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 <net/dst.h>
  93 #include <net/net_namespace.h>
  94 #include <net/protocol.h>
  95 #include <net/ip.h>
  96 #include <net/route.h>
  97 #include <net/inetpeer.h>
  98 #include <net/sock.h>
  99 #include <net/ip_fib.h>
 100 #include <net/arp.h>
 101 #include <net/tcp.h>
 102 #include <net/icmp.h>
 103 #include <net/xfrm.h>
 104 #include <net/netevent.h>
 105 #include <net/rtnetlink.h>
 106 #ifdef CONFIG_SYSCTL
 107 #include <linux/sysctl.h>
 108 #include <linux/kmemleak.h>
 109 #endif
 110 #include <net/secure_seq.h>
 111
 112 #define RT_FL_TOS(oldflp4) \
 113         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
 114
 115 #define RT_GC_TIMEOUT (300*HZ)
 116
 117 static int ip_rt_max_size;
 118 static int ip_rt_redirect_number __read_mostly  = 9;
 119 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 120 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 121 static int ip_rt_error_cost __read_mostly       = HZ;
 122 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 123 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
 124 static int ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
 125 static int ip_rt_min_advmss __read_mostly       = 256;
 126
 127 /*
 128  *      Interface to generic destination cache.
 129  */
 130
 131 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 132 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 133 static unsigned int      ipv4_mtu(const struct dst_entry *dst);
 134 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 135 static void              ipv4_link_failure(struct sk_buff *skb);
 136 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 137                                            struct sk_buff *skb, u32 mtu);
 138 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 139                                         struct sk_buff *skb);
 140 static void             ipv4_dst_destroy(struct dst_entry *dst);
 141
 142 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 143                             int how)
 144 {
 145 }
 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         .protocol =             cpu_to_be16(ETH_P_IP),
 160         .check =                ipv4_dst_check,
 161         .default_advmss =       ipv4_default_advmss,
 162         .mtu =                  ipv4_mtu,
 163         .cow_metrics =          ipv4_cow_metrics,
 164         .destroy =              ipv4_dst_destroy,
 165         .ifdown =               ipv4_dst_ifdown,
 166         .negative_advice =      ipv4_negative_advice,
 167         .link_failure =         ipv4_link_failure,
 168         .update_pmtu =          ip_rt_update_pmtu,
 169         .redirect =             ip_do_redirect,
 170         .local_out =            __ip_local_out,
 171         .neigh_lookup =         ipv4_neigh_lookup,
 172 };
 173
 174 #define ECN_OR_COST(class)      TC_PRIO_##class
 175
 176 const __u8 ip_tos2prio[16] = {
 177         TC_PRIO_BESTEFFORT,
 178         ECN_OR_COST(BESTEFFORT),
 179         TC_PRIO_BESTEFFORT,
 180         ECN_OR_COST(BESTEFFORT),
 181         TC_PRIO_BULK,
 182         ECN_OR_COST(BULK),
 183         TC_PRIO_BULK,
 184         ECN_OR_COST(BULK),
 185         TC_PRIO_INTERACTIVE,
 186         ECN_OR_COST(INTERACTIVE),
 187         TC_PRIO_INTERACTIVE,
 188         ECN_OR_COST(INTERACTIVE),
 189         TC_PRIO_INTERACTIVE_BULK,
 190         ECN_OR_COST(INTERACTIVE_BULK),
 191         TC_PRIO_INTERACTIVE_BULK,
 192         ECN_OR_COST(INTERACTIVE_BULK)
 193 };
 194 EXPORT_SYMBOL(ip_tos2prio);
 195
 196 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 197 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
 198
 199 #ifdef CONFIG_PROC_FS
 200 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 201 {
 202         if (*pos)
 203                 return NULL;
 204         return SEQ_START_TOKEN;
 205 }
 206
 207 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 208 {
 209         ++*pos;
 210         return NULL;
 211 }
 212
 213 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 214 {
 215 }
 216
 217 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 218 {
 219         if (v == SEQ_START_TOKEN)
 220                 seq_printf(seq, "%-127s\n",
 221                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 222                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 223                            "HHUptod\tSpecDst");
 224         return 0;
 225 }
 226
 227 static const struct seq_operations rt_cache_seq_ops = {
 228         .start  = rt_cache_seq_start,
 229         .next   = rt_cache_seq_next,
 230         .stop   = rt_cache_seq_stop,
 231         .show   = rt_cache_seq_show,
 232 };
 233
 234 static int rt_cache_seq_open(struct inode *inode, struct file *file)
 235 {
 236         return seq_open(file, &rt_cache_seq_ops);
 237 }
 238
 239 static const struct file_operations rt_cache_seq_fops = {
 240         .owner   = THIS_MODULE,
 241         .open    = rt_cache_seq_open,
 242         .read    = seq_read,
 243         .llseek  = seq_lseek,
 244         .release = seq_release,
 245 };
 246
 247
 248 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 249 {
 250         int cpu;
 251
 252         if (*pos == 0)
 253                 return SEQ_START_TOKEN;
 254
 255         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 256                 if (!cpu_possible(cpu))
 257                         continue;
 258                 *pos = cpu+1;
 259                 return &per_cpu(rt_cache_stat, cpu);
 260         }
 261         return NULL;
 262 }
 263
 264 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 265 {
 266         int cpu;
 267
 268         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 269                 if (!cpu_possible(cpu))
 270                         continue;
 271                 *pos = cpu+1;
 272                 return &per_cpu(rt_cache_stat, cpu);
 273         }
 274         return NULL;
 275
 276 }
 277
 278 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 279 {
 280
 281 }
 282
 283 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 284 {
 285         struct rt_cache_stat *st = v;
 286
 287         if (v == SEQ_START_TOKEN) {
 288                 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");
 289                 return 0;
 290         }
 291
 292         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 293                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 294                    dst_entries_get_slow(&ipv4_dst_ops),
 295                    0, /* st->in_hit */
 296                    st->in_slow_tot,
 297                    st->in_slow_mc,
 298                    st->in_no_route,
 299                    st->in_brd,
 300                    st->in_martian_dst,
 301                    st->in_martian_src,
 302
 303                    0, /* st->out_hit */
 304                    st->out_slow_tot,
 305                    st->out_slow_mc,
 306
 307                    0, /* st->gc_total */
 308                    0, /* st->gc_ignored */
 309                    0, /* st->gc_goal_miss */
 310                    0, /* st->gc_dst_overflow */
 311                    0, /* st->in_hlist_search */
 312                    0  /* st->out_hlist_search */
 313                 );
 314         return 0;
 315 }
 316
 317 static const struct seq_operations rt_cpu_seq_ops = {
 318         .start  = rt_cpu_seq_start,
 319         .next   = rt_cpu_seq_next,
 320         .stop   = rt_cpu_seq_stop,
 321         .show   = rt_cpu_seq_show,
 322 };
 323
 324
 325 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
 326 {
 327         return seq_open(file, &rt_cpu_seq_ops);
 328 }
 329
 330 static const struct file_operations rt_cpu_seq_fops = {
 331         .owner   = THIS_MODULE,
 332         .open    = rt_cpu_seq_open,
 333         .read    = seq_read,
 334         .llseek  = seq_lseek,
 335         .release = seq_release,
 336 };
 337
 338 #ifdef CONFIG_IP_ROUTE_CLASSID
 339 static int rt_acct_proc_show(struct seq_file *m, void *v)
 340 {
 341         struct ip_rt_acct *dst, *src;
 342         unsigned int i, j;
 343
 344         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 345         if (!dst)
 346                 return -ENOMEM;
 347
 348         for_each_possible_cpu(i) {
 349                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 350                 for (j = 0; j < 256; j++) {
 351                         dst[j].o_bytes   += src[j].o_bytes;
 352                         dst[j].o_packets += src[j].o_packets;
 353                         dst[j].i_bytes   += src[j].i_bytes;
 354                         dst[j].i_packets += src[j].i_packets;
 355                 }
 356         }
 357
 358         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 359         kfree(dst);
 360         return 0;
 361 }
 362
 363 static int rt_acct_proc_open(struct inode *inode, struct file *file)
 364 {
 365         return single_open(file, rt_acct_proc_show, NULL);
 366 }
 367
 368 static const struct file_operations rt_acct_proc_fops = {
 369         .owner          = THIS_MODULE,
 370         .open           = rt_acct_proc_open,
 371         .read           = seq_read,
 372         .llseek         = seq_lseek,
 373         .release        = single_release,
 374 };
 375 #endif
 376
 377 static int __net_init ip_rt_do_proc_init(struct net *net)
 378 {
 379         struct proc_dir_entry *pde;
 380
 381         pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
 382                           &rt_cache_seq_fops);
 383         if (!pde)
 384                 goto err1;
 385
 386         pde = proc_create("rt_cache", S_IRUGO,
 387                           net->proc_net_stat, &rt_cpu_seq_fops);
 388         if (!pde)
 389                 goto err2;
 390
 391 #ifdef CONFIG_IP_ROUTE_CLASSID
 392         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
 393         if (!pde)
 394                 goto err3;
 395 #endif
 396         return 0;
 397
 398 #ifdef CONFIG_IP_ROUTE_CLASSID
 399 err3:
 400         remove_proc_entry("rt_cache", net->proc_net_stat);
 401 #endif
 402 err2:
 403         remove_proc_entry("rt_cache", net->proc_net);
 404 err1:
 405         return -ENOMEM;
 406 }
 407
 408 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 409 {
 410         remove_proc_entry("rt_cache", net->proc_net_stat);
 411         remove_proc_entry("rt_cache", net->proc_net);
 412 #ifdef CONFIG_IP_ROUTE_CLASSID
 413         remove_proc_entry("rt_acct", net->proc_net);
 414 #endif
 415 }
 416
 417 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 418         .init = ip_rt_do_proc_init,
 419         .exit = ip_rt_do_proc_exit,
 420 };
 421
 422 static int __init ip_rt_proc_init(void)
 423 {
 424         return register_pernet_subsys(&ip_rt_proc_ops);
 425 }
 426
 427 #else
 428 static inline int ip_rt_proc_init(void)
 429 {
 430         return 0;
 431 }
 432 #endif /* CONFIG_PROC_FS */
 433
 434 static inline bool rt_is_expired(const struct rtable *rth)
 435 {
 436         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 437 }
 438
 439 void rt_cache_flush(struct net *net)
 440 {
 441         rt_genid_bump_ipv4(net);
 442 }
 443
 444 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 445                                            struct sk_buff *skb,
 446                                            const void *daddr)
 447 {
 448         struct net_device *dev = dst->dev;
 449         const __be32 *pkey = daddr;
 450         const struct rtable *rt;
 451         struct neighbour *n;
 452
 453         rt = (const struct rtable *) dst;
 454         if (rt->rt_gateway)
 455                 pkey = (const __be32 *) &rt->rt_gateway;
 456         else if (skb)
 457                 pkey = &ip_hdr(skb)->daddr;
 458
 459         n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
 460         if (n)
 461                 return n;
 462         return neigh_create(&arp_tbl, pkey, dev);
 463 }
 464
 465 /*
 466  * Peer allocation may fail only in serious out-of-memory conditions.  However
 467  * we still can generate some output.
 468  * Random ID selection looks a bit dangerous because we have no chances to
 469  * select ID being unique in a reasonable period of time.
 470  * But broken packet identifier may be better than no packet at all.
 471  */
 472 static void ip_select_fb_ident(struct iphdr *iph)
 473 {
 474         static DEFINE_SPINLOCK(ip_fb_id_lock);
 475         static u32 ip_fallback_id;
 476         u32 salt;
 477
 478         spin_lock_bh(&ip_fb_id_lock);
 479         salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
 480         iph->id = htons(salt & 0xFFFF);
 481         ip_fallback_id = salt;
 482         spin_unlock_bh(&ip_fb_id_lock);
 483 }
 484
 485 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
 486 {
 487         struct net *net = dev_net(dst->dev);
 488         struct inet_peer *peer;
 489
 490         peer = inet_getpeer_v4(net->ipv4.peers, iph->daddr, 1);
 491         if (peer) {
 492                 iph->id = htons(inet_getid(peer, more));
 493                 inet_putpeer(peer);
 494                 return;
 495         }
 496
 497         ip_select_fb_ident(iph);
 498 }
 499 EXPORT_SYMBOL(__ip_select_ident);
 500
 501 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 502                              const struct iphdr *iph,
 503                              int oif, u8 tos,
 504                              u8 prot, u32 mark, int flow_flags)
 505 {
 506         if (sk) {
 507                 const struct inet_sock *inet = inet_sk(sk);
 508
 509                 oif = sk->sk_bound_dev_if;
 510                 mark = sk->sk_mark;
 511                 tos = RT_CONN_FLAGS(sk);
 512                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 513         }
 514         flowi4_init_output(fl4, oif, mark, tos,
 515                            RT_SCOPE_UNIVERSE, prot,
 516                            flow_flags,
 517                            iph->daddr, iph->saddr, 0, 0);
 518 }
 519
 520 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 521                                const struct sock *sk)
 522 {
 523         const struct iphdr *iph = ip_hdr(skb);
 524         int oif = skb->dev->ifindex;
 525         u8 tos = RT_TOS(iph->tos);
 526         u8 prot = iph->protocol;
 527         u32 mark = skb->mark;
 528
 529         __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
 530 }
 531
 532 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 533 {
 534         const struct inet_sock *inet = inet_sk(sk);
 535         const struct ip_options_rcu *inet_opt;
 536         __be32 daddr = inet->inet_daddr;
 537
 538         rcu_read_lock();
 539         inet_opt = rcu_dereference(inet->inet_opt);
 540         if (inet_opt && inet_opt->opt.srr)
 541                 daddr = inet_opt->opt.faddr;
 542         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 543                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 544                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 545                            inet_sk_flowi_flags(sk),
 546                            daddr, inet->inet_saddr, 0, 0);
 547         rcu_read_unlock();
 548 }
 549
 550 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 551                                  const struct sk_buff *skb)
 552 {
 553         if (skb)
 554                 build_skb_flow_key(fl4, skb, sk);
 555         else
 556                 build_sk_flow_key(fl4, sk);
 557 }
 558
 559 static inline void rt_free(struct rtable *rt)
 560 {
 561         call_rcu(&rt->dst.rcu_head, dst_rcu_free);
 562 }
 563
 564 static DEFINE_SPINLOCK(fnhe_lock);
 565
 566 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 567 {
 568         struct rtable *rt;
 569
 570         rt = rcu_dereference(fnhe->fnhe_rth_input);
 571         if (rt) {
 572                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 573                 rt_free(rt);
 574         }
 575         rt = rcu_dereference(fnhe->fnhe_rth_output);
 576         if (rt) {
 577                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 578                 rt_free(rt);
 579         }
 580 }
 581
 582 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
 583 {
 584         struct fib_nh_exception *fnhe, *oldest;
 585
 586         oldest = rcu_dereference(hash->chain);
 587         for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
 588              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 589                 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
 590                         oldest = fnhe;
 591         }
 592         fnhe_flush_routes(oldest);
 593         return oldest;
 594 }
 595
 596 static inline u32 fnhe_hashfun(__be32 daddr)
 597 {
 598         u32 hval;
 599
 600         hval = (__force u32) daddr;
 601         hval ^= (hval >> 11) ^ (hval >> 22);
 602
 603         return hval & (FNHE_HASH_SIZE - 1);
 604 }
 605
 606 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 607 {
 608         rt->rt_pmtu = fnhe->fnhe_pmtu;
 609         rt->dst.expires = fnhe->fnhe_expires;
 610
 611         if (fnhe->fnhe_gw) {
 612                 rt->rt_flags |= RTCF_REDIRECTED;
 613                 rt->rt_gateway = fnhe->fnhe_gw;
 614                 rt->rt_uses_gateway = 1;
 615         }
 616 }
 617
 618 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
 619                                   u32 pmtu, unsigned long expires)
 620 {
 621         struct fnhe_hash_bucket *hash;
 622         struct fib_nh_exception *fnhe;
 623         struct rtable *rt;
 624         unsigned int i;
 625         int depth;
 626         u32 hval = fnhe_hashfun(daddr);
 627
 628         spin_lock_bh(&fnhe_lock);
 629
 630         hash = nh->nh_exceptions;
 631         if (!hash) {
 632                 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
 633                 if (!hash)
 634                         goto out_unlock;
 635                 nh->nh_exceptions = hash;
 636         }
 637
 638         hash += hval;
 639
 640         depth = 0;
 641         for (fnhe = rcu_dereference(hash->chain); fnhe;
 642              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 643                 if (fnhe->fnhe_daddr == daddr)
 644                         break;
 645                 depth++;
 646         }
 647
 648         if (fnhe) {
 649                 if (gw)
 650                         fnhe->fnhe_gw = gw;
 651                 if (pmtu) {
 652                         fnhe->fnhe_pmtu = pmtu;
 653                         fnhe->fnhe_expires = max(1UL, expires);
 654                 }
 655                 /* Update all cached dsts too */
 656                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 657                 if (rt)
 658                         fill_route_from_fnhe(rt, fnhe);
 659                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 660                 if (rt)
 661                         fill_route_from_fnhe(rt, fnhe);
 662         } else {
 663                 if (depth > FNHE_RECLAIM_DEPTH)
 664                         fnhe = fnhe_oldest(hash);
 665                 else {
 666                         fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 667                         if (!fnhe)
 668                                 goto out_unlock;
 669
 670                         fnhe->fnhe_next = hash->chain;
 671                         rcu_assign_pointer(hash->chain, fnhe);
 672                 }
 673                 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
 674                 fnhe->fnhe_daddr = daddr;
 675                 fnhe->fnhe_gw = gw;
 676                 fnhe->fnhe_pmtu = pmtu;
 677                 fnhe->fnhe_expires = expires;
 678
 679                 /* Exception created; mark the cached routes for the nexthop
 680                  * stale, so anyone caching it rechecks if this exception
 681                  * applies to them.
 682                  */
 683                 rt = rcu_dereference(nh->nh_rth_input);
 684                 if (rt)
 685                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 686
 687                 for_each_possible_cpu(i) {
 688                         struct rtable __rcu **prt;
 689                         prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
 690                         rt = rcu_dereference(*prt);
 691                         if (rt)
 692                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 693                 }
 694         }
 695
 696         fnhe->fnhe_stamp = jiffies;
 697
 698 out_unlock:
 699         spin_unlock_bh(&fnhe_lock);
 700         return;
 701 }
 702
 703 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 704                              bool kill_route)
 705 {
 706         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 707         __be32 old_gw = ip_hdr(skb)->saddr;
 708         struct net_device *dev = skb->dev;
 709         struct in_device *in_dev;
 710         struct fib_result res;
 711         struct neighbour *n;
 712         struct net *net;
 713
 714         switch (icmp_hdr(skb)->code & 7) {
 715         case ICMP_REDIR_NET:
 716         case ICMP_REDIR_NETTOS:
 717         case ICMP_REDIR_HOST:
 718         case ICMP_REDIR_HOSTTOS:
 719                 break;
 720
 721         default:
 722                 return;
 723         }
 724
 725         if (rt->rt_gateway != old_gw)
 726                 return;
 727
 728         in_dev = __in_dev_get_rcu(dev);
 729         if (!in_dev)
 730                 return;
 731
 732         net = dev_net(dev);
 733         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 734             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 735             ipv4_is_zeronet(new_gw))
 736                 goto reject_redirect;
 737
 738         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 739                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 740                         goto reject_redirect;
 741                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 742                         goto reject_redirect;
 743         } else {
 744                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 745                         goto reject_redirect;
 746         }
 747
 748         n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
 749         if (n) {
 750                 if (!(n->nud_state & NUD_VALID)) {
 751                         neigh_event_send(n, NULL);
 752                 } else {
 753                         if (fib_lookup(net, fl4, &res) == 0) {
 754                                 struct fib_nh *nh = &FIB_RES_NH(res);
 755
 756                                 update_or_create_fnhe(nh, fl4->daddr, new_gw,
 757                                                       0, 0);
 758                         }
 759                         if (kill_route)
 760                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 761                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 762                 }
 763                 neigh_release(n);
 764         }
 765         return;
 766
 767 reject_redirect:
 768 #ifdef CONFIG_IP_ROUTE_VERBOSE
 769         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 770                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 771                 __be32 daddr = iph->daddr;
 772                 __be32 saddr = iph->saddr;
 773
 774                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 775                                      "  Advised path = %pI4 -> %pI4\n",
 776                                      &old_gw, dev->name, &new_gw,
 777                                      &saddr, &daddr);
 778         }
 779 #endif
 780         ;
 781 }
 782
 783 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 784 {
 785         struct rtable *rt;
 786         struct flowi4 fl4;
 787         const struct iphdr *iph = (const struct iphdr *) skb->data;
 788         int oif = skb->dev->ifindex;
 789         u8 tos = RT_TOS(iph->tos);
 790         u8 prot = iph->protocol;
 791         u32 mark = skb->mark;
 792
 793         rt = (struct rtable *) dst;
 794
 795         __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
 796         __ip_do_redirect(rt, skb, &fl4, true);
 797 }
 798
 799 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 800 {
 801         struct rtable *rt = (struct rtable *)dst;
 802         struct dst_entry *ret = dst;
 803
 804         if (rt) {
 805                 if (dst->obsolete > 0) {
 806                         ip_rt_put(rt);
 807                         ret = NULL;
 808                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 809                            rt->dst.expires) {
 810                         ip_rt_put(rt);
 811                         ret = NULL;
 812                 }
 813         }
 814         return ret;
 815 }
 816
 817 /*
 818  * Algorithm:
 819  *      1. The first ip_rt_redirect_number redirects are sent
 820  *         with exponential backoff, then we stop sending them at all,
 821  *         assuming that the host ignores our redirects.
 822  *      2. If we did not see packets requiring redirects
 823  *         during ip_rt_redirect_silence, we assume that the host
 824  *         forgot redirected route and start to send redirects again.
 825  *
 826  * This algorithm is much cheaper and more intelligent than dumb load limiting
 827  * in icmp.c.
 828  *
 829  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 830  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 831  */
 832
 833 void ip_rt_send_redirect(struct sk_buff *skb)
 834 {
 835         struct rtable *rt = skb_rtable(skb);
 836         struct in_device *in_dev;
 837         struct inet_peer *peer;
 838         struct net *net;
 839         int log_martians;
 840
 841         rcu_read_lock();
 842         in_dev = __in_dev_get_rcu(rt->dst.dev);
 843         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 844                 rcu_read_unlock();
 845                 return;
 846         }
 847         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 848         rcu_read_unlock();
 849
 850         net = dev_net(rt->dst.dev);
 851         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
 852         if (!peer) {
 853                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 854                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 855                 return;
 856         }
 857
 858         /* No redirected packets during ip_rt_redirect_silence;
 859          * reset the algorithm.
 860          */
 861         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
 862                 peer->rate_tokens = 0;
 863
 864         /* Too many ignored redirects; do not send anything
 865          * set dst.rate_last to the last seen redirected packet.
 866          */
 867         if (peer->rate_tokens >= ip_rt_redirect_number) {
 868                 peer->rate_last = jiffies;
 869                 goto out_put_peer;
 870         }
 871
 872         /* Check for load limit; set rate_last to the latest sent
 873          * redirect.
 874          */
 875         if (peer->rate_tokens == 0 ||
 876             time_after(jiffies,
 877                        (peer->rate_last +
 878                         (ip_rt_redirect_load << peer->rate_tokens)))) {
 879                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 880
 881                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 882                 peer->rate_last = jiffies;
 883                 ++peer->rate_tokens;
 884 #ifdef CONFIG_IP_ROUTE_VERBOSE
 885                 if (log_martians &&
 886                     peer->rate_tokens == ip_rt_redirect_number)
 887                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 888                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 889                                              &ip_hdr(skb)->daddr, &gw);
 890 #endif
 891         }
 892 out_put_peer:
 893         inet_putpeer(peer);
 894 }
 895
 896 static int ip_error(struct sk_buff *skb)
 897 {
 898         struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
 899         struct rtable *rt = skb_rtable(skb);
 900         struct inet_peer *peer;
 901         unsigned long now;
 902         struct net *net;
 903         bool send;
 904         int code;
 905
 906         net = dev_net(rt->dst.dev);
 907         if (!IN_DEV_FORWARD(in_dev)) {
 908                 switch (rt->dst.error) {
 909                 case EHOSTUNREACH:
 910                         IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
 911                         break;
 912
 913                 case ENETUNREACH:
 914                         IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
 915                         break;
 916                 }
 917                 goto out;
 918         }
 919
 920         switch (rt->dst.error) {
 921         case EINVAL:
 922         default:
 923                 goto out;
 924         case EHOSTUNREACH:
 925                 code = ICMP_HOST_UNREACH;
 926                 break;
 927         case ENETUNREACH:
 928                 code = ICMP_NET_UNREACH;
 929                 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
 930                 break;
 931         case EACCES:
 932                 code = ICMP_PKT_FILTERED;
 933                 break;
 934         }
 935
 936         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
 937
 938         send = true;
 939         if (peer) {
 940                 now = jiffies;
 941                 peer->rate_tokens += now - peer->rate_last;
 942                 if (peer->rate_tokens > ip_rt_error_burst)
 943                         peer->rate_tokens = ip_rt_error_burst;
 944                 peer->rate_last = now;
 945                 if (peer->rate_tokens >= ip_rt_error_cost)
 946                         peer->rate_tokens -= ip_rt_error_cost;
 947                 else
 948                         send = false;
 949                 inet_putpeer(peer);
 950         }
 951         if (send)
 952                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
 953
 954 out:    kfree_skb(skb);
 955         return 0;
 956 }
 957
 958 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
 959 {
 960         struct dst_entry *dst = &rt->dst;
 961         struct fib_result res;
 962
 963         if (dst_metric_locked(dst, RTAX_MTU))
 964                 return;
 965
 966         if (dst->dev->mtu < mtu)
 967                 return;
 968
 969         if (mtu < ip_rt_min_pmtu)
 970                 mtu = ip_rt_min_pmtu;
 971
 972         if (rt->rt_pmtu == mtu &&
 973             time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
 974                 return;
 975
 976         rcu_read_lock();
 977         if (fib_lookup(dev_net(dst->dev), fl4, &res) == 0) {
 978                 struct fib_nh *nh = &FIB_RES_NH(res);
 979
 980                 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
 981                                       jiffies + ip_rt_mtu_expires);
 982         }
 983         rcu_read_unlock();
 984 }
 985
 986 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 987                               struct sk_buff *skb, u32 mtu)
 988 {
 989         struct rtable *rt = (struct rtable *) dst;
 990         struct flowi4 fl4;
 991
 992         ip_rt_build_flow_key(&fl4, sk, skb);
 993         __ip_rt_update_pmtu(rt, &fl4, mtu);
 994 }
 995
 996 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
 997                       int oif, u32 mark, u8 protocol, int flow_flags)
 998 {
 999         const struct iphdr *iph = (const struct iphdr *) skb->data;
1000         struct flowi4 fl4;
1001         struct rtable *rt;
1002
1003         __build_flow_key(&fl4, NULL, iph, oif,
1004                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1005         rt = __ip_route_output_key(net, &fl4);
1006         if (!IS_ERR(rt)) {
1007                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1008                 ip_rt_put(rt);
1009         }
1010 }
1011 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1012
1013 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1014 {
1015         const struct iphdr *iph = (const struct iphdr *) skb->data;
1016         struct flowi4 fl4;
1017         struct rtable *rt;
1018
1019         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1020         rt = __ip_route_output_key(sock_net(sk), &fl4);
1021         if (!IS_ERR(rt)) {
1022                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1023                 ip_rt_put(rt);
1024         }
1025 }
1026
1027 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1028 {
1029         const struct iphdr *iph = (const struct iphdr *) skb->data;
1030         struct flowi4 fl4;
1031         struct rtable *rt;
1032         struct dst_entry *odst = NULL;
1033         bool new = false;
1034
1035         bh_lock_sock(sk);
1036
1037         if (!ip_sk_accept_pmtu(sk))
1038                 goto out;
1039
1040         odst = sk_dst_get(sk);
1041
1042         if (sock_owned_by_user(sk) || !odst) {
1043                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1044                 goto out;
1045         }
1046
1047         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1048
1049         rt = (struct rtable *)odst;
1050         if (odst->obsolete && odst->ops->check(odst, 0) == NULL) {
1051                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1052                 if (IS_ERR(rt))
1053                         goto out;
1054
1055                 new = true;
1056         }
1057
1058         __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1059
1060         if (!dst_check(&rt->dst, 0)) {
1061                 if (new)
1062                         dst_release(&rt->dst);
1063
1064                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1065                 if (IS_ERR(rt))
1066                         goto out;
1067
1068                 new = true;
1069         }
1070
1071         if (new)
1072                 sk_dst_set(sk, &rt->dst);
1073
1074 out:
1075         bh_unlock_sock(sk);
1076         dst_release(odst);
1077 }
1078 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1079
1080 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1081                    int oif, u32 mark, u8 protocol, int flow_flags)
1082 {
1083         const struct iphdr *iph = (const struct iphdr *) skb->data;
1084         struct flowi4 fl4;
1085         struct rtable *rt;
1086
1087         __build_flow_key(&fl4, NULL, iph, oif,
1088                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1089         rt = __ip_route_output_key(net, &fl4);
1090         if (!IS_ERR(rt)) {
1091                 __ip_do_redirect(rt, skb, &fl4, false);
1092                 ip_rt_put(rt);
1093         }
1094 }
1095 EXPORT_SYMBOL_GPL(ipv4_redirect);
1096
1097 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1098 {
1099         const struct iphdr *iph = (const struct iphdr *) skb->data;
1100         struct flowi4 fl4;
1101         struct rtable *rt;
1102
1103         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1104         rt = __ip_route_output_key(sock_net(sk), &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_sk_redirect);
1111
1112 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1113 {
1114         struct rtable *rt = (struct rtable *) dst;
1115
1116         /* All IPV4 dsts are created with ->obsolete set to the value
1117          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1118          * into this function always.
1119          *
1120          * When a PMTU/redirect information update invalidates a route,
1121          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1122          * DST_OBSOLETE_DEAD by dst_free().
1123          */
1124         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1125                 return NULL;
1126         return dst;
1127 }
1128
1129 static void ipv4_link_failure(struct sk_buff *skb)
1130 {
1131         struct rtable *rt;
1132
1133         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1134
1135         rt = skb_rtable(skb);
1136         if (rt)
1137                 dst_set_expires(&rt->dst, 0);
1138 }
1139
1140 static int ip_rt_bug(struct sk_buff *skb)
1141 {
1142         pr_debug("%s: %pI4 -> %pI4, %s\n",
1143                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1144                  skb->dev ? skb->dev->name : "?");
1145         kfree_skb(skb);
1146         WARN_ON(1);
1147         return 0;
1148 }
1149
1150 /*
1151    We do not cache source address of outgoing interface,
1152    because it is used only by IP RR, TS and SRR options,
1153    so that it out of fast path.
1154
1155    BTW remember: "addr" is allowed to be not aligned
1156    in IP options!
1157  */
1158
1159 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1160 {
1161         __be32 src;
1162
1163         if (rt_is_output_route(rt))
1164                 src = ip_hdr(skb)->saddr;
1165         else {
1166                 struct fib_result res;
1167                 struct flowi4 fl4;
1168                 struct iphdr *iph;
1169
1170                 iph = ip_hdr(skb);
1171
1172                 memset(&fl4, 0, sizeof(fl4));
1173                 fl4.daddr = iph->daddr;
1174                 fl4.saddr = iph->saddr;
1175                 fl4.flowi4_tos = RT_TOS(iph->tos);
1176                 fl4.flowi4_oif = rt->dst.dev->ifindex;
1177                 fl4.flowi4_iif = skb->dev->ifindex;
1178                 fl4.flowi4_mark = skb->mark;
1179
1180                 rcu_read_lock();
1181                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1182                         src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1183                 else
1184                         src = inet_select_addr(rt->dst.dev,
1185                                                rt_nexthop(rt, iph->daddr),
1186                                                RT_SCOPE_UNIVERSE);
1187                 rcu_read_unlock();
1188         }
1189         memcpy(addr, &src, 4);
1190 }
1191
1192 #ifdef CONFIG_IP_ROUTE_CLASSID
1193 static void set_class_tag(struct rtable *rt, u32 tag)
1194 {
1195         if (!(rt->dst.tclassid & 0xFFFF))
1196                 rt->dst.tclassid |= tag & 0xFFFF;
1197         if (!(rt->dst.tclassid & 0xFFFF0000))
1198                 rt->dst.tclassid |= tag & 0xFFFF0000;
1199 }
1200 #endif
1201
1202 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1203 {
1204         unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1205
1206         if (advmss == 0) {
1207                 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1208                                ip_rt_min_advmss);
1209                 if (advmss > 65535 - 40)
1210                         advmss = 65535 - 40;
1211         }
1212         return advmss;
1213 }
1214
1215 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1216 {
1217         const struct rtable *rt = (const struct rtable *) dst;
1218         unsigned int mtu = rt->rt_pmtu;
1219
1220         if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1221                 mtu = dst_metric_raw(dst, RTAX_MTU);
1222
1223         if (mtu)
1224                 return mtu;
1225
1226         mtu = dst->dev->mtu;
1227
1228         if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1229                 if (rt->rt_uses_gateway && mtu > 576)
1230                         mtu = 576;
1231         }
1232
1233         return min_t(unsigned int, mtu, IP_MAX_MTU);
1234 }
1235
1236 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1237 {
1238         struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1239         struct fib_nh_exception *fnhe;
1240         u32 hval;
1241
1242         if (!hash)
1243                 return NULL;
1244
1245         hval = fnhe_hashfun(daddr);
1246
1247         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1248              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1249                 if (fnhe->fnhe_daddr == daddr)
1250                         return fnhe;
1251         }
1252         return NULL;
1253 }
1254
1255 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1256                               __be32 daddr)
1257 {
1258         bool ret = false;
1259
1260         spin_lock_bh(&fnhe_lock);
1261
1262         if (daddr == fnhe->fnhe_daddr) {
1263                 struct rtable __rcu **porig;
1264                 struct rtable *orig;
1265                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1266
1267                 if (rt_is_input_route(rt))
1268                         porig = &fnhe->fnhe_rth_input;
1269                 else
1270                         porig = &fnhe->fnhe_rth_output;
1271                 orig = rcu_dereference(*porig);
1272
1273                 if (fnhe->fnhe_genid != genid) {
1274                         fnhe->fnhe_genid = genid;
1275                         fnhe->fnhe_gw = 0;
1276                         fnhe->fnhe_pmtu = 0;
1277                         fnhe->fnhe_expires = 0;
1278                         fnhe_flush_routes(fnhe);
1279                         orig = NULL;
1280                 }
1281                 fill_route_from_fnhe(rt, fnhe);
1282                 if (!rt->rt_gateway)
1283                         rt->rt_gateway = daddr;
1284
1285                 if (!(rt->dst.flags & DST_NOCACHE)) {
1286                         rcu_assign_pointer(*porig, rt);
1287                         if (orig)
1288                                 rt_free(orig);
1289                         ret = true;
1290                 }
1291
1292                 fnhe->fnhe_stamp = jiffies;
1293         }
1294         spin_unlock_bh(&fnhe_lock);
1295
1296         return ret;
1297 }
1298
1299 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1300 {
1301         struct rtable *orig, *prev, **p;
1302         bool ret = true;
1303
1304         if (rt_is_input_route(rt)) {
1305                 p = (struct rtable **)&nh->nh_rth_input;
1306         } else {
1307                 p = (struct rtable **)__this_cpu_ptr(nh->nh_pcpu_rth_output);
1308         }
1309         orig = *p;
1310
1311         prev = cmpxchg(p, orig, rt);
1312         if (prev == orig) {
1313                 if (orig)
1314                         rt_free(orig);
1315         } else
1316                 ret = false;
1317
1318         return ret;
1319 }
1320
1321 static DEFINE_SPINLOCK(rt_uncached_lock);
1322 static LIST_HEAD(rt_uncached_list);
1323
1324 static void rt_add_uncached_list(struct rtable *rt)
1325 {
1326         spin_lock_bh(&rt_uncached_lock);
1327         list_add_tail(&rt->rt_uncached, &rt_uncached_list);
1328         spin_unlock_bh(&rt_uncached_lock);
1329 }
1330
1331 static void ipv4_dst_destroy(struct dst_entry *dst)
1332 {
1333         struct rtable *rt = (struct rtable *) dst;
1334
1335         if (!list_empty(&rt->rt_uncached)) {
1336                 spin_lock_bh(&rt_uncached_lock);
1337                 list_del(&rt->rt_uncached);
1338                 spin_unlock_bh(&rt_uncached_lock);
1339         }
1340 }
1341
1342 void rt_flush_dev(struct net_device *dev)
1343 {
1344         if (!list_empty(&rt_uncached_list)) {
1345                 struct net *net = dev_net(dev);
1346                 struct rtable *rt;
1347
1348                 spin_lock_bh(&rt_uncached_lock);
1349                 list_for_each_entry(rt, &rt_uncached_list, rt_uncached) {
1350                         if (rt->dst.dev != dev)
1351                                 continue;
1352                         rt->dst.dev = net->loopback_dev;
1353                         dev_hold(rt->dst.dev);
1354                         dev_put(dev);
1355                 }
1356                 spin_unlock_bh(&rt_uncached_lock);
1357         }
1358 }
1359
1360 static bool rt_cache_valid(const struct rtable *rt)
1361 {
1362         return  rt &&
1363                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1364                 !rt_is_expired(rt);
1365 }
1366
1367 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1368                            const struct fib_result *res,
1369                            struct fib_nh_exception *fnhe,
1370                            struct fib_info *fi, u16 type, u32 itag)
1371 {
1372         bool cached = false;
1373
1374         if (fi) {
1375                 struct fib_nh *nh = &FIB_RES_NH(*res);
1376
1377                 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1378                         rt->rt_gateway = nh->nh_gw;
1379                         rt->rt_uses_gateway = 1;
1380                 }
1381                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1382 #ifdef CONFIG_IP_ROUTE_CLASSID
1383                 rt->dst.tclassid = nh->nh_tclassid;
1384 #endif
1385                 if (unlikely(fnhe))
1386                         cached = rt_bind_exception(rt, fnhe, daddr);
1387                 else if (!(rt->dst.flags & DST_NOCACHE))
1388                         cached = rt_cache_route(nh, rt);
1389                 if (unlikely(!cached)) {
1390                         /* Routes we intend to cache in nexthop exception or
1391                          * FIB nexthop have the DST_NOCACHE bit clear.
1392                          * However, if we are unsuccessful at storing this
1393                          * route into the cache we really need to set it.
1394                          */
1395                         rt->dst.flags |= DST_NOCACHE;
1396                         if (!rt->rt_gateway)
1397                                 rt->rt_gateway = daddr;
1398                         rt_add_uncached_list(rt);
1399                 }
1400         } else
1401                 rt_add_uncached_list(rt);
1402
1403 #ifdef CONFIG_IP_ROUTE_CLASSID
1404 #ifdef CONFIG_IP_MULTIPLE_TABLES
1405         set_class_tag(rt, res->tclassid);
1406 #endif
1407         set_class_tag(rt, itag);
1408 #endif
1409 }
1410
1411 static struct rtable *rt_dst_alloc(struct net_device *dev,
1412                                    bool nopolicy, bool noxfrm, bool will_cache)
1413 {
1414         return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1415                          (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1416                          (nopolicy ? DST_NOPOLICY : 0) |
1417                          (noxfrm ? DST_NOXFRM : 0));
1418 }
1419
1420 /* called in rcu_read_lock() section */
1421 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1422                                 u8 tos, struct net_device *dev, int our)
1423 {
1424         struct rtable *rth;
1425         struct in_device *in_dev = __in_dev_get_rcu(dev);
1426         u32 itag = 0;
1427         int err;
1428
1429         /* Primary sanity checks. */
1430
1431         if (in_dev == NULL)
1432                 return -EINVAL;
1433
1434         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1435             skb->protocol != htons(ETH_P_IP))
1436                 goto e_inval;
1437
1438         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1439                 if (ipv4_is_loopback(saddr))
1440                         goto e_inval;
1441
1442         if (ipv4_is_zeronet(saddr)) {
1443                 if (!ipv4_is_local_multicast(daddr))
1444                         goto e_inval;
1445         } else {
1446                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1447                                           in_dev, &itag);
1448                 if (err < 0)
1449                         goto e_err;
1450         }
1451         rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1452                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1453         if (!rth)
1454                 goto e_nobufs;
1455
1456 #ifdef CONFIG_IP_ROUTE_CLASSID
1457         rth->dst.tclassid = itag;
1458 #endif
1459         rth->dst.output = ip_rt_bug;
1460
1461         rth->rt_genid   = rt_genid_ipv4(dev_net(dev));
1462         rth->rt_flags   = RTCF_MULTICAST;
1463         rth->rt_type    = RTN_MULTICAST;
1464         rth->rt_is_input= 1;
1465         rth->rt_iif     = 0;
1466         rth->rt_pmtu    = 0;
1467         rth->rt_gateway = 0;
1468         rth->rt_uses_gateway = 0;
1469         INIT_LIST_HEAD(&rth->rt_uncached);
1470         if (our) {
1471                 rth->dst.input= ip_local_deliver;
1472                 rth->rt_flags |= RTCF_LOCAL;
1473         }
1474
1475 #ifdef CONFIG_IP_MROUTE
1476         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1477                 rth->dst.input = ip_mr_input;
1478 #endif
1479         RT_CACHE_STAT_INC(in_slow_mc);
1480
1481         skb_dst_set(skb, &rth->dst);
1482         return 0;
1483
1484 e_nobufs:
1485         return -ENOBUFS;
1486 e_inval:
1487         return -EINVAL;
1488 e_err:
1489         return err;
1490 }
1491
1492
1493 static void ip_handle_martian_source(struct net_device *dev,
1494                                      struct in_device *in_dev,
1495                                      struct sk_buff *skb,
1496                                      __be32 daddr,
1497                                      __be32 saddr)
1498 {
1499         RT_CACHE_STAT_INC(in_martian_src);
1500 #ifdef CONFIG_IP_ROUTE_VERBOSE
1501         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1502                 /*
1503                  *      RFC1812 recommendation, if source is martian,
1504                  *      the only hint is MAC header.
1505                  */
1506                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1507                         &daddr, &saddr, dev->name);
1508                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1509                         print_hex_dump(KERN_WARNING, "ll header: ",
1510                                        DUMP_PREFIX_OFFSET, 16, 1,
1511                                        skb_mac_header(skb),
1512                                        dev->hard_header_len, true);
1513                 }
1514         }
1515 #endif
1516 }
1517
1518 /* called in rcu_read_lock() section */
1519 static int __mkroute_input(struct sk_buff *skb,
1520                            const struct fib_result *res,
1521                            struct in_device *in_dev,
1522                            __be32 daddr, __be32 saddr, u32 tos)
1523 {
1524         struct fib_nh_exception *fnhe;
1525         struct rtable *rth;
1526         int err;
1527         struct in_device *out_dev;
1528         unsigned int flags = 0;
1529         bool do_cache;
1530         u32 itag = 0;
1531
1532         /* get a working reference to the output device */
1533         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1534         if (out_dev == NULL) {
1535                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1536                 return -EINVAL;
1537         }
1538
1539         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1540                                   in_dev->dev, in_dev, &itag);
1541         if (err < 0) {
1542                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1543                                          saddr);
1544
1545                 goto cleanup;
1546         }
1547
1548         do_cache = res->fi && !itag;
1549         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1550             (IN_DEV_SHARED_MEDIA(out_dev) ||
1551              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) {
1552                 flags |= RTCF_DOREDIRECT;
1553                 do_cache = false;
1554         }
1555
1556         if (skb->protocol != htons(ETH_P_IP)) {
1557                 /* Not IP (i.e. ARP). Do not create route, if it is
1558                  * invalid for proxy arp. DNAT routes are always valid.
1559                  *
1560                  * Proxy arp feature have been extended to allow, ARP
1561                  * replies back to the same interface, to support
1562                  * Private VLAN switch technologies. See arp.c.
1563                  */
1564                 if (out_dev == in_dev &&
1565                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1566                         err = -EINVAL;
1567                         goto cleanup;
1568                 }
1569         }
1570
1571         fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1572         if (do_cache) {
1573                 if (fnhe != NULL)
1574                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1575                 else
1576                         rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1577
1578                 if (rt_cache_valid(rth)) {
1579                         skb_dst_set_noref(skb, &rth->dst);
1580                         goto out;
1581                 }
1582         }
1583
1584         rth = rt_dst_alloc(out_dev->dev,
1585                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1586                            IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1587         if (!rth) {
1588                 err = -ENOBUFS;
1589                 goto cleanup;
1590         }
1591
1592         rth->rt_genid = rt_genid_ipv4(dev_net(rth->dst.dev));
1593         rth->rt_flags = flags;
1594         rth->rt_type = res->type;
1595         rth->rt_is_input = 1;
1596         rth->rt_iif     = 0;
1597         rth->rt_pmtu    = 0;
1598         rth->rt_gateway = 0;
1599         rth->rt_uses_gateway = 0;
1600         INIT_LIST_HEAD(&rth->rt_uncached);
1601         RT_CACHE_STAT_INC(in_slow_tot);
1602
1603         rth->dst.input = ip_forward;
1604         rth->dst.output = ip_output;
1605
1606         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1607         skb_dst_set(skb, &rth->dst);
1608 out:
1609         err = 0;
1610  cleanup:
1611         return err;
1612 }
1613
1614 static int ip_mkroute_input(struct sk_buff *skb,
1615                             struct fib_result *res,
1616                             const struct flowi4 *fl4,
1617                             struct in_device *in_dev,
1618                             __be32 daddr, __be32 saddr, u32 tos)
1619 {
1620 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1621         if (res->fi && res->fi->fib_nhs > 1)
1622                 fib_select_multipath(res);
1623 #endif
1624
1625         /* create a routing cache entry */
1626         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1627 }
1628
1629 /*
1630  *      NOTE. We drop all the packets that has local source
1631  *      addresses, because every properly looped back packet
1632  *      must have correct destination already attached by output routine.
1633  *
1634  *      Such approach solves two big problems:
1635  *      1. Not simplex devices are handled properly.
1636  *      2. IP spoofing attempts are filtered with 100% of guarantee.
1637  *      called with rcu_read_lock()
1638  */
1639
1640 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1641                                u8 tos, struct net_device *dev)
1642 {
1643         struct fib_result res;
1644         struct in_device *in_dev = __in_dev_get_rcu(dev);
1645         struct flowi4   fl4;
1646         unsigned int    flags = 0;
1647         u32             itag = 0;
1648         struct rtable   *rth;
1649         int             err = -EINVAL;
1650         struct net    *net = dev_net(dev);
1651         bool do_cache;
1652
1653         /* IP on this device is disabled. */
1654
1655         if (!in_dev)
1656                 goto out;
1657
1658         /* Check for the most weird martians, which can be not detected
1659            by fib_lookup.
1660          */
1661
1662         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1663                 goto martian_source;
1664
1665         res.fi = NULL;
1666         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1667                 goto brd_input;
1668
1669         /* Accept zero addresses only to limited broadcast;
1670          * I even do not know to fix it or not. Waiting for complains :-)
1671          */
1672         if (ipv4_is_zeronet(saddr))
1673                 goto martian_source;
1674
1675         if (ipv4_is_zeronet(daddr))
1676                 goto martian_destination;
1677
1678         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1679          * and call it once if daddr or/and saddr are loopback addresses
1680          */
1681         if (ipv4_is_loopback(daddr)) {
1682                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1683                         goto martian_destination;
1684         } else if (ipv4_is_loopback(saddr)) {
1685                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1686                         goto martian_source;
1687         }
1688
1689         /*
1690          *      Now we are ready to route packet.
1691          */
1692         fl4.flowi4_oif = 0;
1693         fl4.flowi4_iif = dev->ifindex;
1694         fl4.flowi4_mark = skb->mark;
1695         fl4.flowi4_tos = tos;
1696         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1697         fl4.daddr = daddr;
1698         fl4.saddr = saddr;
1699         err = fib_lookup(net, &fl4, &res);
1700         if (err != 0) {
1701                 if (!IN_DEV_FORWARD(in_dev))
1702                         err = -EHOSTUNREACH;
1703                 goto no_route;
1704         }
1705
1706         if (res.type == RTN_BROADCAST)
1707                 goto brd_input;
1708
1709         if (res.type == RTN_LOCAL) {
1710                 err = fib_validate_source(skb, saddr, daddr, tos,
1711                                           LOOPBACK_IFINDEX,
1712                                           dev, in_dev, &itag);
1713                 if (err < 0)
1714                         goto martian_source_keep_err;
1715                 goto local_input;
1716         }
1717
1718         if (!IN_DEV_FORWARD(in_dev)) {
1719                 err = -EHOSTUNREACH;
1720                 goto no_route;
1721         }
1722         if (res.type != RTN_UNICAST)
1723                 goto martian_destination;
1724
1725         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1726 out:    return err;
1727
1728 brd_input:
1729         if (skb->protocol != htons(ETH_P_IP))
1730                 goto e_inval;
1731
1732         if (!ipv4_is_zeronet(saddr)) {
1733                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1734                                           in_dev, &itag);
1735                 if (err < 0)
1736                         goto martian_source_keep_err;
1737         }
1738         flags |= RTCF_BROADCAST;
1739         res.type = RTN_BROADCAST;
1740         RT_CACHE_STAT_INC(in_brd);
1741
1742 local_input:
1743         do_cache = false;
1744         if (res.fi) {
1745                 if (!itag) {
1746                         rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1747                         if (rt_cache_valid(rth)) {
1748                                 skb_dst_set_noref(skb, &rth->dst);
1749                                 err = 0;
1750                                 goto out;
1751                         }
1752                         do_cache = true;
1753                 }
1754         }
1755
1756         rth = rt_dst_alloc(net->loopback_dev,
1757                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1758         if (!rth)
1759                 goto e_nobufs;
1760
1761         rth->dst.input= ip_local_deliver;
1762         rth->dst.output= ip_rt_bug;
1763 #ifdef CONFIG_IP_ROUTE_CLASSID
1764         rth->dst.tclassid = itag;
1765 #endif
1766
1767         rth->rt_genid = rt_genid_ipv4(net);
1768         rth->rt_flags   = flags|RTCF_LOCAL;
1769         rth->rt_type    = res.type;
1770         rth->rt_is_input = 1;
1771         rth->rt_iif     = 0;
1772         rth->rt_pmtu    = 0;
1773         rth->rt_gateway = 0;
1774         rth->rt_uses_gateway = 0;
1775         INIT_LIST_HEAD(&rth->rt_uncached);
1776         RT_CACHE_STAT_INC(in_slow_tot);
1777         if (res.type == RTN_UNREACHABLE) {
1778                 rth->dst.input= ip_error;
1779                 rth->dst.error= -err;
1780                 rth->rt_flags   &= ~RTCF_LOCAL;
1781         }
1782         if (do_cache) {
1783                 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1784                         rth->dst.flags |= DST_NOCACHE;
1785                         rt_add_uncached_list(rth);
1786                 }
1787         }
1788         skb_dst_set(skb, &rth->dst);
1789         err = 0;
1790         goto out;
1791
1792 no_route:
1793         RT_CACHE_STAT_INC(in_no_route);
1794         res.type = RTN_UNREACHABLE;
1795         if (err == -ESRCH)
1796                 err = -ENETUNREACH;
1797         goto local_input;
1798
1799         /*
1800          *      Do not cache martian addresses: they should be logged (RFC1812)
1801          */
1802 martian_destination:
1803         RT_CACHE_STAT_INC(in_martian_dst);
1804 #ifdef CONFIG_IP_ROUTE_VERBOSE
1805         if (IN_DEV_LOG_MARTIANS(in_dev))
1806                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1807                                      &daddr, &saddr, dev->name);
1808 #endif
1809
1810 e_inval:
1811         err = -EINVAL;
1812         goto out;
1813
1814 e_nobufs:
1815         err = -ENOBUFS;
1816         goto out;
1817
1818 martian_source:
1819         err = -EINVAL;
1820 martian_source_keep_err:
1821         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1822         goto out;
1823 }
1824
1825 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1826                          u8 tos, struct net_device *dev)
1827 {
1828         int res;
1829
1830         rcu_read_lock();
1831
1832         /* Multicast recognition logic is moved from route cache to here.
1833            The problem was that too many Ethernet cards have broken/missing
1834            hardware multicast filters :-( As result the host on multicasting
1835            network acquires a lot of useless route cache entries, sort of
1836            SDR messages from all the world. Now we try to get rid of them.
1837            Really, provided software IP multicast filter is organized
1838            reasonably (at least, hashed), it does not result in a slowdown
1839            comparing with route cache reject entries.
1840            Note, that multicast routers are not affected, because
1841            route cache entry is created eventually.
1842          */
1843         if (ipv4_is_multicast(daddr)) {
1844                 struct in_device *in_dev = __in_dev_get_rcu(dev);
1845
1846                 if (in_dev) {
1847                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1848                                                   ip_hdr(skb)->protocol);
1849                         if (our
1850 #ifdef CONFIG_IP_MROUTE
1851                                 ||
1852                             (!ipv4_is_local_multicast(daddr) &&
1853                              IN_DEV_MFORWARD(in_dev))
1854 #endif
1855                            ) {
1856                                 int res = ip_route_input_mc(skb, daddr, saddr,
1857                                                             tos, dev, our);
1858                                 rcu_read_unlock();
1859                                 return res;
1860                         }
1861                 }
1862                 rcu_read_unlock();
1863                 return -EINVAL;
1864         }
1865         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1866         rcu_read_unlock();
1867         return res;
1868 }
1869 EXPORT_SYMBOL(ip_route_input_noref);
1870
1871 /* called with rcu_read_lock() */
1872 static struct rtable *__mkroute_output(const struct fib_result *res,
1873                                        const struct flowi4 *fl4, int orig_oif,
1874                                        struct net_device *dev_out,
1875                                        unsigned int flags)
1876 {
1877         struct fib_info *fi = res->fi;
1878         struct fib_nh_exception *fnhe;
1879         struct in_device *in_dev;
1880         u16 type = res->type;
1881         struct rtable *rth;
1882         bool do_cache;
1883
1884         in_dev = __in_dev_get_rcu(dev_out);
1885         if (!in_dev)
1886                 return ERR_PTR(-EINVAL);
1887
1888         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1889                 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1890                         return ERR_PTR(-EINVAL);
1891
1892         if (ipv4_is_lbcast(fl4->daddr))
1893                 type = RTN_BROADCAST;
1894         else if (ipv4_is_multicast(fl4->daddr))
1895                 type = RTN_MULTICAST;
1896         else if (ipv4_is_zeronet(fl4->daddr))
1897                 return ERR_PTR(-EINVAL);
1898
1899         if (dev_out->flags & IFF_LOOPBACK)
1900                 flags |= RTCF_LOCAL;
1901
1902         do_cache = true;
1903         if (type == RTN_BROADCAST) {
1904                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1905                 fi = NULL;
1906         } else if (type == RTN_MULTICAST) {
1907                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1908                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1909                                      fl4->flowi4_proto))
1910                         flags &= ~RTCF_LOCAL;
1911                 else
1912                         do_cache = false;
1913                 /* If multicast route do not exist use
1914                  * default one, but do not gateway in this case.
1915                  * Yes, it is hack.
1916                  */
1917                 if (fi && res->prefixlen < 4)
1918                         fi = NULL;
1919         }
1920
1921         fnhe = NULL;
1922         do_cache &= fi != NULL;
1923         if (do_cache) {
1924                 struct rtable __rcu **prth;
1925                 struct fib_nh *nh = &FIB_RES_NH(*res);
1926
1927                 fnhe = find_exception(nh, fl4->daddr);
1928                 if (fnhe)
1929                         prth = &fnhe->fnhe_rth_output;
1930                 else {
1931                         if (unlikely(fl4->flowi4_flags &
1932                                      FLOWI_FLAG_KNOWN_NH &&
1933                                      !(nh->nh_gw &&
1934                                        nh->nh_scope == RT_SCOPE_LINK))) {
1935                                 do_cache = false;
1936                                 goto add;
1937                         }
1938                         prth = __this_cpu_ptr(nh->nh_pcpu_rth_output);
1939                 }
1940                 rth = rcu_dereference(*prth);
1941                 if (rt_cache_valid(rth)) {
1942                         dst_hold(&rth->dst);
1943                         return rth;
1944                 }
1945         }
1946
1947 add:
1948         rth = rt_dst_alloc(dev_out,
1949                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1950                            IN_DEV_CONF_GET(in_dev, NOXFRM),
1951                            do_cache);
1952         if (!rth)
1953                 return ERR_PTR(-ENOBUFS);
1954
1955         rth->dst.output = ip_output;
1956
1957         rth->rt_genid = rt_genid_ipv4(dev_net(dev_out));
1958         rth->rt_flags   = flags;
1959         rth->rt_type    = type;
1960         rth->rt_is_input = 0;
1961         rth->rt_iif     = orig_oif ? : 0;
1962         rth->rt_pmtu    = 0;
1963         rth->rt_gateway = 0;
1964         rth->rt_uses_gateway = 0;
1965         INIT_LIST_HEAD(&rth->rt_uncached);
1966
1967         RT_CACHE_STAT_INC(out_slow_tot);
1968
1969         if (flags & RTCF_LOCAL)
1970                 rth->dst.input = ip_local_deliver;
1971         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1972                 if (flags & RTCF_LOCAL &&
1973                     !(dev_out->flags & IFF_LOOPBACK)) {
1974                         rth->dst.output = ip_mc_output;
1975                         RT_CACHE_STAT_INC(out_slow_mc);
1976                 }
1977 #ifdef CONFIG_IP_MROUTE
1978                 if (type == RTN_MULTICAST) {
1979                         if (IN_DEV_MFORWARD(in_dev) &&
1980                             !ipv4_is_local_multicast(fl4->daddr)) {
1981                                 rth->dst.input = ip_mr_input;
1982                                 rth->dst.output = ip_mc_output;
1983                         }
1984                 }
1985 #endif
1986         }
1987
1988         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
1989
1990         return rth;
1991 }
1992
1993 /*
1994  * Major route resolver routine.
1995  */
1996
1997 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
1998 {
1999         struct net_device *dev_out = NULL;
2000         __u8 tos = RT_FL_TOS(fl4);
2001         unsigned int flags = 0;
2002         struct fib_result res;
2003         struct rtable *rth;
2004         int orig_oif;
2005
2006         res.tclassid    = 0;
2007         res.fi          = NULL;
2008         res.table       = NULL;
2009
2010         orig_oif = fl4->flowi4_oif;
2011
2012         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2013         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2014         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2015                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2016
2017         rcu_read_lock();
2018         if (fl4->saddr) {
2019                 rth = ERR_PTR(-EINVAL);
2020                 if (ipv4_is_multicast(fl4->saddr) ||
2021                     ipv4_is_lbcast(fl4->saddr) ||
2022                     ipv4_is_zeronet(fl4->saddr))
2023                         goto out;
2024
2025                 /* I removed check for oif == dev_out->oif here.
2026                    It was wrong for two reasons:
2027                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2028                       is assigned to multiple interfaces.
2029                    2. Moreover, we are allowed to send packets with saddr
2030                       of another iface. --ANK
2031                  */
2032
2033                 if (fl4->flowi4_oif == 0 &&
2034                     (ipv4_is_multicast(fl4->daddr) ||
2035                      ipv4_is_lbcast(fl4->daddr))) {
2036                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2037                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2038                         if (dev_out == NULL)
2039                                 goto out;
2040
2041                         /* Special hack: user can direct multicasts
2042                            and limited broadcast via necessary interface
2043                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2044                            This hack is not just for fun, it allows
2045                            vic,vat and friends to work.
2046                            They bind socket to loopback, set ttl to zero
2047                            and expect that it will work.
2048                            From the viewpoint of routing cache they are broken,
2049                            because we are not allowed to build multicast path
2050                            with loopback source addr (look, routing cache
2051                            cannot know, that ttl is zero, so that packet
2052                            will not leave this host and route is valid).
2053                            Luckily, this hack is good workaround.
2054                          */
2055
2056                         fl4->flowi4_oif = dev_out->ifindex;
2057                         goto make_route;
2058                 }
2059
2060                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2061                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2062                         if (!__ip_dev_find(net, fl4->saddr, false))
2063                                 goto out;
2064                 }
2065         }
2066
2067
2068         if (fl4->flowi4_oif) {
2069                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2070                 rth = ERR_PTR(-ENODEV);
2071                 if (dev_out == NULL)
2072                         goto out;
2073
2074                 /* RACE: Check return value of inet_select_addr instead. */
2075                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2076                         rth = ERR_PTR(-ENETUNREACH);
2077                         goto out;
2078                 }
2079                 if (ipv4_is_local_multicast(fl4->daddr) ||
2080                     ipv4_is_lbcast(fl4->daddr)) {
2081                         if (!fl4->saddr)
2082                                 fl4->saddr = inet_select_addr(dev_out, 0,
2083                                                               RT_SCOPE_LINK);
2084                         goto make_route;
2085                 }
2086                 if (!fl4->saddr) {
2087                         if (ipv4_is_multicast(fl4->daddr))
2088                                 fl4->saddr = inet_select_addr(dev_out, 0,
2089                                                               fl4->flowi4_scope);
2090                         else if (!fl4->daddr)
2091                                 fl4->saddr = inet_select_addr(dev_out, 0,
2092                                                               RT_SCOPE_HOST);
2093                 }
2094         }
2095
2096         if (!fl4->daddr) {
2097                 fl4->daddr = fl4->saddr;
2098                 if (!fl4->daddr)
2099                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2100                 dev_out = net->loopback_dev;
2101                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2102                 res.type = RTN_LOCAL;
2103                 flags |= RTCF_LOCAL;
2104                 goto make_route;
2105         }
2106
2107         if (fib_lookup(net, fl4, &res)) {
2108                 res.fi = NULL;
2109                 res.table = NULL;
2110                 if (fl4->flowi4_oif) {
2111                         /* Apparently, routing tables are wrong. Assume,
2112                            that the destination is on link.
2113
2114                            WHY? DW.
2115                            Because we are allowed to send to iface
2116                            even if it has NO routes and NO assigned
2117                            addresses. When oif is specified, routing
2118                            tables are looked up with only one purpose:
2119                            to catch if destination is gatewayed, rather than
2120                            direct. Moreover, if MSG_DONTROUTE is set,
2121                            we send packet, ignoring both routing tables
2122                            and ifaddr state. --ANK
2123
2124
2125                            We could make it even if oif is unknown,
2126                            likely IPv6, but we do not.
2127                          */
2128
2129                         if (fl4->saddr == 0)
2130                                 fl4->saddr = inet_select_addr(dev_out, 0,
2131                                                               RT_SCOPE_LINK);
2132                         res.type = RTN_UNICAST;
2133                         goto make_route;
2134                 }
2135                 rth = ERR_PTR(-ENETUNREACH);
2136                 goto out;
2137         }
2138
2139         if (res.type == RTN_LOCAL) {
2140                 if (!fl4->saddr) {
2141                         if (res.fi->fib_prefsrc)
2142                                 fl4->saddr = res.fi->fib_prefsrc;
2143                         else
2144                                 fl4->saddr = fl4->daddr;
2145                 }
2146                 dev_out = net->loopback_dev;
2147                 fl4->flowi4_oif = dev_out->ifindex;
2148                 flags |= RTCF_LOCAL;
2149                 goto make_route;
2150         }
2151
2152 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2153         if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2154                 fib_select_multipath(&res);
2155         else
2156 #endif
2157         if (!res.prefixlen &&
2158             res.table->tb_num_default > 1 &&
2159             res.type == RTN_UNICAST && !fl4->flowi4_oif)
2160                 fib_select_default(&res);
2161
2162         if (!fl4->saddr)
2163                 fl4->saddr = FIB_RES_PREFSRC(net, res);
2164
2165         dev_out = FIB_RES_DEV(res);
2166         fl4->flowi4_oif = dev_out->ifindex;
2167
2168
2169 make_route:
2170         rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2171
2172 out:
2173         rcu_read_unlock();
2174         return rth;
2175 }
2176 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2177
2178 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2179 {
2180         return NULL;
2181 }
2182
2183 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2184 {
2185         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2186
2187         return mtu ? : dst->dev->mtu;
2188 }
2189
2190 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2191                                           struct sk_buff *skb, u32 mtu)
2192 {
2193 }
2194
2195 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2196                                        struct sk_buff *skb)
2197 {
2198 }
2199
2200 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2201                                           unsigned long old)
2202 {
2203         return NULL;
2204 }
2205
2206 static struct dst_ops ipv4_dst_blackhole_ops = {
2207         .family                 =       AF_INET,
2208         .protocol               =       cpu_to_be16(ETH_P_IP),
2209         .check                  =       ipv4_blackhole_dst_check,
2210         .mtu                    =       ipv4_blackhole_mtu,
2211         .default_advmss         =       ipv4_default_advmss,
2212         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2213         .redirect               =       ipv4_rt_blackhole_redirect,
2214         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
2215         .neigh_lookup           =       ipv4_neigh_lookup,
2216 };
2217
2218 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2219 {
2220         struct rtable *ort = (struct rtable *) dst_orig;
2221         struct rtable *rt;
2222
2223         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2224         if (rt) {
2225                 struct dst_entry *new = &rt->dst;
2226
2227                 new->__use = 1;
2228                 new->input = dst_discard;
2229                 new->output = dst_discard;
2230
2231                 new->dev = ort->dst.dev;
2232                 if (new->dev)
2233                         dev_hold(new->dev);
2234
2235                 rt->rt_is_input = ort->rt_is_input;
2236                 rt->rt_iif = ort->rt_iif;
2237                 rt->rt_pmtu = ort->rt_pmtu;
2238
2239                 rt->rt_genid = rt_genid_ipv4(net);
2240                 rt->rt_flags = ort->rt_flags;
2241                 rt->rt_type = ort->rt_type;
2242                 rt->rt_gateway = ort->rt_gateway;
2243                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2244
2245                 INIT_LIST_HEAD(&rt->rt_uncached);
2246
2247                 dst_free(new);
2248         }
2249
2250         dst_release(dst_orig);
2251
2252         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2253 }
2254
2255 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2256                                     struct sock *sk)
2257 {
2258         struct rtable *rt = __ip_route_output_key(net, flp4);
2259
2260         if (IS_ERR(rt))
2261                 return rt;
2262
2263         if (flp4->flowi4_proto)
2264                 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2265                                                    flowi4_to_flowi(flp4),
2266                                                    sk, 0);
2267
2268         return rt;
2269 }
2270 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2271
2272 static int rt_fill_info(struct net *net,  __be32 dst, __be32 src,
2273                         struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2274                         u32 seq, int event, int nowait, unsigned int flags)
2275 {
2276         struct rtable *rt = skb_rtable(skb);
2277         struct rtmsg *r;
2278         struct nlmsghdr *nlh;
2279         unsigned long expires = 0;
2280         u32 error;
2281         u32 metrics[RTAX_MAX];
2282
2283         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2284         if (nlh == NULL)
2285                 return -EMSGSIZE;
2286
2287         r = nlmsg_data(nlh);
2288         r->rtm_family    = AF_INET;
2289         r->rtm_dst_len  = 32;
2290         r->rtm_src_len  = 0;
2291         r->rtm_tos      = fl4->flowi4_tos;
2292         r->rtm_table    = RT_TABLE_MAIN;
2293         if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2294                 goto nla_put_failure;
2295         r->rtm_type     = rt->rt_type;
2296         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2297         r->rtm_protocol = RTPROT_UNSPEC;
2298         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2299         if (rt->rt_flags & RTCF_NOTIFY)
2300                 r->rtm_flags |= RTM_F_NOTIFY;
2301
2302         if (nla_put_be32(skb, RTA_DST, dst))
2303                 goto nla_put_failure;
2304         if (src) {
2305                 r->rtm_src_len = 32;
2306                 if (nla_put_be32(skb, RTA_SRC, src))
2307                         goto nla_put_failure;
2308         }
2309         if (rt->dst.dev &&
2310             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2311                 goto nla_put_failure;
2312 #ifdef CONFIG_IP_ROUTE_CLASSID
2313         if (rt->dst.tclassid &&
2314             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2315                 goto nla_put_failure;
2316 #endif
2317         if (!rt_is_input_route(rt) &&
2318             fl4->saddr != src) {
2319                 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2320                         goto nla_put_failure;
2321         }
2322         if (rt->rt_uses_gateway &&
2323             nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2324                 goto nla_put_failure;
2325
2326         expires = rt->dst.expires;
2327         if (expires) {
2328                 unsigned long now = jiffies;
2329
2330                 if (time_before(now, expires))
2331                         expires -= now;
2332                 else
2333                         expires = 0;
2334         }
2335
2336         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2337         if (rt->rt_pmtu && expires)
2338                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2339         if (rtnetlink_put_metrics(skb, metrics) < 0)
2340                 goto nla_put_failure;
2341
2342         if (fl4->flowi4_mark &&
2343             nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2344                 goto nla_put_failure;
2345
2346         error = rt->dst.error;
2347
2348         if (rt_is_input_route(rt)) {
2349 #ifdef CONFIG_IP_MROUTE
2350                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2351                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2352                         int err = ipmr_get_route(net, skb,
2353                                                  fl4->saddr, fl4->daddr,
2354                                                  r, nowait);
2355                         if (err <= 0) {
2356                                 if (!nowait) {
2357                                         if (err == 0)
2358                                                 return 0;
2359                                         goto nla_put_failure;
2360                                 } else {
2361                                         if (err == -EMSGSIZE)
2362                                                 goto nla_put_failure;
2363                                         error = err;
2364                                 }
2365                         }
2366                 } else
2367 #endif
2368                         if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2369                                 goto nla_put_failure;
2370         }
2371
2372         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2373                 goto nla_put_failure;
2374
2375         return nlmsg_end(skb, nlh);
2376
2377 nla_put_failure:
2378         nlmsg_cancel(skb, nlh);
2379         return -EMSGSIZE;
2380 }
2381
2382 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2383 {
2384         struct net *net = sock_net(in_skb->sk);
2385         struct rtmsg *rtm;
2386         struct nlattr *tb[RTA_MAX+1];
2387         struct rtable *rt = NULL;
2388         struct flowi4 fl4;
2389         __be32 dst = 0;
2390         __be32 src = 0;
2391         u32 iif;
2392         int err;
2393         int mark;
2394         struct sk_buff *skb;
2395
2396         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2397         if (err < 0)
2398                 goto errout;
2399
2400         rtm = nlmsg_data(nlh);
2401
2402         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2403         if (skb == NULL) {
2404                 err = -ENOBUFS;
2405                 goto errout;
2406         }
2407
2408         /* Reserve room for dummy headers, this skb can pass
2409            through good chunk of routing engine.
2410          */
2411         skb_reset_mac_header(skb);
2412         skb_reset_network_header(skb);
2413
2414         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2415         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2416         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2417
2418         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2419         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2420         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2421         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2422
2423         memset(&fl4, 0, sizeof(fl4));
2424         fl4.daddr = dst;
2425         fl4.saddr = src;
2426         fl4.flowi4_tos = rtm->rtm_tos;
2427         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2428         fl4.flowi4_mark = mark;
2429
2430         if (iif) {
2431                 struct net_device *dev;
2432
2433                 dev = __dev_get_by_index(net, iif);
2434                 if (dev == NULL) {
2435                         err = -ENODEV;
2436                         goto errout_free;
2437                 }
2438
2439                 skb->protocol   = htons(ETH_P_IP);
2440                 skb->dev        = dev;
2441                 skb->mark       = mark;
2442                 local_bh_disable();
2443                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2444                 local_bh_enable();
2445
2446                 rt = skb_rtable(skb);
2447                 if (err == 0 && rt->dst.error)
2448                         err = -rt->dst.error;
2449         } else {
2450                 rt = ip_route_output_key(net, &fl4);
2451
2452                 err = 0;
2453                 if (IS_ERR(rt))
2454                         err = PTR_ERR(rt);
2455         }
2456
2457         if (err)
2458                 goto errout_free;
2459
2460         skb_dst_set(skb, &rt->dst);
2461         if (rtm->rtm_flags & RTM_F_NOTIFY)
2462                 rt->rt_flags |= RTCF_NOTIFY;
2463
2464         err = rt_fill_info(net, dst, src, &fl4, skb,
2465                            NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2466                            RTM_NEWROUTE, 0, 0);
2467         if (err <= 0)
2468                 goto errout_free;
2469
2470         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2471 errout:
2472         return err;
2473
2474 errout_free:
2475         kfree_skb(skb);
2476         goto errout;
2477 }
2478
2479 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
2480 {
2481         return skb->len;
2482 }
2483
2484 void ip_rt_multicast_event(struct in_device *in_dev)
2485 {
2486         rt_cache_flush(dev_net(in_dev->dev));
2487 }
2488
2489 #ifdef CONFIG_SYSCTL
2490 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
2491 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
2492 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
2493 static int ip_rt_gc_elasticity __read_mostly    = 8;
2494
2495 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2496                                         void __user *buffer,
2497                                         size_t *lenp, loff_t *ppos)
2498 {
2499         struct net *net = (struct net *)__ctl->extra1;
2500
2501         if (write) {
2502                 rt_cache_flush(net);
2503                 fnhe_genid_bump(net);
2504                 return 0;
2505         }
2506
2507         return -EINVAL;
2508 }
2509
2510 static struct ctl_table ipv4_route_table[] = {
2511         {
2512                 .procname       = "gc_thresh",
2513                 .data           = &ipv4_dst_ops.gc_thresh,
2514                 .maxlen         = sizeof(int),
2515                 .mode           = 0644,
2516                 .proc_handler   = proc_dointvec,
2517         },
2518         {
2519                 .procname       = "max_size",
2520                 .data           = &ip_rt_max_size,
2521                 .maxlen         = sizeof(int),
2522                 .mode           = 0644,
2523                 .proc_handler   = proc_dointvec,
2524         },
2525         {
2526                 /*  Deprecated. Use gc_min_interval_ms */
2527
2528                 .procname       = "gc_min_interval",
2529                 .data           = &ip_rt_gc_min_interval,
2530                 .maxlen         = sizeof(int),
2531                 .mode           = 0644,
2532                 .proc_handler   = proc_dointvec_jiffies,
2533         },
2534         {
2535                 .procname       = "gc_min_interval_ms",
2536                 .data           = &ip_rt_gc_min_interval,
2537                 .maxlen         = sizeof(int),
2538                 .mode           = 0644,
2539                 .proc_handler   = proc_dointvec_ms_jiffies,
2540         },
2541         {
2542                 .procname       = "gc_timeout",
2543                 .data           = &ip_rt_gc_timeout,
2544                 .maxlen         = sizeof(int),
2545                 .mode           = 0644,
2546                 .proc_handler   = proc_dointvec_jiffies,
2547         },
2548         {
2549                 .procname       = "gc_interval",
2550                 .data           = &ip_rt_gc_interval,
2551                 .maxlen         = sizeof(int),
2552                 .mode           = 0644,
2553                 .proc_handler   = proc_dointvec_jiffies,
2554         },
2555         {
2556                 .procname       = "redirect_load",
2557                 .data           = &ip_rt_redirect_load,
2558                 .maxlen         = sizeof(int),
2559                 .mode           = 0644,
2560                 .proc_handler   = proc_dointvec,
2561         },
2562         {
2563                 .procname       = "redirect_number",
2564                 .data           = &ip_rt_redirect_number,
2565                 .maxlen         = sizeof(int),
2566                 .mode           = 0644,
2567                 .proc_handler   = proc_dointvec,
2568         },
2569         {
2570                 .procname       = "redirect_silence",
2571                 .data           = &ip_rt_redirect_silence,
2572                 .maxlen         = sizeof(int),
2573                 .mode           = 0644,
2574                 .proc_handler   = proc_dointvec,
2575         },
2576         {
2577                 .procname       = "error_cost",
2578                 .data           = &ip_rt_error_cost,
2579                 .maxlen         = sizeof(int),
2580                 .mode           = 0644,
2581                 .proc_handler   = proc_dointvec,
2582         },
2583         {
2584                 .procname       = "error_burst",
2585                 .data           = &ip_rt_error_burst,
2586                 .maxlen         = sizeof(int),
2587                 .mode           = 0644,
2588                 .proc_handler   = proc_dointvec,
2589         },
2590         {
2591                 .procname       = "gc_elasticity",
2592                 .data           = &ip_rt_gc_elasticity,
2593                 .maxlen         = sizeof(int),
2594                 .mode           = 0644,
2595                 .proc_handler   = proc_dointvec,
2596         },
2597         {
2598                 .procname       = "mtu_expires",
2599                 .data           = &ip_rt_mtu_expires,
2600                 .maxlen         = sizeof(int),
2601                 .mode           = 0644,
2602                 .proc_handler   = proc_dointvec_jiffies,
2603         },
2604         {
2605                 .procname       = "min_pmtu",
2606                 .data           = &ip_rt_min_pmtu,
2607                 .maxlen         = sizeof(int),
2608                 .mode           = 0644,
2609                 .proc_handler   = proc_dointvec,
2610         },
2611         {
2612                 .procname       = "min_adv_mss",
2613                 .data           = &ip_rt_min_advmss,
2614                 .maxlen         = sizeof(int),
2615                 .mode           = 0644,
2616                 .proc_handler   = proc_dointvec,
2617         },
2618         { }
2619 };
2620
2621 static struct ctl_table ipv4_route_flush_table[] = {
2622         {
2623                 .procname       = "flush",
2624                 .maxlen         = sizeof(int),
2625                 .mode           = 0200,
2626                 .proc_handler   = ipv4_sysctl_rtcache_flush,
2627         },
2628         { },
2629 };
2630
2631 static __net_init int sysctl_route_net_init(struct net *net)
2632 {
2633         struct ctl_table *tbl;
2634
2635         tbl = ipv4_route_flush_table;
2636         if (!net_eq(net, &init_net)) {
2637                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2638                 if (tbl == NULL)
2639                         goto err_dup;
2640
2641                 /* Don't export sysctls to unprivileged users */
2642                 if (net->user_ns != &init_user_ns)
2643                         tbl[0].procname = NULL;
2644         }
2645         tbl[0].extra1 = net;
2646
2647         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2648         if (net->ipv4.route_hdr == NULL)
2649                 goto err_reg;
2650         return 0;
2651
2652 err_reg:
2653         if (tbl != ipv4_route_flush_table)
2654                 kfree(tbl);
2655 err_dup:
2656         return -ENOMEM;
2657 }
2658
2659 static __net_exit void sysctl_route_net_exit(struct net *net)
2660 {
2661         struct ctl_table *tbl;
2662
2663         tbl = net->ipv4.route_hdr->ctl_table_arg;
2664         unregister_net_sysctl_table(net->ipv4.route_hdr);
2665         BUG_ON(tbl == ipv4_route_flush_table);
2666         kfree(tbl);
2667 }
2668
2669 static __net_initdata struct pernet_operations sysctl_route_ops = {
2670         .init = sysctl_route_net_init,
2671         .exit = sysctl_route_net_exit,
2672 };
2673 #endif
2674
2675 static __net_init int rt_genid_init(struct net *net)
2676 {
2677         atomic_set(&net->ipv4.rt_genid, 0);
2678         atomic_set(&net->fnhe_genid, 0);
2679         get_random_bytes(&net->ipv4.dev_addr_genid,
2680                          sizeof(net->ipv4.dev_addr_genid));
2681         return 0;
2682 }
2683
2684 static __net_initdata struct pernet_operations rt_genid_ops = {
2685         .init = rt_genid_init,
2686 };
2687
2688 static int __net_init ipv4_inetpeer_init(struct net *net)
2689 {
2690         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2691
2692         if (!bp)
2693                 return -ENOMEM;
2694         inet_peer_base_init(bp);
2695         net->ipv4.peers = bp;
2696         return 0;
2697 }
2698
2699 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2700 {
2701         struct inet_peer_base *bp = net->ipv4.peers;
2702
2703         net->ipv4.peers = NULL;
2704         inetpeer_invalidate_tree(bp);
2705         kfree(bp);
2706 }
2707
2708 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2709         .init   =       ipv4_inetpeer_init,
2710         .exit   =       ipv4_inetpeer_exit,
2711 };
2712
2713 #ifdef CONFIG_IP_ROUTE_CLASSID
2714 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2715 #endif /* CONFIG_IP_ROUTE_CLASSID */
2716
2717 int __init ip_rt_init(void)
2718 {
2719         int rc = 0;
2720
2721 #ifdef CONFIG_IP_ROUTE_CLASSID
2722         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2723         if (!ip_rt_acct)
2724                 panic("IP: failed to allocate ip_rt_acct\n");
2725 #endif
2726
2727         ipv4_dst_ops.kmem_cachep =
2728                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2729                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2730
2731         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2732
2733         if (dst_entries_init(&ipv4_dst_ops) < 0)
2734                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2735
2736         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2737                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2738
2739         ipv4_dst_ops.gc_thresh = ~0;
2740         ip_rt_max_size = INT_MAX;
2741
2742         devinet_init();
2743         ip_fib_init();
2744
2745         if (ip_rt_proc_init())
2746                 pr_err("Unable to create route proc files\n");
2747 #ifdef CONFIG_XFRM
2748         xfrm_init();
2749         xfrm4_init();
2750 #endif
2751         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2752
2753 #ifdef CONFIG_SYSCTL
2754         register_pernet_subsys(&sysctl_route_ops);
2755 #endif
2756         register_pernet_subsys(&rt_genid_ops);
2757         register_pernet_subsys(&ipv4_inetpeer_ops);
2758         return rc;
2759 }
2760
2761 #ifdef CONFIG_SYSCTL
2762 /*
2763  * We really need to sanitize the damn ipv4 init order, then all
2764  * this nonsense will go away.
2765  */
2766 void __init ip_static_sysctl_init(void)
2767 {
2768         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
2769 }
2770 #endif