CONFIG_IP_MROUTE
2496                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2497                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2498                         int err = ipmr_get_route(net, skb,
2499                                                  fl4->saddr, fl4->daddr,
2500                                                  r, nowait);
2501                         if (err <= 0) {
2502                                 if (!nowait) {
2503                                         if (err == 0)
2504                                                 return 0;
2505                                         goto nla_put_failure;
2506                                 } else {
2507                                         if (err == -EMSGSIZE)
2508                                                 goto nla_put_failure;
2509                                         error = err;
2510                                 }
2511                         }
2512                 } else
2513 #endif
2514                         if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2515                                 goto nla_put_failure;
2516         }
2517
2518         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2519                 goto nla_put_failure;
2520
2521         nlmsg_end(skb, nlh);
2522         return 0;
2523
2524 nla_put_failure:
2525         nlmsg_cancel(skb, nlh);
2526         return -EMSGSIZE;
2527 }
2528
2529 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2530 {
2531         struct net *net = sock_net(in_skb->sk);
2532         struct rtmsg *rtm;
2533         struct nlattr *tb[RTA_MAX+1];
2534         struct rtable *rt = NULL;
2535         struct flowi4 fl4;
2536         __be32 dst = 0;
2537         __be32 src = 0;
2538         u32 iif;
2539         int err;
2540         int mark;
2541         struct sk_buff *skb;
2542         u32 table_id = RT_TABLE_MAIN;
2543
2544         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2545         if (err < 0)
2546                 goto errout;
2547
2548         rtm = nlmsg_data(nlh);
2549
2550         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2551         if (!skb) {
2552                 err = -ENOBUFS;
2553                 goto errout;
2554         }
2555
2556         /* Reserve room for dummy headers, this skb can pass
2557            through good chunk of routing engine.
2558          */
2559         skb_reset_mac_header(skb);
2560         skb_reset_network_header(skb);
2561
2562         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2563         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2564         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2565
2566         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2567         dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2568         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2569         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2570
2571         memset(&fl4, 0, sizeof(fl4));
2572         fl4.daddr = dst;
2573         fl4.saddr = src;
2574         fl4.flowi4_tos = rtm->rtm_tos;
2575         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2576         fl4.flowi4_mark = mark;
2577
2578         if (netif_index_is_l3_master(net, fl4.flowi4_oif))
2579                 fl4.flowi4_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF;
2580
2581         if (iif) {
2582                 struct net_device *dev;
2583
2584                 dev = __dev_get_by_index(net, iif);
2585                 if (!dev) {
2586                         err = -ENODEV;
2587                         goto errout_free;
2588                 }
2589
2590                 skb->protocol   = htons(ETH_P_IP);
2591                 skb->dev        = dev;
2592                 skb->mark       = mark;
2593                 local_bh_disable();
2594                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2595                 local_bh_enable();
2596
2597                 rt = skb_rtable(skb);
2598                 if (err == 0 && rt->dst.error)
2599                         err = -rt->dst.error;
2600         } else {
2601                 rt = ip_route_output_key(net, &fl4);
2602
2603                 err = 0;
2604                 if (IS_ERR(rt))
2605                         err = PTR_ERR(rt);
2606         }
2607
2608         if (err)
2609                 goto errout_free;
2610
2611         skb_dst_set(skb, &rt->dst);
2612         if (rtm->rtm_flags & RTM_F_NOTIFY)
2613                 rt->rt_flags |= RTCF_NOTIFY;
2614
2615         if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2616                 table_id = rt->rt_table_id;
2617
2618         err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2619                            NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2620                            RTM_NEWROUTE, 0, 0);
2621         if (err < 0)
2622                 goto errout_free;
2623
2624         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2625 errout:
2626         return err;
2627
2628 errout_free:
2629         kfree_skb(skb);
2630         goto errout;
2631 }
2632
2633 void ip_rt_multicast_event(struct in_device *in_dev)
2634 {
2635         rt_cache_flush(dev_net(in_dev->dev));
2636 }
2637
2638 #ifdef CONFIG_SYSCTL
2639 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
2640 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
2641 static int ip_rt_gc_elasticity __read_mostly    = 8;
2642
2643 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2644                                         void __user *buffer,
2645                                         size_t *lenp, loff_t *ppos)
2646 {
2647         struct net *net = (struct net *)__ctl->extra1;
2648
2649         if (write) {
2650                 rt_cache_flush(net);
2651                 fnhe_genid_bump(net);
2652                 return 0;
2653         }
2654
2655         return -EINVAL;
2656 }
2657
2658 static struct ctl_table ipv4_route_table[] = {
2659         {
2660                 .procname       = "gc_thresh",
2661                 .data           = &ipv4_dst_ops.gc_thresh,
2662                 .maxlen         = sizeof(int),
2663                 .mode           = 0644,
2664                 .proc_handler   = proc_dointvec,
2665         },
2666         {
2667                 .procname       = "max_size",
2668                 .data           = &ip_rt_max_size,
2669                 .maxlen         = sizeof(int),
2670                 .mode           = 0644,
2671                 .proc_handler   = proc_dointvec,
2672         },
2673         {
2674                 /*  Deprecated. Use gc_min_interval_ms */
2675
2676                 .procname       = "gc_min_interval",
2677                 .data           = &ip_rt_gc_min_interval,
2678                 .maxlen         = sizeof(int),
2679                 .mode           = 0644,
2680                 .proc_handler   = proc_dointvec_jiffies,
2681         },
2682         {
2683                 .procname       = "gc_min_interval_ms",
2684                 .data           = &ip_rt_gc_min_interval,
2685                 .maxlen         = sizeof(int),
2686                 .mode           = 0644,
2687                 .proc_handler   = proc_dointvec_ms_jiffies,
2688         },
2689         {
2690                 .procname       = "gc_timeout",
2691                 .data           = &ip_rt_gc_timeout,
2692                 .maxlen         = sizeof(int),
2693                 .mode           = 0644,
2694                 .proc_handler   = proc_dointvec_jiffies,
2695         },
2696         {
2697                 .procname       = "gc_interval",
2698                 .data           = &ip_rt_gc_interval,
2699                 .maxlen         = sizeof(int),
2700                 .mode           = 0644,
2701                 .proc_handler   = proc_dointvec_jiffies,
2702         },
2703         {
2704                 .procname       = "redirect_load",
2705                 .data           = &ip_rt_redirect_load,
2706                 .maxlen         = sizeof(int),
2707                 .mode           = 0644,
2708                 .proc_handler   = proc_dointvec,
2709         },
2710         {
2711                 .procname       = "redirect_number",
2712                 .data           = &ip_rt_redirect_number,
2713                 .maxlen         = sizeof(int),
2714                 .mode           = 0644,
2715                 .proc_handler   = proc_dointvec,
2716         },
2717         {
2718                 .procname       = "redirect_silence",
2719                 .data           = &ip_rt_redirect_silence,
2720                 .maxlen         = sizeof(int),
2721                 .mode           = 0644,
2722                 .proc_handler   = proc_dointvec,
2723         },
2724         {
2725                 .procname       = "error_cost",
2726                 .data           = &ip_rt_error_cost,
2727                 .maxlen         = sizeof(int),
2728                 .mode           = 0644,
2729                 .proc_handler   = proc_dointvec,
2730         },
2731         {
2732                 .procname       = "error_burst",
2733                 .data           = &ip_rt_error_burst,
2734                 .maxlen         = sizeof(int),
2735                 .mode           = 0644,
2736                 .proc_handler   = proc_dointvec,
2737         },
2738         {
2739                 .procname       = "gc_elasticity",
2740                 .data           = &ip_rt_gc_elasticity,
2741                 .maxlen         = sizeof(int),
2742                 .mode           = 0644,
2743                 .proc_handler   = proc_dointvec,
2744         },
2745         {
2746                 .procname       = "mtu_expires",
2747                 .data           = &ip_rt_mtu_expires,
2748                 .maxlen         = sizeof(int),
2749                 .mode           = 0644,
2750                 .proc_handler   = proc_dointvec_jiffies,
2751         },
2752         {
2753                 .procname       = "min_pmtu",
2754                 .data           = &ip_rt_min_pmtu,
2755                 .maxlen         = sizeof(int),
2756                 .mode           = 0644,
2757                 .proc_handler   = proc_dointvec,
2758         },
2759         {
2760                 .procname       = "min_adv_mss",
2761                 .data           = &ip_rt_min_advmss,
2762                 .maxlen         = sizeof(int),
2763                 .mode           = 0644,
2764                 .proc_handler   = proc_dointvec,
2765         },
2766         { }
2767 };
2768
2769 static struct ctl_table ipv4_route_flush_table[] = {
2770         {
2771                 .procname       = "flush",
2772                 .maxlen         = sizeof(int),
2773                 .mode           = 0200,
2774                 .proc_handler   = ipv4_sysctl_rtcache_flush,
2775         },
2776         { },
2777 };
2778
2779 static __net_init int sysctl_route_net_init(struct net *net)
2780 {
2781         struct ctl_table *tbl;
2782
2783         tbl = ipv4_route_flush_table;
2784         if (!net_eq(net, &init_net)) {
2785                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2786                 if (!tbl)
2787                         goto err_dup;
2788
2789                 /* Don't export sysctls to unprivileged users */
2790                 if (net->user_ns != &init_user_ns)
2791                         tbl[0].procname = NULL;
2792         }
2793         tbl[0].extra1 = net;
2794
2795         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2796         if (!net->ipv4.route_hdr)
2797                 goto err_reg;
2798         return 0;
2799
2800 err_reg:
2801         if (tbl != ipv4_route_flush_table)
2802                 kfree(tbl);
2803 err_dup:
2804         return -ENOMEM;
2805 }
2806
2807 static __net_exit void sysctl_route_net_exit(struct net *net)
2808 {
2809         struct ctl_table *tbl;
2810
2811         tbl = net->ipv4.route_hdr->ctl_table_arg;
2812         unregister_net_sysctl_table(net->ipv4.route_hdr);
2813         BUG_ON(tbl == ipv4_route_flush_table);
2814         kfree(tbl);
2815 }
2816
2817 static __net_initdata struct pernet_operations sysctl_route_ops = {
2818         .init = sysctl_route_net_init,
2819         .exit = sysctl_route_net_exit,
2820 };
2821 #endif
2822
2823 static __net_init int rt_genid_init(struct net *net)
2824 {
2825         atomic_set(&net->ipv4.rt_genid, 0);
2826         atomic_set(&net->fnhe_genid, 0);
2827         get_random_bytes(&net->ipv4.dev_addr_genid,
2828                          sizeof(net->ipv4.dev_addr_genid));
2829         return 0;
2830 }
2831
2832 static __net_initdata struct pernet_operations rt_genid_ops = {
2833         .init = rt_genid_init,
2834 };
2835
2836 static int __net_init ipv4_inetpeer_init(struct net *net)
2837 {
2838         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2839
2840         if (!bp)
2841                 return -ENOMEM;
2842         inet_peer_base_init(bp);
2843         net->ipv4.peers = bp;
2844         return 0;
2845 }
2846
2847 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2848 {
2849         struct inet_peer_base *bp = net->ipv4.peers;
2850
2851         net->ipv4.peers = NULL;
2852         inetpeer_invalidate_tree(bp);
2853         kfree(bp);
2854 }
2855
2856 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2857         .init   =       ipv4_inetpeer_init,
2858         .exit   =       ipv4_inetpeer_exit,
2859 };
2860
2861 #ifdef CONFIG_IP_ROUTE_CLASSID
2862 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2863 #endif /* CONFIG_IP_ROUTE_CLASSID */
2864
2865 int __init ip_rt_init(void)
2866 {
2867         int rc = 0;
2868         int cpu;
2869
2870         ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2871         if (!ip_idents)
2872                 panic("IP: failed to allocate ip_idents\n");
2873
2874         prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2875
2876         ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
2877         if (!ip_tstamps)
2878                 panic("IP: failed to allocate ip_tstamps\n");
2879
2880         for_each_possible_cpu(cpu) {
2881                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
2882
2883                 INIT_LIST_HEAD(&ul->head);
2884                 spin_lock_init(&ul->lock);
2885         }
2886 #ifdef CONFIG_IP_ROUTE_CLASSID
2887         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2888         if (!ip_rt_acct)
2889                 panic("IP: failed to allocate ip_rt_acct\n");
2890 #endif
2891
2892         ipv4_dst_ops.kmem_cachep =
2893                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2894                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2895
2896         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2897
2898         if (dst_entries_init(&ipv4_dst_ops) < 0)
2899                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2900
2901         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2902                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2903
2904         ipv4_dst_ops.gc_thresh = ~0;
2905         ip_rt_max_size = INT_MAX;
2906
2907         devinet_init();
2908         ip_fib_init();
2909
2910         if (ip_rt_proc_init())
2911                 pr_err("Unable to create route proc files\n");
2912 #ifdef CONFIG_XFRM
2913         xfrm_init();
2914         xfrm4_init();
2915 #endif
2916         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2917
2918 #ifdef CONFIG_SYSCTL
2919         register_pernet_subsys(&sysctl_route_ops);
2920 #endif
2921         register_pernet_subsys(&rt_genid_ops);
2922         register_pernet_subsys(&ipv4_inetpeer_ops);
2923         return rc;
2924 }
2925
2926 #ifdef CONFIG_SYSCTL
2927 /*
2928  * We really need to sanitize the damn ipv4 init order, then all
2929  * this nonsense will go away.
2930  */
2931 void __init ip_static_sysctl_init(void)
2932 {
2933         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
2934 }
2935 #endif