net/ipv4/route.c

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