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             skb->protocol == htons(ETH_P_IP) &&
1558             (IN_DEV_SHARED_MEDIA(out_dev) ||
1559              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1560                 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1561
1562         if (skb->protocol != htons(ETH_P_IP)) {
1563                 /* Not IP (i.e. ARP). Do not create route, if it is
1564                  * invalid for proxy arp. DNAT routes are always valid.
1565                  *
1566                  * Proxy arp feature have been extended to allow, ARP
1567                  * replies back to the same interface, to support
1568                  * Private VLAN switch technologies. See arp.c.
1569                  */
1570                 if (out_dev == in_dev &&
1571                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1572                         err = -EINVAL;
1573                         goto cleanup;
1574                 }
1575         }
1576
1577         fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1578         if (do_cache) {
1579                 if (fnhe != NULL)
1580                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1581                 else
1582                         rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1583
1584                 if (rt_cache_valid(rth)) {
1585                         skb_dst_set_noref(skb, &rth->dst);
1586                         goto out;
1587                 }
1588         }
1589
1590         rth = rt_dst_alloc(out_dev->dev,
1591                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1592                            IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1593         if (!rth) {
1594                 err = -ENOBUFS;
1595                 goto cleanup;
1596         }
1597
1598         rth->rt_genid = rt_genid_ipv4(dev_net(rth->dst.dev));
1599         rth->rt_flags = flags;
1600         rth->rt_type = res->type;
1601         rth->rt_is_input = 1;
1602         rth->rt_iif     = 0;
1603         rth->rt_pmtu    = 0;
1604         rth->rt_gateway = 0;
1605         rth->rt_uses_gateway = 0;
1606         INIT_LIST_HEAD(&rth->rt_uncached);
1607         RT_CACHE_STAT_INC(in_slow_tot);
1608
1609         rth->dst.input = ip_forward;
1610         rth->dst.output = ip_output;
1611
1612         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1613         skb_dst_set(skb, &rth->dst);
1614 out:
1615         err = 0;
1616  cleanup:
1617         return err;
1618 }
1619
1620 static int ip_mkroute_input(struct sk_buff *skb,
1621                             struct fib_result *res,
1622                             const struct flowi4 *fl4,
1623                             struct in_device *in_dev,
1624                             __be32 daddr, __be32 saddr, u32 tos)
1625 {
1626 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1627         if (res->fi && res->fi->fib_nhs > 1)
1628                 fib_select_multipath(res);
1629 #endif
1630
1631         /* create a routing cache entry */
1632         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1633 }
1634
1635 /*
1636  *      NOTE. We drop all the packets that has local source
1637  *      addresses, because every properly looped back packet
1638  *      must have correct destination already attached by output routine.
1639  *
1640  *      Such approach solves two big problems:
1641  *      1. Not simplex devices are handled properly.
1642  *      2. IP spoofing attempts are filtered with 100% of guarantee.
1643  *      called with rcu_read_lock()
1644  */
1645
1646 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1647                                u8 tos, struct net_device *dev)
1648 {
1649         struct fib_result res;
1650         struct in_device *in_dev = __in_dev_get_rcu(dev);
1651         struct flowi4   fl4;
1652         unsigned int    flags = 0;
1653         u32             itag = 0;
1654         struct rtable   *rth;
1655         int             err = -EINVAL;
1656         struct net    *net = dev_net(dev);
1657         bool do_cache;
1658
1659         /* IP on this device is disabled. */
1660
1661         if (!in_dev)
1662                 goto out;
1663
1664         /* Check for the most weird martians, which can be not detected
1665            by fib_lookup.
1666          */
1667
1668         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1669                 goto martian_source;
1670
1671         res.fi = NULL;
1672         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1673                 goto brd_input;
1674
1675         /* Accept zero addresses only to limited broadcast;
1676          * I even do not know to fix it or not. Waiting for complains :-)
1677          */
1678         if (ipv4_is_zeronet(saddr))
1679                 goto martian_source;
1680
1681         if (ipv4_is_zeronet(daddr))
1682                 goto martian_destination;
1683
1684         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1685          * and call it once if daddr or/and saddr are loopback addresses
1686          */
1687         if (ipv4_is_loopback(daddr)) {
1688                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1689                         goto martian_destination;
1690         } else if (ipv4_is_loopback(saddr)) {
1691                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1692                         goto martian_source;
1693         }
1694
1695         /*
1696          *      Now we are ready to route packet.
1697          */
1698         fl4.flowi4_oif = 0;
1699         fl4.flowi4_iif = dev->ifindex;
1700         fl4.flowi4_mark = skb->mark;
1701         fl4.flowi4_tos = tos;
1702         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1703         fl4.daddr = daddr;
1704         fl4.saddr = saddr;
1705         err = fib_lookup(net, &fl4, &res);
1706         if (err != 0) {
1707                 if (!IN_DEV_FORWARD(in_dev))
1708                         err = -EHOSTUNREACH;
1709                 goto no_route;
1710         }
1711
1712         if (res.type == RTN_BROADCAST)
1713                 goto brd_input;
1714
1715         if (res.type == RTN_LOCAL) {
1716                 err = fib_validate_source(skb, saddr, daddr, tos,
1717                                           0, dev, in_dev, &itag);
1718                 if (err < 0)
1719                         goto martian_source_keep_err;
1720                 goto local_input;
1721         }
1722
1723         if (!IN_DEV_FORWARD(in_dev)) {
1724                 err = -EHOSTUNREACH;
1725                 goto no_route;
1726         }
1727         if (res.type != RTN_UNICAST)
1728                 goto martian_destination;
1729
1730         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1731 out:    return err;
1732
1733 brd_input:
1734         if (skb->protocol != htons(ETH_P_IP))
1735                 goto e_inval;
1736
1737         if (!ipv4_is_zeronet(saddr)) {
1738                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1739                                           in_dev, &itag);
1740                 if (err < 0)
1741                         goto martian_source_keep_err;
1742         }
1743         flags |= RTCF_BROADCAST;
1744         res.type = RTN_BROADCAST;
1745         RT_CACHE_STAT_INC(in_brd);
1746
1747 local_input:
1748         do_cache = false;
1749         if (res.fi) {
1750                 if (!itag) {
1751                         rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1752                         if (rt_cache_valid(rth)) {
1753                                 skb_dst_set_noref(skb, &rth->dst);
1754                                 err = 0;
1755                                 goto out;
1756                         }
1757                         do_cache = true;
1758                 }
1759         }
1760
1761         rth = rt_dst_alloc(net->loopback_dev,
1762                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1763         if (!rth)
1764                 goto e_nobufs;
1765
1766         rth->dst.input= ip_local_deliver;
1767         rth->dst.output= ip_rt_bug;
1768 #ifdef CONFIG_IP_ROUTE_CLASSID
1769         rth->dst.tclassid = itag;
1770 #endif
1771
1772         rth->rt_genid = rt_genid_ipv4(net);
1773         rth->rt_flags   = flags|RTCF_LOCAL;
1774         rth->rt_type    = res.type;
1775         rth->rt_is_input = 1;
1776         rth->rt_iif     = 0;
1777         rth->rt_pmtu    = 0;
1778         rth->rt_gateway = 0;
1779         rth->rt_uses_gateway = 0;
1780         INIT_LIST_HEAD(&rth->rt_uncached);
1781         RT_CACHE_STAT_INC(in_slow_tot);
1782         if (res.type == RTN_UNREACHABLE) {
1783                 rth->dst.input= ip_error;
1784                 rth->dst.error= -err;
1785                 rth->rt_flags   &= ~RTCF_LOCAL;
1786         }
1787         if (do_cache) {
1788                 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1789                         rth->dst.flags |= DST_NOCACHE;
1790                         rt_add_uncached_list(rth);
1791                 }
1792         }
1793         skb_dst_set(skb, &rth->dst);
1794         err = 0;
1795         goto out;
1796
1797 no_route:
1798         RT_CACHE_STAT_INC(in_no_route);
1799         res.type = RTN_UNREACHABLE;
1800         res.fi = NULL;
1801         goto local_input;
1802
1803         /*
1804          *      Do not cache martian addresses: they should be logged (RFC1812)
1805          */
1806 martian_destination:
1807         RT_CACHE_STAT_INC(in_martian_dst);
1808 #ifdef CONFIG_IP_ROUTE_VERBOSE
1809         if (IN_DEV_LOG_MARTIANS(in_dev))
1810                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1811                                      &daddr, &saddr, dev->name);
1812 #endif
1813
1814 e_inval:
1815         err = -EINVAL;
1816         goto out;
1817
1818 e_nobufs:
1819         err = -ENOBUFS;
1820         goto out;
1821
1822 martian_source:
1823         err = -EINVAL;
1824 martian_source_keep_err:
1825         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1826         goto out;
1827 }
1828
1829 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1830                          u8 tos, struct net_device *dev)
1831 {
1832         int res;
1833
1834         rcu_read_lock();
1835
1836         /* Multicast recognition logic is moved from route cache to here.
1837            The problem was that too many Ethernet cards have broken/missing
1838            hardware multicast filters :-( As result the host on multicasting
1839            network acquires a lot of useless route cache entries, sort of
1840            SDR messages from all the world. Now we try to get rid of them.
1841            Really, provided software IP multicast filter is organized
1842            reasonably (at least, hashed), it does not result in a slowdown
1843            comparing with route cache reject entries.
1844            Note, that multicast routers are not affected, because
1845            route cache entry is created eventually.
1846          */
1847         if (ipv4_is_multicast(daddr)) {
1848                 struct in_device *in_dev = __in_dev_get_rcu(dev);
1849
1850                 if (in_dev) {
1851                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1852                                                   ip_hdr(skb)->protocol);
1853                         if (our
1854 #ifdef CONFIG_IP_MROUTE
1855                                 ||
1856                             (!ipv4_is_local_multicast(daddr) &&
1857                              IN_DEV_MFORWARD(in_dev))
1858 #endif
1859                            ) {
1860                                 int res = ip_route_input_mc(skb, daddr, saddr,
1861                                                             tos, dev, our);
1862                                 rcu_read_unlock();
1863                                 return res;
1864                         }
1865                 }
1866                 rcu_read_unlock();
1867                 return -EINVAL;
1868         }
1869         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1870         rcu_read_unlock();
1871         return res;
1872 }
1873 EXPORT_SYMBOL(ip_route_input_noref);
1874
1875 /* called with rcu_read_lock() */
1876 static struct rtable *__mkroute_output(const struct fib_result *res,
1877                                        const struct flowi4 *fl4, int orig_oif,
1878                                        struct net_device *dev_out,
1879                                        unsigned int flags)
1880 {
1881         struct fib_info *fi = res->fi;
1882         struct fib_nh_exception *fnhe;
1883         struct in_device *in_dev;
1884         u16 type = res->type;
1885         struct rtable *rth;
1886         bool do_cache;
1887
1888         in_dev = __in_dev_get_rcu(dev_out);
1889         if (!in_dev)
1890                 return ERR_PTR(-EINVAL);
1891
1892         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1893                 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1894                         return ERR_PTR(-EINVAL);
1895
1896         if (ipv4_is_lbcast(fl4->daddr))
1897                 type = RTN_BROADCAST;
1898         else if (ipv4_is_multicast(fl4->daddr))
1899                 type = RTN_MULTICAST;
1900         else if (ipv4_is_zeronet(fl4->daddr))
1901                 return ERR_PTR(-EINVAL);
1902
1903         if (dev_out->flags & IFF_LOOPBACK)
1904                 flags |= RTCF_LOCAL;
1905
1906         do_cache = true;
1907         if (type == RTN_BROADCAST) {
1908                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1909                 fi = NULL;
1910         } else if (type == RTN_MULTICAST) {
1911                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1912                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1913                                      fl4->flowi4_proto))
1914                         flags &= ~RTCF_LOCAL;
1915                 else
1916                         do_cache = false;
1917                 /* If multicast route do not exist use
1918                  * default one, but do not gateway in this case.
1919                  * Yes, it is hack.
1920                  */
1921                 if (fi && res->prefixlen < 4)
1922                         fi = NULL;
1923         }
1924
1925         fnhe = NULL;
1926         do_cache &= fi != NULL;
1927         if (do_cache) {
1928                 struct rtable __rcu **prth;
1929                 struct fib_nh *nh = &FIB_RES_NH(*res);
1930
1931                 fnhe = find_exception(nh, fl4->daddr);
1932                 if (fnhe)
1933                         prth = &fnhe->fnhe_rth_output;
1934                 else {
1935                         if (unlikely(fl4->flowi4_flags &
1936                                      FLOWI_FLAG_KNOWN_NH &&
1937                                      !(nh->nh_gw &&
1938                                        nh->nh_scope == RT_SCOPE_LINK))) {
1939                                 do_cache = false;
1940                                 goto add;
1941                         }
1942                         prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
1943                 }
1944                 rth = rcu_dereference(*prth);
1945                 if (rt_cache_valid(rth)) {
1946                         dst_hold(&rth->dst);
1947                         return rth;
1948                 }
1949         }
1950
1951 add:
1952         rth = rt_dst_alloc(dev_out,
1953                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1954                            IN_DEV_CONF_GET(in_dev, NOXFRM),
1955                            do_cache);
1956         if (!rth)
1957                 return ERR_PTR(-ENOBUFS);
1958
1959         rth->dst.output = ip_output;
1960
1961         rth->rt_genid = rt_genid_ipv4(dev_net(dev_out));
1962         rth->rt_flags   = flags;
1963         rth->rt_type    = type;
1964         rth->rt_is_input = 0;
1965         rth->rt_iif     = orig_oif ? : 0;
1966         rth->rt_pmtu    = 0;
1967         rth->rt_gateway = 0;
1968         rth->rt_uses_gateway = 0;
1969         INIT_LIST_HEAD(&rth->rt_uncached);
1970
1971         RT_CACHE_STAT_INC(out_slow_tot);
1972
1973         if (flags & RTCF_LOCAL)
1974                 rth->dst.input = ip_local_deliver;
1975         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1976                 if (flags & RTCF_LOCAL &&
1977                     !(dev_out->flags & IFF_LOOPBACK)) {
1978                         rth->dst.output = ip_mc_output;
1979                         RT_CACHE_STAT_INC(out_slow_mc);
1980                 }
1981 #ifdef CONFIG_IP_MROUTE
1982                 if (type == RTN_MULTICAST) {
1983                         if (IN_DEV_MFORWARD(in_dev) &&
1984                             !ipv4_is_local_multicast(fl4->daddr)) {
1985                                 rth->dst.input = ip_mr_input;
1986                                 rth->dst.output = ip_mc_output;
1987                         }
1988                 }
1989 #endif
1990         }
1991
1992         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
1993
1994         return rth;
1995 }
1996
1997 /*
1998  * Major route resolver routine.
1999  */
2000
2001 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
2002 {
2003         struct net_device *dev_out = NULL;
2004         __u8 tos = RT_FL_TOS(fl4);
2005         unsigned int flags = 0;
2006         struct fib_result res;
2007         struct rtable *rth;
2008         int orig_oif;
2009
2010         res.tclassid    = 0;
2011         res.fi          = NULL;
2012         res.table       = NULL;
2013
2014         orig_oif = fl4->flowi4_oif;
2015
2016         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2017         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2018         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2019                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2020
2021         rcu_read_lock();
2022         if (fl4->saddr) {
2023                 rth = ERR_PTR(-EINVAL);
2024                 if (ipv4_is_multicast(fl4->saddr) ||
2025                     ipv4_is_lbcast(fl4->saddr) ||
2026                     ipv4_is_zeronet(fl4->saddr))
2027                         goto out;
2028
2029                 /* I removed check for oif == dev_out->oif here.
2030                    It was wrong for two reasons:
2031                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2032                       is assigned to multiple interfaces.
2033                    2. Moreover, we are allowed to send packets with saddr
2034                       of another iface. --ANK
2035                  */
2036
2037                 if (fl4->flowi4_oif == 0 &&
2038                     (ipv4_is_multicast(fl4->daddr) ||
2039                      ipv4_is_lbcast(fl4->daddr))) {
2040                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2041                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2042                         if (dev_out == NULL)
2043                                 goto out;
2044
2045                         /* Special hack: user can direct multicasts
2046                            and limited broadcast via necessary interface
2047                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2048                            This hack is not just for fun, it allows
2049                            vic,vat and friends to work.
2050                            They bind socket to loopback, set ttl to zero
2051                            and expect that it will work.
2052                            From the viewpoint of routing cache they are broken,
2053                            because we are not allowed to build multicast path
2054                            with loopback source addr (look, routing cache
2055                            cannot know, that ttl is zero, so that packet
2056                            will not leave this host and route is valid).
2057                            Luckily, this hack is good workaround.
2058                          */
2059
2060                         fl4->flowi4_oif = dev_out->ifindex;
2061                         goto make_route;
2062                 }
2063
2064                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2065                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2066                         if (!__ip_dev_find(net, fl4->saddr, false))
2067                                 goto out;
2068                 }
2069         }
2070
2071
2072         if (fl4->flowi4_oif) {
2073                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2074                 rth = ERR_PTR(-ENODEV);
2075                 if (dev_out == NULL)
2076                         goto out;
2077
2078                 /* RACE: Check return value of inet_select_addr instead. */
2079                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2080                         rth = ERR_PTR(-ENETUNREACH);
2081                         goto out;
2082                 }
2083                 if (ipv4_is_local_multicast(fl4->daddr) ||
2084                     ipv4_is_lbcast(fl4->daddr)) {
2085                         if (!fl4->saddr)
2086                                 fl4->saddr = inet_select_addr(dev_out, 0,
2087                                                               RT_SCOPE_LINK);
2088                         goto make_route;
2089                 }
2090                 if (!fl4->saddr) {
2091                         if (ipv4_is_multicast(fl4->daddr))
2092                                 fl4->saddr = inet_select_addr(dev_out, 0,
2093                                                               fl4->flowi4_scope);
2094                         else if (!fl4->daddr)
2095                                 fl4->saddr = inet_select_addr(dev_out, 0,
2096                                                               RT_SCOPE_HOST);
2097                 }
2098         }
2099
2100         if (!fl4->daddr) {
2101                 fl4->daddr = fl4->saddr;
2102                 if (!fl4->daddr)
2103                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2104                 dev_out = net->loopback_dev;
2105                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2106                 res.type = RTN_LOCAL;
2107                 flags |= RTCF_LOCAL;
2108                 goto make_route;
2109         }
2110
2111         if (fib_lookup(net, fl4, &res)) {
2112                 res.fi = NULL;
2113                 res.table = NULL;
2114                 if (fl4->flowi4_oif) {
2115                         /* Apparently, routing tables are wrong. Assume,
2116                            that the destination is on link.
2117
2118                            WHY? DW.
2119                            Because we are allowed to send to iface
2120                            even if it has NO routes and NO assigned
2121                            addresses. When oif is specified, routing
2122                            tables are looked up with only one purpose:
2123                            to catch if destination is gatewayed, rather than
2124                            direct. Moreover, if MSG_DONTROUTE is set,
2125                            we send packet, ignoring both routing tables
2126                            and ifaddr state. --ANK
2127
2128
2129                            We could make it even if oif is unknown,
2130                            likely IPv6, but we do not.
2131                          */
2132
2133                         if (fl4->saddr == 0)
2134                                 fl4->saddr = inet_select_addr(dev_out, 0,
2135                                                               RT_SCOPE_LINK);
2136                         res.type = RTN_UNICAST;
2137                         goto make_route;
2138                 }
2139                 rth = ERR_PTR(-ENETUNREACH);
2140                 goto out;
2141         }
2142
2143         if (res.type == RTN_LOCAL) {
2144                 if (!fl4->saddr) {
2145                         if (res.fi->fib_prefsrc)
2146                                 fl4->saddr = res.fi->fib_prefsrc;
2147                         else
2148                                 fl4->saddr = fl4->daddr;
2149                 }
2150                 dev_out = net->loopback_dev;
2151                 fl4->flowi4_oif = dev_out->ifindex;
2152                 flags |= RTCF_LOCAL;
2153                 goto make_route;
2154         }
2155
2156 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2157         if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2158                 fib_select_multipath(&res);
2159         else
2160 #endif
2161         if (!res.prefixlen &&
2162             res.table->tb_num_default > 1 &&
2163             res.type == RTN_UNICAST && !fl4->flowi4_oif)
2164                 fib_select_default(&res);
2165
2166         if (!fl4->saddr)
2167                 fl4->saddr = FIB_RES_PREFSRC(net, res);
2168
2169         dev_out = FIB_RES_DEV(res);
2170         fl4->flowi4_oif = dev_out->ifindex;
2171
2172
2173 make_route:
2174         rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2175
2176 out:
2177         rcu_read_unlock();
2178         return rth;
2179 }
2180 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2181
2182 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2183 {
2184         return NULL;
2185 }
2186
2187 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2188 {
2189         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2190
2191         return mtu ? : dst->dev->mtu;
2192 }
2193
2194 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2195                                           struct sk_buff *skb, u32 mtu)
2196 {
2197 }
2198
2199 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2200                                        struct sk_buff *skb)
2201 {
2202 }
2203
2204 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2205                                           unsigned long old)
2206 {
2207         return NULL;
2208 }
2209
2210 static struct dst_ops ipv4_dst_blackhole_ops = {
2211         .family                 =       AF_INET,
2212         .protocol               =       cpu_to_be16(ETH_P_IP),
2213         .check                  =       ipv4_blackhole_dst_check,
2214         .mtu                    =       ipv4_blackhole_mtu,
2215         .default_advmss         =       ipv4_default_advmss,
2216         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2217         .redirect               =       ipv4_rt_blackhole_redirect,
2218         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
2219         .neigh_lookup           =       ipv4_neigh_lookup,
2220 };
2221
2222 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2223 {
2224         struct rtable *ort = (struct rtable *) dst_orig;
2225         struct rtable *rt;
2226
2227         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2228         if (rt) {
2229                 struct dst_entry *new = &rt->dst;
2230
2231                 new->__use = 1;
2232                 new->input = dst_discard;
2233                 new->output = dst_discard_sk;
2234
2235                 new->dev = ort->dst.dev;
2236                 if (new->dev)
2237                         dev_hold(new->dev);
2238
2239                 rt->rt_is_input = ort->rt_is_input;
2240                 rt->rt_iif = ort->rt_iif;
2241                 rt->rt_pmtu = ort->rt_pmtu;
2242
2243                 rt->rt_genid = rt_genid_ipv4(net);
2244                 rt->rt_flags = ort->rt_flags;
2245                 rt->rt_type = ort->rt_type;
2246                 rt->rt_gateway = ort->rt_gateway;
2247                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2248
2249                 INIT_LIST_HEAD(&rt->rt_uncached);
2250
2251                 dst_free(new);
2252         }
2253
2254         dst_release(dst_orig);
2255
2256         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2257 }
2258
2259 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2260                                     struct sock *sk)
2261 {
2262         struct rtable *rt = __ip_route_output_key(net, flp4);
2263
2264         if (IS_ERR(rt))
2265                 return rt;
2266
2267         if (flp4->flowi4_proto)
2268                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2269                                                         flowi4_to_flowi(flp4),
2270                                                         sk, 0);
2271
2272         return rt;
2273 }
2274 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2275
2276 static int rt_fill_info(struct net *net,  __be32 dst, __be32 src,
2277                         struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2278                         u32 seq, int event, int nowait, unsigned int flags)
2279 {
2280         struct rtable *rt = skb_rtable(skb);
2281         struct rtmsg *r;
2282         struct nlmsghdr *nlh;
2283         unsigned long expires = 0;
2284         u32 error;
2285         u32 metrics[RTAX_MAX];
2286
2287         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2288         if (nlh == NULL)
2289                 return -EMSGSIZE;
2290
2291         r = nlmsg_data(nlh);
2292         r->rtm_family    = AF_INET;
2293         r->rtm_dst_len  = 32;
2294         r->rtm_src_len  = 0;
2295         r->rtm_tos      = fl4->flowi4_tos;
2296         r->rtm_table    = RT_TABLE_MAIN;
2297         if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2298                 goto nla_put_failure;
2299         r->rtm_type     = rt->rt_type;
2300         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2301         r->rtm_protocol = RTPROT_UNSPEC;
2302         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2303         if (rt->rt_flags & RTCF_NOTIFY)
2304                 r->rtm_flags |= RTM_F_NOTIFY;
2305         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2306                 r->rtm_flags |= RTCF_DOREDIRECT;
2307
2308         if (nla_put_be32(skb, RTA_DST, dst))
2309                 goto nla_put_failure;
2310         if (src) {
2311                 r->rtm_src_len = 32;
2312                 if (nla_put_be32(skb, RTA_SRC, src))
2313                         goto nla_put_failure;
2314         }
2315         if (rt->dst.dev &&
2316             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2317                 goto nla_put_failure;
2318 #ifdef CONFIG_IP_ROUTE_CLASSID
2319         if (rt->dst.tclassid &&
2320             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2321                 goto nla_put_failure;
2322 #endif
2323         if (!rt_is_input_route(rt) &&
2324             fl4->saddr != src) {
2325                 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2326                         goto nla_put_failure;
2327         }
2328         if (rt->rt_uses_gateway &&
2329             nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2330                 goto nla_put_failure;
2331
2332         expires = rt->dst.expires;
2333         if (expires) {
2334                 unsigned long now = jiffies;
2335
2336                 if (time_before(now, expires))
2337                         expires -= now;
2338                 else
2339                         expires = 0;
2340         }
2341
2342         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2343         if (rt->rt_pmtu && expires)
2344                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2345         if (rtnetlink_put_metrics(skb, metrics) < 0)
2346                 goto nla_put_failure;
2347
2348         if (fl4->flowi4_mark &&
2349             nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2350                 goto nla_put_failure;
2351
2352         error = rt->dst.error;
2353
2354         if (rt_is_input_route(rt)) {
2355 #ifdef CONFIG_IP_MROUTE
2356                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2357                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2358                         int err = ipmr_get_route(net, skb,
2359                                                  fl4->saddr, fl4->daddr,
2360                                                  r, nowait);
2361                         if (err <= 0) {
2362                                 if (!nowait) {
2363                                         if (err == 0)
2364                                                 return 0;
2365                                         goto nla_put_failure;
2366                                 } else {
2367                                         if (err == -EMSGSIZE)
2368                                                 goto nla_put_failure;
2369                                         error = err;
2370                                 }
2371                         }
2372                 } else
2373 #endif
2374                         if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2375                                 goto nla_put_failure;
2376         }
2377
2378         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2379                 goto nla_put_failure;
2380
2381         return nlmsg_end(skb, nlh);
2382
2383 nla_put_failure:
2384         nlmsg_cancel(skb, nlh);
2385         return -EMSGSIZE;
2386 }
2387
2388 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2389 {
2390         struct net *net = sock_net(in_skb->sk);
2391         struct rtmsg *rtm;
2392         struct nlattr *tb[RTA_MAX+1];
2393         struct rtable *rt = NULL;
2394         struct flowi4 fl4;
2395         __be32 dst = 0;
2396         __be32 src = 0;
2397         u32 iif;
2398         int err;
2399         int mark;
2400         struct sk_buff *skb;
2401
2402         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2403         if (err < 0)
2404                 goto errout;
2405
2406         rtm = nlmsg_data(nlh);
2407
2408         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2409         if (skb == NULL) {
2410                 err = -ENOBUFS;
2411                 goto errout;
2412         }
2413
2414         /* Reserve room for dummy headers, this skb can pass
2415            through good chunk of routing engine.
2416          */
2417         skb_reset_mac_header(skb);
2418         skb_reset_network_header(skb);
2419
2420         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2421         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2422         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2423
2424         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2425         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2426         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2427         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2428
2429         memset(&fl4, 0, sizeof(fl4));
2430         fl4.daddr = dst;
2431         fl4.saddr = src;
2432         fl4.flowi4_tos = rtm->rtm_tos;
2433         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2434         fl4.flowi4_mark = mark;
2435
2436         if (iif) {
2437                 struct net_device *dev;
2438
2439                 dev = __dev_get_by_index(net, iif);
2440                 if (dev == NULL) {
2441                         err = -ENODEV;
2442                         goto errout_free;
2443                 }
2444
2445                 skb->protocol   = htons(ETH_P_IP);
2446                 skb->dev        = dev;
2447                 skb->mark       = mark;
2448                 local_bh_disable();
2449                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2450                 local_bh_enable();
2451
2452                 rt = skb_rtable(skb);
2453                 if (err == 0 && rt->dst.error)
2454                         err = -rt->dst.error;
2455         } else {
2456                 rt = ip_route_output_key(net, &fl4);
2457
2458                 err = 0;
2459                 if (IS_ERR(rt))
2460                         err = PTR_ERR(rt);
2461         }
2462
2463         if (err)
2464                 goto errout_free;
2465
2466         skb_dst_set(skb, &rt->dst);
2467         if (rtm->rtm_flags & RTM_F_NOTIFY)
2468                 rt->rt_flags |= RTCF_NOTIFY;
2469
2470         err = rt_fill_info(net, dst, src, &fl4, skb,
2471                            NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2472                            RTM_NEWROUTE, 0, 0);
2473         if (err <= 0)
2474                 goto errout_free;
2475
2476         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2477 errout:
2478         return err;
2479
2480 errout_free:
2481         kfree_skb(skb);
2482         goto errout;
2483 }
2484
2485 void ip_rt_multicast_event(struct in_device *in_dev)
2486 {
2487         rt_cache_flush(dev_net(in_dev->dev));
2488 }
2489
2490 #ifdef CONFIG_SYSCTL
2491 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
2492 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
2493 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
2494 static int ip_rt_gc_elasticity __read_mostly    = 8;
2495
2496 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2497                                         void __user *buffer,
2498                                         size_t *lenp, loff_t *ppos)
2499 {
2500         struct net *net = (struct net *)__ctl->extra1;
2501
2502         if (write) {
2503                 rt_cache_flush(net);
2504                 fnhe_genid_bump(net);
2505                 return 0;
2506         }
2507
2508         return -EINVAL;
2509 }
2510
2511 static struct ctl_table ipv4_route_table[] = {
2512         {
2513                 .procname       = "gc_thresh",
2514                 .data           = &ipv4_dst_ops.gc_thresh,
2515                 .maxlen         = sizeof(int),
2516                 .mode           = 0644,
2517                 .proc_handler   = proc_dointvec,
2518         },
2519         {
2520                 .procname       = "max_size",
2521                 .data           = &ip_rt_max_size,
2522                 .maxlen         = sizeof(int),
2523                 .mode           = 0644,
2524                 .proc_handler   = proc_dointvec,
2525         },
2526         {
2527                 /*  Deprecated. Use gc_min_interval_ms */
2528
2529                 .procname       = "gc_min_interval",
2530                 .data           = &ip_rt_gc_min_interval,
2531                 .maxlen         = sizeof(int),
2532                 .mode           = 0644,
2533                 .proc_handler   = proc_dointvec_jiffies,
2534         },
2535         {
2536                 .procname       = "gc_min_interval_ms",
2537                 .data           = &ip_rt_gc_min_interval,
2538                 .maxlen         = sizeof(int),
2539                 .mode           = 0644,
2540                 .proc_handler   = proc_dointvec_ms_jiffies,
2541         },
2542         {
2543                 .procname       = "gc_timeout",
2544                 .data           = &ip_rt_gc_timeout,
2545                 .maxlen         = sizeof(int),
2546                 .mode           = 0644,
2547                 .proc_handler   = proc_dointvec_jiffies,
2548         },
2549         {
2550                 .procname       = "gc_interval",
2551                 .data           = &ip_rt_gc_interval,
2552                 .maxlen         = sizeof(int),
2553                 .mode           = 0644,
2554                 .proc_handler   = proc_dointvec_jiffies,
2555         },
2556         {
2557                 .procname       = "redirect_load",
2558                 .data           = &ip_rt_redirect_load,
2559                 .maxlen         = sizeof(int),
2560                 .mode           = 0644,
2561                 .proc_handler   = proc_dointvec,
2562         },
2563         {
2564                 .procname       = "redirect_number",
2565                 .data           = &ip_rt_redirect_number,
2566                 .maxlen         = sizeof(int),
2567                 .mode           = 0644,
2568                 .proc_handler   = proc_dointvec,
2569         },
2570         {
2571                 .procname       = "redirect_silence",
2572                 .data           = &ip_rt_redirect_silence,
2573                 .maxlen         = sizeof(int),
2574                 .mode           = 0644,
2575                 .proc_handler   = proc_dointvec,
2576         },
2577         {
2578                 .procname       = "error_cost",
2579                 .data           = &ip_rt_error_cost,
2580                 .maxlen         = sizeof(int),
2581                 .mode           = 0644,
2582                 .proc_handler   = proc_dointvec,
2583         },
2584         {
2585                 .procname       = "error_burst",
2586                 .data           = &ip_rt_error_burst,
2587                 .maxlen         = sizeof(int),
2588                 .mode           = 0644,
2589                 .proc_handler   = proc_dointvec,
2590         },
2591         {
2592                 .procname       = "gc_elasticity",
2593                 .data           = &ip_rt_gc_elasticity,
2594                 .maxlen         = sizeof(int),
2595                 .mode           = 0644,
2596                 .proc_handler   = proc_dointvec,
2597         },
2598         {
2599                 .procname       = "mtu_expires",
2600                 .data           = &ip_rt_mtu_expires,
2601                 .maxlen         = sizeof(int),
2602                 .mode           = 0644,
2603                 .proc_handler   = proc_dointvec_jiffies,
2604         },
2605         {
2606                 .procname       = "min_pmtu",
2607                 .data           = &ip_rt_min_pmtu,
2608                 .maxlen         = sizeof(int),
2609                 .mode           = 0644,
2610                 .proc_handler   = proc_dointvec,
2611         },
2612         {
2613                 .procname       = "min_adv_mss",
2614                 .data           = &ip_rt_min_advmss,
2615                 .maxlen         = sizeof(int),
2616                 .mode           = 0644,
2617                 .proc_handler   = proc_dointvec,
2618         },
2619         { }
2620 };
2621
2622 static struct ctl_table ipv4_route_flush_table[] = {
2623         {
2624                 .procname       = "flush",
2625                 .maxlen         = sizeof(int),
2626                 .mode           = 0200,
2627                 .proc_handler   = ipv4_sysctl_rtcache_flush,
2628         },
2629         { },
2630 };
2631
2632 static __net_init int sysctl_route_net_init(struct net *net)
2633 {
2634         struct ctl_table *tbl;
2635
2636         tbl = ipv4_route_flush_table;
2637         if (!net_eq(net, &init_net)) {
2638                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2639                 if (tbl == NULL)
2640                         goto err_dup;
2641
2642                 /* Don't export sysctls to unprivileged users */
2643                 if (net->user_ns != &init_user_ns)
2644                         tbl[0].procname = NULL;
2645         }
2646         tbl[0].extra1 = net;
2647
2648         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2649         if (net->ipv4.route_hdr == NULL)
2650                 goto err_reg;
2651         return 0;
2652
2653 err_reg:
2654         if (tbl != ipv4_route_flush_table)
2655                 kfree(tbl);
2656 err_dup:
2657         return -ENOMEM;
2658 }
2659
2660 static __net_exit void sysctl_route_net_exit(struct net *net)
2661 {
2662         struct ctl_table *tbl;
2663
2664         tbl = net->ipv4.route_hdr->ctl_table_arg;
2665         unregister_net_sysctl_table(net->ipv4.route_hdr);
2666         BUG_ON(tbl == ipv4_route_flush_table);
2667         kfree(tbl);
2668 }
2669
2670 static __net_initdata struct pernet_operations sysctl_route_ops = {
2671         .init = sysctl_route_net_init,
2672         .exit = sysctl_route_net_exit,
2673 };
2674 #endif
2675
2676 static __net_init int rt_genid_init(struct net *net)
2677 {
2678         atomic_set(&net->ipv4.rt_genid, 0);
2679         atomic_set(&net->fnhe_genid, 0);
2680         get_random_bytes(&net->ipv4.dev_addr_genid,
2681                          sizeof(net->ipv4.dev_addr_genid));
2682         return 0;
2683 }
2684
2685 static __net_initdata struct pernet_operations rt_genid_ops = {
2686         .init = rt_genid_init,
2687 };
2688
2689 static int __net_init ipv4_inetpeer_init(struct net *net)
2690 {
2691         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2692
2693         if (!bp)
2694                 return -ENOMEM;
2695         inet_peer_base_init(bp);
2696         net->ipv4.peers = bp;
2697         return 0;
2698 }
2699
2700 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2701 {
2702         struct inet_peer_base *bp = net->ipv4.peers;
2703
2704         net->ipv4.peers = NULL;
2705         inetpeer_invalidate_tree(bp);
2706         kfree(bp);
2707 }
2708
2709 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2710         .init   =       ipv4_inetpeer_init,
2711         .exit   =       ipv4_inetpeer_exit,
2712 };
2713
2714 #ifdef CONFIG_IP_ROUTE_CLASSID
2715 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2716 #endif /* CONFIG_IP_ROUTE_CLASSID */
2717
2718 int __init ip_rt_init(void)
2719 {
2720         int rc = 0;
2721
2722         ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2723         if (!ip_idents)
2724                 panic("IP: failed to allocate ip_idents\n");
2725
2726         prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2727
2728 #ifdef CONFIG_IP_ROUTE_CLASSID
2729         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2730         if (!ip_rt_acct)
2731                 panic("IP: failed to allocate ip_rt_acct\n");
2732 #endif
2733
2734         ipv4_dst_ops.kmem_cachep =
2735                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2736                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2737
2738         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2739
2740         if (dst_entries_init(&ipv4_dst_ops) < 0)
2741                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2742
2743         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2744                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2745
2746         ipv4_dst_ops.gc_thresh = ~0;
2747         ip_rt_max_size = INT_MAX;
2748
2749         devinet_init();
2750         ip_fib_init();
2751
2752         if (ip_rt_proc_init())
2753                 pr_err("Unable to create route proc files\n");
2754 #ifdef CONFIG_XFRM
2755         xfrm_init();
2756         xfrm4_init();
2757 #endif
2758         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2759
2760 #ifdef CONFIG_SYSCTL
2761         register_pernet_subsys(&sysctl_route_ops);
2762 #endif
2763         register_pernet_subsys(&rt_genid_ops);
2764         register_pernet_subsys(&ipv4_inetpeer_ops);
2765         return rc;
2766 }
2767
2768 #ifdef CONFIG_SYSCTL
2769 /*
2770  * We really need to sanitize the damn ipv4 init order, then all
2771  * this nonsense will go away.
2772  */
2773 void __init ip_static_sysctl_init(void)
2774 {
2775         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
2776 }
2777 #endif