net/ipv4/route.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              ROUTE - implementation of the IP router.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *              Linus Torvalds, <Linus.Torvalds@helsinki.fi>
  12  *              Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  13  *
  14  * Fixes:
  15  *              Alan Cox        :       Verify area fixes.
  16  *              Alan Cox        :       cli() protects routing changes
  17  *              Rui Oliveira    :       ICMP routing table updates
  18  *              (rco@di.uminho.pt)      Routing table insertion and update
  19  *              Linus Torvalds  :       Rewrote bits to be sensible
  20  *              Alan Cox        :       Added BSD route gw semantics
  21  *              Alan Cox        :       Super /proc >4K
  22  *              Alan Cox        :       MTU in route table
  23  *              Alan Cox        :       MSS actually. Also added the window
  24  *                                      clamper.
  25  *              Sam Lantinga    :       Fixed route matching in rt_del()
  26  *              Alan Cox        :       Routing cache support.
  27  *              Alan Cox        :       Removed compatibility cruft.
  28  *              Alan Cox        :       RTF_REJECT support.
  29  *              Alan Cox        :       TCP irtt support.
  30  *              Jonathan Naylor :       Added Metric support.
  31  *      Miquel van Smoorenburg  :       BSD API fixes.
  32  *      Miquel van Smoorenburg  :       Metrics.
  33  *              Alan Cox        :       Use __u32 properly
  34  *              Alan Cox        :       Aligned routing errors more closely with BSD
  35  *                                      our system is still very different.
  36  *              Alan Cox        :       Faster /proc handling
  37  *      Alexey Kuznetsov        :       Massive rework to support tree based routing,
  38  *                                      routing caches and better behaviour.
  39  *
  40  *              Olaf Erb        :       irtt wasn't being copied right.
  41  *              Bjorn Ekwall    :       Kerneld route support.
  42  *              Alan Cox        :       Multicast fixed (I hope)
  43  *              Pavel Krauz     :       Limited broadcast fixed
  44  *              Mike McLagan    :       Routing by source
  45  *      Alexey Kuznetsov        :       End of old history. Split to fib.c and
  46  *                                      route.c and rewritten from scratch.
  47  *              Andi Kleen      :       Load-limit warning messages.
  48  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  49  *      Vitaly E. Lavrov        :       Race condition in ip_route_input_slow.
  50  *      Tobias Ringstrom        :       Uninitialized res.type in ip_route_output_slow.
  51  *      Vladimir V. Ivanov      :       IP rule info (flowid) is really useful.
  52  *              Marc Boucher    :       routing by fwmark
  53  *      Robert Olsson           :       Added rt_cache statistics
  54  *      Arnaldo C. Melo         :       Convert proc stuff to seq_file
  55  *      Eric Dumazet            :       hashed spinlocks and rt_check_expire() fixes.
  56  *      Ilia Sotnikov           :       Ignore TOS on PMTUD and Redirect
  57  *      Ilia Sotnikov           :       Removed TOS from hash calculations
  58  *
  59  *              This program is free software; you can redistribute it and/or
  60  *              modify it under the terms of the GNU General Public License
  61  *              as published by the Free Software Foundation; either version
  62  *              2 of the License, or (at your option) any later version.
  63  */
  64
  65 #define pr_fmt(fmt) "IPv4: " fmt
  66
  67 #include <linux/module.h>
  68 #include <asm/uaccess.h>
  69 #include <linux/bitops.h>
  70 #include <linux/types.h>
  71 #include <linux/kernel.h>
  72 #include <linux/mm.h>
  73 #include <linux/string.h>
  74 #include <linux/socket.h>
  75 #include <linux/sockios.h>
  76 #include <linux/errno.h>
  77 #include <linux/in.h>
  78 #include <linux/inet.h>
  79 #include <linux/netdevice.h>
  80 #include <linux/proc_fs.h>
  81 #include <linux/init.h>
  82 #include <linux/skbuff.h>
  83 #include <linux/inetdevice.h>
  84 #include <linux/igmp.h>
  85 #include <linux/pkt_sched.h>
  86 #include <linux/mroute.h>
  87 #include <linux/netfilter_ipv4.h>
  88 #include <linux/random.h>
  89 #include <linux/rcupdate.h>
  90 #include <linux/times.h>
  91 #include <linux/slab.h>
  92 #include <linux/jhash.h>
  93 #include <net/dst.h>
  94 #include <net/net_namespace.h>
  95 #include <net/protocol.h>
  96 #include <net/ip.h>
  97 #include <net/route.h>
  98 #include <net/inetpeer.h>
  99 #include <net/sock.h>
 100 #include <net/ip_fib.h>
 101 #include <net/arp.h>
 102 #include <net/tcp.h>
 103 #include <net/icmp.h>
 104 #include <net/xfrm.h>
 105 #include <net/netevent.h>
 106 #include <net/rtnetlink.h>
 107 #ifdef CONFIG_SYSCTL
 108 #include <linux/sysctl.h>
 109 #include <linux/kmemleak.h>
 110 #endif
 111 #include <net/secure_seq.h>
 112
 113 #define RT_FL_TOS(oldflp4) \
 114         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
 115
 116 #define RT_GC_TIMEOUT (300*HZ)
 117
 118 static int ip_rt_max_size;
 119 static int ip_rt_redirect_number __read_mostly  = 9;
 120 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 121 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 122 static int ip_rt_error_cost __read_mostly       = HZ;
 123 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 124 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
 125 static int ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
 126 static int ip_rt_min_advmss __read_mostly       = 256;
 127
 128 /*
 129  *      Interface to generic destination cache.
 130  */
 131
 132 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 133 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 134 static unsigned int      ipv4_mtu(const struct dst_entry *dst);
 135 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 136 static void              ipv4_link_failure(struct sk_buff *skb);
 137 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 138                                            struct sk_buff *skb, u32 mtu);
 139 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 140                                         struct sk_buff *skb);
 141 static void             ipv4_dst_destroy(struct dst_entry *dst);
 142
 143 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 144                             int how)
 145 {
 146 }
 147
 148 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 149 {
 150         WARN_ON(1);
 151         return NULL;
 152 }
 153
 154 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 155                                            struct sk_buff *skb,
 156                                            const void *daddr);
 157
 158 static struct dst_ops ipv4_dst_ops = {
 159         .family =               AF_INET,
 160         .protocol =             cpu_to_be16(ETH_P_IP),
 161         .check =                ipv4_dst_check,
 162         .default_advmss =       ipv4_default_advmss,
 163         .mtu =                  ipv4_mtu,
 164         .cow_metrics =          ipv4_cow_metrics,
 165         .destroy =              ipv4_dst_destroy,
 166         .ifdown =               ipv4_dst_ifdown,
 167         .negative_advice =      ipv4_negative_advice,
 168         .link_failure =         ipv4_link_failure,
 169         .update_pmtu =          ip_rt_update_pmtu,
 170         .redirect =             ip_do_redirect,
 171         .local_out =            __ip_local_out,
 172         .neigh_lookup =         ipv4_neigh_lookup,
 173 };
 174
 175 #define ECN_OR_COST(class)      TC_PRIO_##class
 176
 177 const __u8 ip_tos2prio[16] = {
 178         TC_PRIO_BESTEFFORT,
 179         ECN_OR_COST(BESTEFFORT),
 180         TC_PRIO_BESTEFFORT,
 181         ECN_OR_COST(BESTEFFORT),
 182         TC_PRIO_BULK,
 183         ECN_OR_COST(BULK),
 184         TC_PRIO_BULK,
 185         ECN_OR_COST(BULK),
 186         TC_PRIO_INTERACTIVE,
 187         ECN_OR_COST(INTERACTIVE),
 188         TC_PRIO_INTERACTIVE,
 189         ECN_OR_COST(INTERACTIVE),
 190         TC_PRIO_INTERACTIVE_BULK,
 191         ECN_OR_COST(INTERACTIVE_BULK),
 192         TC_PRIO_INTERACTIVE_BULK,
 193         ECN_OR_COST(INTERACTIVE_BULK)
 194 };
 195 EXPORT_SYMBOL(ip_tos2prio);
 196
 197 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 198 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
 199
 200 #ifdef CONFIG_PROC_FS
 201 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 202 {
 203         if (*pos)
 204                 return NULL;
 205         return SEQ_START_TOKEN;
 206 }
 207
 208 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 209 {
 210         ++*pos;
 211         return NULL;
 212 }
 213
 214 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 215 {
 216 }
 217
 218 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 219 {
 220         if (v == SEQ_START_TOKEN)
 221                 seq_printf(seq, "%-127s\n",
 222                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 223                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 224                            "HHUptod\tSpecDst");
 225         return 0;
 226 }
 227
 228 static const struct seq_operations rt_cache_seq_ops = {
 229         .start  = rt_cache_seq_start,
 230         .next   = rt_cache_seq_next,
 231         .stop   = rt_cache_seq_stop,
 232         .show   = rt_cache_seq_show,
 233 };
 234
 235 static int rt_cache_seq_open(struct inode *inode, struct file *file)
 236 {
 237         return seq_open(file, &rt_cache_seq_ops);
 238 }
 239
 240 static const struct file_operations rt_cache_seq_fops = {
 241         .owner   = THIS_MODULE,
 242         .open    = rt_cache_seq_open,
 243         .read    = seq_read,
 244         .llseek  = seq_lseek,
 245         .release = seq_release,
 246 };
 247
 248
 249 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 250 {
 251         int cpu;
 252
 253         if (*pos == 0)
 254                 return SEQ_START_TOKEN;
 255
 256         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 257                 if (!cpu_possible(cpu))
 258                         continue;
 259                 *pos = cpu+1;
 260                 return &per_cpu(rt_cache_stat, cpu);
 261         }
 262         return NULL;
 263 }
 264
 265 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 266 {
 267         int cpu;
 268
 269         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 270                 if (!cpu_possible(cpu))
 271                         continue;
 272                 *pos = cpu+1;
 273                 return &per_cpu(rt_cache_stat, cpu);
 274         }
 275         return NULL;
 276
 277 }
 278
 279 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 280 {
 281
 282 }
 283
 284 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 285 {
 286         struct rt_cache_stat *st = v;
 287
 288         if (v == SEQ_START_TOKEN) {
 289                 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");
 290                 return 0;
 291         }
 292
 293         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 294                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 295                    dst_entries_get_slow(&ipv4_dst_ops),
 296                    0, /* st->in_hit */
 297                    st->in_slow_tot,
 298                    st->in_slow_mc,
 299                    st->in_no_route,
 300                    st->in_brd,
 301                    st->in_martian_dst,
 302                    st->in_martian_src,
 303
 304                    0, /* st->out_hit */
 305                    st->out_slow_tot,
 306                    st->out_slow_mc,
 307
 308                    0, /* st->gc_total */
 309                    0, /* st->gc_ignored */
 310                    0, /* st->gc_goal_miss */
 311                    0, /* st->gc_dst_overflow */
 312                    0, /* st->in_hlist_search */
 313                    0  /* st->out_hlist_search */
 314                 );
 315         return 0;
 316 }
 317
 318 static const struct seq_operations rt_cpu_seq_ops = {
 319         .start  = rt_cpu_seq_start,
 320         .next   = rt_cpu_seq_next,
 321         .stop   = rt_cpu_seq_stop,
 322         .show   = rt_cpu_seq_show,
 323 };
 324
 325
 326 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
 327 {
 328         return seq_open(file, &rt_cpu_seq_ops);
 329 }
 330
 331 static const struct file_operations rt_cpu_seq_fops = {
 332         .owner   = THIS_MODULE,
 333         .open    = rt_cpu_seq_open,
 334         .read    = seq_read,
 335         .llseek  = seq_lseek,
 336         .release = seq_release,
 337 };
 338
 339 #ifdef CONFIG_IP_ROUTE_CLASSID
 340 static int rt_acct_proc_show(struct seq_file *m, void *v)
 341 {
 342         struct ip_rt_acct *dst, *src;
 343         unsigned int i, j;
 344
 345         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 346         if (!dst)
 347                 return -ENOMEM;
 348
 349         for_each_possible_cpu(i) {
 350                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 351                 for (j = 0; j < 256; j++) {
 352                         dst[j].o_bytes   += src[j].o_bytes;
 353                         dst[j].o_packets += src[j].o_packets;
 354                         dst[j].i_bytes   += src[j].i_bytes;
 355                         dst[j].i_packets += src[j].i_packets;
 356                 }
 357         }
 358
 359         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 360         kfree(dst);
 361         return 0;
 362 }
 363
 364 static int rt_acct_proc_open(struct inode *inode, struct file *file)
 365 {
 366         return single_open(file, rt_acct_proc_show, NULL);
 367 }
 368
 369 static const struct file_operations rt_acct_proc_fops = {
 370         .owner          = THIS_MODULE,
 371         .open           = rt_acct_proc_open,
 372         .read           = seq_read,
 373         .llseek         = seq_lseek,
 374         .release        = single_release,
 375 };
 376 #endif
 377
 378 static int __net_init ip_rt_do_proc_init(struct net *net)
 379 {
 380         struct proc_dir_entry *pde;
 381
 382         pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
 383                           &rt_cache_seq_fops);
 384         if (!pde)
 385                 goto err1;
 386
 387         pde = proc_create("rt_cache", S_IRUGO,
 388                           net->proc_net_stat, &rt_cpu_seq_fops);
 389         if (!pde)
 390                 goto err2;
 391
 392 #ifdef CONFIG_IP_ROUTE_CLASSID
 393         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
 394         if (!pde)
 395                 goto err3;
 396 #endif
 397         return 0;
 398
 399 #ifdef CONFIG_IP_ROUTE_CLASSID
 400 err3:
 401         remove_proc_entry("rt_cache", net->proc_net_stat);
 402 #endif
 403 err2:
 404         remove_proc_entry("rt_cache", net->proc_net);
 405 err1:
 406         return -ENOMEM;
 407 }
 408
 409 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 410 {
 411         remove_proc_entry("rt_cache", net->proc_net_stat);
 412         remove_proc_entry("rt_cache", net->proc_net);
 413 #ifdef CONFIG_IP_ROUTE_CLASSID
 414         remove_proc_entry("rt_acct", net->proc_net);
 415 #endif
 416 }
 417
 418 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 419         .init = ip_rt_do_proc_init,
 420         .exit = ip_rt_do_proc_exit,
 421 };
 422
 423 static int __init ip_rt_proc_init(void)
 424 {
 425         return register_pernet_subsys(&ip_rt_proc_ops);
 426 }
 427
 428 #else
 429 static inline int ip_rt_proc_init(void)
 430 {
 431         return 0;
 432 }
 433 #endif /* CONFIG_PROC_FS */
 434
 435 static inline bool rt_is_expired(const struct rtable *rth)
 436 {
 437         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 438 }
 439
 440 void rt_cache_flush(struct net *net)
 441 {
 442         rt_genid_bump_ipv4(net);
 443 }
 444
 445 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 446                                            struct sk_buff *skb,
 447                                            const void *daddr)
 448 {
 449         struct net_device *dev = dst->dev;
 450         const __be32 *pkey = daddr;
 451         const struct rtable *rt;
 452         struct neighbour *n;
 453
 454         rt = (const struct rtable *) dst;
 455         if (rt->rt_gateway)
 456                 pkey = (const __be32 *) &rt->rt_gateway;
 457         else if (skb)
 458                 pkey = &ip_hdr(skb)->daddr;
 459
 460         n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
 461         if (n)
 462                 return n;
 463         return neigh_create(&arp_tbl, pkey, dev);
 464 }
 465
 466 #define IP_IDENTS_SZ 2048u
 467 struct ip_ident_bucket {
 468         atomic_t        id;
 469         u32             stamp32;
 470 };
 471
 472 static struct ip_ident_bucket *ip_idents __read_mostly;
 473
 474 /* In order to protect privacy, we add a perturbation to identifiers
 475  * if one generator is seldom used. This makes hard for an attacker
 476  * to infer how many packets were sent between two points in time.
 477  */
 478 u32 ip_idents_reserve(u32 hash, int segs)
 479 {
 480         struct ip_ident_bucket *bucket = ip_idents + hash % IP_IDENTS_SZ;
 481         u32 old = ACCESS_ONCE(bucket->stamp32);
 482         u32 now = (u32)jiffies;
 483         u32 delta = 0;
 484
 485         if (old != now && cmpxchg(&bucket->stamp32, old, now) == old)
 486                 delta = prandom_u32_max(now - old);
 487
 488         return atomic_add_return(segs + delta, &bucket->id) - segs;
 489 }
 490 EXPORT_SYMBOL(ip_idents_reserve);
 491
 492 void __ip_select_ident(struct iphdr *iph, int segs)
 493 {
 494         static u32 ip_idents_hashrnd __read_mostly;
 495         u32 hash, id;
 496
 497         net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
 498
 499         hash = jhash_3words((__force u32)iph->daddr,
 500                             (__force u32)iph->saddr,
 501                             iph->protocol,
 502                             ip_idents_hashrnd);
 503         id = ip_idents_reserve(hash, segs);
 504         iph->id = htons(id);
 505 }
 506 EXPORT_SYMBOL(__ip_select_ident);
 507
 508 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 509                              const struct iphdr *iph,
 510                              int oif, u8 tos,
 511                              u8 prot, u32 mark, int flow_flags)
 512 {
 513         if (sk) {
 514                 const struct inet_sock *inet = inet_sk(sk);
 515
 516                 oif = sk->sk_bound_dev_if;
 517                 mark = sk->sk_mark;
 518                 tos = RT_CONN_FLAGS(sk);
 519                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 520         }
 521         flowi4_init_output(fl4, oif, mark, tos,
 522                            RT_SCOPE_UNIVERSE, prot,
 523                            flow_flags,
 524                            iph->daddr, iph->saddr, 0, 0);
 525 }
 526
 527 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 528                                const struct sock *sk)
 529 {
 530         const struct iphdr *iph = ip_hdr(skb);
 531         int oif = skb->dev->ifindex;
 532         u8 tos = RT_TOS(iph->tos);
 533         u8 prot = iph->protocol;
 534         u32 mark = skb->mark;
 535
 536         __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
 537 }
 538
 539 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 540 {
 541         const struct inet_sock *inet = inet_sk(sk);
 542         const struct ip_options_rcu *inet_opt;
 543         __be32 daddr = inet->inet_daddr;
 544
 545         rcu_read_lock();
 546         inet_opt = rcu_dereference(inet->inet_opt);
 547         if (inet_opt && inet_opt->opt.srr)
 548                 daddr = inet_opt->opt.faddr;
 549         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 550                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 551                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 552                            inet_sk_flowi_flags(sk),
 553                            daddr, inet->inet_saddr, 0, 0);
 554         rcu_read_unlock();
 555 }
 556
 557 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 558                                  const struct sk_buff *skb)
 559 {
 560         if (skb)
 561                 build_skb_flow_key(fl4, skb, sk);
 562         else
 563                 build_sk_flow_key(fl4, sk);
 564 }
 565
 566 static inline void rt_free(struct rtable *rt)
 567 {
 568         call_rcu(&rt->dst.rcu_head, dst_rcu_free);
 569 }
 570
 571 static DEFINE_SPINLOCK(fnhe_lock);
 572
 573 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 574 {
 575         struct rtable *rt;
 576
 577         rt = rcu_dereference(fnhe->fnhe_rth_input);
 578         if (rt) {
 579                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 580                 rt_free(rt);
 581         }
 582         rt = rcu_dereference(fnhe->fnhe_rth_output);
 583         if (rt) {
 584                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 585                 rt_free(rt);
 586         }
 587 }
 588
 589 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
 590 {
 591         struct fib_nh_exception *fnhe, *oldest;
 592
 593         oldest = rcu_dereference(hash->chain);
 594         for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
 595              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 596                 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
 597                         oldest = fnhe;
 598         }
 599         fnhe_flush_routes(oldest);
 600         return oldest;
 601 }
 602
 603 static inline u32 fnhe_hashfun(__be32 daddr)
 604 {
 605         u32 hval;
 606
 607         hval = (__force u32) daddr;
 608         hval ^= (hval >> 11) ^ (hval >> 22);
 609
 610         return hval & (FNHE_HASH_SIZE - 1);
 611 }
 612
 613 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 614 {
 615         rt->rt_pmtu = fnhe->fnhe_pmtu;
 616         rt->dst.expires = fnhe->fnhe_expires;
 617
 618         if (fnhe->fnhe_gw) {
 619                 rt->rt_flags |= RTCF_REDIRECTED;
 620                 rt->rt_gateway = fnhe->fnhe_gw;
 621                 rt->rt_uses_gateway = 1;
 622         }
 623 }
 624
 625 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
 626                                   u32 pmtu, unsigned long expires)
 627 {
 628         struct fnhe_hash_bucket *hash;
 629         struct fib_nh_exception *fnhe;
 630         struct rtable *rt;
 631         unsigned int i;
 632         int depth;
 633         u32 hval = fnhe_hashfun(daddr);
 634
 635         spin_lock_bh(&fnhe_lock);
 636
 637         hash = nh->nh_exceptions;
 638         if (!hash) {
 639                 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
 640                 if (!hash)
 641                         goto out_unlock;
 642                 nh->nh_exceptions = hash;
 643         }
 644
 645         hash += hval;
 646
 647         depth = 0;
 648         for (fnhe = rcu_dereference(hash->chain); fnhe;
 649              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 650                 if (fnhe->fnhe_daddr == daddr)
 651                         break;
 652                 depth++;
 653         }
 654
 655         if (fnhe) {
 656                 if (gw)
 657                         fnhe->fnhe_gw = gw;
 658                 if (pmtu) {
 659                         fnhe->fnhe_pmtu = pmtu;
 660                         fnhe->fnhe_expires = max(1UL, expires);
 661                 }
 662                 /* Update all cached dsts too */
 663                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 664                 if (rt)
 665                         fill_route_from_fnhe(rt, fnhe);
 666                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 667                 if (rt)
 668                         fill_route_from_fnhe(rt, fnhe);
 669         } else {
 670                 if (depth > FNHE_RECLAIM_DEPTH)
 671                         fnhe = fnhe_oldest(hash);
 672                 else {
 673                         fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 674                         if (!fnhe)
 675                                 goto out_unlock;
 676
 677                         fnhe->fnhe_next = hash->chain;
 678                         rcu_assign_pointer(hash->chain, fnhe);
 679                 }
 680                 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
 681                 fnhe->fnhe_daddr = daddr;
 682                 fnhe->fnhe_gw = gw;
 683                 fnhe->fnhe_pmtu = pmtu;
 684                 fnhe->fnhe_expires = expires;
 685
 686                 /* Exception created; mark the cached routes for the nexthop
 687                  * stale, so anyone caching it rechecks if this exception
 688                  * applies to them.
 689                  */
 690                 rt = rcu_dereference(nh->nh_rth_input);
 691                 if (rt)
 692                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 693
 694                 for_each_possible_cpu(i) {
 695                         struct rtable __rcu **prt;
 696                         prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
 697                         rt = rcu_dereference(*prt);
 698                         if (rt)
 699                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 700                 }
 701         }
 702
 703         fnhe->fnhe_stamp = jiffies;
 704
 705 out_unlock:
 706         spin_unlock_bh(&fnhe_lock);
 707         return;
 708 }
 709
 710 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 711                              bool kill_route)
 712 {
 713         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 714         __be32 old_gw = ip_hdr(skb)->saddr;
 715         struct net_device *dev = skb->dev;
 716         struct in_device *in_dev;
 717         struct fib_result res;
 718         struct neighbour *n;
 719         struct net *net;
 720
 721         switch (icmp_hdr(skb)->code & 7) {
 722         case ICMP_REDIR_NET:
 723         case ICMP_REDIR_NETTOS:
 724         case ICMP_REDIR_HOST:
 725         case ICMP_REDIR_HOSTTOS:
 726                 break;
 727
 728         default:
 729                 return;
 730         }
 731
 732         if (rt->rt_gateway != old_gw)
 733                 return;
 734
 735         in_dev = __in_dev_get_rcu(dev);
 736         if (!in_dev)
 737                 return;
 738
 739         net = dev_net(dev);
 740         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 741             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 742             ipv4_is_zeronet(new_gw))
 743                 goto reject_redirect;
 744
 745         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 746                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 747                         goto reject_redirect;
 748                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 749                         goto reject_redirect;
 750         } else {
 751                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 752                         goto reject_redirect;
 753         }
 754
 755         n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
 756         if (n) {
 757                 if (!(n->nud_state & NUD_VALID)) {
 758                         neigh_event_send(n, NULL);
 759                 } else {
 760                         if (fib_lookup(net, fl4, &res) == 0) {
 761                                 struct fib_nh *nh = &FIB_RES_NH(res);
 762
 763                                 update_or_create_fnhe(nh, fl4->daddr, new_gw,
 764                                                       0, 0);
 765                         }
 766                         if (kill_route)
 767                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 768                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 769                 }
 770                 neigh_release(n);
 771         }
 772         return;
 773
 774 reject_redirect:
 775 #ifdef CONFIG_IP_ROUTE_VERBOSE
 776         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 777                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 778                 __be32 daddr = iph->daddr;
 779                 __be32 saddr = iph->saddr;
 780
 781                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 782                                      "  Advised path = %pI4 -> %pI4\n",
 783                                      &old_gw, dev->name, &new_gw,
 784                                      &saddr, &daddr);
 785         }
 786 #endif
 787         ;
 788 }
 789
 790 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 791 {
 792         struct rtable *rt;
 793         struct flowi4 fl4;
 794         const struct iphdr *iph = (const struct iphdr *) skb->data;
 795         int oif = skb->dev->ifindex;
 796         u8 tos = RT_TOS(iph->tos);
 797         u8 prot = iph->protocol;
 798         u32 mark = skb->mark;
 799
 800         rt = (struct rtable *) dst;
 801
 802         __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
 803         __ip_do_redirect(rt, skb, &fl4, true);
 804 }
 805
 806 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 807 {
 808         struct rtable *rt = (struct rtable *)dst;
 809         struct dst_entry *ret = dst;
 810
 811         if (rt) {
 812                 if (dst->obsolete > 0) {
 813                         ip_rt_put(rt);
 814                         ret = NULL;
 815                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 816                            rt->dst.expires) {
 817                         ip_rt_put(rt);
 818                         ret = NULL;
 819                 }
 820         }
 821         return ret;
 822 }
 823
 824 /*
 825  * Algorithm:
 826  *      1. The first ip_rt_redirect_number redirects are sent
 827  *         with exponential backoff, then we stop sending them at all,
 828  *         assuming that the host ignores our redirects.
 829  *      2. If we did not see packets requiring redirects
 830  *         during ip_rt_redirect_silence, we assume that the host
 831  *         forgot redirected route and start to send redirects again.
 832  *
 833  * This algorithm is much cheaper and more intelligent than dumb load limiting
 834  * in icmp.c.
 835  *
 836  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 837  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 838  */
 839
 840 void ip_rt_send_redirect(struct sk_buff *skb)
 841 {
 842         struct rtable *rt = skb_rtable(skb);
 843         struct in_device *in_dev;
 844         struct inet_peer *peer;
 845         struct net *net;
 846         int log_martians;
 847
 848         rcu_read_lock();
 849         in_dev = __in_dev_get_rcu(rt->dst.dev);
 850         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 851                 rcu_read_unlock();
 852                 return;
 853         }
 854         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 855         rcu_read_unlock();
 856
 857         net = dev_net(rt->dst.dev);
 858         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
 859         if (!peer) {
 860                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 861                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 862                 return;
 863         }
 864
 865         /* No redirected packets during ip_rt_redirect_silence;
 866          * reset the algorithm.
 867          */
 868         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
 869                 peer->rate_tokens = 0;
 870
 871         /* Too many ignored redirects; do not send anything
 872          * set dst.rate_last to the last seen redirected packet.
 873          */
 874         if (peer->rate_tokens >= ip_rt_redirect_number) {
 875                 peer->rate_last = jiffies;
 876                 goto out_put_peer;
 877         }
 878
 879         /* Check for load limit; set rate_last to the latest sent
 880          * redirect.
 881          */
 882         if (peer->rate_tokens == 0 ||
 883             time_after(jiffies,
 884                        (peer->rate_last +
 885                         (ip_rt_redirect_load << peer->rate_tokens)))) {
 886                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 887
 888                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 889                 peer->rate_last = jiffies;
 890                 ++peer->rate_tokens;
 891 #ifdef CONFIG_IP_ROUTE_VERBOSE
 892                 if (log_martians &&
 893                     peer->rate_tokens == ip_rt_redirect_number)
 894                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 895                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 896                                              &ip_hdr(skb)->daddr, &gw);
 897 #endif
 898         }
 899 out_put_peer:
 900         inet_putpeer(peer);
 901 }
 902
 903 static int ip_error(struct sk_buff *skb)
 904 {
 905         struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
 906         struct rtable *rt = skb_rtable(skb);
 907         struct inet_peer *peer;
 908         unsigned long now;
 909         struct net *net;
 910         bool send;
 911         int code;
 912
 913         net = dev_net(rt->dst.dev);
 914         if (!IN_DEV_FORWARD(in_dev)) {
 915                 switch (rt->dst.error) {
 916                 case EHOSTUNREACH:
 917                         IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
 918                         break;
 919
 920                 case ENETUNREACH:
 921                         IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
 922                         break;
 923                 }
 924                 goto out;
 925         }
 926
 927         switch (rt->dst.error) {
 928         case EINVAL:
 929         default:
 930                 goto out;
 931         case EHOSTUNREACH:
 932                 code = ICMP_HOST_UNREACH;
 933                 break;
 934         case ENETUNREACH:
 935                 code = ICMP_NET_UNREACH;
 936                 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
 937                 break;
 938         case EACCES:
 939                 code = ICMP_PKT_FILTERED;
 940                 break;
 941         }
 942
 943         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
 944
 945         send = true;
 946         if (peer) {
 947                 now = jiffies;
 948                 peer->rate_tokens += now - peer->rate_last;
 949                 if (peer->rate_tokens > ip_rt_error_burst)
 950                         peer->rate_tokens = ip_rt_error_burst;
 951                 peer->rate_last = now;
 952                 if (peer->rate_tokens >= ip_rt_error_cost)
 953                         peer->rate_tokens -= ip_rt_error_cost;
 954                 else
 955                         send = false;
 956                 inet_putpeer(peer);
 957         }
 958         if (send)
 959                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
 960
 961 out:    kfree_skb(skb);
 962         return 0;
 963 }
 964
 965 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
 966 {
 967         struct dst_entry *dst = &rt->dst;
 968         struct fib_result res;
 969
 970         if (dst_metric_locked(dst, RTAX_MTU))
 971                 return;
 972
 973         if (dst->dev->mtu < mtu)
 974                 return;
 975
 976         if (mtu < ip_rt_min_pmtu)
 977                 mtu = ip_rt_min_pmtu;
 978
 979         if (rt->rt_pmtu == mtu &&
 980             time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
 981                 return;
 982
 983         rcu_read_lock();
 984         if (fib_lookup(dev_net(dst->dev), fl4, &res) == 0) {
 985                 struct fib_nh *nh = &FIB_RES_NH(res);
 986
 987                 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
 988                                       jiffies + ip_rt_mtu_expires);
 989         }
 990         rcu_read_unlock();
 991 }
 992
 993 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 994                               struct sk_buff *skb, u32 mtu)
 995 {
 996         struct rtable *rt = (struct rtable *) dst;
 997         struct flowi4 fl4;
 998
 999         ip_rt_build_flow_key(&fl4, sk, skb);
1000         __ip_rt_update_pmtu(rt, &fl4, mtu);
1001 }
1002
1003 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1004                       int oif, u32 mark, u8 protocol, int flow_flags)
1005 {
1006         const struct iphdr *iph = (const struct iphdr *) skb->data;
1007         struct flowi4 fl4;
1008         struct rtable *rt;
1009
1010         __build_flow_key(&fl4, NULL, iph, oif,
1011                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1012         rt = __ip_route_output_key(net, &fl4);
1013         if (!IS_ERR(rt)) {
1014                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1015                 ip_rt_put(rt);
1016         }
1017 }
1018 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1019
1020 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1021 {
1022         const struct iphdr *iph = (const struct iphdr *) skb->data;
1023         struct flowi4 fl4;
1024         struct rtable *rt;
1025
1026         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1027         rt = __ip_route_output_key(sock_net(sk), &fl4);
1028         if (!IS_ERR(rt)) {
1029                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1030                 ip_rt_put(rt);
1031         }
1032 }
1033
1034 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1035 {
1036         const struct iphdr *iph = (const struct iphdr *) skb->data;
1037         struct flowi4 fl4;
1038         struct rtable *rt;
1039         struct dst_entry *odst = NULL;
1040         bool new = false;
1041
1042         bh_lock_sock(sk);
1043
1044         if (!ip_sk_accept_pmtu(sk))
1045                 goto out;
1046
1047         odst = sk_dst_get(sk);
1048
1049         if (sock_owned_by_user(sk) || !odst) {
1050                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1051                 goto out;
1052         }
1053
1054         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1055
1056         rt = (struct rtable *)odst;
1057         if (odst->obsolete && odst->ops->check(odst, 0) == NULL) {
1058                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1059                 if (IS_ERR(rt))
1060                         goto out;
1061
1062                 new = true;
1063         }
1064
1065         __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1066
1067         if (!dst_check(&rt->dst, 0)) {
1068                 if (new)
1069                         dst_release(&rt->dst);
1070
1071                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1072                 if (IS_ERR(rt))
1073                         goto out;
1074
1075                 new = true;
1076         }
1077
1078         if (new)
1079                 sk_dst_set(sk, &rt->dst);
1080
1081 out:
1082         bh_unlock_sock(sk);
1083         dst_release(odst);
1084 }
1085 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1086
1087 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1088                    int oif, u32 mark, u8 protocol, int flow_flags)
1089 {
1090         const struct iphdr *iph = (const struct iphdr *) skb->data;
1091         struct flowi4 fl4;
1092         struct rtable *rt;
1093
1094         __build_flow_key(&fl4, NULL, iph, oif,
1095                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1096         rt = __ip_route_output_key(net, &fl4);
1097         if (!IS_ERR(rt)) {
1098                 __ip_do_redirect(rt, skb, &fl4, false);
1099                 ip_rt_put(rt);
1100         }
1101 }
1102 EXPORT_SYMBOL_GPL(ipv4_redirect);
1103
1104 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1105 {
1106         const struct iphdr *iph = (const struct iphdr *) skb->data;
1107         struct flowi4 fl4;
1108         struct rtable *rt;
1109
1110         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1111         rt = __ip_route_output_key(sock_net(sk), &fl4);
1112         if (!IS_ERR(rt)) {
1113                 __ip_do_redirect(rt, skb, &fl4, false);
1114                 ip_rt_put(rt);
1115         }
1116 }
1117 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1118
1119 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1120 {
1121         struct rtable *rt = (struct rtable *) dst;
1122
1123         /* All IPV4 dsts are created with ->obsolete set to the value
1124          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1125          * into this function always.
1126          *
1127          * When a PMTU/redirect information update invalidates a route,
1128          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1129          * DST_OBSOLETE_DEAD by dst_free().
1130          */
1131         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1132                 return NULL;
1133         return dst;
1134 }
1135
1136 static void ipv4_link_failure(struct sk_buff *skb)
1137 {
1138         struct rtable *rt;
1139
1140         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1141
1142         rt = skb_rtable(skb);
1143         if (rt)
1144                 dst_set_expires(&rt->dst, 0);
1145 }
1146
1147 static int ip_rt_bug(struct sk_buff *skb)
1148 {
1149         pr_debug("%s: %pI4 -> %pI4, %s\n",
1150                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1151                  skb->dev ? skb->dev->name : "?");
1152         kfree_skb(skb);
1153         WARN_ON(1);
1154         return 0;
1155 }
1156
1157 /*
1158    We do not cache source address of outgoing interface,
1159    because it is used only by IP RR, TS and SRR options,
1160    so that it out of fast path.
1161
1162    BTW remember: "addr" is allowed to be not aligned
1163    in IP options!
1164  */
1165
1166 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1167 {
1168         __be32 src;
1169
1170         if (rt_is_output_route(rt))
1171                 src = ip_hdr(skb)->saddr;
1172         else {
1173                 struct fib_result res;
1174                 struct flowi4 fl4;
1175                 struct iphdr *iph;
1176
1177                 iph = ip_hdr(skb);
1178
1179                 memset(&fl4, 0, sizeof(fl4));
1180                 fl4.daddr = iph->daddr;
1181                 fl4.saddr = iph->saddr;
1182                 fl4.flowi4_tos = RT_TOS(iph->tos);
1183                 fl4.flowi4_oif = rt->dst.dev->ifindex;
1184                 fl4.flowi4_iif = skb->dev->ifindex;
1185                 fl4.flowi4_mark = skb->mark;
1186
1187                 rcu_read_lock();
1188                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1189                         src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1190                 else
1191                         src = inet_select_addr(rt->dst.dev,
1192                                                rt_nexthop(rt, iph->daddr),
1193                                                RT_SCOPE_UNIVERSE);
1194                 rcu_read_unlock();
1195         }
1196         memcpy(addr, &src, 4);
1197 }
1198
1199 #ifdef CONFIG_IP_ROUTE_CLASSID
1200 static void set_class_tag(struct rtable *rt, u32 tag)
1201 {
1202         if (!(rt->dst.tclassid & 0xFFFF))
1203                 rt->dst.tclassid |= tag & 0xFFFF;
1204         if (!(rt->dst.tclassid & 0xFFFF0000))
1205                 rt->dst.tclassid |= tag & 0xFFFF0000;
1206 }
1207 #endif
1208
1209 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1210 {
1211         unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1212
1213         if (advmss == 0) {
1214                 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1215                                ip_rt_min_advmss);
1216                 if (advmss > 65535 - 40)
1217                         advmss = 65535 - 40;
1218         }
1219         return advmss;
1220 }
1221
1222 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1223 {
1224         const struct rtable *rt = (const struct rtable *) dst;
1225         unsigned int mtu = rt->rt_pmtu;
1226
1227         if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1228                 mtu = dst_metric_raw(dst, RTAX_MTU);
1229
1230         if (mtu)
1231                 return mtu;
1232
1233         mtu = dst->dev->mtu;
1234
1235         if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1236                 if (rt->rt_uses_gateway && mtu > 576)
1237                         mtu = 576;
1238         }
1239
1240         return min_t(unsigned int, mtu, IP_MAX_MTU);
1241 }
1242
1243 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1244 {
1245         struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1246         struct fib_nh_exception *fnhe;
1247         u32 hval;
1248
1249         if (!hash)
1250                 return NULL;
1251
1252         hval = fnhe_hashfun(daddr);
1253
1254         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1255              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1256                 if (fnhe->fnhe_daddr == daddr)
1257                         return fnhe;
1258         }
1259         return NULL;
1260 }
1261
1262 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1263                               __be32 daddr)
1264 {
1265         bool ret = false;
1266
1267         spin_lock_bh(&fnhe_lock);
1268
1269         if (daddr == fnhe->fnhe_daddr) {
1270                 struct rtable __rcu **porig;
1271                 struct rtable *orig;
1272                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1273
1274                 if (rt_is_input_route(rt))
1275                         porig = &fnhe->fnhe_rth_input;
1276                 else
1277                         porig = &fnhe->fnhe_rth_output;
1278                 orig = rcu_dereference(*porig);
1279
1280                 if (fnhe->fnhe_genid != genid) {
1281                         fnhe->fnhe_genid = genid;
1282                         fnhe->fnhe_gw = 0;
1283                         fnhe->fnhe_pmtu = 0;
1284                         fnhe->fnhe_expires = 0;
1285                         fnhe_flush_routes(fnhe);
1286                         orig = NULL;
1287                 }
1288                 fill_route_from_fnhe(rt, fnhe);
1289                 if (!rt->rt_gateway)
1290                         rt->rt_gateway = daddr;
1291
1292                 if (!(rt->dst.flags & DST_NOCACHE)) {
1293                         rcu_assign_pointer(*porig, rt);
1294                         if (orig)
1295                                 rt_free(orig);
1296                         ret = true;
1297                 }
1298
1299                 fnhe->fnhe_stamp = jiffies;
1300         }
1301         spin_unlock_bh(&fnhe_lock);
1302
1303         return ret;
1304 }
1305
1306 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1307 {
1308         struct rtable *orig, *prev, **p;
1309         bool ret = true;
1310
1311         if (rt_is_input_route(rt)) {
1312                 p = (struct rtable **)&nh->nh_rth_input;
1313         } else {
1314                 p = (struct rtable **)__this_cpu_ptr(nh->nh_pcpu_rth_output);
1315         }
1316         orig = *p;
1317
1318         prev = cmpxchg(p, orig, rt);
1319         if (prev == orig) {
1320                 if (orig)
1321                         rt_free(orig);
1322         } else
1323                 ret = false;
1324
1325         return ret;
1326 }
1327
1328 static DEFINE_SPINLOCK(rt_uncached_lock);
1329 static LIST_HEAD(rt_uncached_list);
1330
1331 static void rt_add_uncached_list(struct rtable *rt)
1332 {
1333         spin_lock_bh(&rt_uncached_lock);
1334         list_add_tail(&rt->rt_uncached, &rt_uncached_list);
1335         spin_unlock_bh(&rt_uncached_lock);
1336 }
1337
1338 static void ipv4_dst_destroy(struct dst_entry *dst)
1339 {
1340         struct rtable *rt = (struct rtable *) dst;
1341
1342         if (!list_empty(&rt->rt_uncached)) {
1343                 spin_lock_bh(&rt_uncached_lock);
1344                 list_del(&rt->rt_uncached);
1345                 spin_unlock_bh(&rt_uncached_lock);
1346         }
1347 }
1348
1349 void rt_flush_dev(struct net_device *dev)
1350 {
1351         if (!list_empty(&rt_uncached_list)) {
1352                 struct net *net = dev_net(dev);
1353                 struct rtable *rt;
1354
1355                 spin_lock_bh(&rt_uncached_lock);
1356                 list_for_each_entry(rt, &rt_uncached_list, rt_uncached) {
1357                         if (rt->dst.dev != dev)
1358                                 continue;
1359                         rt->dst.dev = net->loopback_dev;
1360                         dev_hold(rt->dst.dev);
1361                         dev_put(dev);
1362                 }
1363                 spin_unlock_bh(&rt_uncached_lock);
1364         }
1365 }
1366
1367 static bool rt_cache_valid(const struct rtable *rt)
1368 {
1369         return  rt &&
1370                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1371                 !rt_is_expired(rt);
1372 }
1373
1374 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1375                            const struct fib_result *res,
1376                            struct fib_nh_exception *fnhe,
1377                            struct fib_info *fi, u16 type, u32 itag)
1378 {
1379         bool cached = false;
1380
1381         if (fi) {
1382                 struct fib_nh *nh = &FIB_RES_NH(*res);
1383
1384                 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1385                         rt->rt_gateway = nh->nh_gw;
1386                         rt->rt_uses_gateway = 1;
1387                 }
1388                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1389 #ifdef CONFIG_IP_ROUTE_CLASSID
1390                 rt->dst.tclassid = nh->nh_tclassid;
1391 #endif
1392                 if (unlikely(fnhe))
1393                         cached = rt_bind_exception(rt, fnhe, daddr);
1394                 else if (!(rt->dst.flags & DST_NOCACHE))
1395                         cached = rt_cache_route(nh, rt);
1396                 if (unlikely(!cached)) {
1397                         /* Routes we intend to cache in nexthop exception or
1398                          * FIB nexthop have the DST_NOCACHE bit clear.
1399                          * However, if we are unsuccessful at storing this
1400                          * route into the cache we really need to set it.
1401                          */
1402                         rt->dst.flags |= DST_NOCACHE;
1403                         if (!rt->rt_gateway)
1404                                 rt->rt_gateway = daddr;
1405                         rt_add_uncached_list(rt);
1406                 }
1407         } else
1408                 rt_add_uncached_list(rt);
1409
1410 #ifdef CONFIG_IP_ROUTE_CLASSID
1411 #ifdef CONFIG_IP_MULTIPLE_TABLES
1412         set_class_tag(rt, res->tclassid);
1413 #endif
1414         set_class_tag(rt, itag);
1415 #endif
1416 }
1417
1418 static struct rtable *rt_dst_alloc(struct net_device *dev,
1419                                    bool nopolicy, bool noxfrm, bool will_cache)
1420 {
1421         return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1422                          (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1423                          (nopolicy ? DST_NOPOLICY : 0) |
1424                          (noxfrm ? DST_NOXFRM : 0));
1425 }
1426
1427 /* called in rcu_read_lock() section */
1428 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1429                                 u8 tos, struct net_device *dev, int our)
1430 {
1431         struct rtable *rth;
1432         struct in_device *in_dev = __in_dev_get_rcu(dev);
1433         u32 itag = 0;
1434         int err;
1435
1436         /* Primary sanity checks. */
1437
1438         if (in_dev == NULL)
1439                 return -EINVAL;
1440
1441         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1442             skb->protocol != htons(ETH_P_IP))
1443                 goto e_inval;
1444
1445         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1446                 if (ipv4_is_loopback(saddr))
1447                         goto e_inval;
1448
1449         if (ipv4_is_zeronet(saddr)) {
1450                 if (!ipv4_is_local_multicast(daddr))
1451                         goto e_inval;
1452         } else {
1453                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1454                                           in_dev, &itag);
1455                 if (err < 0)
1456                         goto e_err;
1457         }
1458         rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1459                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1460         if (!rth)
1461                 goto e_nobufs;
1462
1463 #ifdef CONFIG_IP_ROUTE_CLASSID
1464         rth->dst.tclassid = itag;
1465 #endif
1466         rth->dst.output = ip_rt_bug;
1467
1468         rth->rt_genid   = rt_genid_ipv4(dev_net(dev));
1469         rth->rt_flags   = RTCF_MULTICAST;
1470         rth->rt_type    = RTN_MULTICAST;
1471         rth->rt_is_input= 1;
1472         rth->rt_iif     = 0;
1473         rth->rt_pmtu    = 0;
1474         rth->rt_gateway = 0;
1475         rth->rt_uses_gateway = 0;
1476         INIT_LIST_HEAD(&rth->rt_uncached);
1477         if (our) {
1478                 rth->dst.input= ip_local_deliver;
1479                 rth->rt_flags |= RTCF_LOCAL;
1480         }
1481
1482 #ifdef CONFIG_IP_MROUTE
1483         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1484                 rth->dst.input = ip_mr_input;
1485 #endif
1486         RT_CACHE_STAT_INC(in_slow_mc);
1487
1488         skb_dst_set(skb, &rth->dst);
1489         return 0;
1490
1491 e_nobufs:
1492         return -ENOBUFS;
1493 e_inval:
1494         return -EINVAL;
1495 e_err:
1496         return err;
1497 }
1498
1499
1500 static void ip_handle_martian_source(struct net_device *dev,
1501                                      struct in_device *in_dev,
1502                                      struct sk_buff *skb,
1503                                      __be32 daddr,
1504                                      __be32 saddr)
1505 {
1506         RT_CACHE_STAT_INC(in_martian_src);
1507 #ifdef CONFIG_IP_ROUTE_VERBOSE
1508         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1509                 /*
1510                  *      RFC1812 recommendation, if source is martian,
1511                  *      the only hint is MAC header.
1512                  */
1513                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1514                         &daddr, &saddr, dev->name);
1515                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1516                         print_hex_dump(KERN_WARNING, "ll header: ",
1517                                        DUMP_PREFIX_OFFSET, 16, 1,
1518                                        skb_mac_header(skb),
1519                                        dev->hard_header_len, true);
1520                 }
1521         }
1522 #endif
1523 }
1524
1525 /* called in rcu_read_lock() section */
1526 static int __mkroute_input(struct sk_buff *skb,
1527                            const struct fib_result *res,
1528                            struct in_device *in_dev,
1529                            __be32 daddr, __be32 saddr, u32 tos)
1530 {
1531         struct fib_nh_exception *fnhe;
1532         struct rtable *rth;
1533         int err;
1534         struct in_device *out_dev;
1535         unsigned int flags = 0;
1536         bool do_cache;
1537         u32 itag = 0;
1538
1539         /* get a working reference to the output device */
1540         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1541         if (out_dev == NULL) {
1542                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1543                 return -EINVAL;
1544         }
1545
1546         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1547                                   in_dev->dev, in_dev, &itag);
1548         if (err < 0) {
1549                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1550                                          saddr);
1551
1552                 goto cleanup;
1553         }
1554
1555         do_cache = res->fi && !itag;
1556         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1557             skb->protocol == htons(ETH_P_IP) &&
1558             (IN_DEV_SHARED_MEDIA(out_dev) ||
1559              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1560                 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1561
1562         if (skb->protocol != htons(ETH_P_IP)) {
1563                 /* Not IP (i.e. ARP). Do not create route, if it is
1564                  * invalid for proxy arp. DNAT routes are always valid.
1565                  *
1566                  * Proxy arp feature have been extended to allow, ARP
1567                  * replies back to the same interface, to support
1568                  * Private VLAN switch technologies. See arp.c.
1569                  */
1570                 if (out_dev == in_dev &&
1571                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1572                         err = -EINVAL;
1573                         goto cleanup;
1574                 }
1575         }
1576
1577         fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1578         if (do_cache) {
1579                 if (fnhe != NULL)
1580                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1581                 else
1582                         rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1583
1584                 if (rt_cache_valid(rth)) {
1585                         skb_dst_set_noref(skb, &rth->dst);
1586                         goto out;
1587                 }
1588         }
1589
1590         rth = rt_dst_alloc(out_dev->dev,
1591                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1592                            IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1593         if (!rth) {
1594                 err = -ENOBUFS;
1595                 goto cleanup;
1596         }
1597
1598         rth->rt_genid = rt_genid_ipv4(dev_net(rth->dst.dev));
1599         rth->rt_flags = flags;
1600         rth->rt_type = res->type;
1601         rth->rt_is_input = 1;
1602         rth->rt_iif     = 0;
1603         rth->rt_pmtu    = 0;
1604         rth->rt_gateway = 0;
1605         rth->rt_uses_gateway = 0;
1606         INIT_LIST_HEAD(&rth->rt_uncached);
1607         RT_CACHE_STAT_INC(in_slow_tot);
1608
1609         rth->dst.input = ip_forward;
1610         rth->dst.output = ip_output;
1611
1612         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1613         skb_dst_set(skb, &rth->dst);
1614 out:
1615         err = 0;
1616  cleanup:
1617         return err;
1618 }
1619
1620 static int ip_mkroute_input(struct sk_buff *skb,
1621                             struct fib_result *res,
1622                             const struct flowi4 *fl4,
1623                             struct in_device *in_dev,
1624                             __be32 daddr, __be32 saddr, u32 tos)
1625 {
1626 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1627         if (res->fi && res->fi->fib_nhs > 1)
1628                 fib_select_multipath(res);
1629 #endif
1630
1631         /* create a routing cache entry */
1632         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1633 }
1634
1635 /*
1636  *      NOTE. We drop all the packets that has local source
1637  *      addresses, because every properly looped back packet
1638  *      must have correct destination already attached by output routine.
1639  *
1640  *      Such approach solves two big problems:
1641  *      1. Not simplex devices are handled properly.
1642  *      2. IP spoofing attempts are filtered with 100% of guarantee.
1643  *      called with rcu_read_lock()
1644  */
1645
1646 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1647                                u8 tos, struct net_device *dev)
1648 {
1649         struct fib_result res;
1650         struct in_device *in_dev = __in_dev_get_rcu(dev);
1651         struct flowi4   fl4;
1652         unsigned int    flags = 0;
1653         u32             itag = 0;
1654         struct rtable   *rth;
1655         int             err = -EINVAL;
1656         struct net    *net = dev_net(dev);
1657         bool do_cache;
1658
1659         /* IP on this device is disabled. */
1660
1661         if (!in_dev)
1662                 goto out;
1663
1664         /* Check for the most weird martians, which can be not detected
1665            by fib_lookup.
1666          */
1667
1668         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1669                 goto martian_source;
1670
1671         res.fi = NULL;
1672         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1673                 goto brd_input;
1674
1675         /* Accept zero addresses only to limited broadcast;
1676          * I even do not know to fix it or not. Waiting for complains :-)
1677          */
1678         if (ipv4_is_zeronet(saddr))
1679                 goto martian_source;
1680
1681         if (ipv4_is_zeronet(daddr))
1682                 goto martian_destination;
1683
1684         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1685          * and call it once if daddr or/and saddr are loopback addresses
1686          */
1687         if (ipv4_is_loopback(daddr)) {
1688                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1689                         goto martian_destination;
1690         } else if (ipv4_is_loopback(saddr)) {
1691                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1692                         goto martian_source;
1693         }
1694
1695         /*
1696          *      Now we are ready to route packet.
1697          */
1698         fl4.flowi4_oif = 0;
1699         fl4.flowi4_iif = dev->ifindex;
1700         fl4.flowi4_mark = skb->mark;
1701         fl4.flowi4_tos = tos;
1702         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1703         fl4.daddr = daddr;
1704         fl4.saddr = saddr;
1705         err = fib_lookup(net, &fl4, &res);
1706         if (err != 0) {
1707                 if (!IN_DEV_FORWARD(in_dev))
1708                         err = -EHOSTUNREACH;
1709                 goto no_route;
1710         }
1711
1712         if (res.type == RTN_BROADCAST)
1713                 goto brd_input;
1714
1715         if (res.type == RTN_LOCAL) {
1716                 err = fib_validate_source(skb, saddr, daddr, tos,
1717                                           LOOPBACK_IFINDEX,
1718                                           dev, in_dev, &itag);
1719                 if (err < 0)
1720                         goto martian_source_keep_err;
1721                 goto local_input;
1722         }
1723
1724         if (!IN_DEV_FORWARD(in_dev)) {
1725                 err = -EHOSTUNREACH;
1726                 goto no_route;
1727         }
1728         if (res.type != RTN_UNICAST)
1729                 goto martian_destination;
1730
1731         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1732 out:    return err;
1733
1734 brd_input:
1735         if (skb->protocol != htons(ETH_P_IP))
1736                 goto e_inval;
1737
1738         if (!ipv4_is_zeronet(saddr)) {
1739                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1740                                           in_dev, &itag);
1741                 if (err < 0)
1742                         goto martian_source_keep_err;
1743         }
1744         flags |= RTCF_BROADCAST;
1745         res.type = RTN_BROADCAST;
1746         RT_CACHE_STAT_INC(in_brd);
1747
1748 local_input:
1749         do_cache = false;
1750         if (res.fi) {
1751                 if (!itag) {
1752                         rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1753                         if (rt_cache_valid(rth)) {
1754                                 skb_dst_set_noref(skb, &rth->dst);
1755                                 err = 0;
1756                                 goto out;
1757                         }
1758                         do_cache = true;
1759                 }
1760         }
1761
1762         rth = rt_dst_alloc(net->loopback_dev,
1763                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1764         if (!rth)
1765                 goto e_nobufs;
1766
1767         rth->dst.input= ip_local_deliver;
1768         rth->dst.output= ip_rt_bug;
1769 #ifdef CONFIG_IP_ROUTE_CLASSID
1770         rth->dst.tclassid = itag;
1771 #endif
1772
1773         rth->rt_genid = rt_genid_ipv4(net);
1774         rth->rt_flags   = flags|RTCF_LOCAL;
1775         rth->rt_type    = res.type;
1776         rth->rt_is_input = 1;
1777         rth->rt_iif     = 0;
1778         rth->rt_pmtu    = 0;
1779         rth->rt_gateway = 0;
1780         rth->rt_uses_gateway = 0;
1781         INIT_LIST_HEAD(&rth->rt_uncached);
1782         RT_CACHE_STAT_INC(in_slow_tot);
1783         if (res.type == RTN_UNREACHABLE) {
1784                 rth->dst.input= ip_error;
1785                 rth->dst.error= -err;
1786                 rth->rt_flags   &= ~RTCF_LOCAL;
1787         }
1788         if (do_cache) {
1789                 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1790                         rth->dst.flags |= DST_NOCACHE;
1791                         rt_add_uncached_list(rth);
1792                 }
1793         }
1794         skb_dst_set(skb, &rth->dst);
1795         err = 0;
1796         goto out;
1797
1798 no_route:
1799         RT_CACHE_STAT_INC(in_no_route);
1800         res.type = RTN_UNREACHABLE;
1801         if (err == -ESRCH)
1802                 err = -ENETUNREACH;
1803         goto local_input;
1804
1805         /*
1806          *      Do not cache martian addresses: they should be logged (RFC1812)
1807          */
1808 martian_destination:
1809         RT_CACHE_STAT_INC(in_martian_dst);
1810 #ifdef CONFIG_IP_ROUTE_VERBOSE
1811         if (IN_DEV_LOG_MARTIANS(in_dev))
1812                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1813                                      &daddr, &saddr, dev->name);
1814 #endif
1815
1816 e_inval:
1817         err = -EINVAL;
1818         goto out;
1819
1820 e_nobufs:
1821         err = -ENOBUFS;
1822         goto out;
1823
1824 martian_source:
1825         err = -EINVAL;
1826 martian_source_keep_err:
1827         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1828         goto out;
1829 }
1830
1831 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1832                          u8 tos, struct net_device *dev)
1833 {
1834         int res;
1835
1836         rcu_read_lock();
1837
1838         /* Multicast recognition logic is moved from route cache to here.
1839            The problem was that too many Ethernet cards have broken/missing
1840            hardware multicast filters :-( As result the host on multicasting
1841            network acquires a lot of useless route cache entries, sort of
1842            SDR messages from all the world. Now we try to get rid of them.
1843            Really, provided software IP multicast filter is organized
1844            reasonably (at least, hashed), it does not result in a slowdown
1845            comparing with route cache reject entries.
1846            Note, that multicast routers are not affected, because
1847            route cache entry is created eventually.
1848          */
1849         if (ipv4_is_multicast(daddr)) {
1850                 struct in_device *in_dev = __in_dev_get_rcu(dev);
1851
1852                 if (in_dev) {
1853                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1854                                                   ip_hdr(skb)->protocol);
1855                         if (our
1856 #ifdef CONFIG_IP_MROUTE
1857                                 ||
1858                             (!ipv4_is_local_multicast(daddr) &&
1859                              IN_DEV_MFORWARD(in_dev))
1860 #endif
1861                            ) {
1862                                 int res = ip_route_input_mc(skb, daddr, saddr,
1863                                                             tos, dev, our);
1864                                 rcu_read_unlock();
1865                                 return res;
1866                         }
1867                 }
1868                 rcu_read_unlock();
1869                 return -EINVAL;
1870         }
1871         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1872         rcu_read_unlock();
1873         return res;
1874 }
1875 EXPORT_SYMBOL(ip_route_input_noref);
1876
1877 /* called with rcu_read_lock() */
1878 static struct rtable *__mkroute_output(const struct fib_result *res,
1879                                        const struct flowi4 *fl4, int orig_oif,
1880                                        struct net_device *dev_out,
1881                                        unsigned int flags)
1882 {
1883         struct fib_info *fi = res->fi;
1884         struct fib_nh_exception *fnhe;
1885         struct in_device *in_dev;
1886         u16 type = res->type;
1887         struct rtable *rth;
1888         bool do_cache;
1889
1890         in_dev = __in_dev_get_rcu(dev_out);
1891         if (!in_dev)
1892                 return ERR_PTR(-EINVAL);
1893
1894         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1895                 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1896                         return ERR_PTR(-EINVAL);
1897
1898         if (ipv4_is_lbcast(fl4->daddr))
1899                 type = RTN_BROADCAST;
1900         else if (ipv4_is_multicast(fl4->daddr))
1901                 type = RTN_MULTICAST;
1902         else if (ipv4_is_zeronet(fl4->daddr))
1903                 return ERR_PTR(-EINVAL);
1904
1905         if (dev_out->flags & IFF_LOOPBACK)
1906                 flags |= RTCF_LOCAL;
1907
1908         do_cache = true;
1909         if (type == RTN_BROADCAST) {
1910                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1911                 fi = NULL;
1912         } else if (type == RTN_MULTICAST) {
1913                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1914                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1915                                      fl4->flowi4_proto))
1916                         flags &= ~RTCF_LOCAL;
1917                 else
1918                         do_cache = false;
1919                 /* If multicast route do not exist use
1920                  * default one, but do not gateway in this case.
1921                  * Yes, it is hack.
1922                  */
1923                 if (fi && res->prefixlen < 4)
1924                         fi = NULL;
1925         }
1926
1927         fnhe = NULL;
1928         do_cache &= fi != NULL;
1929         if (do_cache) {
1930                 struct rtable __rcu **prth;
1931                 struct fib_nh *nh = &FIB_RES_NH(*res);
1932
1933                 fnhe = find_exception(nh, fl4->daddr);
1934                 if (fnhe)
1935                         prth = &fnhe->fnhe_rth_output;
1936                 else {
1937                         if (unlikely(fl4->flowi4_flags &
1938                                      FLOWI_FLAG_KNOWN_NH &&
1939                                      !(nh->nh_gw &&
1940                                        nh->nh_scope == RT_SCOPE_LINK))) {
1941                                 do_cache = false;
1942                                 goto add;
1943                         }
1944                         prth = __this_cpu_ptr(nh->nh_pcpu_rth_output);
1945                 }
1946                 rth = rcu_dereference(*prth);
1947                 if (rt_cache_valid(rth)) {
1948                         dst_hold(&rth->dst);
1949                         return rth;
1950                 }
1951         }
1952
1953 add:
1954         rth = rt_dst_alloc(dev_out,
1955                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1956                            IN_DEV_CONF_GET(in_dev, NOXFRM),
1957                            do_cache);
1958         if (!rth)
1959                 return ERR_PTR(-ENOBUFS);
1960
1961         rth->dst.output = ip_output;
1962
1963         rth->rt_genid = rt_genid_ipv4(dev_net(dev_out));
1964         rth->rt_flags   = flags;
1965         rth->rt_type    = type;
1966         rth->rt_is_input = 0;
1967         rth->rt_iif     = orig_oif ? : 0;
1968         rth->rt_pmtu    = 0;
1969         rth->rt_gateway = 0;
1970         rth->rt_uses_gateway = 0;
1971         INIT_LIST_HEAD(&rth->rt_uncached);
1972
1973         RT_CACHE_STAT_INC(out_slow_tot);
1974
1975         if (flags & RTCF_LOCAL)
1976                 rth->dst.input = ip_local_deliver;
1977         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1978                 if (flags & RTCF_LOCAL &&
1979                     !(dev_out->flags & IFF_LOOPBACK)) {
1980                         rth->dst.output = ip_mc_output;
1981                         RT_CACHE_STAT_INC(out_slow_mc);
1982                 }
1983 #ifdef CONFIG_IP_MROUTE
1984                 if (type == RTN_MULTICAST) {
1985                         if (IN_DEV_MFORWARD(in_dev) &&
1986                             !ipv4_is_local_multicast(fl4->daddr)) {
1987                                 rth->dst.input = ip_mr_input;
1988                                 rth->dst.output = ip_mc_output;
1989                         }
1990                 }
1991 #endif
1992         }
1993
1994         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
1995
1996         return rth;
1997 }
1998
1999 /*
2000  * Major route resolver routine.
2001  */
2002
2003 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
2004 {
2005         struct net_device *dev_out = NULL;
2006         __u8 tos = RT_FL_TOS(fl4);
2007         unsigned int flags = 0;
2008         struct fib_result res;
2009         struct rtable *rth;
2010         int orig_oif;
2011
2012         res.tclassid    = 0;
2013         res.fi          = NULL;
2014         res.table       = NULL;
2015
2016         orig_oif = fl4->flowi4_oif;
2017
2018         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2019         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2020         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2021                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2022
2023         rcu_read_lock();
2024         if (fl4->saddr) {
2025                 rth = ERR_PTR(-EINVAL);
2026                 if (ipv4_is_multicast(fl4->saddr) ||
2027                     ipv4_is_lbcast(fl4->saddr) ||
2028                     ipv4_is_zeronet(fl4->saddr))
2029                         goto out;
2030
2031                 /* I removed check for oif == dev_out->oif here.
2032                    It was wrong for two reasons:
2033                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2034                       is assigned to multiple interfaces.
2035                    2. Moreover, we are allowed to send packets with saddr
2036                       of another iface. --ANK
2037                  */
2038
2039                 if (fl4->flowi4_oif == 0 &&
2040                     (ipv4_is_multicast(fl4->daddr) ||
2041                      ipv4_is_lbcast(fl4->daddr))) {
2042                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2043                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2044                         if (dev_out == NULL)
2045                                 goto out;
2046
2047                         /* Special hack: user can direct multicasts
2048                            and limited broadcast via necessary interface
2049                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2050                            This hack is not just for fun, it allows
2051                            vic,vat and friends to work.
2052                            They bind socket to loopback, set ttl to zero
2053                            and expect that it will work.
2054                            From the viewpoint of routing cache they are broken,
2055                            because we are not allowed to build multicast path
2056                            with loopback source addr (look, routing cache
2057                            cannot know, that ttl is zero, so that packet
2058                            will not leave this host and route is valid).
2059                            Luckily, this hack is good workaround.
2060                          */
2061
2062                         fl4->flowi4_oif = dev_out->ifindex;
2063                         goto make_route;
2064                 }
2065
2066                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2067                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2068                         if (!__ip_dev_find(net, fl4->saddr, false))
2069                                 goto out;
2070                 }
2071         }
2072
2073
2074         if (fl4->flowi4_oif) {
2075                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2076                 rth = ERR_PTR(-ENODEV);
2077                 if (dev_out == NULL)
2078                         goto out;
2079
2080                 /* RACE: Check return value of inet_select_addr instead. */
2081                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2082                         rth = ERR_PTR(-ENETUNREACH);
2083                         goto out;
2084                 }
2085                 if (ipv4_is_local_multicast(fl4->daddr) ||
2086                     ipv4_is_lbcast(fl4->daddr)) {
2087                         if (!fl4->saddr)
2088                                 fl4->saddr = inet_select_addr(dev_out, 0,
2089                                                               RT_SCOPE_LINK);
2090                         goto make_route;
2091                 }
2092                 if (!fl4->saddr) {
2093                         if (ipv4_is_multicast(fl4->daddr))
2094                                 fl4->saddr = inet_select_addr(dev_out, 0,
2095                                                               fl4->flowi4_scope);
2096                         else if (!fl4->daddr)
2097                                 fl4->saddr = inet_select_addr(dev_out, 0,
2098                                                               RT_SCOPE_HOST);
2099                 }
2100         }
2101
2102         if (!fl4->daddr) {
2103                 fl4->daddr = fl4->saddr;
2104                 if (!fl4->daddr)
2105                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2106                 dev_out = net->loopback_dev;
2107                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2108                 res.type = RTN_LOCAL;
2109                 flags |= RTCF_LOCAL;
2110                 goto make_route;
2111         }
2112
2113         if (fib_lookup(net, fl4, &res)) {
2114                 res.fi = NULL;
2115                 res.table = NULL;
2116                 if (fl4->flowi4_oif) {
2117                         /* Apparently, routing tables are wrong. Assume,
2118                            that the destination is on link.
2119
2120                            WHY? DW.
2121                            Because we are allowed to send to iface
2122                            even if it has NO routes and NO assigned
2123                            addresses. When oif is specified, routing
2124                            tables are looked up with only one purpose:
2125                            to catch if destination is gatewayed, rather than
2126                            direct. Moreover, if MSG_DONTROUTE is set,
2127                            we send packet, ignoring both routing tables
2128                            and ifaddr state. --ANK
2129
2130
2131                            We could make it even if oif is unknown,
2132                            likely IPv6, but we do not.
2133                          */
2134
2135                         if (fl4->saddr == 0)
2136                                 fl4->saddr = inet_select_addr(dev_out, 0,
2137                                                               RT_SCOPE_LINK);
2138                         res.type = RTN_UNICAST;
2139                         goto make_route;
2140                 }
2141                 rth = ERR_PTR(-ENETUNREACH);
2142                 goto out;
2143         }
2144
2145         if (res.type == RTN_LOCAL) {
2146                 if (!fl4->saddr) {
2147                         if (res.fi->fib_prefsrc)
2148                                 fl4->saddr = res.fi->fib_prefsrc;
2149                         else
2150                                 fl4->saddr = fl4->daddr;
2151                 }
2152                 dev_out = net->loopback_dev;
2153                 fl4->flowi4_oif = dev_out->ifindex;
2154                 flags |= RTCF_LOCAL;
2155                 goto make_route;
2156         }
2157
2158 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2159         if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2160                 fib_select_multipath(&res);
2161         else
2162 #endif
2163         if (!res.prefixlen &&
2164             res.table->tb_num_default > 1 &&
2165             res.type == RTN_UNICAST && !fl4->flowi4_oif)
2166                 fib_select_default(&res);
2167
2168         if (!fl4->saddr)
2169                 fl4->saddr = FIB_RES_PREFSRC(net, res);
2170
2171         dev_out = FIB_RES_DEV(res);
2172         fl4->flowi4_oif = dev_out->ifindex;
2173
2174
2175 make_route:
2176         rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2177
2178 out:
2179         rcu_read_unlock();
2180         return rth;
2181 }
2182 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2183
2184 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2185 {
2186         return NULL;
2187 }
2188
2189 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2190 {
2191         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2192
2193         return mtu ? : dst->dev->mtu;
2194 }
2195
2196 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2197                                           struct sk_buff *skb, u32 mtu)
2198 {
2199 }
2200
2201 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2202                                        struct sk_buff *skb)
2203 {
2204 }
2205
2206 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2207                                           unsigned long old)
2208 {
2209         return NULL;
2210 }
2211
2212 static struct dst_ops ipv4_dst_blackhole_ops = {
2213         .family                 =       AF_INET,
2214         .protocol               =       cpu_to_be16(ETH_P_IP),
2215         .check                  =       ipv4_blackhole_dst_check,
2216         .mtu                    =       ipv4_blackhole_mtu,
2217         .default_advmss         =       ipv4_default_advmss,
2218         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2219         .redirect               =       ipv4_rt_blackhole_redirect,
2220         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
2221         .neigh_lookup           =       ipv4_neigh_lookup,
2222 };
2223
2224 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2225 {
2226         struct rtable *ort = (struct rtable *) dst_orig;
2227         struct rtable *rt;
2228
2229         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2230         if (rt) {
2231                 struct dst_entry *new = &rt->dst;
2232
2233                 new->__use = 1;
2234                 new->input = dst_discard;
2235                 new->output = dst_discard;
2236
2237                 new->dev = ort->dst.dev;
2238                 if (new->dev)
2239                         dev_hold(new->dev);
2240
2241                 rt->rt_is_input = ort->rt_is_input;
2242                 rt->rt_iif = ort->rt_iif;
2243                 rt->rt_pmtu = ort->rt_pmtu;
2244
2245                 rt->rt_genid = rt_genid_ipv4(net);
2246                 rt->rt_flags = ort->rt_flags;
2247                 rt->rt_type = ort->rt_type;
2248                 rt->rt_gateway = ort->rt_gateway;
2249                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2250
2251                 INIT_LIST_HEAD(&rt->rt_uncached);
2252
2253                 dst_free(new);
2254         }
2255
2256         dst_release(dst_orig);
2257
2258         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2259 }
2260
2261 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2262                                     struct sock *sk)
2263 {
2264         struct rtable *rt = __ip_route_output_key(net, flp4);
2265
2266         if (IS_ERR(rt))
2267                 return rt;
2268
2269         if (flp4->flowi4_proto)
2270                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2271                                                         flowi4_to_flowi(flp4),
2272                                                         sk, 0);
2273
2274         return rt;
2275 }
2276 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2277
2278 static int rt_fill_info(struct net *net,  __be32 dst, __be32 src,
2279                         struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2280                         u32 seq, int event, int nowait, unsigned int flags)
2281 {
2282         struct rtable *rt = skb_rtable(skb);
2283         struct rtmsg *r;
2284         struct nlmsghdr *nlh;
2285         unsigned long expires = 0;
2286         u32 error;
2287         u32 metrics[RTAX_MAX];
2288
2289         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2290         if (nlh == NULL)
2291                 return -EMSGSIZE;
2292
2293         r = nlmsg_data(nlh);
2294         r->rtm_family    = AF_INET;
2295         r->rtm_dst_len  = 32;
2296         r->rtm_src_len  = 0;
2297         r->rtm_tos      = fl4->flowi4_tos;
2298         r->rtm_table    = RT_TABLE_MAIN;
2299         if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2300                 goto nla_put_failure;
2301         r->rtm_type     = rt->rt_type;
2302         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2303         r->rtm_protocol = RTPROT_UNSPEC;
2304         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2305         if (rt->rt_flags & RTCF_NOTIFY)
2306                 r->rtm_flags |= RTM_F_NOTIFY;
2307         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2308                 r->rtm_flags |= RTCF_DOREDIRECT;
2309
2310         if (nla_put_be32(skb, RTA_DST, dst))
2311                 goto nla_put_failure;
2312         if (src) {
2313                 r->rtm_src_len = 32;
2314                 if (nla_put_be32(skb, RTA_SRC, src))
2315                         goto nla_put_failure;
2316         }
2317         if (rt->dst.dev &&
2318             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2319                 goto nla_put_failure;
2320 #ifdef CONFIG_IP_ROUTE_CLASSID
2321         if (rt->dst.tclassid &&
2322             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2323                 goto nla_put_failure;
2324 #endif
2325         if (!rt_is_input_route(rt) &&
2326             fl4->saddr != src) {
2327                 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2328                         goto nla_put_failure;
2329         }
2330         if (rt->rt_uses_gateway &&
2331             nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2332                 goto nla_put_failure;
2333
2334         expires = rt->dst.expires;
2335         if (expires) {
2336                 unsigned long now = jiffies;
2337
2338                 if (time_before(now, expires))
2339                         expires -= now;
2340                 else
2341                         expires = 0;
2342         }
2343
2344         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2345         if (rt->rt_pmtu && expires)
2346                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2347         if (rtnetlink_put_metrics(skb, metrics) < 0)
2348                 goto nla_put_failure;
2349
2350         if (fl4->flowi4_mark &&
2351             nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2352                 goto nla_put_failure;
2353
2354         error = rt->dst.error;
2355
2356         if (rt_is_input_route(rt)) {
2357 #ifdef CONFIG_IP_MROUTE
2358                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2359                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2360                         int err = ipmr_get_route(net, skb,
2361                                                  fl4->saddr, fl4->daddr,
2362                                                  r, nowait);
2363                         if (err <= 0) {
2364                                 if (!nowait) {
2365                                         if (err == 0)
2366                                                 return 0;
2367                                         goto nla_put_failure;
2368                                 } else {
2369                                         if (err == -EMSGSIZE)
2370                                                 goto nla_put_failure;
2371                                         error = err;
2372                                 }
2373                         }
2374                 } else
2375 #endif
2376                         if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2377                                 goto nla_put_failure;
2378         }
2379
2380         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2381                 goto nla_put_failure;
2382
2383         return nlmsg_end(skb, nlh);
2384
2385 nla_put_failure:
2386         nlmsg_cancel(skb, nlh);
2387         return -EMSGSIZE;
2388 }
2389
2390 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2391 {
2392         struct net *net = sock_net(in_skb->sk);
2393         struct rtmsg *rtm;
2394         struct nlattr *tb[RTA_MAX+1];
2395         struct rtable *rt = NULL;
2396         struct flowi4 fl4;
2397         __be32 dst = 0;
2398         __be32 src = 0;
2399         u32 iif;
2400         int err;
2401         int mark;
2402         struct sk_buff *skb;
2403
2404         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2405         if (err < 0)
2406                 goto errout;
2407
2408         rtm = nlmsg_data(nlh);
2409
2410         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2411         if (skb == NULL) {
2412                 err = -ENOBUFS;
2413                 goto errout;
2414         }
2415
2416         /* Reserve room for dummy headers, this skb can pass
2417            through good chunk of routing engine.
2418          */
2419         skb_reset_mac_header(skb);
2420         skb_reset_network_header(skb);
2421
2422         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2423         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2424         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2425
2426         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2427         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2428         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2429         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2430
2431         memset(&fl4, 0, sizeof(fl4));
2432         fl4.daddr = dst;
2433         fl4.saddr = src;
2434         fl4.flowi4_tos = rtm->rtm_tos;
2435         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2436         fl4.flowi4_mark = mark;
2437
2438         if (iif) {
2439                 struct net_device *dev;
2440
2441                 dev = __dev_get_by_index(net, iif);
2442                 if (dev == NULL) {
2443                         err = -ENODEV;
2444                         goto errout_free;
2445                 }
2446
2447                 skb->protocol   = htons(ETH_P_IP);
2448                 skb->dev        = dev;
2449                 skb->mark       = mark;
2450                 local_bh_disable();
2451                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2452                 local_bh_enable();
2453
2454                 rt = skb_rtable(skb);
2455                 if (err == 0 && rt->dst.error)
2456                         err = -rt->dst.error;
2457         } else {
2458                 rt = ip_route_output_key(net, &fl4);
2459
2460                 err = 0;
2461                 if (IS_ERR(rt))
2462                         err = PTR_ERR(rt);
2463         }
2464
2465         if (err)
2466                 goto errout_free;
2467
2468         skb_dst_set(skb, &rt->dst);
2469         if (rtm->rtm_flags & RTM_F_NOTIFY)
2470                 rt->rt_flags |= RTCF_NOTIFY;
2471
2472         err = rt_fill_info(net, dst, src, &fl4, skb,
2473                            NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2474                            RTM_NEWROUTE, 0, 0);
2475         if (err <= 0)
2476                 goto errout_free;
2477
2478         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2479 errout:
2480         return err;
2481
2482 errout_free:
2483         kfree_skb(skb);
2484         goto errout;
2485 }
2486
2487 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
2488 {
2489         return skb->len;
2490 }
2491
2492 void ip_rt_multicast_event(struct in_device *in_dev)
2493 {
2494         rt_cache_flush(dev_net(in_dev->dev));
2495 }
2496
2497 #ifdef CONFIG_SYSCTL
2498 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
2499 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
2500 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
2501 static int ip_rt_gc_elasticity __read_mostly    = 8;
2502
2503 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2504                                         void __user *buffer,
2505                                         size_t *lenp, loff_t *ppos)
2506 {
2507         struct net *net = (struct net *)__ctl->extra1;
2508
2509         if (write) {
2510                 rt_cache_flush(net);
2511                 fnhe_genid_bump(net);
2512                 return 0;
2513         }
2514
2515         return -EINVAL;
2516 }
2517
2518 static struct ctl_table ipv4_route_table[] = {
2519         {
2520                 .procname       = "gc_thresh",
2521                 .data           = &ipv4_dst_ops.gc_thresh,
2522                 .maxlen         = sizeof(int),
2523                 .mode           = 0644,
2524                 .proc_handler   = proc_dointvec,
2525         },
2526         {
2527                 .procname       = "max_size",
2528                 .data           = &ip_rt_max_size,
2529                 .maxlen         = sizeof(int),
2530                 .mode           = 0644,
2531                 .proc_handler   = proc_dointvec,
2532         },
2533         {
2534                 /*  Deprecated. Use gc_min_interval_ms */
2535
2536                 .procname       = "gc_min_interval",
2537                 .data           = &ip_rt_gc_min_interval,
2538                 .maxlen         = sizeof(int),
2539                 .mode           = 0644,
2540                 .proc_handler   = proc_dointvec_jiffies,
2541         },
2542         {
2543                 .procname       = "gc_min_interval_ms",
2544                 .data           = &ip_rt_gc_min_interval,
2545                 .maxlen         = sizeof(int),
2546                 .mode           = 0644,
2547                 .proc_handler   = proc_dointvec_ms_jiffies,
2548         },
2549         {
2550                 .procname       = "gc_timeout",
2551                 .data           = &ip_rt_gc_timeout,
2552                 .maxlen         = sizeof(int),
2553                 .mode           = 0644,
2554                 .proc_handler   = proc_dointvec_jiffies,
2555         },
2556         {
2557                 .procname       = "gc_interval",
2558                 .data           = &ip_rt_gc_interval,
2559                 .maxlen         = sizeof(int),
2560                 .mode           = 0644,
2561                 .proc_handler   = proc_dointvec_jiffies,
2562         },
2563         {
2564                 .procname       = "redirect_load",
2565                 .data           = &ip_rt_redirect_load,
2566                 .maxlen         = sizeof(int),
2567                 .mode           = 0644,
2568                 .proc_handler   = proc_dointvec,
2569         },
2570         {
2571                 .procname       = "redirect_number",
2572                 .data           = &ip_rt_redirect_number,
2573                 .maxlen         = sizeof(int),
2574                 .mode           = 0644,
2575                 .proc_handler   = proc_dointvec,
2576         },
2577         {
2578                 .procname       = "redirect_silence",
2579                 .data           = &ip_rt_redirect_silence,
2580                 .maxlen         = sizeof(int),
2581                 .mode           = 0644,
2582                 .proc_handler   = proc_dointvec,
2583         },
2584         {
2585                 .procname       = "error_cost",
2586                 .data           = &ip_rt_error_cost,
2587                 .maxlen         = sizeof(int),
2588                 .mode           = 0644,
2589                 .proc_handler   = proc_dointvec,
2590         },
2591         {
2592                 .procname       = "error_burst",
2593                 .data           = &ip_rt_error_burst,
2594                 .maxlen         = sizeof(int),
2595                 .mode           = 0644,
2596                 .proc_handler   = proc_dointvec,
2597         },
2598         {
2599                 .procname       = "gc_elasticity",
2600                 .data           = &ip_rt_gc_elasticity,
2601                 .maxlen         = sizeof(int),
2602                 .mode           = 0644,
2603                 .proc_handler   = proc_dointvec,
2604         },
2605         {
2606                 .procname       = "mtu_expires",
2607                 .data           = &ip_rt_mtu_expires,
2608                 .maxlen         = sizeof(int),
2609                 .mode           = 0644,
2610                 .proc_handler   = proc_dointvec_jiffies,
2611         },
2612         {
2613                 .procname       = "min_pmtu",
2614                 .data           = &ip_rt_min_pmtu,
2615                 .maxlen         = sizeof(int),
2616                 .mode           = 0644,
2617                 .proc_handler   = proc_dointvec,
2618         },
2619         {
2620                 .procname       = "min_adv_mss",
2621                 .data           = &ip_rt_min_advmss,
2622                 .maxlen         = sizeof(int),
2623                 .mode           = 0644,
2624                 .proc_handler   = proc_dointvec,
2625         },
2626         { }
2627 };
2628
2629 static struct ctl_table ipv4_route_flush_table[] = {
2630         {
2631                 .procname       = "flush",
2632                 .maxlen         = sizeof(int),
2633                 .mode           = 0200,
2634                 .proc_handler   = ipv4_sysctl_rtcache_flush,
2635         },
2636         { },
2637 };
2638
2639 static __net_init int sysctl_route_net_init(struct net *net)
2640 {
2641         struct ctl_table *tbl;
2642
2643         tbl = ipv4_route_flush_table;
2644         if (!net_eq(net, &init_net)) {
2645                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2646                 if (tbl == NULL)
2647                         goto err_dup;
2648
2649                 /* Don't export sysctls to unprivileged users */
2650                 if (net->user_ns != &init_user_ns)
2651                         tbl[0].procname = NULL;
2652         }
2653         tbl[0].extra1 = net;
2654
2655         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2656         if (net->ipv4.route_hdr == NULL)
2657                 goto err_reg;
2658         return 0;
2659
2660 err_reg:
2661         if (tbl != ipv4_route_flush_table)
2662                 kfree(tbl);
2663 err_dup:
2664         return -ENOMEM;
2665 }
2666
2667 static __net_exit void sysctl_route_net_exit(struct net *net)
2668 {
2669         struct ctl_table *tbl;
2670
2671         tbl = net->ipv4.route_hdr->ctl_table_arg;
2672         unregister_net_sysctl_table(net->ipv4.route_hdr);
2673         BUG_ON(tbl == ipv4_route_flush_table);
2674         kfree(tbl);
2675 }
2676
2677 static __net_initdata struct pernet_operations sysctl_route_ops = {
2678         .init = sysctl_route_net_init,
2679         .exit = sysctl_route_net_exit,
2680 };
2681 #endif
2682
2683 static __net_init int rt_genid_init(struct net *net)
2684 {
2685         atomic_set(&net->ipv4.rt_genid, 0);
2686         atomic_set(&net->fnhe_genid, 0);
2687         get_random_bytes(&net->ipv4.dev_addr_genid,
2688                          sizeof(net->ipv4.dev_addr_genid));
2689         return 0;
2690 }
2691
2692 static __net_initdata struct pernet_operations rt_genid_ops = {
2693         .init = rt_genid_init,
2694 };
2695
2696 static int __net_init ipv4_inetpeer_init(struct net *net)
2697 {
2698         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2699
2700         if (!bp)
2701                 return -ENOMEM;
2702         inet_peer_base_init(bp);
2703         net->ipv4.peers = bp;
2704         return 0;
2705 }
2706
2707 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2708 {
2709         struct inet_peer_base *bp = net->ipv4.peers;
2710
2711         net->ipv4.peers = NULL;
2712         inetpeer_invalidate_tree(bp);
2713         kfree(bp);
2714 }
2715
2716 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2717         .init   =       ipv4_inetpeer_init,
2718         .exit   =       ipv4_inetpeer_exit,
2719 };
2720
2721 #ifdef CONFIG_IP_ROUTE_CLASSID
2722 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2723 #endif /* CONFIG_IP_ROUTE_CLASSID */
2724
2725 int __init ip_rt_init(void)
2726 {
2727         int rc = 0;
2728
2729         ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2730         if (!ip_idents)
2731                 panic("IP: failed to allocate ip_idents\n");
2732
2733         prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2734
2735 #ifdef CONFIG_IP_ROUTE_CLASSID
2736         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2737         if (!ip_rt_acct)
2738                 panic("IP: failed to allocate ip_rt_acct\n");
2739 #endif
2740
2741         ipv4_dst_ops.kmem_cachep =
2742                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2743                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2744
2745         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2746
2747         if (dst_entries_init(&ipv4_dst_ops) < 0)
2748                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2749
2750         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2751                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2752
2753         ipv4_dst_ops.gc_thresh = ~0;
2754         ip_rt_max_size = INT_MAX;
2755
2756         devinet_init();
2757         ip_fib_init();
2758
2759         if (ip_rt_proc_init())
2760                 pr_err("Unable to create route proc files\n");
2761 #ifdef CONFIG_XFRM
2762         xfrm_init();
2763         xfrm4_init();
2764 #endif
2765         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2766
2767 #ifdef CONFIG_SYSCTL
2768         register_pernet_subsys(&sysctl_route_ops);
2769 #endif
2770         register_pernet_subsys(&rt_genid_ops);
2771         register_pernet_subsys(&ipv4_inetpeer_ops);
2772         return rc;
2773 }
2774
2775 #ifdef CONFIG_SYSCTL
2776 /*
2777  * We really need to sanitize the damn ipv4 init order, then all
2778  * this nonsense will go away.
2779  */
2780 void __init ip_static_sysctl_init(void)
2781 {
2782         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
2783 }
2784 #endif