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/net_namespace.h>
  95 #include <net/protocol.h>
  96 #include <net/ip.h>
  97 #include <net/route.h>
  98 #include <net/inetpeer.h>
  99 #include <net/sock.h>
 100 #include <net/ip_fib.h>
 101 #include <net/arp.h>
 102 #include <net/tcp.h>
 103 #include <net/icmp.h>
 104 #include <net/xfrm.h>
 105 #include <net/netevent.h>
 106 #include <net/rtnetlink.h>
 107 #ifdef CONFIG_SYSCTL
 108 #include <linux/sysctl.h>
 109 #include <linux/kmemleak.h>
 110 #endif
 111 #include <net/secure_seq.h>
 112
 113 #define RT_FL_TOS(oldflp4) \
 114         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
 115
 116 #define RT_GC_TIMEOUT (300*HZ)
 117
 118 static int ip_rt_max_size;
 119 static int ip_rt_redirect_number __read_mostly  = 9;
 120 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 121 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 122 static int ip_rt_error_cost __read_mostly       = HZ;
 123 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 124 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
 125 static int ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
 126 static int ip_rt_min_advmss __read_mostly       = 256;
 127
 128 /*
 129  *      Interface to generic destination cache.
 130  */
 131
 132 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 133 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 134 static unsigned int      ipv4_mtu(const struct dst_entry *dst);
 135 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 136 static void              ipv4_link_failure(struct sk_buff *skb);
 137 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 138                                            struct sk_buff *skb, u32 mtu);
 139 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 140                                         struct sk_buff *skb);
 141 static void             ipv4_dst_destroy(struct dst_entry *dst);
 142
 143 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 144 {
 145         WARN_ON(1);
 146         return NULL;
 147 }
 148
 149 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 150                                            struct sk_buff *skb,
 151                                            const void *daddr);
 152
 153 static struct dst_ops ipv4_dst_ops = {
 154         .family =               AF_INET,
 155         .protocol =             cpu_to_be16(ETH_P_IP),
 156         .check =                ipv4_dst_check,
 157         .default_advmss =       ipv4_default_advmss,
 158         .mtu =                  ipv4_mtu,
 159         .cow_metrics =          ipv4_cow_metrics,
 160         .destroy =              ipv4_dst_destroy,
 161         .negative_advice =      ipv4_negative_advice,
 162         .link_failure =         ipv4_link_failure,
 163         .update_pmtu =          ip_rt_update_pmtu,
 164         .redirect =             ip_do_redirect,
 165         .local_out =            __ip_local_out,
 166         .neigh_lookup =         ipv4_neigh_lookup,
 167 };
 168
 169 #define ECN_OR_COST(class)      TC_PRIO_##class
 170
 171 const __u8 ip_tos2prio[16] = {
 172         TC_PRIO_BESTEFFORT,
 173         ECN_OR_COST(BESTEFFORT),
 174         TC_PRIO_BESTEFFORT,
 175         ECN_OR_COST(BESTEFFORT),
 176         TC_PRIO_BULK,
 177         ECN_OR_COST(BULK),
 178         TC_PRIO_BULK,
 179         ECN_OR_COST(BULK),
 180         TC_PRIO_INTERACTIVE,
 181         ECN_OR_COST(INTERACTIVE),
 182         TC_PRIO_INTERACTIVE,
 183         ECN_OR_COST(INTERACTIVE),
 184         TC_PRIO_INTERACTIVE_BULK,
 185         ECN_OR_COST(INTERACTIVE_BULK),
 186         TC_PRIO_INTERACTIVE_BULK,
 187         ECN_OR_COST(INTERACTIVE_BULK)
 188 };
 189 EXPORT_SYMBOL(ip_tos2prio);
 190
 191 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 192 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
 193
 194 #ifdef CONFIG_PROC_FS
 195 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 196 {
 197         if (*pos)
 198                 return NULL;
 199         return SEQ_START_TOKEN;
 200 }
 201
 202 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 203 {
 204         ++*pos;
 205         return NULL;
 206 }
 207
 208 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 209 {
 210 }
 211
 212 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 213 {
 214         if (v == SEQ_START_TOKEN)
 215                 seq_printf(seq, "%-127s\n",
 216                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 217                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 218                            "HHUptod\tSpecDst");
 219         return 0;
 220 }
 221
 222 static const struct seq_operations rt_cache_seq_ops = {
 223         .start  = rt_cache_seq_start,
 224         .next   = rt_cache_seq_next,
 225         .stop   = rt_cache_seq_stop,
 226         .show   = rt_cache_seq_show,
 227 };
 228
 229 static int rt_cache_seq_open(struct inode *inode, struct file *file)
 230 {
 231         return seq_open(file, &rt_cache_seq_ops);
 232 }
 233
 234 static const struct file_operations rt_cache_seq_fops = {
 235         .owner   = THIS_MODULE,
 236         .open    = rt_cache_seq_open,
 237         .read    = seq_read,
 238         .llseek  = seq_lseek,
 239         .release = seq_release,
 240 };
 241
 242
 243 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 244 {
 245         int cpu;
 246
 247         if (*pos == 0)
 248                 return SEQ_START_TOKEN;
 249
 250         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 251                 if (!cpu_possible(cpu))
 252                         continue;
 253                 *pos = cpu+1;
 254                 return &per_cpu(rt_cache_stat, cpu);
 255         }
 256         return NULL;
 257 }
 258
 259 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 260 {
 261         int cpu;
 262
 263         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 264                 if (!cpu_possible(cpu))
 265                         continue;
 266                 *pos = cpu+1;
 267                 return &per_cpu(rt_cache_stat, cpu);
 268         }
 269         return NULL;
 270
 271 }
 272
 273 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 274 {
 275
 276 }
 277
 278 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 279 {
 280         struct rt_cache_stat *st = v;
 281
 282         if (v == SEQ_START_TOKEN) {
 283                 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");
 284                 return 0;
 285         }
 286
 287         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 288                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 289                    dst_entries_get_slow(&ipv4_dst_ops),
 290                    0, /* st->in_hit */
 291                    st->in_slow_tot,
 292                    st->in_slow_mc,
 293                    st->in_no_route,
 294                    st->in_brd,
 295                    st->in_martian_dst,
 296                    st->in_martian_src,
 297
 298                    0, /* st->out_hit */
 299                    st->out_slow_tot,
 300                    st->out_slow_mc,
 301
 302                    0, /* st->gc_total */
 303                    0, /* st->gc_ignored */
 304                    0, /* st->gc_goal_miss */
 305                    0, /* st->gc_dst_overflow */
 306                    0, /* st->in_hlist_search */
 307                    0  /* st->out_hlist_search */
 308                 );
 309         return 0;
 310 }
 311
 312 static const struct seq_operations rt_cpu_seq_ops = {
 313         .start  = rt_cpu_seq_start,
 314         .next   = rt_cpu_seq_next,
 315         .stop   = rt_cpu_seq_stop,
 316         .show   = rt_cpu_seq_show,
 317 };
 318
 319
 320 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
 321 {
 322         return seq_open(file, &rt_cpu_seq_ops);
 323 }
 324
 325 static const struct file_operations rt_cpu_seq_fops = {
 326         .owner   = THIS_MODULE,
 327         .open    = rt_cpu_seq_open,
 328         .read    = seq_read,
 329         .llseek  = seq_lseek,
 330         .release = seq_release,
 331 };
 332
 333 #ifdef CONFIG_IP_ROUTE_CLASSID
 334 static int rt_acct_proc_show(struct seq_file *m, void *v)
 335 {
 336         struct ip_rt_acct *dst, *src;
 337         unsigned int i, j;
 338
 339         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 340         if (!dst)
 341                 return -ENOMEM;
 342
 343         for_each_possible_cpu(i) {
 344                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 345                 for (j = 0; j < 256; j++) {
 346                         dst[j].o_bytes   += src[j].o_bytes;
 347                         dst[j].o_packets += src[j].o_packets;
 348                         dst[j].i_bytes   += src[j].i_bytes;
 349                         dst[j].i_packets += src[j].i_packets;
 350                 }
 351         }
 352
 353         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 354         kfree(dst);
 355         return 0;
 356 }
 357
 358 static int rt_acct_proc_open(struct inode *inode, struct file *file)
 359 {
 360         return single_open(file, rt_acct_proc_show, NULL);
 361 }
 362
 363 static const struct file_operations rt_acct_proc_fops = {
 364         .owner          = THIS_MODULE,
 365         .open           = rt_acct_proc_open,
 366         .read           = seq_read,
 367         .llseek         = seq_lseek,
 368         .release        = single_release,
 369 };
 370 #endif
 371
 372 static int __net_init ip_rt_do_proc_init(struct net *net)
 373 {
 374         struct proc_dir_entry *pde;
 375
 376         pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
 377                           &rt_cache_seq_fops);
 378         if (!pde)
 379                 goto err1;
 380
 381         pde = proc_create("rt_cache", S_IRUGO,
 382                           net->proc_net_stat, &rt_cpu_seq_fops);
 383         if (!pde)
 384                 goto err2;
 385
 386 #ifdef CONFIG_IP_ROUTE_CLASSID
 387         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
 388         if (!pde)
 389                 goto err3;
 390 #endif
 391         return 0;
 392
 393 #ifdef CONFIG_IP_ROUTE_CLASSID
 394 err3:
 395         remove_proc_entry("rt_cache", net->proc_net_stat);
 396 #endif
 397 err2:
 398         remove_proc_entry("rt_cache", net->proc_net);
 399 err1:
 400         return -ENOMEM;
 401 }
 402
 403 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 404 {
 405         remove_proc_entry("rt_cache", net->proc_net_stat);
 406         remove_proc_entry("rt_cache", net->proc_net);
 407 #ifdef CONFIG_IP_ROUTE_CLASSID
 408         remove_proc_entry("rt_acct", net->proc_net);
 409 #endif
 410 }
 411
 412 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 413         .init = ip_rt_do_proc_init,
 414         .exit = ip_rt_do_proc_exit,
 415 };
 416
 417 static int __init ip_rt_proc_init(void)
 418 {
 419         return register_pernet_subsys(&ip_rt_proc_ops);
 420 }
 421
 422 #else
 423 static inline int ip_rt_proc_init(void)
 424 {
 425         return 0;
 426 }
 427 #endif /* CONFIG_PROC_FS */
 428
 429 static inline bool rt_is_expired(const struct rtable *rth)
 430 {
 431         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 432 }
 433
 434 void rt_cache_flush(struct net *net)
 435 {
 436         rt_genid_bump_ipv4(net);
 437 }
 438
 439 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 440                                            struct sk_buff *skb,
 441                                            const void *daddr)
 442 {
 443         struct net_device *dev = dst->dev;
 444         const __be32 *pkey = daddr;
 445         const struct rtable *rt;
 446         struct neighbour *n;
 447
 448         rt = (const struct rtable *) dst;
 449         if (rt->rt_gateway)
 450                 pkey = (const __be32 *) &rt->rt_gateway;
 451         else if (skb)
 452                 pkey = &ip_hdr(skb)->daddr;
 453
 454         n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
 455         if (n)
 456                 return n;
 457         return neigh_create(&arp_tbl, pkey, dev);
 458 }
 459
 460 #define IP_IDENTS_SZ 2048u
 461 struct ip_ident_bucket {
 462         atomic_t        id;
 463         u32             stamp32;
 464 };
 465
 466 static struct ip_ident_bucket *ip_idents __read_mostly;
 467
 468 /* In order to protect privacy, we add a perturbation to identifiers
 469  * if one generator is seldom used. This makes hard for an attacker
 470  * to infer how many packets were sent between two points in time.
 471  */
 472 u32 ip_idents_reserve(u32 hash, int segs)
 473 {
 474         struct ip_ident_bucket *bucket = ip_idents + hash % IP_IDENTS_SZ;
 475         u32 old = ACCESS_ONCE(bucket->stamp32);
 476         u32 now = (u32)jiffies;
 477         u32 delta = 0;
 478
 479         if (old != now && cmpxchg(&bucket->stamp32, old, now) == old)
 480                 delta = prandom_u32_max(now - old);
 481
 482         return atomic_add_return(segs + delta, &bucket->id) - segs;
 483 }
 484 EXPORT_SYMBOL(ip_idents_reserve);
 485
 486 void __ip_select_ident(struct iphdr *iph, int segs)
 487 {
 488         static u32 ip_idents_hashrnd __read_mostly;
 489         u32 hash, id;
 490
 491         net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
 492
 493         hash = jhash_3words((__force u32)iph->daddr,
 494                             (__force u32)iph->saddr,
 495                             iph->protocol,
 496                             ip_idents_hashrnd);
 497         id = ip_idents_reserve(hash, segs);
 498         iph->id = htons(id);
 499 }
 500 EXPORT_SYMBOL(__ip_select_ident);
 501
 502 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 503                              const struct iphdr *iph,
 504                              int oif, u8 tos,
 505                              u8 prot, u32 mark, int flow_flags)
 506 {
 507         if (sk) {
 508                 const struct inet_sock *inet = inet_sk(sk);
 509
 510                 oif = sk->sk_bound_dev_if;
 511                 mark = sk->sk_mark;
 512                 tos = RT_CONN_FLAGS(sk);
 513                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 514         }
 515         flowi4_init_output(fl4, oif, mark, tos,
 516                            RT_SCOPE_UNIVERSE, prot,
 517                            flow_flags,
 518                            iph->daddr, iph->saddr, 0, 0);
 519 }
 520
 521 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 522                                const struct sock *sk)
 523 {
 524         const struct iphdr *iph = ip_hdr(skb);
 525         int oif = skb->dev->ifindex;
 526         u8 tos = RT_TOS(iph->tos);
 527         u8 prot = iph->protocol;
 528         u32 mark = skb->mark;
 529
 530         __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
 531 }
 532
 533 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 534 {
 535         const struct inet_sock *inet = inet_sk(sk);
 536         const struct ip_options_rcu *inet_opt;
 537         __be32 daddr = inet->inet_daddr;
 538
 539         rcu_read_lock();
 540         inet_opt = rcu_dereference(inet->inet_opt);
 541         if (inet_opt && inet_opt->opt.srr)
 542                 daddr = inet_opt->opt.faddr;
 543         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 544                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 545                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 546                            inet_sk_flowi_flags(sk),
 547                            daddr, inet->inet_saddr, 0, 0);
 548         rcu_read_unlock();
 549 }
 550
 551 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 552                                  const struct sk_buff *skb)
 553 {
 554         if (skb)
 555                 build_skb_flow_key(fl4, skb, sk);
 556         else
 557                 build_sk_flow_key(fl4, sk);
 558 }
 559
 560 static inline void rt_free(struct rtable *rt)
 561 {
 562         call_rcu(&rt->dst.rcu_head, dst_rcu_free);
 563 }
 564
 565 static DEFINE_SPINLOCK(fnhe_lock);
 566
 567 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 568 {
 569         struct rtable *rt;
 570
 571         rt = rcu_dereference(fnhe->fnhe_rth_input);
 572         if (rt) {
 573                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 574                 rt_free(rt);
 575         }
 576         rt = rcu_dereference(fnhe->fnhe_rth_output);
 577         if (rt) {
 578                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 579                 rt_free(rt);
 580         }
 581 }
 582
 583 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
 584 {
 585         struct fib_nh_exception *fnhe, *oldest;
 586
 587         oldest = rcu_dereference(hash->chain);
 588         for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
 589              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 590                 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
 591                         oldest = fnhe;
 592         }
 593         fnhe_flush_routes(oldest);
 594         return oldest;
 595 }
 596
 597 static inline u32 fnhe_hashfun(__be32 daddr)
 598 {
 599         static u32 fnhe_hashrnd __read_mostly;
 600         u32 hval;
 601
 602         net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
 603         hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
 604         return hash_32(hval, FNHE_HASH_SHIFT);
 605 }
 606
 607 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 608 {
 609         rt->rt_pmtu = fnhe->fnhe_pmtu;
 610         rt->dst.expires = fnhe->fnhe_expires;
 611
 612         if (fnhe->fnhe_gw) {
 613                 rt->rt_flags |= RTCF_REDIRECTED;
 614                 rt->rt_gateway = fnhe->fnhe_gw;
 615                 rt->rt_uses_gateway = 1;
 616         }
 617 }
 618
 619 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
 620                                   u32 pmtu, unsigned long expires)
 621 {
 622         struct fnhe_hash_bucket *hash;
 623         struct fib_nh_exception *fnhe;
 624         struct rtable *rt;
 625         unsigned int i;
 626         int depth;
 627         u32 hval = fnhe_hashfun(daddr);
 628
 629         spin_lock_bh(&fnhe_lock);
 630
 631         hash = rcu_dereference(nh->nh_exceptions);
 632         if (!hash) {
 633                 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
 634                 if (!hash)
 635                         goto out_unlock;
 636                 rcu_assign_pointer(nh->nh_exceptions, hash);
 637         }
 638
 639         hash += hval;
 640
 641         depth = 0;
 642         for (fnhe = rcu_dereference(hash->chain); fnhe;
 643              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 644                 if (fnhe->fnhe_daddr == daddr)
 645                         break;
 646                 depth++;
 647         }
 648
 649         if (fnhe) {
 650                 if (gw)
 651                         fnhe->fnhe_gw = gw;
 652                 if (pmtu) {
 653                         fnhe->fnhe_pmtu = pmtu;
 654                         fnhe->fnhe_expires = max(1UL, expires);
 655                 }
 656                 /* Update all cached dsts too */
 657                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 658                 if (rt)
 659                         fill_route_from_fnhe(rt, fnhe);
 660                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 661                 if (rt)
 662                         fill_route_from_fnhe(rt, fnhe);
 663         } else {
 664                 if (depth > FNHE_RECLAIM_DEPTH)
 665                         fnhe = fnhe_oldest(hash);
 666                 else {
 667                         fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 668                         if (!fnhe)
 669                                 goto out_unlock;
 670
 671                         fnhe->fnhe_next = hash->chain;
 672                         rcu_assign_pointer(hash->chain, fnhe);
 673                 }
 674                 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
 675                 fnhe->fnhe_daddr = daddr;
 676                 fnhe->fnhe_gw = gw;
 677                 fnhe->fnhe_pmtu = pmtu;
 678                 fnhe->fnhe_expires = expires;
 679
 680                 /* Exception created; mark the cached routes for the nexthop
 681                  * stale, so anyone caching it rechecks if this exception
 682                  * applies to them.
 683                  */
 684                 rt = rcu_dereference(nh->nh_rth_input);
 685                 if (rt)
 686                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 687
 688                 for_each_possible_cpu(i) {
 689                         struct rtable __rcu **prt;
 690                         prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
 691                         rt = rcu_dereference(*prt);
 692                         if (rt)
 693                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 694                 }
 695         }
 696
 697         fnhe->fnhe_stamp = jiffies;
 698
 699 out_unlock:
 700         spin_unlock_bh(&fnhe_lock);
 701 }
 702
 703 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 704                              bool kill_route)
 705 {
 706         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 707         __be32 old_gw = ip_hdr(skb)->saddr;
 708         struct net_device *dev = skb->dev;
 709         struct in_device *in_dev;
 710         struct fib_result res;
 711         struct neighbour *n;
 712         struct net *net;
 713
 714         switch (icmp_hdr(skb)->code & 7) {
 715         case ICMP_REDIR_NET:
 716         case ICMP_REDIR_NETTOS:
 717         case ICMP_REDIR_HOST:
 718         case ICMP_REDIR_HOSTTOS:
 719                 break;
 720
 721         default:
 722                 return;
 723         }
 724
 725         if (rt->rt_gateway != old_gw)
 726                 return;
 727
 728         in_dev = __in_dev_get_rcu(dev);
 729         if (!in_dev)
 730                 return;
 731
 732         net = dev_net(dev);
 733         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 734             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 735             ipv4_is_zeronet(new_gw))
 736                 goto reject_redirect;
 737
 738         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 739                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 740                         goto reject_redirect;
 741                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 742                         goto reject_redirect;
 743         } else {
 744                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 745                         goto reject_redirect;
 746         }
 747
 748         n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
 749         if (!IS_ERR(n)) {
 750                 if (!(n->nud_state & NUD_VALID)) {
 751                         neigh_event_send(n, NULL);
 752                 } else {
 753                         if (fib_lookup(net, fl4, &res) == 0) {
 754                                 struct fib_nh *nh = &FIB_RES_NH(res);
 755
 756                                 update_or_create_fnhe(nh, fl4->daddr, new_gw,
 757                                                       0, 0);
 758                         }
 759                         if (kill_route)
 760                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 761                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 762                 }
 763                 neigh_release(n);
 764         }
 765         return;
 766
 767 reject_redirect:
 768 #ifdef CONFIG_IP_ROUTE_VERBOSE
 769         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 770                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 771                 __be32 daddr = iph->daddr;
 772                 __be32 saddr = iph->saddr;
 773
 774                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 775                                      "  Advised path = %pI4 -> %pI4\n",
 776                                      &old_gw, dev->name, &new_gw,
 777                                      &saddr, &daddr);
 778         }
 779 #endif
 780         ;
 781 }
 782
 783 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 784 {
 785         struct rtable *rt;
 786         struct flowi4 fl4;
 787         const struct iphdr *iph = (const struct iphdr *) skb->data;
 788         int oif = skb->dev->ifindex;
 789         u8 tos = RT_TOS(iph->tos);
 790         u8 prot = iph->protocol;
 791         u32 mark = skb->mark;
 792
 793         rt = (struct rtable *) dst;
 794
 795         __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
 796         __ip_do_redirect(rt, skb, &fl4, true);
 797 }
 798
 799 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 800 {
 801         struct rtable *rt = (struct rtable *)dst;
 802         struct dst_entry *ret = dst;
 803
 804         if (rt) {
 805                 if (dst->obsolete > 0) {
 806                         ip_rt_put(rt);
 807                         ret = NULL;
 808                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 809                            rt->dst.expires) {
 810                         ip_rt_put(rt);
 811                         ret = NULL;
 812                 }
 813         }
 814         return ret;
 815 }
 816
 817 /*
 818  * Algorithm:
 819  *      1. The first ip_rt_redirect_number redirects are sent
 820  *         with exponential backoff, then we stop sending them at all,
 821  *         assuming that the host ignores our redirects.
 822  *      2. If we did not see packets requiring redirects
 823  *         during ip_rt_redirect_silence, we assume that the host
 824  *         forgot redirected route and start to send redirects again.
 825  *
 826  * This algorithm is much cheaper and more intelligent than dumb load limiting
 827  * in icmp.c.
 828  *
 829  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 830  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 831  */
 832
 833 void ip_rt_send_redirect(struct sk_buff *skb)
 834 {
 835         struct rtable *rt = skb_rtable(skb);
 836         struct in_device *in_dev;
 837         struct inet_peer *peer;
 838         struct net *net;
 839         int log_martians;
 840
 841         rcu_read_lock();
 842         in_dev = __in_dev_get_rcu(rt->dst.dev);
 843         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 844                 rcu_read_unlock();
 845                 return;
 846         }
 847         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 848         rcu_read_unlock();
 849
 850         net = dev_net(rt->dst.dev);
 851         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
 852         if (!peer) {
 853                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 854                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 855                 return;
 856         }
 857
 858         /* No redirected packets during ip_rt_redirect_silence;
 859          * reset the algorithm.
 860          */
 861         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
 862                 peer->rate_tokens = 0;
 863
 864         /* Too many ignored redirects; do not send anything
 865          * set dst.rate_last to the last seen redirected packet.
 866          */
 867         if (peer->rate_tokens >= ip_rt_redirect_number) {
 868                 peer->rate_last = jiffies;
 869                 goto out_put_peer;
 870         }
 871
 872         /* Check for load limit; set rate_last to the latest sent
 873          * redirect.
 874          */
 875         if (peer->rate_tokens == 0 ||
 876             time_after(jiffies,
 877                        (peer->rate_last +
 878                         (ip_rt_redirect_load << peer->rate_tokens)))) {
 879                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 880
 881                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 882                 peer->rate_last = jiffies;
 883                 ++peer->rate_tokens;
 884 #ifdef CONFIG_IP_ROUTE_VERBOSE
 885                 if (log_martians &&
 886                     peer->rate_tokens == ip_rt_redirect_number)
 887                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 888                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 889                                              &ip_hdr(skb)->daddr, &gw);
 890 #endif
 891         }
 892 out_put_peer:
 893         inet_putpeer(peer);
 894 }
 895
 896 static int ip_error(struct sk_buff *skb)
 897 {
 898         struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
 899         struct rtable *rt = skb_rtable(skb);
 900         struct inet_peer *peer;
 901         unsigned long now;
 902         struct net *net;
 903         bool send;
 904         int code;
 905
 906         net = dev_net(rt->dst.dev);
 907         if (!IN_DEV_FORWARD(in_dev)) {
 908                 switch (rt->dst.error) {
 909                 case EHOSTUNREACH:
 910                         IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
 911                         break;
 912
 913                 case ENETUNREACH:
 914                         IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
 915                         break;
 916                 }
 917                 goto out;
 918         }
 919
 920         switch (rt->dst.error) {
 921         case EINVAL:
 922         default:
 923                 goto out;
 924         case EHOSTUNREACH:
 925                 code = ICMP_HOST_UNREACH;
 926                 break;
 927         case ENETUNREACH:
 928                 code = ICMP_NET_UNREACH;
 929                 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
 930                 break;
 931         case EACCES:
 932                 code = ICMP_PKT_FILTERED;
 933                 break;
 934         }
 935
 936         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
 937
 938         send = true;
 939         if (peer) {
 940                 now = jiffies;
 941                 peer->rate_tokens += now - peer->rate_last;
 942                 if (peer->rate_tokens > ip_rt_error_burst)
 943                         peer->rate_tokens = ip_rt_error_burst;
 944                 peer->rate_last = now;
 945                 if (peer->rate_tokens >= ip_rt_error_cost)
 946                         peer->rate_tokens -= ip_rt_error_cost;
 947                 else
 948                         send = false;
 949                 inet_putpeer(peer);
 950         }
 951         if (send)
 952                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
 953
 954 out:    kfree_skb(skb);
 955         return 0;
 956 }
 957
 958 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
 959 {
 960         struct dst_entry *dst = &rt->dst;
 961         struct fib_result res;
 962
 963         if (dst_metric_locked(dst, RTAX_MTU))
 964                 return;
 965
 966         if (dst->dev->mtu < mtu)
 967                 return;
 968
 969         if (mtu < ip_rt_min_pmtu)
 970                 mtu = ip_rt_min_pmtu;
 971
 972         if (rt->rt_pmtu == mtu &&
 973             time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
 974                 return;
 975
 976         rcu_read_lock();
 977         if (fib_lookup(dev_net(dst->dev), fl4, &res) == 0) {
 978                 struct fib_nh *nh = &FIB_RES_NH(res);
 979
 980                 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
 981                                       jiffies + ip_rt_mtu_expires);
 982         }
 983         rcu_read_unlock();
 984 }
 985
 986 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 987                               struct sk_buff *skb, u32 mtu)
 988 {
 989         struct rtable *rt = (struct rtable *) dst;
 990         struct flowi4 fl4;
 991
 992         ip_rt_build_flow_key(&fl4, sk, skb);
 993         __ip_rt_update_pmtu(rt, &fl4, mtu);
 994 }
 995
 996 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
 997                       int oif, u32 mark, u8 protocol, int flow_flags)
 998 {
 999         const struct iphdr *iph = (const struct iphdr *) skb->data;
1000         struct flowi4 fl4;
1001         struct rtable *rt;
1002
1003         if (!mark)
1004                 mark = IP4_REPLY_MARK(net, skb->mark);
1005
1006         __build_flow_key(&fl4, NULL, iph, oif,
1007                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1008         rt = __ip_route_output_key(net, &fl4);
1009         if (!IS_ERR(rt)) {
1010                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1011                 ip_rt_put(rt);
1012         }
1013 }
1014 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1015
1016 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1017 {
1018         const struct iphdr *iph = (const struct iphdr *) skb->data;
1019         struct flowi4 fl4;
1020         struct rtable *rt;
1021
1022         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1023
1024         if (!fl4.flowi4_mark)
1025                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1026
1027         rt = __ip_route_output_key(sock_net(sk), &fl4);
1028         if (!IS_ERR(rt)) {
1029                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1030                 ip_rt_put(rt);
1031         }
1032 }
1033
1034 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1035 {
1036         const struct iphdr *iph = (const struct iphdr *) skb->data;
1037         struct flowi4 fl4;
1038         struct rtable *rt;
1039         struct dst_entry *odst = NULL;
1040         bool new = false;
1041
1042         bh_lock_sock(sk);
1043
1044         if (!ip_sk_accept_pmtu(sk))
1045                 goto out;
1046
1047         odst = sk_dst_get(sk);
1048
1049         if (sock_owned_by_user(sk) || !odst) {
1050                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1051                 goto out;
1052         }
1053
1054         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1055
1056         rt = (struct rtable *)odst;
1057         if (odst->obsolete && odst->ops->check(odst, 0) == NULL) {
1058                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1059                 if (IS_ERR(rt))
1060                         goto out;
1061
1062                 new = true;
1063         }
1064
1065         __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1066
1067         if (!dst_check(&rt->dst, 0)) {
1068                 if (new)
1069                         dst_release(&rt->dst);
1070
1071                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1072                 if (IS_ERR(rt))
1073                         goto out;
1074
1075                 new = true;
1076         }
1077
1078         if (new)
1079                 sk_dst_set(sk, &rt->dst);
1080
1081 out:
1082         bh_unlock_sock(sk);
1083         dst_release(odst);
1084 }
1085 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1086
1087 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1088                    int oif, u32 mark, u8 protocol, int flow_flags)
1089 {
1090         const struct iphdr *iph = (const struct iphdr *) skb->data;
1091         struct flowi4 fl4;
1092         struct rtable *rt;
1093
1094         __build_flow_key(&fl4, NULL, iph, oif,
1095                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1096         rt = __ip_route_output_key(net, &fl4);
1097         if (!IS_ERR(rt)) {
1098                 __ip_do_redirect(rt, skb, &fl4, false);
1099                 ip_rt_put(rt);
1100         }
1101 }
1102 EXPORT_SYMBOL_GPL(ipv4_redirect);
1103
1104 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1105 {
1106         const struct iphdr *iph = (const struct iphdr *) skb->data;
1107         struct flowi4 fl4;
1108         struct rtable *rt;
1109
1110         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1111         rt = __ip_route_output_key(sock_net(sk), &fl4);
1112         if (!IS_ERR(rt)) {
1113                 __ip_do_redirect(rt, skb, &fl4, false);
1114                 ip_rt_put(rt);
1115         }
1116 }
1117 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1118
1119 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1120 {
1121         struct rtable *rt = (struct rtable *) dst;
1122
1123         /* All IPV4 dsts are created with ->obsolete set to the value
1124          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1125          * into this function always.
1126          *
1127          * When a PMTU/redirect information update invalidates a route,
1128          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1129          * DST_OBSOLETE_DEAD by dst_free().
1130          */
1131         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1132                 return NULL;
1133         return dst;
1134 }
1135
1136 static void ipv4_link_failure(struct sk_buff *skb)
1137 {
1138         struct rtable *rt;
1139
1140         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1141
1142         rt = skb_rtable(skb);
1143         if (rt)
1144                 dst_set_expires(&rt->dst, 0);
1145 }
1146
1147 static int ip_rt_bug(struct sock *sk, struct sk_buff *skb)
1148 {
1149         pr_debug("%s: %pI4 -> %pI4, %s\n",
1150                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1151                  skb->dev ? skb->dev->name : "?");
1152         kfree_skb(skb);
1153         WARN_ON(1);
1154         return 0;
1155 }
1156
1157 /*
1158    We do not cache source address of outgoing interface,
1159    because it is used only by IP RR, TS and SRR options,
1160    so that it out of fast path.
1161
1162    BTW remember: "addr" is allowed to be not aligned
1163    in IP options!
1164  */
1165
1166 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1167 {
1168         __be32 src;
1169
1170         if (rt_is_output_route(rt))
1171                 src = ip_hdr(skb)->saddr;
1172         else {
1173                 struct fib_result res;
1174                 struct flowi4 fl4;
1175                 struct iphdr *iph;
1176
1177                 iph = ip_hdr(skb);
1178
1179                 memset(&fl4, 0, sizeof(fl4));
1180                 fl4.daddr = iph->daddr;
1181                 fl4.saddr = iph->saddr;
1182                 fl4.flowi4_tos = RT_TOS(iph->tos);
1183                 fl4.flowi4_oif = rt->dst.dev->ifindex;
1184                 fl4.flowi4_iif = skb->dev->ifindex;
1185                 fl4.flowi4_mark = skb->mark;
1186
1187                 rcu_read_lock();
1188                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1189                         src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1190                 else
1191                         src = inet_select_addr(rt->dst.dev,
1192                                                rt_nexthop(rt, iph->daddr),
1193                                                RT_SCOPE_UNIVERSE);
1194                 rcu_read_unlock();
1195         }
1196         memcpy(addr, &src, 4);
1197 }
1198
1199 #ifdef CONFIG_IP_ROUTE_CLASSID
1200 static void set_class_tag(struct rtable *rt, u32 tag)
1201 {
1202         if (!(rt->dst.tclassid & 0xFFFF))
1203                 rt->dst.tclassid |= tag & 0xFFFF;
1204         if (!(rt->dst.tclassid & 0xFFFF0000))
1205                 rt->dst.tclassid |= tag & 0xFFFF0000;
1206 }
1207 #endif
1208
1209 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1210 {
1211         unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1212
1213         if (advmss == 0) {
1214                 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1215                                ip_rt_min_advmss);
1216                 if (advmss > 65535 - 40)
1217                         advmss = 65535 - 40;
1218         }
1219         return advmss;
1220 }
1221
1222 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1223 {
1224         const struct rtable *rt = (const struct rtable *) dst;
1225         unsigned int mtu = rt->rt_pmtu;
1226
1227         if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1228                 mtu = dst_metric_raw(dst, RTAX_MTU);
1229
1230         if (mtu)
1231                 return mtu;
1232
1233         mtu = dst->dev->mtu;
1234
1235         if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1236                 if (rt->rt_uses_gateway && mtu > 576)
1237                         mtu = 576;
1238         }
1239
1240         return min_t(unsigned int, mtu, IP_MAX_MTU);
1241 }
1242
1243 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1244 {
1245         struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1246         struct fib_nh_exception *fnhe;
1247         u32 hval;
1248
1249         if (!hash)
1250                 return NULL;
1251
1252         hval = fnhe_hashfun(daddr);
1253
1254         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1255              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1256                 if (fnhe->fnhe_daddr == daddr)
1257                         return fnhe;
1258         }
1259         return NULL;
1260 }
1261
1262 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1263                               __be32 daddr)
1264 {
1265         bool ret = false;
1266
1267         spin_lock_bh(&fnhe_lock);
1268
1269         if (daddr == fnhe->fnhe_daddr) {
1270                 struct rtable __rcu **porig;
1271                 struct rtable *orig;
1272                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1273
1274                 if (rt_is_input_route(rt))
1275                         porig = &fnhe->fnhe_rth_input;
1276                 else
1277                         porig = &fnhe->fnhe_rth_output;
1278                 orig = rcu_dereference(*porig);
1279
1280                 if (fnhe->fnhe_genid != genid) {
1281                         fnhe->fnhe_genid = genid;
1282                         fnhe->fnhe_gw = 0;
1283                         fnhe->fnhe_pmtu = 0;
1284                         fnhe->fnhe_expires = 0;
1285                         fnhe_flush_routes(fnhe);
1286                         orig = NULL;
1287                 }
1288                 fill_route_from_fnhe(rt, fnhe);
1289                 if (!rt->rt_gateway)
1290                         rt->rt_gateway = daddr;
1291
1292                 if (!(rt->dst.flags & DST_NOCACHE)) {
1293                         rcu_assign_pointer(*porig, rt);
1294                         if (orig)
1295                                 rt_free(orig);
1296                         ret = true;
1297                 }
1298
1299                 fnhe->fnhe_stamp = jiffies;
1300         }
1301         spin_unlock_bh(&fnhe_lock);
1302
1303         return ret;
1304 }
1305
1306 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1307 {
1308         struct rtable *orig, *prev, **p;
1309         bool ret = true;
1310
1311         if (rt_is_input_route(rt)) {
1312                 p = (struct rtable **)&nh->nh_rth_input;
1313         } else {
1314                 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1315         }
1316         orig = *p;
1317
1318         prev = cmpxchg(p, orig, rt);
1319         if (prev == orig) {
1320                 if (orig)
1321                         rt_free(orig);
1322         } else
1323                 ret = false;
1324
1325         return ret;
1326 }
1327
1328 static DEFINE_SPINLOCK(rt_uncached_lock);
1329 static LIST_HEAD(rt_uncached_list);
1330
1331 static void rt_add_uncached_list(struct rtable *rt)
1332 {
1333         spin_lock_bh(&rt_uncached_lock);
1334         list_add_tail(&rt->rt_uncached, &rt_uncached_list);
1335         spin_unlock_bh(&rt_uncached_lock);
1336 }
1337
1338 static void ipv4_dst_destroy(struct dst_entry *dst)
1339 {
1340         struct rtable *rt = (struct rtable *) dst;
1341
1342         if (!list_empty(&rt->rt_uncached)) {
1343                 spin_lock_bh(&rt_uncached_lock);
1344                 list_del(&rt->rt_uncached);
1345                 spin_unlock_bh(&rt_uncached_lock);
1346         }
1347 }
1348
1349 void rt_flush_dev(struct net_device *dev)
1350 {
1351         if (!list_empty(&rt_uncached_list)) {
1352                 struct net *net = dev_net(dev);
1353                 struct rtable *rt;
1354
1355                 spin_lock_bh(&rt_uncached_lock);
1356                 list_for_each_entry(rt, &rt_uncached_list, rt_uncached) {
1357                         if (rt->dst.dev != dev)
1358                                 continue;
1359                         rt->dst.dev = net->loopback_dev;
1360                         dev_hold(rt->dst.dev);
1361                         dev_put(dev);
1362                 }
1363                 spin_unlock_bh(&rt_uncached_lock);
1364         }
1365 }
1366
1367 static bool rt_cache_valid(const struct rtable *rt)
1368 {
1369         return  rt &&
1370                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1371                 !rt_is_expired(rt);
1372 }
1373
1374 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1375                            const struct fib_result *res,
1376                            struct fib_nh_exception *fnhe,
1377                            struct fib_info *fi, u16 type, u32 itag)
1378 {
1379         bool cached = false;
1380
1381         if (fi) {
1382                 struct fib_nh *nh = &FIB_RES_NH(*res);
1383
1384                 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1385                         rt->rt_gateway = nh->nh_gw;
1386                         rt->rt_uses_gateway = 1;
1387                 }
1388                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1389 #ifdef CONFIG_IP_ROUTE_CLASSID
1390                 rt->dst.tclassid = nh->nh_tclassid;
1391 #endif
1392                 if (unlikely(fnhe))
1393                         cached = rt_bind_exception(rt, fnhe, daddr);
1394                 else if (!(rt->dst.flags & DST_NOCACHE))
1395                         cached = rt_cache_route(nh, rt);
1396                 if (unlikely(!cached)) {
1397                         /* Routes we intend to cache in nexthop exception or
1398                          * FIB nexthop have the DST_NOCACHE bit clear.
1399                          * However, if we are unsuccessful at storing this
1400                          * route into the cache we really need to set it.
1401                          */
1402                         rt->dst.flags |= DST_NOCACHE;
1403                         if (!rt->rt_gateway)
1404                                 rt->rt_gateway = daddr;
1405                         rt_add_uncached_list(rt);
1406                 }
1407         } else
1408                 rt_add_uncached_list(rt);
1409
1410 #ifdef CONFIG_IP_ROUTE_CLASSID
1411 #ifdef CONFIG_IP_MULTIPLE_TABLES
1412         set_class_tag(rt, res->tclassid);
1413 #endif
1414         set_class_tag(rt, itag);
1415 #endif
1416 }
1417
1418 static struct rtable *rt_dst_alloc(struct net_device *dev,
1419                                    bool nopolicy, bool noxfrm, bool will_cache)
1420 {
1421         return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1422                          (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1423                          (nopolicy ? DST_NOPOLICY : 0) |
1424                          (noxfrm ? DST_NOXFRM : 0));
1425 }
1426
1427 /* called in rcu_read_lock() section */
1428 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1429                                 u8 tos, struct net_device *dev, int our)
1430 {
1431         struct rtable *rth;
1432         struct in_device *in_dev = __in_dev_get_rcu(dev);
1433         u32 itag = 0;
1434         int err;
1435
1436         /* Primary sanity checks. */
1437
1438         if (in_dev == NULL)
1439                 return -EINVAL;
1440
1441         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1442             skb->protocol != htons(ETH_P_IP))
1443                 goto e_inval;
1444
1445         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1446                 if (ipv4_is_loopback(saddr))
1447                         goto e_inval;
1448
1449         if (ipv4_is_zeronet(saddr)) {
1450                 if (!ipv4_is_local_multicast(daddr))
1451                         goto e_inval;
1452         } else {
1453                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1454                                           in_dev, &itag);
1455                 if (err < 0)
1456                         goto e_err;
1457         }
1458         rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1459                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1460         if (!rth)
1461                 goto e_nobufs;
1462
1463 #ifdef CONFIG_IP_ROUTE_CLASSID
1464         rth->dst.tclassid = itag;
1465 #endif
1466         rth->dst.output = ip_rt_bug;
1467
1468         rth->rt_genid   = rt_genid_ipv4(dev_net(dev));
1469         rth->rt_flags   = RTCF_MULTICAST;
1470         rth->rt_type    = RTN_MULTICAST;
1471         rth->rt_is_input= 1;
1472         rth->rt_iif     = 0;
1473         rth->rt_pmtu    = 0;
1474         rth->rt_gateway = 0;
1475         rth->rt_uses_gateway = 0;
1476         INIT_LIST_HEAD(&rth->rt_uncached);
1477         if (our) {
1478                 rth->dst.input= ip_local_deliver;
1479                 rth->rt_flags |= RTCF_LOCAL;
1480         }
1481
1482 #ifdef CONFIG_IP_MROUTE
1483         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1484                 rth->dst.input = ip_mr_input;
1485 #endif
1486         RT_CACHE_STAT_INC(in_slow_mc);
1487
1488         skb_dst_set(skb, &rth->dst);
1489         return 0;
1490
1491 e_nobufs:
1492         return -ENOBUFS;
1493 e_inval:
1494         return -EINVAL;
1495 e_err:
1496         return err;
1497 }
1498
1499
1500 static void ip_handle_martian_source(struct net_device *dev,
1501                                      struct in_device *in_dev,
1502                                      struct sk_buff *skb,
1503                                      __be32 daddr,
1504                                      __be32 saddr)
1505 {
1506         RT_CACHE_STAT_INC(in_martian_src);
1507 #ifdef CONFIG_IP_ROUTE_VERBOSE
1508         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1509                 /*
1510                  *      RFC1812 recommendation, if source is martian,
1511                  *      the only hint is MAC header.
1512                  */
1513                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1514                         &daddr, &saddr, dev->name);
1515                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1516                         print_hex_dump(KERN_WARNING, "ll header: ",
1517                                        DUMP_PREFIX_OFFSET, 16, 1,
1518                                        skb_mac_header(skb),
1519                                        dev->hard_header_len, true);
1520                 }
1521         }
1522 #endif
1523 }
1524
1525 /* called in rcu_read_lock() section */
1526 static int __mkroute_input(struct sk_buff *skb,
1527                            const struct fib_result *res,
1528                            struct in_device *in_dev,
1529                            __be32 daddr, __be32 saddr, u32 tos)
1530 {
1531         struct fib_nh_exception *fnhe;
1532         struct rtable *rth;
1533         int err;
1534         struct in_device *out_dev;
1535         unsigned int flags = 0;
1536         bool do_cache;
1537         u32 itag = 0;
1538
1539         /* get a working reference to the output device */
1540         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1541         if (out_dev == NULL) {
1542                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1543                 return -EINVAL;
1544         }
1545
1546         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1547                                   in_dev->dev, in_dev, &itag);
1548         if (err < 0) {
1549                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1550                                          saddr);
1551
1552                 goto cleanup;
1553         }
1554
1555         do_cache = res->fi && !itag;
1556         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1557             (IN_DEV_SHARED_MEDIA(out_dev) ||
1558              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) {
1559                 flags |= RTCF_DOREDIRECT;
1560                 do_cache = false;
1561         }
1562
1563         if (skb->protocol != htons(ETH_P_IP)) {
1564                 /* Not IP (i.e. ARP). Do not create route, if it is
1565                  * invalid for proxy arp. DNAT routes are always valid.
1566                  *
1567                  * Proxy arp feature have been extended to allow, ARP
1568                  * replies back to the same interface, to support
1569                  * Private VLAN switch technologies. See arp.c.
1570                  */
1571                 if (out_dev == in_dev &&
1572                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1573                         err = -EINVAL;
1574                         goto cleanup;
1575                 }
1576         }
1577
1578         fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1579         if (do_cache) {
1580                 if (fnhe != NULL)
1581                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1582                 else
1583                         rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1584
1585                 if (rt_cache_valid(rth)) {
1586                         skb_dst_set_noref(skb, &rth->dst);
1587                         goto out;
1588                 }
1589         }
1590
1591         rth = rt_dst_alloc(out_dev->dev,
1592                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1593                            IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1594         if (!rth) {
1595                 err = -ENOBUFS;
1596                 goto cleanup;
1597         }
1598
1599         rth->rt_genid = rt_genid_ipv4(dev_net(rth->dst.dev));
1600         rth->rt_flags = flags;
1601         rth->rt_type = res->type;
1602         rth->rt_is_input = 1;
1603         rth->rt_iif     = 0;
1604         rth->rt_pmtu    = 0;
1605         rth->rt_gateway = 0;
1606         rth->rt_uses_gateway = 0;
1607         INIT_LIST_HEAD(&rth->rt_uncached);
1608         RT_CACHE_STAT_INC(in_slow_tot);
1609
1610         rth->dst.input = ip_forward;
1611         rth->dst.output = ip_output;
1612
1613         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1614         skb_dst_set(skb, &rth->dst);
1615 out:
1616         err = 0;
1617  cleanup:
1618         return err;
1619 }
1620
1621 static int ip_mkroute_input(struct sk_buff *skb,
1622                             struct fib_result *res,
1623                             const struct flowi4 *fl4,
1624                             struct in_device *in_dev,
1625                             __be32 daddr, __be32 saddr, u32 tos)
1626 {
1627 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1628         if (res->fi && res->fi->fib_nhs > 1)
1629                 fib_select_multipath(res);
1630 #endif
1631
1632         /* create a routing cache entry */
1633         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1634 }
1635
1636 /*
1637  *      NOTE. We drop all the packets that has local source
1638  *      addresses, because every properly looped back packet
1639  *      must have correct destination already attached by output routine.
1640  *
1641  *      Such approach solves two big problems:
1642  *      1. Not simplex devices are handled properly.
1643  *      2. IP spoofing attempts are filtered with 100% of guarantee.
1644  *      called with rcu_read_lock()
1645  */
1646
1647 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1648                                u8 tos, struct net_device *dev)
1649 {
1650         struct fib_result res;
1651         struct in_device *in_dev = __in_dev_get_rcu(dev);
1652         struct flowi4   fl4;
1653         unsigned int    flags = 0;
1654         u32             itag = 0;
1655         struct rtable   *rth;
1656         int             err = -EINVAL;
1657         struct net    *net = dev_net(dev);
1658         bool do_cache;
1659
1660         /* IP on this device is disabled. */
1661
1662         if (!in_dev)
1663                 goto out;
1664
1665         /* Check for the most weird martians, which can be not detected
1666            by fib_lookup.
1667          */
1668
1669         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1670                 goto martian_source;
1671
1672         res.fi = NULL;
1673         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1674                 goto brd_input;
1675
1676         /* Accept zero addresses only to limited broadcast;
1677          * I even do not know to fix it or not. Waiting for complains :-)
1678          */
1679         if (ipv4_is_zeronet(saddr))
1680                 goto martian_source;
1681
1682         if (ipv4_is_zeronet(daddr))
1683                 goto martian_destination;
1684
1685         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1686          * and call it once if daddr or/and saddr are loopback addresses
1687          */
1688         if (ipv4_is_loopback(daddr)) {
1689                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1690                         goto martian_destination;
1691         } else if (ipv4_is_loopback(saddr)) {
1692                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1693                         goto martian_source;
1694         }
1695
1696         /*
1697          *      Now we are ready to route packet.
1698          */
1699         fl4.flowi4_oif = 0;
1700         fl4.flowi4_iif = dev->ifindex;
1701         fl4.flowi4_mark = skb->mark;
1702         fl4.flowi4_tos = tos;
1703         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1704         fl4.daddr = daddr;
1705         fl4.saddr = saddr;
1706         err = fib_lookup(net, &fl4, &res);
1707         if (err != 0) {
1708                 if (!IN_DEV_FORWARD(in_dev))
1709                         err = -EHOSTUNREACH;
1710                 goto no_route;
1711         }
1712
1713         if (res.type == RTN_BROADCAST)
1714                 goto brd_input;
1715
1716         if (res.type == RTN_LOCAL) {
1717                 err = fib_validate_source(skb, saddr, daddr, tos,
1718                                           0, dev, in_dev, &itag);
1719                 if (err < 0)
1720                         goto martian_source_keep_err;
1721                 goto local_input;
1722         }
1723
1724         if (!IN_DEV_FORWARD(in_dev)) {
1725                 err = -EHOSTUNREACH;
1726                 goto no_route;
1727         }
1728         if (res.type != RTN_UNICAST)
1729                 goto martian_destination;
1730
1731         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1732 out:    return err;
1733
1734 brd_input:
1735         if (skb->protocol != htons(ETH_P_IP))
1736                 goto e_inval;
1737
1738         if (!ipv4_is_zeronet(saddr)) {
1739                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1740                                           in_dev, &itag);
1741                 if (err < 0)
1742                         goto martian_source_keep_err;
1743         }
1744         flags |= RTCF_BROADCAST;
1745         res.type = RTN_BROADCAST;
1746         RT_CACHE_STAT_INC(in_brd);
1747
1748 local_input:
1749         do_cache = false;
1750         if (res.fi) {
1751                 if (!itag) {
1752                         rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1753                         if (rt_cache_valid(rth)) {
1754                                 skb_dst_set_noref(skb, &rth->dst);
1755                                 err = 0;
1756                                 goto out;
1757                         }
1758                         do_cache = true;
1759                 }
1760         }
1761
1762         rth = rt_dst_alloc(net->loopback_dev,
1763                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1764         if (!rth)
1765                 goto e_nobufs;
1766
1767         rth->dst.input= ip_local_deliver;
1768         rth->dst.output= ip_rt_bug;
1769 #ifdef CONFIG_IP_ROUTE_CLASSID
1770         rth->dst.tclassid = itag;
1771 #endif
1772
1773         rth->rt_genid = rt_genid_ipv4(net);
1774         rth->rt_flags   = flags|RTCF_LOCAL;
1775         rth->rt_type    = res.type;
1776         rth->rt_is_input = 1;
1777         rth->rt_iif     = 0;
1778         rth->rt_pmtu    = 0;
1779         rth->rt_gateway = 0;
1780         rth->rt_uses_gateway = 0;
1781         INIT_LIST_HEAD(&rth->rt_uncached);
1782         RT_CACHE_STAT_INC(in_slow_tot);
1783         if (res.type == RTN_UNREACHABLE) {
1784                 rth->dst.input= ip_error;
1785                 rth->dst.error= -err;
1786                 rth->rt_flags   &= ~RTCF_LOCAL;
1787         }
1788         if (do_cache) {
1789                 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1790                         rth->dst.flags |= DST_NOCACHE;
1791                         rt_add_uncached_list(rth);
1792                 }
1793         }
1794         skb_dst_set(skb, &rth->dst);
1795         err = 0;
1796         goto out;
1797
1798 no_route:
1799         RT_CACHE_STAT_INC(in_no_route);
1800         res.type = RTN_UNREACHABLE;
1801         res.fi = NULL;
1802         goto local_input;
1803
1804         /*
1805          *      Do not cache martian addresses: they should be logged (RFC1812)
1806          */
1807 martian_destination:
1808         RT_CACHE_STAT_INC(in_martian_dst);
1809 #ifdef CONFIG_IP_ROUTE_VERBOSE
1810         if (IN_DEV_LOG_MARTIANS(in_dev))
1811                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1812                                      &daddr, &saddr, dev->name);
1813 #endif
1814
1815 e_inval:
1816         err = -EINVAL;
1817         goto out;
1818
1819 e_nobufs:
1820         err = -ENOBUFS;
1821         goto out;
1822
1823 martian_source:
1824         err = -EINVAL;
1825 martian_source_keep_err:
1826         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1827         goto out;
1828 }
1829
1830 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1831                          u8 tos, struct net_device *dev)
1832 {
1833         int res;
1834
1835         rcu_read_lock();
1836
1837         /* Multicast recognition logic is moved from route cache to here.
1838            The problem was that too many Ethernet cards have broken/missing
1839            hardware multicast filters :-( As result the host on multicasting
1840            network acquires a lot of useless route cache entries, sort of
1841            SDR messages from all the world. Now we try to get rid of them.
1842            Really, provided software IP multicast filter is organized
1843            reasonably (at least, hashed), it does not result in a slowdown
1844            comparing with route cache reject entries.
1845            Note, that multicast routers are not affected, because
1846            route cache entry is created eventually.
1847          */
1848         if (ipv4_is_multicast(daddr)) {
1849                 struct in_device *in_dev = __in_dev_get_rcu(dev);
1850
1851                 if (in_dev) {
1852                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1853                                                   ip_hdr(skb)->protocol);
1854                         if (our
1855 #ifdef CONFIG_IP_MROUTE
1856                                 ||
1857                             (!ipv4_is_local_multicast(daddr) &&
1858                              IN_DEV_MFORWARD(in_dev))
1859 #endif
1860                            ) {
1861                                 int res = ip_route_input_mc(skb, daddr, saddr,
1862                                                             tos, dev, our);
1863                                 rcu_read_unlock();
1864                                 return res;
1865                         }
1866                 }
1867                 rcu_read_unlock();
1868                 return -EINVAL;
1869         }
1870         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1871         rcu_read_unlock();
1872         return res;
1873 }
1874 EXPORT_SYMBOL(ip_route_input_noref);
1875
1876 /* called with rcu_read_lock() */
1877 static struct rtable *__mkroute_output(const struct fib_result *res,
1878                                        const struct flowi4 *fl4, int orig_oif,
1879                                        struct net_device *dev_out,
1880                                        unsigned int flags)
1881 {
1882         struct fib_info *fi = res->fi;
1883         struct fib_nh_exception *fnhe;
1884         struct in_device *in_dev;
1885         u16 type = res->type;
1886         struct rtable *rth;
1887         bool do_cache;
1888
1889         in_dev = __in_dev_get_rcu(dev_out);
1890         if (!in_dev)
1891                 return ERR_PTR(-EINVAL);
1892
1893         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1894                 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1895                         return ERR_PTR(-EINVAL);
1896
1897         if (ipv4_is_lbcast(fl4->daddr))
1898                 type = RTN_BROADCAST;
1899         else if (ipv4_is_multicast(fl4->daddr))
1900                 type = RTN_MULTICAST;
1901         else if (ipv4_is_zeronet(fl4->daddr))
1902                 return ERR_PTR(-EINVAL);
1903
1904         if (dev_out->flags & IFF_LOOPBACK)
1905                 flags |= RTCF_LOCAL;
1906
1907         do_cache = true;
1908         if (type == RTN_BROADCAST) {
1909                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1910                 fi = NULL;
1911         } else if (type == RTN_MULTICAST) {
1912                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1913                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1914                                      fl4->flowi4_proto))
1915                         flags &= ~RTCF_LOCAL;
1916                 else
1917                         do_cache = false;
1918                 /* If multicast route do not exist use
1919                  * default one, but do not gateway in this case.
1920                  * Yes, it is hack.
1921                  */
1922                 if (fi && res->prefixlen < 4)
1923                         fi = NULL;
1924         }
1925
1926         fnhe = NULL;
1927         do_cache &= fi != NULL;
1928         if (do_cache) {
1929                 struct rtable __rcu **prth;
1930                 struct fib_nh *nh = &FIB_RES_NH(*res);
1931
1932                 fnhe = find_exception(nh, fl4->daddr);
1933                 if (fnhe)
1934                         prth = &fnhe->fnhe_rth_output;
1935                 else {
1936                         if (unlikely(fl4->flowi4_flags &
1937                                      FLOWI_FLAG_KNOWN_NH &&
1938                                      !(nh->nh_gw &&
1939                                        nh->nh_scope == RT_SCOPE_LINK))) {
1940                                 do_cache = false;
1941                                 goto add;
1942                         }
1943                         prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
1944                 }
1945                 rth = rcu_dereference(*prth);
1946                 if (rt_cache_valid(rth)) {
1947                         dst_hold(&rth->dst);
1948                         return rth;
1949                 }
1950         }
1951
1952 add:
1953         rth = rt_dst_alloc(dev_out,
1954                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1955                            IN_DEV_CONF_GET(in_dev, NOXFRM),
1956                            do_cache);
1957         if (!rth)
1958                 return ERR_PTR(-ENOBUFS);
1959
1960         rth->dst.output = ip_output;
1961
1962         rth->rt_genid = rt_genid_ipv4(dev_net(dev_out));
1963         rth->rt_flags   = flags;
1964         rth->rt_type    = type;
1965         rth->rt_is_input = 0;
1966         rth->rt_iif     = orig_oif ? : 0;
1967         rth->rt_pmtu    = 0;
1968         rth->rt_gateway = 0;
1969         rth->rt_uses_gateway = 0;
1970         INIT_LIST_HEAD(&rth->rt_uncached);
1971
1972         RT_CACHE_STAT_INC(out_slow_tot);
1973
1974         if (flags & RTCF_LOCAL)
1975                 rth->dst.input = ip_local_deliver;
1976         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1977                 if (flags & RTCF_LOCAL &&
1978                     !(dev_out->flags & IFF_LOOPBACK)) {
1979                         rth->dst.output = ip_mc_output;
1980                         RT_CACHE_STAT_INC(out_slow_mc);
1981                 }
1982 #ifdef CONFIG_IP_MROUTE
1983                 if (type == RTN_MULTICAST) {
1984                         if (IN_DEV_MFORWARD(in_dev) &&
1985                             !ipv4_is_local_multicast(fl4->daddr)) {
1986                                 rth->dst.input = ip_mr_input;
1987                                 rth->dst.output = ip_mc_output;
1988                         }
1989                 }
1990 #endif
1991         }
1992
1993         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
1994
1995         return rth;
1996 }
1997
1998 /*
1999  * Major route resolver routine.
2000  */
2001
2002 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
2003 {
2004         struct net_device *dev_out = NULL;
2005         __u8 tos = RT_FL_TOS(fl4);
2006         unsigned int flags = 0;
2007         struct fib_result res;
2008         struct rtable *rth;
2009         int orig_oif;
2010
2011         res.tclassid    = 0;
2012         res.fi          = NULL;
2013         res.table       = NULL;
2014
2015         orig_oif = fl4->flowi4_oif;
2016
2017         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2018         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2019         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2020                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2021
2022         rcu_read_lock();
2023         if (fl4->saddr) {
2024                 rth = ERR_PTR(-EINVAL);
2025                 if (ipv4_is_multicast(fl4->saddr) ||
2026                     ipv4_is_lbcast(fl4->saddr) ||
2027                     ipv4_is_zeronet(fl4->saddr))
2028                         goto out;
2029
2030                 /* I removed check for oif == dev_out->oif here.
2031                    It was wrong for two reasons:
2032                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2033                       is assigned to multiple interfaces.
2034                    2. Moreover, we are allowed to send packets with saddr
2035                       of another iface. --ANK
2036                  */
2037
2038                 if (fl4->flowi4_oif == 0 &&
2039                     (ipv4_is_multicast(fl4->daddr) ||
2040                      ipv4_is_lbcast(fl4->daddr))) {
2041                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2042                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2043                         if (dev_out == NULL)
2044                                 goto out;
2045
2046                         /* Special hack: user can direct multicasts
2047                            and limited broadcast via necessary interface
2048                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2049                            This hack is not just for fun, it allows
2050                            vic,vat and friends to work.
2051                            They bind socket to loopback, set ttl to zero
2052                            and expect that it will work.
2053                            From the viewpoint of routing cache they are broken,
2054                            because we are not allowed to build multicast path
2055                            with loopback source addr (look, routing cache
2056                            cannot know, that ttl is zero, so that packet
2057                            will not leave this host and route is valid).
2058                            Luckily, this hack is good workaround.
2059                          */
2060
2061                         fl4->flowi4_oif = dev_out->ifindex;
2062                         goto make_route;
2063                 }
2064
2065                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2066                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2067                         if (!__ip_dev_find(net, fl4->saddr, false))
2068                                 goto out;
2069                 }
2070         }
2071
2072
2073         if (fl4->flowi4_oif) {
2074                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2075                 rth = ERR_PTR(-ENODEV);
2076                 if (dev_out == NULL)
2077                         goto out;
2078
2079                 /* RACE: Check return value of inet_select_addr instead. */
2080                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2081                         rth = ERR_PTR(-ENETUNREACH);
2082                         goto out;
2083                 }
2084                 if (ipv4_is_local_multicast(fl4->daddr) ||
2085                     ipv4_is_lbcast(fl4->daddr)) {
2086                         if (!fl4->saddr)
2087                                 fl4->saddr = inet_select_addr(dev_out, 0,
2088                                                               RT_SCOPE_LINK);
2089                         goto make_route;
2090                 }
2091                 if (!fl4->saddr) {
2092                         if (ipv4_is_multicast(fl4->daddr))
2093                                 fl4->saddr = inet_select_addr(dev_out, 0,
2094                                                               fl4->flowi4_scope);
2095                         else if (!fl4->daddr)
2096                                 fl4->saddr = inet_select_addr(dev_out, 0,
2097                                                               RT_SCOPE_HOST);
2098                 }
2099         }
2100
2101         if (!fl4->daddr) {
2102                 fl4->daddr = fl4->saddr;
2103                 if (!fl4->daddr)
2104                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2105                 dev_out = net->loopback_dev;
2106                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2107                 res.type = RTN_LOCAL;
2108                 flags |= RTCF_LOCAL;
2109                 goto make_route;
2110         }
2111
2112         if (fib_lookup(net, fl4, &res)) {
2113                 res.fi = NULL;
2114                 res.table = NULL;
2115                 if (fl4->flowi4_oif) {
2116                         /* Apparently, routing tables are wrong. Assume,
2117                            that the destination is on link.
2118
2119                            WHY? DW.
2120                            Because we are allowed to send to iface
2121                            even if it has NO routes and NO assigned
2122                            addresses. When oif is specified, routing
2123                            tables are looked up with only one purpose:
2124                            to catch if destination is gatewayed, rather than
2125                            direct. Moreover, if MSG_DONTROUTE is set,
2126                            we send packet, ignoring both routing tables
2127                            and ifaddr state. --ANK
2128
2129
2130                            We could make it even if oif is unknown,
2131                            likely IPv6, but we do not.
2132                          */
2133
2134                         if (fl4->saddr == 0)
2135                                 fl4->saddr = inet_select_addr(dev_out, 0,
2136                                                               RT_SCOPE_LINK);
2137                         res.type = RTN_UNICAST;
2138                         goto make_route;
2139                 }
2140                 rth = ERR_PTR(-ENETUNREACH);
2141                 goto out;
2142         }
2143
2144         if (res.type == RTN_LOCAL) {
2145                 if (!fl4->saddr) {
2146                         if (res.fi->fib_prefsrc)
2147                                 fl4->saddr = res.fi->fib_prefsrc;
2148                         else
2149                                 fl4->saddr = fl4->daddr;
2150                 }
2151                 dev_out = net->loopback_dev;
2152                 fl4->flowi4_oif = dev_out->ifindex;
2153                 flags |= RTCF_LOCAL;
2154                 goto make_route;
2155         }
2156
2157 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2158         if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2159                 fib_select_multipath(&res);
2160         else
2161 #endif
2162         if (!res.prefixlen &&
2163             res.table->tb_num_default > 1 &&
2164             res.type == RTN_UNICAST && !fl4->flowi4_oif)
2165                 fib_select_default(&res);
2166
2167         if (!fl4->saddr)
2168                 fl4->saddr = FIB_RES_PREFSRC(net, res);
2169
2170         dev_out = FIB_RES_DEV(res);
2171         fl4->flowi4_oif = dev_out->ifindex;
2172
2173
2174 make_route:
2175         rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2176
2177 out:
2178         rcu_read_unlock();
2179         return rth;
2180 }
2181 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2182
2183 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2184 {
2185         return NULL;
2186 }
2187
2188 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2189 {
2190         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2191
2192         return mtu ? : dst->dev->mtu;
2193 }
2194
2195 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2196                                           struct sk_buff *skb, u32 mtu)
2197 {
2198 }
2199
2200 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2201                                        struct sk_buff *skb)
2202 {
2203 }
2204
2205 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2206                                           unsigned long old)
2207 {
2208         return NULL;
2209 }
2210
2211 static struct dst_ops ipv4_dst_blackhole_ops = {
2212         .family                 =       AF_INET,
2213         .protocol               =       cpu_to_be16(ETH_P_IP),
2214         .check                  =       ipv4_blackhole_dst_check,
2215         .mtu                    =       ipv4_blackhole_mtu,
2216         .default_advmss         =       ipv4_default_advmss,
2217         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2218         .redirect               =       ipv4_rt_blackhole_redirect,
2219         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
2220         .neigh_lookup           =       ipv4_neigh_lookup,
2221 };
2222
2223 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2224 {
2225         struct rtable *ort = (struct rtable *) dst_orig;
2226         struct rtable *rt;
2227
2228         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2229         if (rt) {
2230                 struct dst_entry *new = &rt->dst;
2231
2232                 new->__use = 1;
2233                 new->input = dst_discard;
2234                 new->output = dst_discard_sk;
2235
2236                 new->dev = ort->dst.dev;
2237                 if (new->dev)
2238                         dev_hold(new->dev);
2239
2240                 rt->rt_is_input = ort->rt_is_input;
2241                 rt->rt_iif = ort->rt_iif;
2242                 rt->rt_pmtu = ort->rt_pmtu;
2243
2244                 rt->rt_genid = rt_genid_ipv4(net);
2245                 rt->rt_flags = ort->rt_flags;
2246                 rt->rt_type = ort->rt_type;
2247                 rt->rt_gateway = ort->rt_gateway;
2248                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2249
2250                 INIT_LIST_HEAD(&rt->rt_uncached);
2251
2252                 dst_free(new);
2253         }
2254
2255         dst_release(dst_orig);
2256
2257         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2258 }
2259
2260 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2261                                     struct sock *sk)
2262 {
2263         struct rtable *rt = __ip_route_output_key(net, flp4);
2264
2265         if (IS_ERR(rt))
2266                 return rt;
2267
2268         if (flp4->flowi4_proto)
2269                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2270                                                         flowi4_to_flowi(flp4),
2271                                                         sk, 0);
2272
2273         return rt;
2274 }
2275 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2276
2277 static int rt_fill_info(struct net *net,  __be32 dst, __be32 src,
2278                         struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2279                         u32 seq, int event, int nowait, unsigned int flags)
2280 {
2281         struct rtable *rt = skb_rtable(skb);
2282         struct rtmsg *r;
2283         struct nlmsghdr *nlh;
2284         unsigned long expires = 0;
2285         u32 error;
2286         u32 metrics[RTAX_MAX];
2287
2288         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2289         if (nlh == NULL)
2290                 return -EMSGSIZE;
2291
2292         r = nlmsg_data(nlh);
2293         r->rtm_family    = AF_INET;
2294         r->rtm_dst_len  = 32;
2295         r->rtm_src_len  = 0;
2296         r->rtm_tos      = fl4->flowi4_tos;
2297         r->rtm_table    = RT_TABLE_MAIN;
2298         if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2299                 goto nla_put_failure;
2300         r->rtm_type     = rt->rt_type;
2301         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2302         r->rtm_protocol = RTPROT_UNSPEC;
2303         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2304         if (rt->rt_flags & RTCF_NOTIFY)
2305                 r->rtm_flags |= RTM_F_NOTIFY;
2306
2307         if (nla_put_be32(skb, RTA_DST, dst))
2308                 goto nla_put_failure;
2309         if (src) {
2310                 r->rtm_src_len = 32;
2311                 if (nla_put_be32(skb, RTA_SRC, src))
2312                         goto nla_put_failure;
2313         }
2314         if (rt->dst.dev &&
2315             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2316                 goto nla_put_failure;
2317 #ifdef CONFIG_IP_ROUTE_CLASSID
2318         if (rt->dst.tclassid &&
2319             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2320                 goto nla_put_failure;
2321 #endif
2322         if (!rt_is_input_route(rt) &&
2323             fl4->saddr != src) {
2324                 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2325                         goto nla_put_failure;
2326         }
2327         if (rt->rt_uses_gateway &&
2328             nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2329                 goto nla_put_failure;
2330
2331         expires = rt->dst.expires;
2332         if (expires) {
2333                 unsigned long now = jiffies;
2334
2335                 if (time_before(now, expires))
2336                         expires -= now;
2337                 else
2338                         expires = 0;
2339         }
2340
2341         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2342         if (rt->rt_pmtu && expires)
2343                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2344         if (rtnetlink_put_metrics(skb, metrics) < 0)
2345                 goto nla_put_failure;
2346
2347         if (fl4->flowi4_mark &&
2348             nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2349                 goto nla_put_failure;
2350
2351         error = rt->dst.error;
2352
2353         if (rt_is_input_route(rt)) {
2354 #ifdef CONFIG_IP_MROUTE
2355                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2356                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2357                         int err = ipmr_get_route(net, skb,
2358                                                  fl4->saddr, fl4->daddr,
2359                                                  r, nowait);
2360                         if (err <= 0) {
2361                                 if (!nowait) {
2362                                         if (err == 0)
2363                                                 return 0;
2364                                         goto nla_put_failure;
2365                                 } else {
2366                                         if (err == -EMSGSIZE)
2367                                                 goto nla_put_failure;
2368                                         error = err;
2369                                 }
2370                         }
2371                 } else
2372 #endif
2373                         if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2374                                 goto nla_put_failure;
2375         }
2376
2377         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2378                 goto nla_put_failure;
2379
2380         return nlmsg_end(skb, nlh);
2381
2382 nla_put_failure:
2383         nlmsg_cancel(skb, nlh);
2384         return -EMSGSIZE;
2385 }
2386
2387 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2388 {
2389         struct net *net = sock_net(in_skb->sk);
2390         struct rtmsg *rtm;
2391         struct nlattr *tb[RTA_MAX+1];
2392         struct rtable *rt = NULL;
2393         struct flowi4 fl4;
2394         __be32 dst = 0;
2395         __be32 src = 0;
2396         u32 iif;
2397         int err;
2398         int mark;
2399         struct sk_buff *skb;
2400
2401         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2402         if (err < 0)
2403                 goto errout;
2404
2405         rtm = nlmsg_data(nlh);
2406
2407         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2408         if (skb == NULL) {
2409                 err = -ENOBUFS;
2410                 goto errout;
2411         }
2412
2413         /* Reserve room for dummy headers, this skb can pass
2414            through good chunk of routing engine.
2415          */
2416         skb_reset_mac_header(skb);
2417         skb_reset_network_header(skb);
2418
2419         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2420         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2421         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2422
2423         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2424         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2425         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2426         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2427
2428         memset(&fl4, 0, sizeof(fl4));
2429         fl4.daddr = dst;
2430         fl4.saddr = src;
2431         fl4.flowi4_tos = rtm->rtm_tos;
2432         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2433         fl4.flowi4_mark = mark;
2434
2435         if (iif) {
2436                 struct net_device *dev;
2437
2438                 dev = __dev_get_by_index(net, iif);
2439                 if (dev == NULL) {
2440                         err = -ENODEV;
2441                         goto errout_free;
2442                 }
2443
2444                 skb->protocol   = htons(ETH_P_IP);
2445                 skb->dev        = dev;
2446                 skb->mark       = mark;
2447                 local_bh_disable();
2448                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2449                 local_bh_enable();
2450
2451                 rt = skb_rtable(skb);
2452                 if (err == 0 && rt->dst.error)
2453                         err = -rt->dst.error;
2454         } else {
2455                 rt = ip_route_output_key(net, &fl4);
2456
2457                 err = 0;
2458                 if (IS_ERR(rt))
2459                         err = PTR_ERR(rt);
2460         }
2461
2462         if (err)
2463                 goto errout_free;
2464
2465         skb_dst_set(skb, &rt->dst);
2466         if (rtm->rtm_flags & RTM_F_NOTIFY)
2467                 rt->rt_flags |= RTCF_NOTIFY;
2468
2469         err = rt_fill_info(net, dst, src, &fl4, skb,
2470                            NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2471                            RTM_NEWROUTE, 0, 0);
2472         if (err <= 0)
2473                 goto errout_free;
2474
2475         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2476 errout:
2477         return err;
2478
2479 errout_free:
2480         kfree_skb(skb);
2481         goto errout;
2482 }
2483
2484 void ip_rt_multicast_event(struct in_device *in_dev)
2485 {
2486         rt_cache_flush(dev_net(in_dev->dev));
2487 }
2488
2489 #ifdef CONFIG_SYSCTL
2490 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
2491 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
2492 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
2493 static int ip_rt_gc_elasticity __read_mostly    = 8;
2494
2495 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2496                                         void __user *buffer,
2497                                         size_t *lenp, loff_t *ppos)
2498 {
2499         struct net *net = (struct net *)__ctl->extra1;
2500
2501         if (write) {
2502                 rt_cache_flush(net);
2503                 fnhe_genid_bump(net);
2504                 return 0;
2505         }
2506
2507         return -EINVAL;
2508 }
2509
2510 static struct ctl_table ipv4_route_table[] = {
2511         {
2512                 .procname       = "gc_thresh",
2513                 .data           = &ipv4_dst_ops.gc_thresh,
2514                 .maxlen         = sizeof(int),
2515                 .mode           = 0644,
2516                 .proc_handler   = proc_dointvec,
2517         },
2518         {
2519                 .procname       = "max_size",
2520                 .data           = &ip_rt_max_size,
2521                 .maxlen         = sizeof(int),
2522                 .mode           = 0644,
2523                 .proc_handler   = proc_dointvec,
2524         },
2525         {
2526                 /*  Deprecated. Use gc_min_interval_ms */
2527
2528                 .procname       = "gc_min_interval",
2529                 .data           = &ip_rt_gc_min_interval,
2530                 .maxlen         = sizeof(int),
2531                 .mode           = 0644,
2532                 .proc_handler   = proc_dointvec_jiffies,
2533         },
2534         {
2535                 .procname       = "gc_min_interval_ms",
2536                 .data           = &ip_rt_gc_min_interval,
2537                 .maxlen         = sizeof(int),
2538                 .mode           = 0644,
2539                 .proc_handler   = proc_dointvec_ms_jiffies,
2540         },
2541         {
2542                 .procname       = "gc_timeout",
2543                 .data           = &ip_rt_gc_timeout,
2544                 .maxlen         = sizeof(int),
2545                 .mode           = 0644,
2546                 .proc_handler   = proc_dointvec_jiffies,
2547         },
2548         {
2549                 .procname       = "gc_interval",
2550                 .data           = &ip_rt_gc_interval,
2551                 .maxlen         = sizeof(int),
2552                 .mode           = 0644,
2553                 .proc_handler   = proc_dointvec_jiffies,
2554         },
2555         {
2556                 .procname       = "redirect_load",
2557                 .data           = &ip_rt_redirect_load,
2558                 .maxlen         = sizeof(int),
2559                 .mode           = 0644,
2560                 .proc_handler   = proc_dointvec,
2561         },
2562         {
2563                 .procname       = "redirect_number",
2564                 .data           = &ip_rt_redirect_number,
2565                 .maxlen         = sizeof(int),
2566                 .mode           = 0644,
2567                 .proc_handler   = proc_dointvec,
2568         },
2569         {
2570                 .procname       = "redirect_silence",
2571                 .data           = &ip_rt_redirect_silence,
2572                 .maxlen         = sizeof(int),
2573                 .mode           = 0644,
2574                 .proc_handler   = proc_dointvec,
2575         },
2576         {
2577                 .procname       = "error_cost",
2578                 .data           = &ip_rt_error_cost,
2579                 .maxlen         = sizeof(int),
2580                 .mode           = 0644,
2581                 .proc_handler   = proc_dointvec,
2582         },
2583         {
2584                 .procname       = "error_burst",
2585                 .data           = &ip_rt_error_burst,
2586                 .maxlen         = sizeof(int),
2587                 .mode           = 0644,
2588                 .proc_handler   = proc_dointvec,
2589         },
2590         {
2591                 .procname       = "gc_elasticity",
2592                 .data           = &ip_rt_gc_elasticity,
2593                 .maxlen         = sizeof(int),
2594                 .mode           = 0644,
2595                 .proc_handler   = proc_dointvec,
2596         },
2597         {
2598                 .procname       = "mtu_expires",
2599                 .data           = &ip_rt_mtu_expires,
2600                 .maxlen         = sizeof(int),
2601                 .mode           = 0644,
2602                 .proc_handler   = proc_dointvec_jiffies,
2603         },
2604         {
2605                 .procname       = "min_pmtu",
2606                 .data           = &ip_rt_min_pmtu,
2607                 .maxlen         = sizeof(int),
2608                 .mode           = 0644,
2609                 .proc_handler   = proc_dointvec,
2610         },
2611         {
2612                 .procname       = "min_adv_mss",
2613                 .data           = &ip_rt_min_advmss,
2614                 .maxlen         = sizeof(int),
2615                 .mode           = 0644,
2616                 .proc_handler   = proc_dointvec,
2617         },
2618         { }
2619 };
2620
2621 static struct ctl_table ipv4_route_flush_table[] = {
2622         {
2623                 .procname       = "flush",
2624                 .maxlen         = sizeof(int),
2625                 .mode           = 0200,
2626                 .proc_handler   = ipv4_sysctl_rtcache_flush,
2627         },
2628         { },
2629 };
2630
2631 static __net_init int sysctl_route_net_init(struct net *net)
2632 {
2633         struct ctl_table *tbl;
2634
2635         tbl = ipv4_route_flush_table;
2636         if (!net_eq(net, &init_net)) {
2637                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2638                 if (tbl == NULL)
2639                         goto err_dup;
2640
2641                 /* Don't export sysctls to unprivileged users */
2642                 if (net->user_ns != &init_user_ns)
2643                         tbl[0].procname = NULL;
2644         }
2645         tbl[0].extra1 = net;
2646
2647         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2648         if (net->ipv4.route_hdr == NULL)
2649                 goto err_reg;
2650         return 0;
2651
2652 err_reg:
2653         if (tbl != ipv4_route_flush_table)
2654                 kfree(tbl);
2655 err_dup:
2656         return -ENOMEM;
2657 }
2658
2659 static __net_exit void sysctl_route_net_exit(struct net *net)
2660 {
2661         struct ctl_table *tbl;
2662
2663         tbl = net->ipv4.route_hdr->ctl_table_arg;
2664         unregister_net_sysctl_table(net->ipv4.route_hdr);
2665         BUG_ON(tbl == ipv4_route_flush_table);
2666         kfree(tbl);
2667 }
2668
2669 static __net_initdata struct pernet_operations sysctl_route_ops = {
2670         .init = sysctl_route_net_init,
2671         .exit = sysctl_route_net_exit,
2672 };
2673 #endif
2674
2675 static __net_init int rt_genid_init(struct net *net)
2676 {
2677         atomic_set(&net->ipv4.rt_genid, 0);
2678         atomic_set(&net->fnhe_genid, 0);
2679         get_random_bytes(&net->ipv4.dev_addr_genid,
2680                          sizeof(net->ipv4.dev_addr_genid));
2681         return 0;
2682 }
2683
2684 static __net_initdata struct pernet_operations rt_genid_ops = {
2685         .init = rt_genid_init,
2686 };
2687
2688 static int __net_init ipv4_inetpeer_init(struct net *net)
2689 {
2690         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2691
2692         if (!bp)
2693                 return -ENOMEM;
2694         inet_peer_base_init(bp);
2695         net->ipv4.peers = bp;
2696         return 0;
2697 }
2698
2699 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2700 {
2701         struct inet_peer_base *bp = net->ipv4.peers;
2702
2703         net->ipv4.peers = NULL;
2704         inetpeer_invalidate_tree(bp);
2705         kfree(bp);
2706 }
2707
2708 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2709         .init   =       ipv4_inetpeer_init,
2710         .exit   =       ipv4_inetpeer_exit,
2711 };
2712
2713 #ifdef CONFIG_IP_ROUTE_CLASSID
2714 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2715 #endif /* CONFIG_IP_ROUTE_CLASSID */
2716
2717 int __init ip_rt_init(void)
2718 {
2719         int rc = 0;
2720
2721         ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2722         if (!ip_idents)
2723                 panic("IP: failed to allocate ip_idents\n");
2724
2725         prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2726
2727 #ifdef CONFIG_IP_ROUTE_CLASSID
2728         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2729         if (!ip_rt_acct)
2730                 panic("IP: failed to allocate ip_rt_acct\n");
2731 #endif
2732
2733         ipv4_dst_ops.kmem_cachep =
2734                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2735                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2736
2737         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2738
2739         if (dst_entries_init(&ipv4_dst_ops) < 0)
2740                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2741
2742         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2743                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2744
2745         ipv4_dst_ops.gc_thresh = ~0;
2746         ip_rt_max_size = INT_MAX;
2747
2748         devinet_init();
2749         ip_fib_init();
2750
2751         if (ip_rt_proc_init())
2752                 pr_err("Unable to create route proc files\n");
2753 #ifdef CONFIG_XFRM
2754         xfrm_init();
2755         xfrm4_init();
2756 #endif
2757         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2758
2759 #ifdef CONFIG_SYSCTL
2760         register_pernet_subsys(&sysctl_route_ops);
2761 #endif
2762         register_pernet_subsys(&rt_genid_ops);
2763         register_pernet_subsys(&ipv4_inetpeer_ops);
2764         return rc;
2765 }
2766
2767 #ifdef CONFIG_SYSCTL
2768 /*
2769  * We really need to sanitize the damn ipv4 init order, then all
2770  * this nonsense will go away.
2771  */
2772 void __init ip_static_sysctl_init(void)
2773 {
2774         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
2775 }
2776 #endif