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 struct uncached_list {
1329         spinlock_t              lock;
1330         struct list_head        head;
1331 };
1332
1333 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1334
1335 static void rt_add_uncached_list(struct rtable *rt)
1336 {
1337         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1338
1339         rt->rt_uncached_list = ul;
1340
1341         spin_lock_bh(&ul->lock);
1342         list_add_tail(&rt->rt_uncached, &ul->head);
1343         spin_unlock_bh(&ul->lock);
1344 }
1345
1346 static void ipv4_dst_destroy(struct dst_entry *dst)
1347 {
1348         struct rtable *rt = (struct rtable *) dst;
1349
1350         if (!list_empty(&rt->rt_uncached)) {
1351                 struct uncached_list *ul = rt->rt_uncached_list;
1352
1353                 spin_lock_bh(&ul->lock);
1354                 list_del(&rt->rt_uncached);
1355                 spin_unlock_bh(&ul->lock);
1356         }
1357 }
1358
1359 void rt_flush_dev(struct net_device *dev)
1360 {
1361         struct net *net = dev_net(dev);
1362         struct rtable *rt;
1363         int cpu;
1364
1365         for_each_possible_cpu(cpu) {
1366                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1367
1368                 spin_lock_bh(&ul->lock);
1369                 list_for_each_entry(rt, &ul->head, rt_uncached) {
1370                         if (rt->dst.dev != dev)
1371                                 continue;
1372                         rt->dst.dev = net->loopback_dev;
1373                         dev_hold(rt->dst.dev);
1374                         dev_put(dev);
1375                 }
1376                 spin_unlock_bh(&ul->lock);
1377         }
1378 }
1379
1380 static bool rt_cache_valid(const struct rtable *rt)
1381 {
1382         return  rt &&
1383                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1384                 !rt_is_expired(rt);
1385 }
1386
1387 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1388                            const struct fib_result *res,
1389                            struct fib_nh_exception *fnhe,
1390                            struct fib_info *fi, u16 type, u32 itag)
1391 {
1392         bool cached = false;
1393
1394         if (fi) {
1395                 struct fib_nh *nh = &FIB_RES_NH(*res);
1396
1397                 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1398                         rt->rt_gateway = nh->nh_gw;
1399                         rt->rt_uses_gateway = 1;
1400                 }
1401                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1402 #ifdef CONFIG_IP_ROUTE_CLASSID
1403                 rt->dst.tclassid = nh->nh_tclassid;
1404 #endif
1405                 if (unlikely(fnhe))
1406                         cached = rt_bind_exception(rt, fnhe, daddr);
1407                 else if (!(rt->dst.flags & DST_NOCACHE))
1408                         cached = rt_cache_route(nh, rt);
1409                 if (unlikely(!cached)) {
1410                         /* Routes we intend to cache in nexthop exception or
1411                          * FIB nexthop have the DST_NOCACHE bit clear.
1412                          * However, if we are unsuccessful at storing this
1413                          * route into the cache we really need to set it.
1414                          */
1415                         rt->dst.flags |= DST_NOCACHE;
1416                         if (!rt->rt_gateway)
1417                                 rt->rt_gateway = daddr;
1418                         rt_add_uncached_list(rt);
1419                 }
1420         } else
1421                 rt_add_uncached_list(rt);
1422
1423 #ifdef CONFIG_IP_ROUTE_CLASSID
1424 #ifdef CONFIG_IP_MULTIPLE_TABLES
1425         set_class_tag(rt, res->tclassid);
1426 #endif
1427         set_class_tag(rt, itag);
1428 #endif
1429 }
1430
1431 static struct rtable *rt_dst_alloc(struct net_device *dev,
1432                                    bool nopolicy, bool noxfrm, bool will_cache)
1433 {
1434         return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1435                          (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1436                          (nopolicy ? DST_NOPOLICY : 0) |
1437                          (noxfrm ? DST_NOXFRM : 0));
1438 }
1439
1440 /* called in rcu_read_lock() section */
1441 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1442                                 u8 tos, struct net_device *dev, int our)
1443 {
1444         struct rtable *rth;
1445         struct in_device *in_dev = __in_dev_get_rcu(dev);
1446         u32 itag = 0;
1447         int err;
1448
1449         /* Primary sanity checks. */
1450
1451         if (in_dev == NULL)
1452                 return -EINVAL;
1453
1454         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1455             skb->protocol != htons(ETH_P_IP))
1456                 goto e_inval;
1457
1458         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1459                 if (ipv4_is_loopback(saddr))
1460                         goto e_inval;
1461
1462         if (ipv4_is_zeronet(saddr)) {
1463                 if (!ipv4_is_local_multicast(daddr))
1464                         goto e_inval;
1465         } else {
1466                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1467                                           in_dev, &itag);
1468                 if (err < 0)
1469                         goto e_err;
1470         }
1471         rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1472                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1473         if (!rth)
1474                 goto e_nobufs;
1475
1476 #ifdef CONFIG_IP_ROUTE_CLASSID
1477         rth->dst.tclassid = itag;
1478 #endif
1479         rth->dst.output = ip_rt_bug;
1480
1481         rth->rt_genid   = rt_genid_ipv4(dev_net(dev));
1482         rth->rt_flags   = RTCF_MULTICAST;
1483         rth->rt_type    = RTN_MULTICAST;
1484         rth->rt_is_input= 1;
1485         rth->rt_iif     = 0;
1486         rth->rt_pmtu    = 0;
1487         rth->rt_gateway = 0;
1488         rth->rt_uses_gateway = 0;
1489         INIT_LIST_HEAD(&rth->rt_uncached);
1490         if (our) {
1491                 rth->dst.input= ip_local_deliver;
1492                 rth->rt_flags |= RTCF_LOCAL;
1493         }
1494
1495 #ifdef CONFIG_IP_MROUTE
1496         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1497                 rth->dst.input = ip_mr_input;
1498 #endif
1499         RT_CACHE_STAT_INC(in_slow_mc);
1500
1501         skb_dst_set(skb, &rth->dst);
1502         return 0;
1503
1504 e_nobufs:
1505         return -ENOBUFS;
1506 e_inval:
1507         return -EINVAL;
1508 e_err:
1509         return err;
1510 }
1511
1512
1513 static void ip_handle_martian_source(struct net_device *dev,
1514                                      struct in_device *in_dev,
1515                                      struct sk_buff *skb,
1516                                      __be32 daddr,
1517                                      __be32 saddr)
1518 {
1519         RT_CACHE_STAT_INC(in_martian_src);
1520 #ifdef CONFIG_IP_ROUTE_VERBOSE
1521         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1522                 /*
1523                  *      RFC1812 recommendation, if source is martian,
1524                  *      the only hint is MAC header.
1525                  */
1526                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1527                         &daddr, &saddr, dev->name);
1528                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1529                         print_hex_dump(KERN_WARNING, "ll header: ",
1530                                        DUMP_PREFIX_OFFSET, 16, 1,
1531                                        skb_mac_header(skb),
1532                                        dev->hard_header_len, true);
1533                 }
1534         }
1535 #endif
1536 }
1537
1538 /* called in rcu_read_lock() section */
1539 static int __mkroute_input(struct sk_buff *skb,
1540                            const struct fib_result *res,
1541                            struct in_device *in_dev,
1542                            __be32 daddr, __be32 saddr, u32 tos)
1543 {
1544         struct fib_nh_exception *fnhe;
1545         struct rtable *rth;
1546         int err;
1547         struct in_device *out_dev;
1548         unsigned int flags = 0;
1549         bool do_cache;
1550         u32 itag = 0;
1551
1552         /* get a working reference to the output device */
1553         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1554         if (out_dev == NULL) {
1555                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1556                 return -EINVAL;
1557         }
1558
1559         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1560                                   in_dev->dev, in_dev, &itag);
1561         if (err < 0) {
1562                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1563                                          saddr);
1564
1565                 goto cleanup;
1566         }
1567
1568         do_cache = res->fi && !itag;
1569         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1570             (IN_DEV_SHARED_MEDIA(out_dev) ||
1571              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) {
1572                 flags |= RTCF_DOREDIRECT;
1573                 do_cache = false;
1574         }
1575
1576         if (skb->protocol != htons(ETH_P_IP)) {
1577                 /* Not IP (i.e. ARP). Do not create route, if it is
1578                  * invalid for proxy arp. DNAT routes are always valid.
1579                  *
1580                  * Proxy arp feature have been extended to allow, ARP
1581                  * replies back to the same interface, to support
1582                  * Private VLAN switch technologies. See arp.c.
1583                  */
1584                 if (out_dev == in_dev &&
1585                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1586                         err = -EINVAL;
1587                         goto cleanup;
1588                 }
1589         }
1590
1591         fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1592         if (do_cache) {
1593                 if (fnhe != NULL)
1594                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1595                 else
1596                         rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1597
1598                 if (rt_cache_valid(rth)) {
1599                         skb_dst_set_noref(skb, &rth->dst);
1600                         goto out;
1601                 }
1602         }
1603
1604         rth = rt_dst_alloc(out_dev->dev,
1605                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1606                            IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1607         if (!rth) {
1608                 err = -ENOBUFS;
1609                 goto cleanup;
1610         }
1611
1612         rth->rt_genid = rt_genid_ipv4(dev_net(rth->dst.dev));
1613         rth->rt_flags = flags;
1614         rth->rt_type = res->type;
1615         rth->rt_is_input = 1;
1616         rth->rt_iif     = 0;
1617         rth->rt_pmtu    = 0;
1618         rth->rt_gateway = 0;
1619         rth->rt_uses_gateway = 0;
1620         INIT_LIST_HEAD(&rth->rt_uncached);
1621         RT_CACHE_STAT_INC(in_slow_tot);
1622
1623         rth->dst.input = ip_forward;
1624         rth->dst.output = ip_output;
1625
1626         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1627         skb_dst_set(skb, &rth->dst);
1628 out:
1629         err = 0;
1630  cleanup:
1631         return err;
1632 }
1633
1634 static int ip_mkroute_input(struct sk_buff *skb,
1635                             struct fib_result *res,
1636                             const struct flowi4 *fl4,
1637                             struct in_device *in_dev,
1638                             __be32 daddr, __be32 saddr, u32 tos)
1639 {
1640 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1641         if (res->fi && res->fi->fib_nhs > 1)
1642                 fib_select_multipath(res);
1643 #endif
1644
1645         /* create a routing cache entry */
1646         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1647 }
1648
1649 /*
1650  *      NOTE. We drop all the packets that has local source
1651  *      addresses, because every properly looped back packet
1652  *      must have correct destination already attached by output routine.
1653  *
1654  *      Such approach solves two big problems:
1655  *      1. Not simplex devices are handled properly.
1656  *      2. IP spoofing attempts are filtered with 100% of guarantee.
1657  *      called with rcu_read_lock()
1658  */
1659
1660 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1661                                u8 tos, struct net_device *dev)
1662 {
1663         struct fib_result res;
1664         struct in_device *in_dev = __in_dev_get_rcu(dev);
1665         struct flowi4   fl4;
1666         unsigned int    flags = 0;
1667         u32             itag = 0;
1668         struct rtable   *rth;
1669         int             err = -EINVAL;
1670         struct net    *net = dev_net(dev);
1671         bool do_cache;
1672
1673         /* IP on this device is disabled. */
1674
1675         if (!in_dev)
1676                 goto out;
1677
1678         /* Check for the most weird martians, which can be not detected
1679            by fib_lookup.
1680          */
1681
1682         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1683                 goto martian_source;
1684
1685         res.fi = NULL;
1686         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1687                 goto brd_input;
1688
1689         /* Accept zero addresses only to limited broadcast;
1690          * I even do not know to fix it or not. Waiting for complains :-)
1691          */
1692         if (ipv4_is_zeronet(saddr))
1693                 goto martian_source;
1694
1695         if (ipv4_is_zeronet(daddr))
1696                 goto martian_destination;
1697
1698         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1699          * and call it once if daddr or/and saddr are loopback addresses
1700          */
1701         if (ipv4_is_loopback(daddr)) {
1702                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1703                         goto martian_destination;
1704         } else if (ipv4_is_loopback(saddr)) {
1705                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1706                         goto martian_source;
1707         }
1708
1709         /*
1710          *      Now we are ready to route packet.
1711          */
1712         fl4.flowi4_oif = 0;
1713         fl4.flowi4_iif = dev->ifindex;
1714         fl4.flowi4_mark = skb->mark;
1715         fl4.flowi4_tos = tos;
1716         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1717         fl4.daddr = daddr;
1718         fl4.saddr = saddr;
1719         err = fib_lookup(net, &fl4, &res);
1720         if (err != 0) {
1721                 if (!IN_DEV_FORWARD(in_dev))
1722                         err = -EHOSTUNREACH;
1723                 goto no_route;
1724         }
1725
1726         if (res.type == RTN_BROADCAST)
1727                 goto brd_input;
1728
1729         if (res.type == RTN_LOCAL) {
1730                 err = fib_validate_source(skb, saddr, daddr, tos,
1731                                           0, dev, in_dev, &itag);
1732                 if (err < 0)
1733                         goto martian_source_keep_err;
1734                 goto local_input;
1735         }
1736
1737         if (!IN_DEV_FORWARD(in_dev)) {
1738                 err = -EHOSTUNREACH;
1739                 goto no_route;
1740         }
1741         if (res.type != RTN_UNICAST)
1742                 goto martian_destination;
1743
1744         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1745 out:    return err;
1746
1747 brd_input:
1748         if (skb->protocol != htons(ETH_P_IP))
1749                 goto e_inval;
1750
1751         if (!ipv4_is_zeronet(saddr)) {
1752                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1753                                           in_dev, &itag);
1754                 if (err < 0)
1755                         goto martian_source_keep_err;
1756         }
1757         flags |= RTCF_BROADCAST;
1758         res.type = RTN_BROADCAST;
1759         RT_CACHE_STAT_INC(in_brd);
1760
1761 local_input:
1762         do_cache = false;
1763         if (res.fi) {
1764                 if (!itag) {
1765                         rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1766                         if (rt_cache_valid(rth)) {
1767                                 skb_dst_set_noref(skb, &rth->dst);
1768                                 err = 0;
1769                                 goto out;
1770                         }
1771                         do_cache = true;
1772                 }
1773         }
1774
1775         rth = rt_dst_alloc(net->loopback_dev,
1776                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1777         if (!rth)
1778                 goto e_nobufs;
1779
1780         rth->dst.input= ip_local_deliver;
1781         rth->dst.output= ip_rt_bug;
1782 #ifdef CONFIG_IP_ROUTE_CLASSID
1783         rth->dst.tclassid = itag;
1784 #endif
1785
1786         rth->rt_genid = rt_genid_ipv4(net);
1787         rth->rt_flags   = flags|RTCF_LOCAL;
1788         rth->rt_type    = res.type;
1789         rth->rt_is_input = 1;
1790         rth->rt_iif     = 0;
1791         rth->rt_pmtu    = 0;
1792         rth->rt_gateway = 0;
1793         rth->rt_uses_gateway = 0;
1794         INIT_LIST_HEAD(&rth->rt_uncached);
1795         RT_CACHE_STAT_INC(in_slow_tot);
1796         if (res.type == RTN_UNREACHABLE) {
1797                 rth->dst.input= ip_error;
1798                 rth->dst.error= -err;
1799                 rth->rt_flags   &= ~RTCF_LOCAL;
1800         }
1801         if (do_cache) {
1802                 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1803                         rth->dst.flags |= DST_NOCACHE;
1804                         rt_add_uncached_list(rth);
1805                 }
1806         }
1807         skb_dst_set(skb, &rth->dst);
1808         err = 0;
1809         goto out;
1810
1811 no_route:
1812         RT_CACHE_STAT_INC(in_no_route);
1813         res.type = RTN_UNREACHABLE;
1814         res.fi = NULL;
1815         goto local_input;
1816
1817         /*
1818          *      Do not cache martian addresses: they should be logged (RFC1812)
1819          */
1820 martian_destination:
1821         RT_CACHE_STAT_INC(in_martian_dst);
1822 #ifdef CONFIG_IP_ROUTE_VERBOSE
1823         if (IN_DEV_LOG_MARTIANS(in_dev))
1824                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1825                                      &daddr, &saddr, dev->name);
1826 #endif
1827
1828 e_inval:
1829         err = -EINVAL;
1830         goto out;
1831
1832 e_nobufs:
1833         err = -ENOBUFS;
1834         goto out;
1835
1836 martian_source:
1837         err = -EINVAL;
1838 martian_source_keep_err:
1839         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1840         goto out;
1841 }
1842
1843 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1844                          u8 tos, struct net_device *dev)
1845 {
1846         int res;
1847
1848         rcu_read_lock();
1849
1850         /* Multicast recognition logic is moved from route cache to here.
1851            The problem was that too many Ethernet cards have broken/missing
1852            hardware multicast filters :-( As result the host on multicasting
1853            network acquires a lot of useless route cache entries, sort of
1854            SDR messages from all the world. Now we try to get rid of them.
1855            Really, provided software IP multicast filter is organized
1856            reasonably (at least, hashed), it does not result in a slowdown
1857            comparing with route cache reject entries.
1858            Note, that multicast routers are not affected, because
1859            route cache entry is created eventually.
1860          */
1861         if (ipv4_is_multicast(daddr)) {
1862                 struct in_device *in_dev = __in_dev_get_rcu(dev);
1863
1864                 if (in_dev) {
1865                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1866                                                   ip_hdr(skb)->protocol);
1867                         if (our
1868 #ifdef CONFIG_IP_MROUTE
1869                                 ||
1870                             (!ipv4_is_local_multicast(daddr) &&
1871                              IN_DEV_MFORWARD(in_dev))
1872 #endif
1873                            ) {
1874                                 int res = ip_route_input_mc(skb, daddr, saddr,
1875                                                             tos, dev, our);
1876                                 rcu_read_unlock();
1877                                 return res;
1878                         }
1879                 }
1880                 rcu_read_unlock();
1881                 return -EINVAL;
1882         }
1883         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1884         rcu_read_unlock();
1885         return res;
1886 }
1887 EXPORT_SYMBOL(ip_route_input_noref);
1888
1889 /* called with rcu_read_lock() */
1890 static struct rtable *__mkroute_output(const struct fib_result *res,
1891                                        const struct flowi4 *fl4, int orig_oif,
1892                                        struct net_device *dev_out,
1893                                        unsigned int flags)
1894 {
1895         struct fib_info *fi = res->fi;
1896         struct fib_nh_exception *fnhe;
1897         struct in_device *in_dev;
1898         u16 type = res->type;
1899         struct rtable *rth;
1900         bool do_cache;
1901
1902         in_dev = __in_dev_get_rcu(dev_out);
1903         if (!in_dev)
1904                 return ERR_PTR(-EINVAL);
1905
1906         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1907                 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1908                         return ERR_PTR(-EINVAL);
1909
1910         if (ipv4_is_lbcast(fl4->daddr))
1911                 type = RTN_BROADCAST;
1912         else if (ipv4_is_multicast(fl4->daddr))
1913                 type = RTN_MULTICAST;
1914         else if (ipv4_is_zeronet(fl4->daddr))
1915                 return ERR_PTR(-EINVAL);
1916
1917         if (dev_out->flags & IFF_LOOPBACK)
1918                 flags |= RTCF_LOCAL;
1919
1920         do_cache = true;
1921         if (type == RTN_BROADCAST) {
1922                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1923                 fi = NULL;
1924         } else if (type == RTN_MULTICAST) {
1925                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1926                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1927                                      fl4->flowi4_proto))
1928                         flags &= ~RTCF_LOCAL;
1929                 else
1930                         do_cache = false;
1931                 /* If multicast route do not exist use
1932                  * default one, but do not gateway in this case.
1933                  * Yes, it is hack.
1934                  */
1935                 if (fi && res->prefixlen < 4)
1936                         fi = NULL;
1937         }
1938
1939         fnhe = NULL;
1940         do_cache &= fi != NULL;
1941         if (do_cache) {
1942                 struct rtable __rcu **prth;
1943                 struct fib_nh *nh = &FIB_RES_NH(*res);
1944
1945                 fnhe = find_exception(nh, fl4->daddr);
1946                 if (fnhe)
1947                         prth = &fnhe->fnhe_rth_output;
1948                 else {
1949                         if (unlikely(fl4->flowi4_flags &
1950                                      FLOWI_FLAG_KNOWN_NH &&
1951                                      !(nh->nh_gw &&
1952                                        nh->nh_scope == RT_SCOPE_LINK))) {
1953                                 do_cache = false;
1954                                 goto add;
1955                         }
1956                         prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
1957                 }
1958                 rth = rcu_dereference(*prth);
1959                 if (rt_cache_valid(rth)) {
1960                         dst_hold(&rth->dst);
1961                         return rth;
1962                 }
1963         }
1964
1965 add:
1966         rth = rt_dst_alloc(dev_out,
1967                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1968                            IN_DEV_CONF_GET(in_dev, NOXFRM),
1969                            do_cache);
1970         if (!rth)
1971                 return ERR_PTR(-ENOBUFS);
1972
1973         rth->dst.output = ip_output;
1974
1975         rth->rt_genid = rt_genid_ipv4(dev_net(dev_out));
1976         rth->rt_flags   = flags;
1977         rth->rt_type    = type;
1978         rth->rt_is_input = 0;
1979         rth->rt_iif     = orig_oif ? : 0;
1980         rth->rt_pmtu    = 0;
1981         rth->rt_gateway = 0;
1982         rth->rt_uses_gateway = 0;
1983         INIT_LIST_HEAD(&rth->rt_uncached);
1984
1985         RT_CACHE_STAT_INC(out_slow_tot);
1986
1987         if (flags & RTCF_LOCAL)
1988                 rth->dst.input = ip_local_deliver;
1989         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1990                 if (flags & RTCF_LOCAL &&
1991                     !(dev_out->flags & IFF_LOOPBACK)) {
1992                         rth->dst.output = ip_mc_output;
1993                         RT_CACHE_STAT_INC(out_slow_mc);
1994                 }
1995 #ifdef CONFIG_IP_MROUTE
1996                 if (type == RTN_MULTICAST) {
1997                         if (IN_DEV_MFORWARD(in_dev) &&
1998                             !ipv4_is_local_multicast(fl4->daddr)) {
1999                                 rth->dst.input = ip_mr_input;
2000                                 rth->dst.output = ip_mc_output;
2001                         }
2002                 }
2003 #endif
2004         }
2005
2006         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2007
2008         return rth;
2009 }
2010
2011 /*
2012  * Major route resolver routine.
2013  */
2014
2015 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
2016 {
2017         struct net_device *dev_out = NULL;
2018         __u8 tos = RT_FL_TOS(fl4);
2019         unsigned int flags = 0;
2020         struct fib_result res;
2021         struct rtable *rth;
2022         int orig_oif;
2023
2024         res.tclassid    = 0;
2025         res.fi          = NULL;
2026         res.table       = NULL;
2027
2028         orig_oif = fl4->flowi4_oif;
2029
2030         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2031         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2032         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2033                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2034
2035         rcu_read_lock();
2036         if (fl4->saddr) {
2037                 rth = ERR_PTR(-EINVAL);
2038                 if (ipv4_is_multicast(fl4->saddr) ||
2039                     ipv4_is_lbcast(fl4->saddr) ||
2040                     ipv4_is_zeronet(fl4->saddr))
2041                         goto out;
2042
2043                 /* I removed check for oif == dev_out->oif here.
2044                    It was wrong for two reasons:
2045                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2046                       is assigned to multiple interfaces.
2047                    2. Moreover, we are allowed to send packets with saddr
2048                       of another iface. --ANK
2049                  */
2050
2051                 if (fl4->flowi4_oif == 0 &&
2052                     (ipv4_is_multicast(fl4->daddr) ||
2053                      ipv4_is_lbcast(fl4->daddr))) {
2054                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2055                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2056                         if (dev_out == NULL)
2057                                 goto out;
2058
2059                         /* Special hack: user can direct multicasts
2060                            and limited broadcast via necessary interface
2061                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2062                            This hack is not just for fun, it allows
2063                            vic,vat and friends to work.
2064                            They bind socket to loopback, set ttl to zero
2065                            and expect that it will work.
2066                            From the viewpoint of routing cache they are broken,
2067                            because we are not allowed to build multicast path
2068                            with loopback source addr (look, routing cache
2069                            cannot know, that ttl is zero, so that packet
2070                            will not leave this host and route is valid).
2071                            Luckily, this hack is good workaround.
2072                          */
2073
2074                         fl4->flowi4_oif = dev_out->ifindex;
2075                         goto make_route;
2076                 }
2077
2078                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2079                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2080                         if (!__ip_dev_find(net, fl4->saddr, false))
2081                                 goto out;
2082                 }
2083         }
2084
2085
2086         if (fl4->flowi4_oif) {
2087                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2088                 rth = ERR_PTR(-ENODEV);
2089                 if (dev_out == NULL)
2090                         goto out;
2091
2092                 /* RACE: Check return value of inet_select_addr instead. */
2093                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2094                         rth = ERR_PTR(-ENETUNREACH);
2095                         goto out;
2096                 }
2097                 if (ipv4_is_local_multicast(fl4->daddr) ||
2098                     ipv4_is_lbcast(fl4->daddr)) {
2099                         if (!fl4->saddr)
2100                                 fl4->saddr = inet_select_addr(dev_out, 0,
2101                                                               RT_SCOPE_LINK);
2102                         goto make_route;
2103                 }
2104                 if (!fl4->saddr) {
2105                         if (ipv4_is_multicast(fl4->daddr))
2106                                 fl4->saddr = inet_select_addr(dev_out, 0,
2107                                                               fl4->flowi4_scope);
2108                         else if (!fl4->daddr)
2109                                 fl4->saddr = inet_select_addr(dev_out, 0,
2110                                                               RT_SCOPE_HOST);
2111                 }
2112         }
2113
2114         if (!fl4->daddr) {
2115                 fl4->daddr = fl4->saddr;
2116                 if (!fl4->daddr)
2117                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2118                 dev_out = net->loopback_dev;
2119                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2120                 res.type = RTN_LOCAL;
2121                 flags |= RTCF_LOCAL;
2122                 goto make_route;
2123         }
2124
2125         if (fib_lookup(net, fl4, &res)) {
2126                 res.fi = NULL;
2127                 res.table = NULL;
2128                 if (fl4->flowi4_oif) {
2129                         /* Apparently, routing tables are wrong. Assume,
2130                            that the destination is on link.
2131
2132                            WHY? DW.
2133                            Because we are allowed to send to iface
2134                            even if it has NO routes and NO assigned
2135                            addresses. When oif is specified, routing
2136                            tables are looked up with only one purpose:
2137                            to catch if destination is gatewayed, rather than
2138                            direct. Moreover, if MSG_DONTROUTE is set,
2139                            we send packet, ignoring both routing tables
2140                            and ifaddr state. --ANK
2141
2142
2143                            We could make it even if oif is unknown,
2144                            likely IPv6, but we do not.
2145                          */
2146
2147                         if (fl4->saddr == 0)
2148                                 fl4->saddr = inet_select_addr(dev_out, 0,
2149                                                               RT_SCOPE_LINK);
2150                         res.type = RTN_UNICAST;
2151                         goto make_route;
2152                 }
2153                 rth = ERR_PTR(-ENETUNREACH);
2154                 goto out;
2155         }
2156
2157         if (res.type == RTN_LOCAL) {
2158                 if (!fl4->saddr) {
2159                         if (res.fi->fib_prefsrc)
2160                                 fl4->saddr = res.fi->fib_prefsrc;
2161                         else
2162                                 fl4->saddr = fl4->daddr;
2163                 }
2164                 dev_out = net->loopback_dev;
2165                 fl4->flowi4_oif = dev_out->ifindex;
2166                 flags |= RTCF_LOCAL;
2167                 goto make_route;
2168         }
2169
2170 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2171         if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2172                 fib_select_multipath(&res);
2173         else
2174 #endif
2175         if (!res.prefixlen &&
2176             res.table->tb_num_default > 1 &&
2177             res.type == RTN_UNICAST && !fl4->flowi4_oif)
2178                 fib_select_default(&res);
2179
2180         if (!fl4->saddr)
2181                 fl4->saddr = FIB_RES_PREFSRC(net, res);
2182
2183         dev_out = FIB_RES_DEV(res);
2184         fl4->flowi4_oif = dev_out->ifindex;
2185
2186
2187 make_route:
2188         rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2189
2190 out:
2191         rcu_read_unlock();
2192         return rth;
2193 }
2194 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2195
2196 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2197 {
2198         return NULL;
2199 }
2200
2201 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2202 {
2203         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2204
2205         return mtu ? : dst->dev->mtu;
2206 }
2207
2208 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2209                                           struct sk_buff *skb, u32 mtu)
2210 {
2211 }
2212
2213 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2214                                        struct sk_buff *skb)
2215 {
2216 }
2217
2218 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2219                                           unsigned long old)
2220 {
2221         return NULL;
2222 }
2223
2224 static struct dst_ops ipv4_dst_blackhole_ops = {
2225         .family                 =       AF_INET,
2226         .protocol               =       cpu_to_be16(ETH_P_IP),
2227         .check                  =       ipv4_blackhole_dst_check,
2228         .mtu                    =       ipv4_blackhole_mtu,
2229         .default_advmss         =       ipv4_default_advmss,
2230         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2231         .redirect               =       ipv4_rt_blackhole_redirect,
2232         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
2233         .neigh_lookup           =       ipv4_neigh_lookup,
2234 };
2235
2236 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2237 {
2238         struct rtable *ort = (struct rtable *) dst_orig;
2239         struct rtable *rt;
2240
2241         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2242         if (rt) {
2243                 struct dst_entry *new = &rt->dst;
2244
2245                 new->__use = 1;
2246                 new->input = dst_discard;
2247                 new->output = dst_discard_sk;
2248
2249                 new->dev = ort->dst.dev;
2250                 if (new->dev)
2251                         dev_hold(new->dev);
2252
2253                 rt->rt_is_input = ort->rt_is_input;
2254                 rt->rt_iif = ort->rt_iif;
2255                 rt->rt_pmtu = ort->rt_pmtu;
2256
2257                 rt->rt_genid = rt_genid_ipv4(net);
2258                 rt->rt_flags = ort->rt_flags;
2259                 rt->rt_type = ort->rt_type;
2260                 rt->rt_gateway = ort->rt_gateway;
2261                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2262
2263                 INIT_LIST_HEAD(&rt->rt_uncached);
2264
2265                 dst_free(new);
2266         }
2267
2268         dst_release(dst_orig);
2269
2270         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2271 }
2272
2273 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2274                                     struct sock *sk)
2275 {
2276         struct rtable *rt = __ip_route_output_key(net, flp4);
2277
2278         if (IS_ERR(rt))
2279                 return rt;
2280
2281         if (flp4->flowi4_proto)
2282                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2283                                                         flowi4_to_flowi(flp4),
2284                                                         sk, 0);
2285
2286         return rt;
2287 }
2288 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2289
2290 static int rt_fill_info(struct net *net,  __be32 dst, __be32 src,
2291                         struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2292                         u32 seq, int event, int nowait, unsigned int flags)
2293 {
2294         struct rtable *rt = skb_rtable(skb);
2295         struct rtmsg *r;
2296         struct nlmsghdr *nlh;
2297         unsigned long expires = 0;
2298         u32 error;
2299         u32 metrics[RTAX_MAX];
2300
2301         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2302         if (nlh == NULL)
2303                 return -EMSGSIZE;
2304
2305         r = nlmsg_data(nlh);
2306         r->rtm_family    = AF_INET;
2307         r->rtm_dst_len  = 32;
2308         r->rtm_src_len  = 0;
2309         r->rtm_tos      = fl4->flowi4_tos;
2310         r->rtm_table    = RT_TABLE_MAIN;
2311         if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2312                 goto nla_put_failure;
2313         r->rtm_type     = rt->rt_type;
2314         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2315         r->rtm_protocol = RTPROT_UNSPEC;
2316         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2317         if (rt->rt_flags & RTCF_NOTIFY)
2318                 r->rtm_flags |= RTM_F_NOTIFY;
2319
2320         if (nla_put_be32(skb, RTA_DST, dst))
2321                 goto nla_put_failure;
2322         if (src) {
2323                 r->rtm_src_len = 32;
2324                 if (nla_put_be32(skb, RTA_SRC, src))
2325                         goto nla_put_failure;
2326         }
2327         if (rt->dst.dev &&
2328             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2329                 goto nla_put_failure;
2330 #ifdef CONFIG_IP_ROUTE_CLASSID
2331         if (rt->dst.tclassid &&
2332             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2333                 goto nla_put_failure;
2334 #endif
2335         if (!rt_is_input_route(rt) &&
2336             fl4->saddr != src) {
2337                 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2338                         goto nla_put_failure;
2339         }
2340         if (rt->rt_uses_gateway &&
2341             nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2342                 goto nla_put_failure;
2343
2344         expires = rt->dst.expires;
2345         if (expires) {
2346                 unsigned long now = jiffies;
2347
2348                 if (time_before(now, expires))
2349                         expires -= now;
2350                 else
2351                         expires = 0;
2352         }
2353
2354         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2355         if (rt->rt_pmtu && expires)
2356                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2357         if (rtnetlink_put_metrics(skb, metrics) < 0)
2358                 goto nla_put_failure;
2359
2360         if (fl4->flowi4_mark &&
2361             nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2362                 goto nla_put_failure;
2363
2364         error = rt->dst.error;
2365
2366         if (rt_is_input_route(rt)) {
2367 #ifdef CONFIG_IP_MROUTE
2368                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2369                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2370                         int err = ipmr_get_route(net, skb,
2371                                                  fl4->saddr, fl4->daddr,
2372                                                  r, nowait);
2373                         if (err <= 0) {
2374                                 if (!nowait) {
2375                                         if (err == 0)
2376                                                 return 0;
2377                                         goto nla_put_failure;
2378                                 } else {
2379                                         if (err == -EMSGSIZE)
2380                                                 goto nla_put_failure;
2381                                         error = err;
2382                                 }
2383                         }
2384                 } else
2385 #endif
2386                         if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2387                                 goto nla_put_failure;
2388         }
2389
2390         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2391                 goto nla_put_failure;
2392
2393         nlmsg_end(skb, nlh);
2394         return 0;
2395
2396 nla_put_failure:
2397         nlmsg_cancel(skb, nlh);
2398         return -EMSGSIZE;
2399 }
2400
2401 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2402 {
2403         struct net *net = sock_net(in_skb->sk);
2404         struct rtmsg *rtm;
2405         struct nlattr *tb[RTA_MAX+1];
2406         struct rtable *rt = NULL;
2407         struct flowi4 fl4;
2408         __be32 dst = 0;
2409         __be32 src = 0;
2410         u32 iif;
2411         int err;
2412         int mark;
2413         struct sk_buff *skb;
2414
2415         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2416         if (err < 0)
2417                 goto errout;
2418
2419         rtm = nlmsg_data(nlh);
2420
2421         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2422         if (skb == NULL) {
2423                 err = -ENOBUFS;
2424                 goto errout;
2425         }
2426
2427         /* Reserve room for dummy headers, this skb can pass
2428            through good chunk of routing engine.
2429          */
2430         skb_reset_mac_header(skb);
2431         skb_reset_network_header(skb);
2432
2433         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2434         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2435         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2436
2437         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2438         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2439         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2440         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2441
2442         memset(&fl4, 0, sizeof(fl4));
2443         fl4.daddr = dst;
2444         fl4.saddr = src;
2445         fl4.flowi4_tos = rtm->rtm_tos;
2446         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2447         fl4.flowi4_mark = mark;
2448
2449         if (iif) {
2450                 struct net_device *dev;
2451
2452                 dev = __dev_get_by_index(net, iif);
2453                 if (dev == NULL) {
2454                         err = -ENODEV;
2455                         goto errout_free;
2456                 }
2457
2458                 skb->protocol   = htons(ETH_P_IP);
2459                 skb->dev        = dev;
2460                 skb->mark       = mark;
2461                 local_bh_disable();
2462                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2463                 local_bh_enable();
2464
2465                 rt = skb_rtable(skb);
2466                 if (err == 0 && rt->dst.error)
2467                         err = -rt->dst.error;
2468         } else {
2469                 rt = ip_route_output_key(net, &fl4);
2470
2471                 err = 0;
2472                 if (IS_ERR(rt))
2473                         err = PTR_ERR(rt);
2474         }
2475
2476         if (err)
2477                 goto errout_free;
2478
2479         skb_dst_set(skb, &rt->dst);
2480         if (rtm->rtm_flags & RTM_F_NOTIFY)
2481                 rt->rt_flags |= RTCF_NOTIFY;
2482
2483         err = rt_fill_info(net, dst, src, &fl4, skb,
2484                            NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2485                            RTM_NEWROUTE, 0, 0);
2486         if (err < 0)
2487                 goto errout_free;
2488
2489         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2490 errout:
2491         return err;
2492
2493 errout_free:
2494         kfree_skb(skb);
2495         goto errout;
2496 }
2497
2498 void ip_rt_multicast_event(struct in_device *in_dev)
2499 {
2500         rt_cache_flush(dev_net(in_dev->dev));
2501 }
2502
2503 #ifdef CONFIG_SYSCTL
2504 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
2505 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
2506 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
2507 static int ip_rt_gc_elasticity __read_mostly    = 8;
2508
2509 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2510                                         void __user *buffer,
2511                                         size_t *lenp, loff_t *ppos)
2512 {
2513         struct net *net = (struct net *)__ctl->extra1;
2514
2515         if (write) {
2516                 rt_cache_flush(net);
2517                 fnhe_genid_bump(net);
2518                 return 0;
2519         }
2520
2521         return -EINVAL;
2522 }
2523
2524 static struct ctl_table ipv4_route_table[] = {
2525         {
2526                 .procname       = "gc_thresh",
2527                 .data           = &ipv4_dst_ops.gc_thresh,
2528                 .maxlen         = sizeof(int),
2529                 .mode           = 0644,
2530                 .proc_handler   = proc_dointvec,
2531         },
2532         {
2533                 .procname       = "max_size",
2534                 .data           = &ip_rt_max_size,
2535                 .maxlen         = sizeof(int),
2536                 .mode           = 0644,
2537                 .proc_handler   = proc_dointvec,
2538         },
2539         {
2540                 /*  Deprecated. Use gc_min_interval_ms */
2541
2542                 .procname       = "gc_min_interval",
2543                 .data           = &ip_rt_gc_min_interval,
2544                 .maxlen         = sizeof(int),
2545                 .mode           = 0644,
2546                 .proc_handler   = proc_dointvec_jiffies,
2547         },
2548         {
2549                 .procname       = "gc_min_interval_ms",
2550                 .data           = &ip_rt_gc_min_interval,
2551                 .maxlen         = sizeof(int),
2552                 .mode           = 0644,
2553                 .proc_handler   = proc_dointvec_ms_jiffies,
2554         },
2555         {
2556                 .procname       = "gc_timeout",
2557                 .data           = &ip_rt_gc_timeout,
2558                 .maxlen         = sizeof(int),
2559                 .mode           = 0644,
2560                 .proc_handler   = proc_dointvec_jiffies,
2561         },
2562         {
2563                 .procname       = "gc_interval",
2564                 .data           = &ip_rt_gc_interval,
2565                 .maxlen         = sizeof(int),
2566                 .mode           = 0644,
2567                 .proc_handler   = proc_dointvec_jiffies,
2568         },
2569         {
2570                 .procname       = "redirect_load",
2571                 .data           = &ip_rt_redirect_load,
2572                 .maxlen         = sizeof(int),
2573                 .mode           = 0644,
2574                 .proc_handler   = proc_dointvec,
2575         },
2576         {
2577                 .procname       = "redirect_number",
2578                 .data           = &ip_rt_redirect_number,
2579                 .maxlen         = sizeof(int),
2580                 .mode           = 0644,
2581                 .proc_handler   = proc_dointvec,
2582         },
2583         {
2584                 .procname       = "redirect_silence",
2585                 .data           = &ip_rt_redirect_silence,
2586                 .maxlen         = sizeof(int),
2587                 .mode           = 0644,
2588                 .proc_handler   = proc_dointvec,
2589         },
2590         {
2591                 .procname       = "error_cost",
2592                 .data           = &ip_rt_error_cost,
2593                 .maxlen         = sizeof(int),
2594                 .mode           = 0644,
2595                 .proc_handler   = proc_dointvec,
2596         },
2597         {
2598                 .procname       = "error_burst",
2599                 .data           = &ip_rt_error_burst,
2600                 .maxlen         = sizeof(int),
2601                 .mode           = 0644,
2602                 .proc_handler   = proc_dointvec,
2603         },
2604         {
2605                 .procname       = "gc_elasticity",
2606                 .data           = &ip_rt_gc_elasticity,
2607                 .maxlen         = sizeof(int),
2608                 .mode           = 0644,
2609                 .proc_handler   = proc_dointvec,
2610         },
2611         {
2612                 .procname       = "mtu_expires",
2613                 .data           = &ip_rt_mtu_expires,
2614                 .maxlen         = sizeof(int),
2615                 .mode           = 0644,
2616                 .proc_handler   = proc_dointvec_jiffies,
2617         },
2618         {
2619                 .procname       = "min_pmtu",
2620                 .data           = &ip_rt_min_pmtu,
2621                 .maxlen         = sizeof(int),
2622                 .mode           = 0644,
2623                 .proc_handler   = proc_dointvec,
2624         },
2625         {
2626                 .procname       = "min_adv_mss",
2627                 .data           = &ip_rt_min_advmss,
2628                 .maxlen         = sizeof(int),
2629                 .mode           = 0644,
2630                 .proc_handler   = proc_dointvec,
2631         },
2632         { }
2633 };
2634
2635 static struct ctl_table ipv4_route_flush_table[] = {
2636         {
2637                 .procname       = "flush",
2638                 .maxlen         = sizeof(int),
2639                 .mode           = 0200,
2640                 .proc_handler   = ipv4_sysctl_rtcache_flush,
2641         },
2642         { },
2643 };
2644
2645 static __net_init int sysctl_route_net_init(struct net *net)
2646 {
2647         struct ctl_table *tbl;
2648
2649         tbl = ipv4_route_flush_table;
2650         if (!net_eq(net, &init_net)) {
2651                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2652                 if (tbl == NULL)
2653                         goto err_dup;
2654
2655                 /* Don't export sysctls to unprivileged users */
2656                 if (net->user_ns != &init_user_ns)
2657                         tbl[0].procname = NULL;
2658         }
2659         tbl[0].extra1 = net;
2660
2661         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2662         if (net->ipv4.route_hdr == NULL)
2663                 goto err_reg;
2664         return 0;
2665
2666 err_reg:
2667         if (tbl != ipv4_route_flush_table)
2668                 kfree(tbl);
2669 err_dup:
2670         return -ENOMEM;
2671 }
2672
2673 static __net_exit void sysctl_route_net_exit(struct net *net)
2674 {
2675         struct ctl_table *tbl;
2676
2677         tbl = net->ipv4.route_hdr->ctl_table_arg;
2678         unregister_net_sysctl_table(net->ipv4.route_hdr);
2679         BUG_ON(tbl == ipv4_route_flush_table);
2680         kfree(tbl);
2681 }
2682
2683 static __net_initdata struct pernet_operations sysctl_route_ops = {
2684         .init = sysctl_route_net_init,
2685         .exit = sysctl_route_net_exit,
2686 };
2687 #endif
2688
2689 static __net_init int rt_genid_init(struct net *net)
2690 {
2691         atomic_set(&net->ipv4.rt_genid, 0);
2692         atomic_set(&net->fnhe_genid, 0);
2693         get_random_bytes(&net->ipv4.dev_addr_genid,
2694                          sizeof(net->ipv4.dev_addr_genid));
2695         return 0;
2696 }
2697
2698 static __net_initdata struct pernet_operations rt_genid_ops = {
2699         .init = rt_genid_init,
2700 };
2701
2702 static int __net_init ipv4_inetpeer_init(struct net *net)
2703 {
2704         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2705
2706         if (!bp)
2707                 return -ENOMEM;
2708         inet_peer_base_init(bp);
2709         net->ipv4.peers = bp;
2710         return 0;
2711 }
2712
2713 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2714 {
2715         struct inet_peer_base *bp = net->ipv4.peers;
2716
2717         net->ipv4.peers = NULL;
2718         inetpeer_invalidate_tree(bp);
2719         kfree(bp);
2720 }
2721
2722 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2723         .init   =       ipv4_inetpeer_init,
2724         .exit   =       ipv4_inetpeer_exit,
2725 };
2726
2727 #ifdef CONFIG_IP_ROUTE_CLASSID
2728 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2729 #endif /* CONFIG_IP_ROUTE_CLASSID */
2730
2731 int __init ip_rt_init(void)
2732 {
2733         int rc = 0;
2734         int cpu;
2735
2736         ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2737         if (!ip_idents)
2738                 panic("IP: failed to allocate ip_idents\n");
2739
2740         prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2741
2742         for_each_possible_cpu(cpu) {
2743                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
2744
2745                 INIT_LIST_HEAD(&ul->head);
2746                 spin_lock_init(&ul->lock);
2747         }
2748 #ifdef CONFIG_IP_ROUTE_CLASSID
2749         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2750         if (!ip_rt_acct)
2751                 panic("IP: failed to allocate ip_rt_acct\n");
2752 #endif
2753
2754         ipv4_dst_ops.kmem_cachep =
2755                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2756                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2757
2758         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2759
2760         if (dst_entries_init(&ipv4_dst_ops) < 0)
2761                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2762
2763         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2764                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2765
2766         ipv4_dst_ops.gc_thresh = ~0;
2767         ip_rt_max_size = INT_MAX;
2768
2769         devinet_init();
2770         ip_fib_init();
2771
2772         if (ip_rt_proc_init())
2773                 pr_err("Unable to create route proc files\n");
2774 #ifdef CONFIG_XFRM
2775         xfrm_init();
2776         xfrm4_init();
2777 #endif
2778         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2779
2780 #ifdef CONFIG_SYSCTL
2781         register_pernet_subsys(&sysctl_route_ops);
2782 #endif
2783         register_pernet_subsys(&rt_genid_ops);
2784         register_pernet_subsys(&ipv4_inetpeer_ops);
2785         return rc;
2786 }
2787
2788 #ifdef CONFIG_SYSCTL
2789 /*
2790  * We really need to sanitize the damn ipv4 init order, then all
2791  * this nonsense will go away.
2792  */
2793 void __init ip_static_sysctl_init(void)
2794 {
2795         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
2796 }
2797 #endif