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, <[email protected]>
  10  *              Alan Cox, <[email protected]>
  11  *              Linus Torvalds, <[email protected]>
  12  *              Alexey Kuznetsov, <[email protected]>
  13  *
  14  * Fixes:
  15  *              Alan Cox        :       Verify area fixes.
  16  *              Alan Cox        :       cli() protects routing changes
  17  *              Rui Oliveira    :       ICMP routing table updates
  18  *              ([email protected])      Routing table insertion and update
  19  *              Linus Torvalds  :       Rewrote bits to be sensible
  20  *              Alan Cox        :       Added BSD route gw semantics
  21  *              Alan Cox        :       Super /proc >4K
  22  *              Alan Cox        :       MTU in route table
  23  *              Alan Cox        :       MSS actually. Also added the window
  24  *                                      clamper.
  25  *              Sam Lantinga    :       Fixed route matching in rt_del()
  26  *              Alan Cox        :       Routing cache support.
  27  *              Alan Cox        :       Removed compatibility cruft.
  28  *              Alan Cox        :       RTF_REJECT support.
  29  *              Alan Cox        :       TCP irtt support.
  30  *              Jonathan Naylor :       Added Metric support.
  31  *      Miquel van Smoorenburg  :       BSD API fixes.
  32  *      Miquel van Smoorenburg  :       Metrics.
  33  *              Alan Cox        :       Use __u32 properly
  34  *              Alan Cox        :       Aligned routing errors more closely with BSD
  35  *                                      our system is still very different.
  36  *              Alan Cox        :       Faster /proc handling
  37  *      Alexey Kuznetsov        :       Massive rework to support tree based routing,
  38  *                                      routing caches and better behaviour.
  39  *
  40  *              Olaf Erb        :       irtt wasn't being copied right.
  41  *              Bjorn Ekwall    :       Kerneld route support.
  42  *              Alan Cox        :       Multicast fixed (I hope)
  43  *              Pavel Krauz     :       Limited broadcast fixed
  44  *              Mike McLagan    :       Routing by source
  45  *      Alexey Kuznetsov        :       End of old history. Split to fib.c and
  46  *                                      route.c and rewritten from scratch.
  47  *              Andi Kleen      :       Load-limit warning messages.
  48  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  49  *      Vitaly E. Lavrov        :       Race condition in ip_route_input_slow.
  50  *      Tobias Ringstrom        :       Uninitialized res.type in ip_route_output_slow.
  51  *      Vladimir V. Ivanov      :       IP rule info (flowid) is really useful.
  52  *              Marc Boucher    :       routing by fwmark
  53  *      Robert Olsson           :       Added rt_cache statistics
  54  *      Arnaldo C. Melo         :       Convert proc stuff to seq_file
  55  *      Eric Dumazet            :       hashed spinlocks and rt_check_expire() fixes.
  56  *      Ilia Sotnikov           :       Ignore TOS on PMTUD and Redirect
  57  *      Ilia Sotnikov           :       Removed TOS from hash calculations
  58  *
  59  *              This program is free software; you can redistribute it and/or
  60  *              modify it under the terms of the GNU General Public License
  61  *              as published by the Free Software Foundation; either version
  62  *              2 of the License, or (at your option) any later version.
  63  */
  64
  65 #define pr_fmt(fmt) "IPv4: " fmt
  66
  67 #include <linux/module.h>
  68 #include <asm/uaccess.h>
  69 #include <linux/bitops.h>
  70 #include <linux/types.h>
  71 #include <linux/kernel.h>
  72 #include <linux/mm.h>
  73 #include <linux/string.h>
  74 #include <linux/socket.h>
  75 #include <linux/sockios.h>
  76 #include <linux/errno.h>
  77 #include <linux/in.h>
  78 #include <linux/inet.h>
  79 #include <linux/netdevice.h>
  80 #include <linux/proc_fs.h>
  81 #include <linux/init.h>
  82 #include <linux/skbuff.h>
  83 #include <linux/inetdevice.h>
  84 #include <linux/igmp.h>
  85 #include <linux/pkt_sched.h>
  86 #include <linux/mroute.h>
  87 #include <linux/netfilter_ipv4.h>
  88 #include <linux/random.h>
  89 #include <linux/rcupdate.h>
  90 #include <linux/times.h>
  91 #include <linux/slab.h>
  92 #include <linux/jhash.h>
  93 #include <net/dst.h>
  94 #include <net/net_namespace.h>
  95 #include <net/protocol.h>
  96 #include <net/ip.h>
  97 #include <net/route.h>
  98 #include <net/inetpeer.h>
  99 #include <net/sock.h>
 100 #include <net/ip_fib.h>
 101 #include <net/arp.h>
 102 #include <net/tcp.h>
 103 #include <net/icmp.h>
 104 #include <net/xfrm.h>
 105 #include <net/netevent.h>
 106 #include <net/rtnetlink.h>
 107 #ifdef CONFIG_SYSCTL
 108 #include <linux/sysctl.h>
 109 #include <linux/kmemleak.h>
 110 #endif
 111 #include <net/secure_seq.h>
 112
 113 #define RT_FL_TOS(oldflp4) \
 114         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
 115
 116 #define RT_GC_TIMEOUT (300*HZ)
 117
 118 static int ip_rt_max_size;
 119 static int ip_rt_redirect_number __read_mostly  = 9;
 120 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 121 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 122 static int ip_rt_error_cost __read_mostly       = HZ;
 123 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 124 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
 125 static int ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
 126 static int ip_rt_min_advmss __read_mostly       = 256;
 127
 128 /*
 129  *      Interface to generic destination cache.
 130  */
 131
 132 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 133 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 134 static unsigned int      ipv4_mtu(const struct dst_entry *dst);
 135 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 136 static void              ipv4_link_failure(struct sk_buff *skb);
 137 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 138                                            struct sk_buff *skb, u32 mtu);
 139 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 140                                         struct sk_buff *skb);
 141 static void             ipv4_dst_destroy(struct dst_entry *dst);
 142
 143 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 144 {
 145         WARN_ON(1);
 146         return NULL;
 147 }
 148
 149 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 150                                            struct sk_buff *skb,
 151                                            const void *daddr);
 152
 153 static struct dst_ops ipv4_dst_ops = {
 154         .family =               AF_INET,
 155         .check =                ipv4_dst_check,
 156         .default_advmss =       ipv4_default_advmss,
 157         .mtu =                  ipv4_mtu,
 158         .cow_metrics =          ipv4_cow_metrics,
 159         .destroy =              ipv4_dst_destroy,
 160         .negative_advice =      ipv4_negative_advice,
 161         .link_failure =         ipv4_link_failure,
 162         .update_pmtu =          ip_rt_update_pmtu,
 163         .redirect =             ip_do_redirect,
 164         .local_out =            __ip_local_out,
 165         .neigh_lookup =         ipv4_neigh_lookup,
 166 };
 167
 168 #define ECN_OR_COST(class)      TC_PRIO_##class
 169
 170 const __u8 ip_tos2prio[16] = {
 171         TC_PRIO_BESTEFFORT,
 172         ECN_OR_COST(BESTEFFORT),
 173         TC_PRIO_BESTEFFORT,
 174         ECN_OR_COST(BESTEFFORT),
 175         TC_PRIO_BULK,
 176         ECN_OR_COST(BULK),
 177         TC_PRIO_BULK,
 178         ECN_OR_COST(BULK),
 179         TC_PRIO_INTERACTIVE,
 180         ECN_OR_COST(INTERACTIVE),
 181         TC_PRIO_INTERACTIVE,
 182         ECN_OR_COST(INTERACTIVE),
 183         TC_PRIO_INTERACTIVE_BULK,
 184         ECN_OR_COST(INTERACTIVE_BULK),
 185         TC_PRIO_INTERACTIVE_BULK,
 186         ECN_OR_COST(INTERACTIVE_BULK)
 187 };
 188 EXPORT_SYMBOL(ip_tos2prio);
 189
 190 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 191 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
 192
 193 #ifdef CONFIG_PROC_FS
 194 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 195 {
 196         if (*pos)
 197                 return NULL;
 198         return SEQ_START_TOKEN;
 199 }
 200
 201 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 202 {
 203         ++*pos;
 204         return NULL;
 205 }
 206
 207 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 208 {
 209 }
 210
 211 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 212 {
 213         if (v == SEQ_START_TOKEN)
 214                 seq_printf(seq, "%-127s\n",
 215                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 216                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 217                            "HHUptod\tSpecDst");
 218         return 0;
 219 }
 220
 221 static const struct seq_operations rt_cache_seq_ops = {
 222         .start  = rt_cache_seq_start,
 223         .next   = rt_cache_seq_next,
 224         .stop   = rt_cache_seq_stop,
 225         .show   = rt_cache_seq_show,
 226 };
 227
 228 static int rt_cache_seq_open(struct inode *inode, struct file *file)
 229 {
 230         return seq_open(file, &rt_cache_seq_ops);
 231 }
 232
 233 static const struct file_operations rt_cache_seq_fops = {
 234         .owner   = THIS_MODULE,
 235         .open    = rt_cache_seq_open,
 236         .read    = seq_read,
 237         .llseek  = seq_lseek,
 238         .release = seq_release,
 239 };
 240
 241
 242 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 243 {
 244         int cpu;
 245
 246         if (*pos == 0)
 247                 return SEQ_START_TOKEN;
 248
 249         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 250                 if (!cpu_possible(cpu))
 251                         continue;
 252                 *pos = cpu+1;
 253                 return &per_cpu(rt_cache_stat, cpu);
 254         }
 255         return NULL;
 256 }
 257
 258 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 259 {
 260         int cpu;
 261
 262         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 263                 if (!cpu_possible(cpu))
 264                         continue;
 265                 *pos = cpu+1;
 266                 return &per_cpu(rt_cache_stat, cpu);
 267         }
 268         return NULL;
 269
 270 }
 271
 272 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 273 {
 274
 275 }
 276
 277 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 278 {
 279         struct rt_cache_stat *st = v;
 280
 281         if (v == SEQ_START_TOKEN) {
 282                 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");
 283                 return 0;
 284         }
 285
 286         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 287                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 288                    dst_entries_get_slow(&ipv4_dst_ops),
 289                    0, /* st->in_hit */
 290                    st->in_slow_tot,
 291                    st->in_slow_mc,
 292                    st->in_no_route,
 293                    st->in_brd,
 294                    st->in_martian_dst,
 295                    st->in_martian_src,
 296
 297                    0, /* st->out_hit */
 298                    st->out_slow_tot,
 299                    st->out_slow_mc,
 300
 301                    0, /* st->gc_total */
 302                    0, /* st->gc_ignored */
 303                    0, /* st->gc_goal_miss */
 304                    0, /* st->gc_dst_overflow */
 305                    0, /* st->in_hlist_search */
 306                    0  /* st->out_hlist_search */
 307                 );
 308         return 0;
 309 }
 310
 311 static const struct seq_operations rt_cpu_seq_ops = {
 312         .start  = rt_cpu_seq_start,
 313         .next   = rt_cpu_seq_next,
 314         .stop   = rt_cpu_seq_stop,
 315         .show   = rt_cpu_seq_show,
 316 };
 317
 318
 319 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
 320 {
 321         return seq_open(file, &rt_cpu_seq_ops);
 322 }
 323
 324 static const struct file_operations rt_cpu_seq_fops = {
 325         .owner   = THIS_MODULE,
 326         .open    = rt_cpu_seq_open,
 327         .read    = seq_read,
 328         .llseek  = seq_lseek,
 329         .release = seq_release,
 330 };
 331
 332 #ifdef CONFIG_IP_ROUTE_CLASSID
 333 static int rt_acct_proc_show(struct seq_file *m, void *v)
 334 {
 335         struct ip_rt_acct *dst, *src;
 336         unsigned int i, j;
 337
 338         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 339         if (!dst)
 340                 return -ENOMEM;
 341
 342         for_each_possible_cpu(i) {
 343                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 344                 for (j = 0; j < 256; j++) {
 345                         dst[j].o_bytes   += src[j].o_bytes;
 346                         dst[j].o_packets += src[j].o_packets;
 347                         dst[j].i_bytes   += src[j].i_bytes;
 348                         dst[j].i_packets += src[j].i_packets;
 349                 }
 350         }
 351
 352         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 353         kfree(dst);
 354         return 0;
 355 }
 356
 357 static int rt_acct_proc_open(struct inode *inode, struct file *file)
 358 {
 359         return single_open(file, rt_acct_proc_show, NULL);
 360 }
 361
 362 static const struct file_operations rt_acct_proc_fops = {
 363         .owner          = THIS_MODULE,
 364         .open           = rt_acct_proc_open,
 365         .read           = seq_read,
 366         .llseek         = seq_lseek,
 367         .release        = single_release,
 368 };
 369 #endif
 370
 371 static int __net_init ip_rt_do_proc_init(struct net *net)
 372 {
 373         struct proc_dir_entry *pde;
 374
 375         pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
 376                           &rt_cache_seq_fops);
 377         if (!pde)
 378                 goto err1;
 379
 380         pde = proc_create("rt_cache", S_IRUGO,
 381                           net->proc_net_stat, &rt_cpu_seq_fops);
 382         if (!pde)
 383                 goto err2;
 384
 385 #ifdef CONFIG_IP_ROUTE_CLASSID
 386         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
 387         if (!pde)
 388                 goto err3;
 389 #endif
 390         return 0;
 391
 392 #ifdef CONFIG_IP_ROUTE_CLASSID
 393 err3:
 394         remove_proc_entry("rt_cache", net->proc_net_stat);
 395 #endif
 396 err2:
 397         remove_proc_entry("rt_cache", net->proc_net);
 398 err1:
 399         return -ENOMEM;
 400 }
 401
 402 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 403 {
 404         remove_proc_entry("rt_cache", net->proc_net_stat);
 405         remove_proc_entry("rt_cache", net->proc_net);
 406 #ifdef CONFIG_IP_ROUTE_CLASSID
 407         remove_proc_entry("rt_acct", net->proc_net);
 408 #endif
 409 }
 410
 411 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 412         .init = ip_rt_do_proc_init,
 413         .exit = ip_rt_do_proc_exit,
 414 };
 415
 416 static int __init ip_rt_proc_init(void)
 417 {
 418         return register_pernet_subsys(&ip_rt_proc_ops);
 419 }
 420
 421 #else
 422 static inline int ip_rt_proc_init(void)
 423 {
 424         return 0;
 425 }
 426 #endif /* CONFIG_PROC_FS */
 427
 428 static inline bool rt_is_expired(const struct rtable *rth)
 429 {
 430         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 431 }
 432
 433 void rt_cache_flush(struct net *net)
 434 {
 435         rt_genid_bump_ipv4(net);
 436 }
 437
 438 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 439                                            struct sk_buff *skb,
 440                                            const void *daddr)
 441 {
 442         struct net_device *dev = dst->dev;
 443         const __be32 *pkey = daddr;
 444         const struct rtable *rt;
 445         struct neighbour *n;
 446
 447         rt = (const struct rtable *) dst;
 448         if (rt->rt_gateway)
 449                 pkey = (const __be32 *) &rt->rt_gateway;
 450         else if (skb)
 451                 pkey = &ip_hdr(skb)->daddr;
 452
 453         n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
 454         if (n)
 455                 return n;
 456         return neigh_create(&arp_tbl, pkey, dev);
 457 }
 458
 459 #define IP_IDENTS_SZ 2048u
 460 struct ip_ident_bucket {
 461         atomic_t        id;
 462         u32             stamp32;
 463 };
 464
 465 static struct ip_ident_bucket *ip_idents __read_mostly;
 466
 467 /* In order to protect privacy, we add a perturbation to identifiers
 468  * if one generator is seldom used. This makes hard for an attacker
 469  * to infer how many packets were sent between two points in time.
 470  */
 471 u32 ip_idents_reserve(u32 hash, int segs)
 472 {
 473         struct ip_ident_bucket *bucket = ip_idents + hash % IP_IDENTS_SZ;
 474         u32 old = ACCESS_ONCE(bucket->stamp32);
 475         u32 now = (u32)jiffies;
 476         u32 delta = 0;
 477
 478         if (old != now && cmpxchg(&bucket->stamp32, old, now) == old)
 479                 delta = prandom_u32_max(now - old);
 480
 481         return atomic_add_return(segs + delta, &bucket->id) - segs;
 482 }
 483 EXPORT_SYMBOL(ip_idents_reserve);
 484
 485 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
 486 {
 487         static u32 ip_idents_hashrnd __read_mostly;
 488         u32 hash, id;
 489
 490         net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
 491
 492         hash = jhash_3words((__force u32)iph->daddr,
 493                             (__force u32)iph->saddr,
 494                             iph->protocol ^ net_hash_mix(net),
 495                             ip_idents_hashrnd);
 496         id = ip_idents_reserve(hash, segs);
 497         iph->id = htons(id);
 498 }
 499 EXPORT_SYMBOL(__ip_select_ident);
 500
 501 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 502                              const struct iphdr *iph,
 503                              int oif, u8 tos,
 504                              u8 prot, u32 mark, int flow_flags)
 505 {
 506         if (sk) {
 507                 const struct inet_sock *inet = inet_sk(sk);
 508
 509                 oif = sk->sk_bound_dev_if;
 510                 mark = sk->sk_mark;
 511                 tos = RT_CONN_FLAGS(sk);
 512                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 513         }
 514         flowi4_init_output(fl4, oif, mark, tos,
 515                            RT_SCOPE_UNIVERSE, prot,
 516                            flow_flags,
 517                            iph->daddr, iph->saddr, 0, 0);
 518 }
 519
 520 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 521                                const struct sock *sk)
 522 {
 523         const struct iphdr *iph = ip_hdr(skb);
 524         int oif = skb->dev->ifindex;
 525         u8 tos = RT_TOS(iph->tos);
 526         u8 prot = iph->protocol;
 527         u32 mark = skb->mark;
 528
 529         __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
 530 }
 531
 532 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 533 {
 534         const struct inet_sock *inet = inet_sk(sk);
 535         const struct ip_options_rcu *inet_opt;
 536         __be32 daddr = inet->inet_daddr;
 537
 538         rcu_read_lock();
 539         inet_opt = rcu_dereference(inet->inet_opt);
 540         if (inet_opt && inet_opt->opt.srr)
 541                 daddr = inet_opt->opt.faddr;
 542         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 543                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 544                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 545                            inet_sk_flowi_flags(sk),
 546                            daddr, inet->inet_saddr, 0, 0);
 547         rcu_read_unlock();
 548 }
 549
 550 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 551                                  const struct sk_buff *skb)
 552 {
 553         if (skb)
 554                 build_skb_flow_key(fl4, skb, sk);
 555         else
 556                 build_sk_flow_key(fl4, sk);
 557 }
 558
 559 static inline void rt_free(struct rtable *rt)
 560 {
 561         call_rcu(&rt->dst.rcu_head, dst_rcu_free);
 562 }
 563
 564 static DEFINE_SPINLOCK(fnhe_lock);
 565
 566 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 567 {
 568         struct rtable *rt;
 569
 570         rt = rcu_dereference(fnhe->fnhe_rth_input);
 571         if (rt) {
 572                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 573                 rt_free(rt);
 574         }
 575         rt = rcu_dereference(fnhe->fnhe_rth_output);
 576         if (rt) {
 577                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 578                 rt_free(rt);
 579         }
 580 }
 581
 582 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
 583 {
 584         struct fib_nh_exception *fnhe, *oldest;
 585
 586         oldest = rcu_dereference(hash->chain);
 587         for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
 588              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 589                 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
 590                         oldest = fnhe;
 591         }
 592         fnhe_flush_routes(oldest);
 593         return oldest;
 594 }
 595
 596 static inline u32 fnhe_hashfun(__be32 daddr)
 597 {
 598         static u32 fnhe_hashrnd __read_mostly;
 599         u32 hval;
 600
 601         net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
 602         hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
 603         return hash_32(hval, FNHE_HASH_SHIFT);
 604 }
 605
 606 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 607 {
 608         rt->rt_pmtu = fnhe->fnhe_pmtu;
 609         rt->dst.expires = fnhe->fnhe_expires;
 610
 611         if (fnhe->fnhe_gw) {
 612                 rt->rt_flags |= RTCF_REDIRECTED;
 613                 rt->rt_gateway = fnhe->fnhe_gw;
 614                 rt->rt_uses_gateway = 1;
 615         }
 616 }
 617
 618 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
 619                                   u32 pmtu, unsigned long expires)
 620 {
 621         struct fnhe_hash_bucket *hash;
 622         struct fib_nh_exception *fnhe;
 623         struct rtable *rt;
 624         unsigned int i;
 625         int depth;
 626         u32 hval = fnhe_hashfun(daddr);
 627
 628         spin_lock_bh(&fnhe_lock);
 629
 630         hash = rcu_dereference(nh->nh_exceptions);
 631         if (!hash) {
 632                 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
 633                 if (!hash)
 634                         goto out_unlock;
 635                 rcu_assign_pointer(nh->nh_exceptions, hash);
 636         }
 637
 638         hash += hval;
 639
 640         depth = 0;
 641         for (fnhe = rcu_dereference(hash->chain); fnhe;
 642              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 643                 if (fnhe->fnhe_daddr == daddr)
 644                         break;
 645                 depth++;
 646         }
 647
 648         if (fnhe) {
 649                 if (gw)
 650                         fnhe->fnhe_gw = gw;
 651                 if (pmtu) {
 652                         fnhe->fnhe_pmtu = pmtu;
 653                         fnhe->fnhe_expires = max(1UL, expires);
 654                 }
 655                 /* Update all cached dsts too */
 656                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 657                 if (rt)
 658                         fill_route_from_fnhe(rt, fnhe);
 659                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 660                 if (rt)
 661                         fill_route_from_fnhe(rt, fnhe);
 662         } else {
 663                 if (depth > FNHE_RECLAIM_DEPTH)
 664                         fnhe = fnhe_oldest(hash);
 665                 else {
 666                         fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 667                         if (!fnhe)
 668                                 goto out_unlock;
 669
 670                         fnhe->fnhe_next = hash->chain;
 671                         rcu_assign_pointer(hash->chain, fnhe);
 672                 }
 673                 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
 674                 fnhe->fnhe_daddr = daddr;
 675                 fnhe->fnhe_gw = gw;
 676                 fnhe->fnhe_pmtu = pmtu;
 677                 fnhe->fnhe_expires = expires;
 678
 679                 /* Exception created; mark the cached routes for the nexthop
 680                  * stale, so anyone caching it rechecks if this exception
 681                  * applies to them.
 682                  */
 683                 rt = rcu_dereference(nh->nh_rth_input);
 684                 if (rt)
 685                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 686
 687                 for_each_possible_cpu(i) {
 688                         struct rtable __rcu **prt;
 689                         prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
 690                         rt = rcu_dereference(*prt);
 691                         if (rt)
 692                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 693                 }
 694         }
 695
 696         fnhe->fnhe_stamp = jiffies;
 697
 698 out_unlock:
 699         spin_unlock_bh(&fnhe_lock);
 700 }
 701
 702 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 703                              bool kill_route)
 704 {
 705         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 706         __be32 old_gw = ip_hdr(skb)->saddr;
 707         struct net_device *dev = skb->dev;
 708         struct in_device *in_dev;
 709         struct fib_result res;
 710         struct neighbour *n;
 711         struct net *net;
 712
 713         switch (icmp_hdr(skb)->code & 7) {
 714         case ICMP_REDIR_NET:
 715         case ICMP_REDIR_NETTOS:
 716         case ICMP_REDIR_HOST:
 717         case ICMP_REDIR_HOSTTOS:
 718                 break;
 719
 720         default:
 721                 return;
 722         }
 723
 724         if (rt->rt_gateway != old_gw)
 725                 return;
 726
 727         in_dev = __in_dev_get_rcu(dev);
 728         if (!in_dev)
 729                 return;
 730
 731         net = dev_net(dev);
 732         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 733             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 734             ipv4_is_zeronet(new_gw))
 735                 goto reject_redirect;
 736
 737         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 738                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 739                         goto reject_redirect;
 740                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 741                         goto reject_redirect;
 742         } else {
 743                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 744                         goto reject_redirect;
 745         }
 746
 747         n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
 748         if (!IS_ERR(n)) {
 749                 if (!(n->nud_state & NUD_VALID)) {
 750                         neigh_event_send(n, NULL);
 751                 } else {
 752                         if (fib_lookup(net, fl4, &res) == 0) {
 753                                 struct fib_nh *nh = &FIB_RES_NH(res);
 754
 755                                 update_or_create_fnhe(nh, fl4->daddr, new_gw,
 756                                                       0, 0);
 757                         }
 758                         if (kill_route)
 759                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 760                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 761                 }
 762                 neigh_release(n);
 763         }
 764         return;
 765
 766 reject_redirect:
 767 #ifdef CONFIG_IP_ROUTE_VERBOSE
 768         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 769                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 770                 __be32 daddr = iph->daddr;
 771                 __be32 saddr = iph->saddr;
 772
 773                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 774                                      "  Advised path = %pI4 -> %pI4\n",
 775                                      &old_gw, dev->name, &new_gw,
 776                                      &saddr, &daddr);
 777         }
 778 #endif
 779         ;
 780 }
 781
 782 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 783 {
 784         struct rtable *rt;
 785         struct flowi4 fl4;
 786         const struct iphdr *iph = (const struct iphdr *) skb->data;
 787         int oif = skb->dev->ifindex;
 788         u8 tos = RT_TOS(iph->tos);
 789         u8 prot = iph->protocol;
 790         u32 mark = skb->mark;
 791
 792         rt = (struct rtable *) dst;
 793
 794         __build_flow_key(&fl4, sk, iph, oif, tos, prot, mark, 0);
 795         __ip_do_redirect(rt, skb, &fl4, true);
 796 }
 797
 798 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 799 {
 800         struct rtable *rt = (struct rtable *)dst;
 801         struct dst_entry *ret = dst;
 802
 803         if (rt) {
 804                 if (dst->obsolete > 0) {
 805                         ip_rt_put(rt);
 806                         ret = NULL;
 807                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 808                            rt->dst.expires) {
 809                         ip_rt_put(rt);
 810                         ret = NULL;
 811                 }
 812         }
 813         return ret;
 814 }
 815
 816 /*
 817  * Algorithm:
 818  *      1. The first ip_rt_redirect_number redirects are sent
 819  *         with exponential backoff, then we stop sending them at all,
 820  *         assuming that the host ignores our redirects.
 821  *      2. If we did not see packets requiring redirects
 822  *         during ip_rt_redirect_silence, we assume that the host
 823  *         forgot redirected route and start to send redirects again.
 824  *
 825  * This algorithm is much cheaper and more intelligent than dumb load limiting
 826  * in icmp.c.
 827  *
 828  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 829  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 830  */
 831
 832 void ip_rt_send_redirect(struct sk_buff *skb)
 833 {
 834         struct rtable *rt = skb_rtable(skb);
 835         struct in_device *in_dev;
 836         struct inet_peer *peer;
 837         struct net *net;
 838         int log_martians;
 839
 840         rcu_read_lock();
 841         in_dev = __in_dev_get_rcu(rt->dst.dev);
 842         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 843                 rcu_read_unlock();
 844                 return;
 845         }
 846         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 847         rcu_read_unlock();
 848
 849         net = dev_net(rt->dst.dev);
 850         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
 851         if (!peer) {
 852                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 853                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 854                 return;
 855         }
 856
 857         /* No redirected packets during ip_rt_redirect_silence;
 858          * reset the algorithm.
 859          */
 860         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
 861                 peer->rate_tokens = 0;
 862
 863         /* Too many ignored redirects; do not send anything
 864          * set dst.rate_last to the last seen redirected packet.
 865          */
 866         if (peer->rate_tokens >= ip_rt_redirect_number) {
 867                 peer->rate_last = jiffies;
 868                 goto out_put_peer;
 869         }
 870
 871         /* Check for load limit; set rate_last to the latest sent
 872          * redirect.
 873          */
 874         if (peer->rate_tokens == 0 ||
 875             time_after(jiffies,
 876                        (peer->rate_last +
 877                         (ip_rt_redirect_load << peer->rate_tokens)))) {
 878                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 879
 880                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 881                 peer->rate_last = jiffies;
 882                 ++peer->rate_tokens;
 883 #ifdef CONFIG_IP_ROUTE_VERBOSE
 884                 if (log_martians &&
 885                     peer->rate_tokens == ip_rt_redirect_number)
 886                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 887                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 888                                              &ip_hdr(skb)->daddr, &gw);
 889 #endif
 890         }
 891 out_put_peer:
 892         inet_putpeer(peer);
 893 }
 894
 895 static int ip_error(struct sk_buff *skb)
 896 {
 897         struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
 898         struct rtable *rt = skb_rtable(skb);
 899         struct inet_peer *peer;
 900         unsigned long now;
 901         struct net *net;
 902         bool send;
 903         int code;
 904
 905         /* IP on this device is disabled. */
 906         if (!in_dev)
 907                 goto out;
 908
 909         net = dev_net(rt->dst.dev);
 910         if (!IN_DEV_FORWARD(in_dev)) {
 911                 switch (rt->dst.error) {
 912                 case EHOSTUNREACH:
 913                         IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
 914                         break;
 915
 916                 case ENETUNREACH:
 917                         IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
 918                         break;
 919                 }
 920                 goto out;
 921         }
 922
 923         switch (rt->dst.error) {
 924         case EINVAL:
 925         default:
 926                 goto out;
 927         case EHOSTUNREACH:
 928                 code = ICMP_HOST_UNREACH;
 929                 break;
 930         case ENETUNREACH:
 931                 code = ICMP_NET_UNREACH;
 932                 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
 933                 break;
 934         case EACCES:
 935                 code = ICMP_PKT_FILTERED;
 936                 break;
 937         }
 938
 939         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
 940
 941         send = true;
 942         if (peer) {
 943                 now = jiffies;
 944                 peer->rate_tokens += now - peer->rate_last;
 945                 if (peer->rate_tokens > ip_rt_error_burst)
 946                         peer->rate_tokens = ip_rt_error_burst;
 947                 peer->rate_last = now;
 948                 if (peer->rate_tokens >= ip_rt_error_cost)
 949                         peer->rate_tokens -= ip_rt_error_cost;
 950                 else
 951                         send = false;
 952                 inet_putpeer(peer);
 953         }
 954         if (send)
 955                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
 956
 957 out:    kfree_skb(skb);
 958         return 0;
 959 }
 960
 961 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
 962 {
 963         struct dst_entry *dst = &rt->dst;
 964         struct fib_result res;
 965
 966         if (dst_metric_locked(dst, RTAX_MTU))
 967                 return;
 968
 969         if (ipv4_mtu(dst) < mtu)
 970                 return;
 971
 972         if (mtu < ip_rt_min_pmtu)
 973                 mtu = ip_rt_min_pmtu;
 974
 975         if (rt->rt_pmtu == mtu &&
 976             time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
 977                 return;
 978
 979         rcu_read_lock();
 980         if (fib_lookup(dev_net(dst->dev), fl4, &res) == 0) {
 981                 struct fib_nh *nh = &FIB_RES_NH(res);
 982
 983                 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
 984                                       jiffies + ip_rt_mtu_expires);
 985         }
 986         rcu_read_unlock();
 987 }
 988
 989 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 990                               struct sk_buff *skb, u32 mtu)
 991 {
 992         struct rtable *rt = (struct rtable *) dst;
 993         struct flowi4 fl4;
 994
 995         ip_rt_build_flow_key(&fl4, sk, skb);
 996         __ip_rt_update_pmtu(rt, &fl4, mtu);
 997 }
 998
 999 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1000                       int oif, u32 mark, u8 protocol, int flow_flags)
1001 {
1002         const struct iphdr *iph = (const struct iphdr *) skb->data;
1003         struct flowi4 fl4;
1004         struct rtable *rt;
1005
1006         if (!mark)
1007                 mark = IP4_REPLY_MARK(net, skb->mark);
1008
1009         __build_flow_key(&fl4, NULL, iph, oif,
1010                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1011         rt = __ip_route_output_key(net, &fl4);
1012         if (!IS_ERR(rt)) {
1013                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1014                 ip_rt_put(rt);
1015         }
1016 }
1017 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1018
1019 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1020 {
1021         const struct iphdr *iph = (const struct iphdr *) skb->data;
1022         struct flowi4 fl4;
1023         struct rtable *rt;
1024
1025         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1026
1027         if (!fl4.flowi4_mark)
1028                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1029
1030         rt = __ip_route_output_key(sock_net(sk), &fl4);
1031         if (!IS_ERR(rt)) {
1032                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1033                 ip_rt_put(rt);
1034         }
1035 }
1036
1037 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1038 {
1039         const struct iphdr *iph = (const struct iphdr *) skb->data;
1040         struct flowi4 fl4;
1041         struct rtable *rt;
1042         struct dst_entry *odst = NULL;
1043         bool new = false;
1044
1045         bh_lock_sock(sk);
1046
1047         if (!ip_sk_accept_pmtu(sk))
1048                 goto out;
1049
1050         odst = sk_dst_get(sk);
1051
1052         if (sock_owned_by_user(sk) || !odst) {
1053                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1054                 goto out;
1055         }
1056
1057         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1058
1059         rt = (struct rtable *)odst;
1060         if (odst->obsolete && !odst->ops->check(odst, 0)) {
1061                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1062                 if (IS_ERR(rt))
1063                         goto out;
1064
1065                 new = true;
1066         }
1067
1068         __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1069
1070         if (!dst_check(&rt->dst, 0)) {
1071                 if (new)
1072                         dst_release(&rt->dst);
1073
1074                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1075                 if (IS_ERR(rt))
1076                         goto out;
1077
1078                 new = true;
1079         }
1080
1081         if (new)
1082                 sk_dst_set(sk, &rt->dst);
1083
1084 out:
1085         bh_unlock_sock(sk);
1086         dst_release(odst);
1087 }
1088 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1089
1090 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1091                    int oif, u32 mark, u8 protocol, int flow_flags)
1092 {
1093         const struct iphdr *iph = (const struct iphdr *) skb->data;
1094         struct flowi4 fl4;
1095         struct rtable *rt;
1096
1097         __build_flow_key(&fl4, NULL, iph, oif,
1098                          RT_TOS(iph->tos), protocol, mark, flow_flags);
1099         rt = __ip_route_output_key(net, &fl4);
1100         if (!IS_ERR(rt)) {
1101                 __ip_do_redirect(rt, skb, &fl4, false);
1102                 ip_rt_put(rt);
1103         }
1104 }
1105 EXPORT_SYMBOL_GPL(ipv4_redirect);
1106
1107 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1108 {
1109         const struct iphdr *iph = (const struct iphdr *) skb->data;
1110         struct flowi4 fl4;
1111         struct rtable *rt;
1112
1113         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1114         rt = __ip_route_output_key(sock_net(sk), &fl4);
1115         if (!IS_ERR(rt)) {
1116                 __ip_do_redirect(rt, skb, &fl4, false);
1117                 ip_rt_put(rt);
1118         }
1119 }
1120 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1121
1122 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1123 {
1124         struct rtable *rt = (struct rtable *) dst;
1125
1126         /* All IPV4 dsts are created with ->obsolete set to the value
1127          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1128          * into this function always.
1129          *
1130          * When a PMTU/redirect information update invalidates a route,
1131          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1132          * DST_OBSOLETE_DEAD by dst_free().
1133          */
1134         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1135                 return NULL;
1136         return dst;
1137 }
1138
1139 static void ipv4_link_failure(struct sk_buff *skb)
1140 {
1141         struct rtable *rt;
1142
1143         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1144
1145         rt = skb_rtable(skb);
1146         if (rt)
1147                 dst_set_expires(&rt->dst, 0);
1148 }
1149
1150 static int ip_rt_bug(struct sock *sk, struct sk_buff *skb)
1151 {
1152         pr_debug("%s: %pI4 -> %pI4, %s\n",
1153                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1154                  skb->dev ? skb->dev->name : "?");
1155         kfree_skb(skb);
1156         WARN_ON(1);
1157         return 0;
1158 }
1159
1160 /*
1161    We do not cache source address of outgoing interface,
1162    because it is used only by IP RR, TS and SRR options,
1163    so that it out of fast path.
1164
1165    BTW remember: "addr" is allowed to be not aligned
1166    in IP options!
1167  */
1168
1169 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1170 {
1171         __be32 src;
1172
1173         if (rt_is_output_route(rt))
1174                 src = ip_hdr(skb)->saddr;
1175         else {
1176                 struct fib_result res;
1177                 struct flowi4 fl4;
1178                 struct iphdr *iph;
1179
1180                 iph = ip_hdr(skb);
1181
1182                 memset(&fl4, 0, sizeof(fl4));
1183                 fl4.daddr = iph->daddr;
1184                 fl4.saddr = iph->saddr;
1185                 fl4.flowi4_tos = RT_TOS(iph->tos);
1186                 fl4.flowi4_oif = rt->dst.dev->ifindex;
1187                 fl4.flowi4_iif = skb->dev->ifindex;
1188                 fl4.flowi4_mark = skb->mark;
1189
1190                 rcu_read_lock();
1191                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1192                         src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1193                 else
1194                         src = inet_select_addr(rt->dst.dev,
1195                                                rt_nexthop(rt, iph->daddr),
1196                                                RT_SCOPE_UNIVERSE);
1197                 rcu_read_unlock();
1198         }
1199         memcpy(addr, &src, 4);
1200 }
1201
1202 #ifdef CONFIG_IP_ROUTE_CLASSID
1203 static void set_class_tag(struct rtable *rt, u32 tag)
1204 {
1205         if (!(rt->dst.tclassid & 0xFFFF))
1206                 rt->dst.tclassid |= tag & 0xFFFF;
1207         if (!(rt->dst.tclassid & 0xFFFF0000))
1208                 rt->dst.tclassid |= tag & 0xFFFF0000;
1209 }
1210 #endif
1211
1212 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1213 {
1214         unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1215
1216         if (advmss == 0) {
1217                 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1218                                ip_rt_min_advmss);
1219                 if (advmss > 65535 - 40)
1220                         advmss = 65535 - 40;
1221         }
1222         return advmss;
1223 }
1224
1225 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1226 {
1227         const struct rtable *rt = (const struct rtable *) dst;
1228         unsigned int mtu = rt->rt_pmtu;
1229
1230         if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1231                 mtu = dst_metric_raw(dst, RTAX_MTU);
1232
1233         if (mtu)
1234                 return mtu;
1235
1236         mtu = dst->dev->mtu;
1237
1238         if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1239                 if (rt->rt_uses_gateway && mtu > 576)
1240                         mtu = 576;
1241         }
1242
1243         return min_t(unsigned int, mtu, IP_MAX_MTU);
1244 }
1245
1246 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1247 {
1248         struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1249         struct fib_nh_exception *fnhe;
1250         u32 hval;
1251
1252         if (!hash)
1253                 return NULL;
1254
1255         hval = fnhe_hashfun(daddr);
1256
1257         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1258              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1259                 if (fnhe->fnhe_daddr == daddr)
1260                         return fnhe;
1261         }
1262         return NULL;
1263 }
1264
1265 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1266                               __be32 daddr)
1267 {
1268         bool ret = false;
1269
1270         spin_lock_bh(&fnhe_lock);
1271
1272         if (daddr == fnhe->fnhe_daddr) {
1273                 struct rtable __rcu **porig;
1274                 struct rtable *orig;
1275                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1276
1277                 if (rt_is_input_route(rt))
1278                         porig = &fnhe->fnhe_rth_input;
1279                 else
1280                         porig = &fnhe->fnhe_rth_output;
1281                 orig = rcu_dereference(*porig);
1282
1283                 if (fnhe->fnhe_genid != genid) {
1284                         fnhe->fnhe_genid = genid;
1285                         fnhe->fnhe_gw = 0;
1286                         fnhe->fnhe_pmtu = 0;
1287                         fnhe->fnhe_expires = 0;
1288                         fnhe_flush_routes(fnhe);
1289                         orig = NULL;
1290                 }
1291                 fill_route_from_fnhe(rt, fnhe);
1292                 if (!rt->rt_gateway)
1293                         rt->rt_gateway = daddr;
1294
1295                 if (!(rt->dst.flags & DST_NOCACHE)) {
1296                         rcu_assign_pointer(*porig, rt);
1297                         if (orig)
1298                                 rt_free(orig);
1299                         ret = true;
1300                 }
1301
1302                 fnhe->fnhe_stamp = jiffies;
1303         }
1304         spin_unlock_bh(&fnhe_lock);
1305
1306         return ret;
1307 }
1308
1309 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1310 {
1311         struct rtable *orig, *prev, **p;
1312         bool ret = true;
1313
1314         if (rt_is_input_route(rt)) {
1315                 p = (struct rtable **)&nh->nh_rth_input;
1316         } else {
1317                 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1318         }
1319         orig = *p;
1320
1321         prev = cmpxchg(p, orig, rt);
1322         if (prev == orig) {
1323                 if (orig)
1324                         rt_free(orig);
1325         } else
1326                 ret = false;
1327
1328         return ret;
1329 }
1330
1331 struct uncached_list {
1332         spinlock_t              lock;
1333         struct list_head        head;
1334 };
1335
1336 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1337
1338 static void rt_add_uncached_list(struct rtable *rt)
1339 {
1340         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1341
1342         rt->rt_uncached_list = ul;
1343
1344         spin_lock_bh(&ul->lock);
1345         list_add_tail(&rt->rt_uncached, &ul->head);
1346         spin_unlock_bh(&ul->lock);
1347 }
1348
1349 static void ipv4_dst_destroy(struct dst_entry *dst)
1350 {
1351         struct rtable *rt = (struct rtable *) dst;
1352
1353         if (!list_empty(&rt->rt_uncached)) {
1354                 struct uncached_list *ul = rt->rt_uncached_list;
1355
1356                 spin_lock_bh(&ul->lock);
1357                 list_del(&rt->rt_uncached);
1358                 spin_unlock_bh(&ul->lock);
1359         }
1360 }
1361
1362 void rt_flush_dev(struct net_device *dev)
1363 {
1364         struct net *net = dev_net(dev);
1365         struct rtable *rt;
1366         int cpu;
1367
1368         for_each_possible_cpu(cpu) {
1369                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1370
1371                 spin_lock_bh(&ul->lock);
1372                 list_for_each_entry(rt, &ul->head, rt_uncached) {
1373                         if (rt->dst.dev != dev)
1374                                 continue;
1375                         rt->dst.dev = net->loopback_dev;
1376                         dev_hold(rt->dst.dev);
1377                         dev_put(dev);
1378                 }
1379                 spin_unlock_bh(&ul->lock);
1380         }
1381 }
1382
1383 static bool rt_cache_valid(const struct rtable *rt)
1384 {
1385         return  rt &&
1386                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1387                 !rt_is_expired(rt);
1388 }
1389
1390 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1391                            const struct fib_result *res,
1392                            struct fib_nh_exception *fnhe,
1393                            struct fib_info *fi, u16 type, u32 itag)
1394 {
1395         bool cached = false;
1396
1397         if (fi) {
1398                 struct fib_nh *nh = &FIB_RES_NH(*res);
1399
1400                 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1401                         rt->rt_gateway = nh->nh_gw;
1402                         rt->rt_uses_gateway = 1;
1403                 }
1404                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1405 #ifdef CONFIG_IP_ROUTE_CLASSID
1406                 rt->dst.tclassid = nh->nh_tclassid;
1407 #endif
1408                 if (unlikely(fnhe))
1409                         cached = rt_bind_exception(rt, fnhe, daddr);
1410                 else if (!(rt->dst.flags & DST_NOCACHE))
1411                         cached = rt_cache_route(nh, rt);
1412                 if (unlikely(!cached)) {
1413                         /* Routes we intend to cache in nexthop exception or
1414                          * FIB nexthop have the DST_NOCACHE bit clear.
1415                          * However, if we are unsuccessful at storing this
1416                          * route into the cache we really need to set it.
1417                          */
1418                         rt->dst.flags |= DST_NOCACHE;
1419                         if (!rt->rt_gateway)
1420                                 rt->rt_gateway = daddr;
1421                         rt_add_uncached_list(rt);
1422                 }
1423         } else
1424                 rt_add_uncached_list(rt);
1425
1426 #ifdef CONFIG_IP_ROUTE_CLASSID
1427 #ifdef CONFIG_IP_MULTIPLE_TABLES
1428         set_class_tag(rt, res->tclassid);
1429 #endif
1430         set_class_tag(rt, itag);
1431 #endif
1432 }
1433
1434 static struct rtable *rt_dst_alloc(struct net_device *dev,
1435                                    bool nopolicy, bool noxfrm, bool will_cache)
1436 {
1437         return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1438                          (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1439                          (nopolicy ? DST_NOPOLICY : 0) |
1440                          (noxfrm ? DST_NOXFRM : 0));
1441 }
1442
1443 /* called in rcu_read_lock() section */
1444 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1445                                 u8 tos, struct net_device *dev, int our)
1446 {
1447         struct rtable *rth;
1448         struct in_device *in_dev = __in_dev_get_rcu(dev);
1449         u32 itag = 0;
1450         int err;
1451
1452         /* Primary sanity checks. */
1453
1454         if (!in_dev)
1455                 return -EINVAL;
1456
1457         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1458             skb->protocol != htons(ETH_P_IP))
1459                 goto e_inval;
1460
1461         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1462                 if (ipv4_is_loopback(saddr))
1463                         goto e_inval;
1464
1465         if (ipv4_is_zeronet(saddr)) {
1466                 if (!ipv4_is_local_multicast(daddr))
1467                         goto e_inval;
1468         } else {
1469                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1470                                           in_dev, &itag);
1471                 if (err < 0)
1472                         goto e_err;
1473         }
1474         rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1475                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1476         if (!rth)
1477                 goto e_nobufs;
1478
1479 #ifdef CONFIG_IP_ROUTE_CLASSID
1480         rth->dst.tclassid = itag;
1481 #endif
1482         rth->dst.output = ip_rt_bug;
1483
1484         rth->rt_genid   = rt_genid_ipv4(dev_net(dev));
1485         rth->rt_flags   = RTCF_MULTICAST;
1486         rth->rt_type    = RTN_MULTICAST;
1487         rth->rt_is_input= 1;
1488         rth->rt_iif     = 0;
1489         rth->rt_pmtu    = 0;
1490         rth->rt_gateway = 0;
1491         rth->rt_uses_gateway = 0;
1492         INIT_LIST_HEAD(&rth->rt_uncached);
1493         if (our) {
1494                 rth->dst.input= ip_local_deliver;
1495                 rth->rt_flags |= RTCF_LOCAL;
1496         }
1497
1498 #ifdef CONFIG_IP_MROUTE
1499         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1500                 rth->dst.input = ip_mr_input;
1501 #endif
1502         RT_CACHE_STAT_INC(in_slow_mc);
1503
1504         skb_dst_set(skb, &rth->dst);
1505         return 0;
1506
1507 e_nobufs:
1508         return -ENOBUFS;
1509 e_inval:
1510         return -EINVAL;
1511 e_err:
1512         return err;
1513 }
1514
1515
1516 static void ip_handle_martian_source(struct net_device *dev,
1517                                      struct in_device *in_dev,
1518                                      struct sk_buff *skb,
1519                                      __be32 daddr,
1520                                      __be32 saddr)
1521 {
1522         RT_CACHE_STAT_INC(in_martian_src);
1523 #ifdef CONFIG_IP_ROUTE_VERBOSE
1524         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1525                 /*
1526                  *      RFC1812 recommendation, if source is martian,
1527                  *      the only hint is MAC header.
1528                  */
1529                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1530                         &daddr, &saddr, dev->name);
1531                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1532                         print_hex_dump(KERN_WARNING, "ll header: ",
1533                                        DUMP_PREFIX_OFFSET, 16, 1,
1534                                        skb_mac_header(skb),
1535                                        dev->hard_header_len, true);
1536                 }
1537         }
1538 #endif
1539 }
1540
1541 /* called in rcu_read_lock() section */
1542 static int __mkroute_input(struct sk_buff *skb,
1543                            const struct fib_result *res,
1544                            struct in_device *in_dev,
1545                            __be32 daddr, __be32 saddr, u32 tos)
1546 {
1547         struct fib_nh_exception *fnhe;
1548         struct rtable *rth;
1549         int err;
1550         struct in_device *out_dev;
1551         unsigned int flags = 0;
1552         bool do_cache;
1553         u32 itag = 0;
1554
1555         /* get a working reference to the output device */
1556         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1557         if (!out_dev) {
1558                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1559                 return -EINVAL;
1560         }
1561
1562         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1563                                   in_dev->dev, in_dev, &itag);
1564         if (err < 0) {
1565                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1566                                          saddr);
1567
1568                 goto cleanup;
1569         }
1570
1571         do_cache = res->fi && !itag;
1572         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1573             skb->protocol == htons(ETH_P_IP) &&
1574             (IN_DEV_SHARED_MEDIA(out_dev) ||
1575              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1576                 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1577
1578         if (skb->protocol != htons(ETH_P_IP)) {
1579                 /* Not IP (i.e. ARP). Do not create route, if it is
1580                  * invalid for proxy arp. DNAT routes are always valid.
1581                  *
1582                  * Proxy arp feature have been extended to allow, ARP
1583                  * replies back to the same interface, to support
1584                  * Private VLAN switch technologies. See arp.c.
1585                  */
1586                 if (out_dev == in_dev &&
1587                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1588                         err = -EINVAL;
1589                         goto cleanup;
1590                 }
1591         }
1592
1593         fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1594         if (do_cache) {
1595                 if (fnhe)
1596                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1597                 else
1598                         rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1599
1600                 if (rt_cache_valid(rth)) {
1601                         skb_dst_set_noref(skb, &rth->dst);
1602                         goto out;
1603                 }
1604         }
1605
1606         rth = rt_dst_alloc(out_dev->dev,
1607                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1608                            IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1609         if (!rth) {
1610                 err = -ENOBUFS;
1611                 goto cleanup;
1612         }
1613
1614         rth->rt_genid = rt_genid_ipv4(dev_net(rth->dst.dev));
1615         rth->rt_flags = flags;
1616         rth->rt_type = res->type;
1617         rth->rt_is_input = 1;
1618         rth->rt_iif     = 0;
1619         rth->rt_pmtu    = 0;
1620         rth->rt_gateway = 0;
1621         rth->rt_uses_gateway = 0;
1622         INIT_LIST_HEAD(&rth->rt_uncached);
1623         RT_CACHE_STAT_INC(in_slow_tot);
1624
1625         rth->dst.input = ip_forward;
1626         rth->dst.output = ip_output;
1627
1628         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1629         skb_dst_set(skb, &rth->dst);
1630 out:
1631         err = 0;
1632  cleanup:
1633         return err;
1634 }
1635
1636 static int ip_mkroute_input(struct sk_buff *skb,
1637                             struct fib_result *res,
1638                             const struct flowi4 *fl4,
1639                             struct in_device *in_dev,
1640                             __be32 daddr, __be32 saddr, u32 tos)
1641 {
1642 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1643         if (res->fi && res->fi->fib_nhs > 1)
1644                 fib_select_multipath(res);
1645 #endif
1646
1647         /* create a routing cache entry */
1648         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1649 }
1650
1651 /*
1652  *      NOTE. We drop all the packets that has local source
1653  *      addresses, because every properly looped back packet
1654  *      must have correct destination already attached by output routine.
1655  *
1656  *      Such approach solves two big problems:
1657  *      1. Not simplex devices are handled properly.
1658  *      2. IP spoofing attempts are filtered with 100% of guarantee.
1659  *      called with rcu_read_lock()
1660  */
1661
1662 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1663                                u8 tos, struct net_device *dev)
1664 {
1665         struct fib_result res;
1666         struct in_device *in_dev = __in_dev_get_rcu(dev);
1667         struct flowi4   fl4;
1668         unsigned int    flags = 0;
1669         u32             itag = 0;
1670         struct rtable   *rth;
1671         int             err = -EINVAL;
1672         struct net    *net = dev_net(dev);
1673         bool do_cache;
1674
1675         /* IP on this device is disabled. */
1676
1677         if (!in_dev)
1678                 goto out;
1679
1680         /* Check for the most weird martians, which can be not detected
1681            by fib_lookup.
1682          */
1683
1684         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1685                 goto martian_source;
1686
1687         res.fi = NULL;
1688         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1689                 goto brd_input;
1690
1691         /* Accept zero addresses only to limited broadcast;
1692          * I even do not know to fix it or not. Waiting for complains :-)
1693          */
1694         if (ipv4_is_zeronet(saddr))
1695                 goto martian_source;
1696
1697         if (ipv4_is_zeronet(daddr))
1698                 goto martian_destination;
1699
1700         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1701          * and call it once if daddr or/and saddr are loopback addresses
1702          */
1703         if (ipv4_is_loopback(daddr)) {
1704                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1705                         goto martian_destination;
1706         } else if (ipv4_is_loopback(saddr)) {
1707                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1708                         goto martian_source;
1709         }
1710
1711         /*
1712          *      Now we are ready to route packet.
1713          */
1714         fl4.flowi4_oif = 0;
1715         fl4.flowi4_iif = dev->ifindex;
1716         fl4.flowi4_mark = skb->mark;
1717         fl4.flowi4_tos = tos;
1718         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1719         fl4.daddr = daddr;
1720         fl4.saddr = saddr;
1721         err = fib_lookup(net, &fl4, &res);
1722         if (err != 0) {
1723                 if (!IN_DEV_FORWARD(in_dev))
1724                         err = -EHOSTUNREACH;
1725                 goto no_route;
1726         }
1727
1728         if (res.type == RTN_BROADCAST)
1729                 goto brd_input;
1730
1731         if (res.type == RTN_LOCAL) {
1732                 err = fib_validate_source(skb, saddr, daddr, tos,
1733                                           0, dev, in_dev, &itag);
1734                 if (err < 0)
1735                         goto martian_source_keep_err;
1736                 goto local_input;
1737         }
1738
1739         if (!IN_DEV_FORWARD(in_dev)) {
1740                 err = -EHOSTUNREACH;
1741                 goto no_route;
1742         }
1743         if (res.type != RTN_UNICAST)
1744                 goto martian_destination;
1745
1746         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1747 out:    return err;
1748
1749 brd_input:
1750         if (skb->protocol != htons(ETH_P_IP))
1751                 goto e_inval;
1752
1753         if (!ipv4_is_zeronet(saddr)) {
1754                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1755                                           in_dev, &itag);
1756                 if (err < 0)
1757                         goto martian_source_keep_err;
1758         }
1759         flags |= RTCF_BROADCAST;
1760         res.type = RTN_BROADCAST;
1761         RT_CACHE_STAT_INC(in_brd);
1762
1763 local_input:
1764         do_cache = false;
1765         if (res.fi) {
1766                 if (!itag) {
1767                         rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1768                         if (rt_cache_valid(rth)) {
1769                                 skb_dst_set_noref(skb, &rth->dst);
1770                                 err = 0;
1771                                 goto out;
1772                         }
1773                         do_cache = true;
1774                 }
1775         }
1776
1777         rth = rt_dst_alloc(net->loopback_dev,
1778                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1779         if (!rth)
1780                 goto e_nobufs;
1781
1782         rth->dst.input= ip_local_deliver;
1783         rth->dst.output= ip_rt_bug;
1784 #ifdef CONFIG_IP_ROUTE_CLASSID
1785         rth->dst.tclassid = itag;
1786 #endif
1787
1788         rth->rt_genid = rt_genid_ipv4(net);
1789         rth->rt_flags   = flags|RTCF_LOCAL;
1790         rth->rt_type    = res.type;
1791         rth->rt_is_input = 1;
1792         rth->rt_iif     = 0;
1793         rth->rt_pmtu    = 0;
1794         rth->rt_gateway = 0;
1795         rth->rt_uses_gateway = 0;
1796         INIT_LIST_HEAD(&rth->rt_uncached);
1797         RT_CACHE_STAT_INC(in_slow_tot);
1798         if (res.type == RTN_UNREACHABLE) {
1799                 rth->dst.input= ip_error;
1800                 rth->dst.error= -err;
1801                 rth->rt_flags   &= ~RTCF_LOCAL;
1802         }
1803         if (do_cache) {
1804                 if (unlikely(!rt_cache_route(&FIB_RES_NH(res), rth))) {
1805                         rth->dst.flags |= DST_NOCACHE;
1806                         rt_add_uncached_list(rth);
1807                 }
1808         }
1809         skb_dst_set(skb, &rth->dst);
1810         err = 0;
1811         goto out;
1812
1813 no_route:
1814         RT_CACHE_STAT_INC(in_no_route);
1815         res.type = RTN_UNREACHABLE;
1816         res.fi = NULL;
1817         goto local_input;
1818
1819         /*
1820          *      Do not cache martian addresses: they should be logged (RFC1812)
1821          */
1822 martian_destination:
1823         RT_CACHE_STAT_INC(in_martian_dst);
1824 #ifdef CONFIG_IP_ROUTE_VERBOSE
1825         if (IN_DEV_LOG_MARTIANS(in_dev))
1826                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1827                                      &daddr, &saddr, dev->name);
1828 #endif
1829
1830 e_inval:
1831         err = -EINVAL;
1832         goto out;
1833
1834 e_nobufs:
1835         err = -ENOBUFS;
1836         goto out;
1837
1838 martian_source:
1839         err = -EINVAL;
1840 martian_source_keep_err:
1841         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1842         goto out;
1843 }
1844
1845 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1846                          u8 tos, struct net_device *dev)
1847 {
1848         int res;
1849
1850         rcu_read_lock();
1851
1852         /* Multicast recognition logic is moved from route cache to here.
1853            The problem was that too many Ethernet cards have broken/missing
1854            hardware multicast filters :-( As result the host on multicasting
1855            network acquires a lot of useless route cache entries, sort of
1856            SDR messages from all the world. Now we try to get rid of them.
1857            Really, provided software IP multicast filter is organized
1858            reasonably (at least, hashed), it does not result in a slowdown
1859            comparing with route cache reject entries.
1860            Note, that multicast routers are not affected, because
1861            route cache entry is created eventually.
1862          */
1863         if (ipv4_is_multicast(daddr)) {
1864                 struct in_device *in_dev = __in_dev_get_rcu(dev);
1865
1866                 if (in_dev) {
1867                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1868                                                   ip_hdr(skb)->protocol);
1869                         if (our
1870 #ifdef CONFIG_IP_MROUTE
1871                                 ||
1872                             (!ipv4_is_local_multicast(daddr) &&
1873                              IN_DEV_MFORWARD(in_dev))
1874 #endif
1875                            ) {
1876                                 int res = ip_route_input_mc(skb, daddr, saddr,
1877                                                             tos, dev, our);
1878                                 rcu_read_unlock();
1879                                 return res;
1880                         }
1881                 }
1882                 rcu_read_unlock();
1883                 return -EINVAL;
1884         }
1885         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1886         rcu_read_unlock();
1887         return res;
1888 }
1889 EXPORT_SYMBOL(ip_route_input_noref);
1890
1891 /* called with rcu_read_lock() */
1892 static struct rtable *__mkroute_output(const struct fib_result *res,
1893                                        const struct flowi4 *fl4, int orig_oif,
1894                                        struct net_device *dev_out,
1895                                        unsigned int flags)
1896 {
1897         struct fib_info *fi = res->fi;
1898         struct fib_nh_exception *fnhe;
1899         struct in_device *in_dev;
1900         u16 type = res->type;
1901         struct rtable *rth;
1902         bool do_cache;
1903
1904         in_dev = __in_dev_get_rcu(dev_out);
1905         if (!in_dev)
1906                 return ERR_PTR(-EINVAL);
1907
1908         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1909                 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1910                         return ERR_PTR(-EINVAL);
1911
1912         if (ipv4_is_lbcast(fl4->daddr))
1913                 type = RTN_BROADCAST;
1914         else if (ipv4_is_multicast(fl4->daddr))
1915                 type = RTN_MULTICAST;
1916         else if (ipv4_is_zeronet(fl4->daddr))
1917                 return ERR_PTR(-EINVAL);
1918
1919         if (dev_out->flags & IFF_LOOPBACK)
1920                 flags |= RTCF_LOCAL;
1921
1922         do_cache = true;
1923         if (type == RTN_BROADCAST) {
1924                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1925                 fi = NULL;
1926         } else if (type == RTN_MULTICAST) {
1927                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1928                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1929                                      fl4->flowi4_proto))
1930                         flags &= ~RTCF_LOCAL;
1931                 else
1932                         do_cache = false;
1933                 /* If multicast route do not exist use
1934                  * default one, but do not gateway in this case.
1935                  * Yes, it is hack.
1936                  */
1937                 if (fi && res->prefixlen < 4)
1938                         fi = NULL;
1939         }
1940
1941         fnhe = NULL;
1942         do_cache &= fi != NULL;
1943         if (do_cache) {
1944                 struct rtable __rcu **prth;
1945                 struct fib_nh *nh = &FIB_RES_NH(*res);
1946
1947                 fnhe = find_exception(nh, fl4->daddr);
1948                 if (fnhe)
1949                         prth = &fnhe->fnhe_rth_output;
1950                 else {
1951                         if (unlikely(fl4->flowi4_flags &
1952                                      FLOWI_FLAG_KNOWN_NH &&
1953                                      !(nh->nh_gw &&
1954                                        nh->nh_scope == RT_SCOPE_LINK))) {
1955                                 do_cache = false;
1956                                 goto add;
1957                         }
1958                         prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
1959                 }
1960                 rth = rcu_dereference(*prth);
1961                 if (rt_cache_valid(rth)) {
1962                         dst_hold(&rth->dst);
1963                         return rth;
1964                 }
1965         }
1966
1967 add:
1968         rth = rt_dst_alloc(dev_out,
1969                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1970                            IN_DEV_CONF_GET(in_dev, NOXFRM),
1971                            do_cache);
1972         if (!rth)
1973                 return ERR_PTR(-ENOBUFS);
1974
1975         rth->dst.output = ip_output;
1976
1977         rth->rt_genid = rt_genid_ipv4(dev_net(dev_out));
1978         rth->rt_flags   = flags;
1979         rth->rt_type    = type;
1980         rth->rt_is_input = 0;
1981         rth->rt_iif     = orig_oif ? : 0;
1982         rth->rt_pmtu    = 0;
1983         rth->rt_gateway = 0;
1984         rth->rt_uses_gateway = 0;
1985         INIT_LIST_HEAD(&rth->rt_uncached);
1986
1987         RT_CACHE_STAT_INC(out_slow_tot);
1988
1989         if (flags & RTCF_LOCAL)
1990                 rth->dst.input = ip_local_deliver;
1991         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1992                 if (flags & RTCF_LOCAL &&
1993                     !(dev_out->flags & IFF_LOOPBACK)) {
1994                         rth->dst.output = ip_mc_output;
1995                         RT_CACHE_STAT_INC(out_slow_mc);
1996                 }
1997 #ifdef CONFIG_IP_MROUTE
1998                 if (type == RTN_MULTICAST) {
1999                         if (IN_DEV_MFORWARD(in_dev) &&
2000                             !ipv4_is_local_multicast(fl4->daddr)) {
2001                                 rth->dst.input = ip_mr_input;
2002                                 rth->dst.output = ip_mc_output;
2003                         }
2004                 }
2005 #endif
2006         }
2007
2008         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2009
2010         return rth;
2011 }
2012
2013 /*
2014  * Major route resolver routine.
2015  */
2016
2017 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
2018 {
2019         struct net_device *dev_out = NULL;
2020         __u8 tos = RT_FL_TOS(fl4);
2021         unsigned int flags = 0;
2022         struct fib_result res;
2023         struct rtable *rth;
2024         int orig_oif;
2025
2026         res.tclassid    = 0;
2027         res.fi          = NULL;
2028         res.table       = NULL;
2029
2030         orig_oif = fl4->flowi4_oif;
2031
2032         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2033         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2034         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2035                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2036
2037         rcu_read_lock();
2038         if (fl4->saddr) {
2039                 rth = ERR_PTR(-EINVAL);
2040                 if (ipv4_is_multicast(fl4->saddr) ||
2041                     ipv4_is_lbcast(fl4->saddr) ||
2042                     ipv4_is_zeronet(fl4->saddr))
2043                         goto out;
2044
2045                 /* I removed check for oif == dev_out->oif here.
2046                    It was wrong for two reasons:
2047                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2048                       is assigned to multiple interfaces.
2049                    2. Moreover, we are allowed to send packets with saddr
2050                       of another iface. --ANK
2051                  */
2052
2053                 if (fl4->flowi4_oif == 0 &&
2054                     (ipv4_is_multicast(fl4->daddr) ||
2055                      ipv4_is_lbcast(fl4->daddr))) {
2056                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2057                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2058                         if (!dev_out)
2059                                 goto out;
2060
2061                         /* Special hack: user can direct multicasts
2062                            and limited broadcast via necessary interface
2063                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2064                            This hack is not just for fun, it allows
2065                            vic,vat and friends to work.
2066                            They bind socket to loopback, set ttl to zero
2067                            and expect that it will work.
2068                            From the viewpoint of routing cache they are broken,
2069                            because we are not allowed to build multicast path
2070                            with loopback source addr (look, routing cache
2071                            cannot know, that ttl is zero, so that packet
2072                            will not leave this host and route is valid).
2073                            Luckily, this hack is good workaround.
2074                          */
2075
2076                         fl4->flowi4_oif = dev_out->ifindex;
2077                         goto make_route;
2078                 }
2079
2080                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2081                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2082                         if (!__ip_dev_find(net, fl4->saddr, false))
2083                                 goto out;
2084                 }
2085         }
2086
2087
2088         if (fl4->flowi4_oif) {
2089                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2090                 rth = ERR_PTR(-ENODEV);
2091                 if (!dev_out)
2092                         goto out;
2093
2094                 /* RACE: Check return value of inet_select_addr instead. */
2095                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2096                         rth = ERR_PTR(-ENETUNREACH);
2097                         goto out;
2098                 }
2099                 if (ipv4_is_local_multicast(fl4->daddr) ||
2100                     ipv4_is_lbcast(fl4->daddr)) {
2101                         if (!fl4->saddr)
2102                                 fl4->saddr = inet_select_addr(dev_out, 0,
2103                                                               RT_SCOPE_LINK);
2104                         goto make_route;
2105                 }
2106                 if (!fl4->saddr) {
2107                         if (ipv4_is_multicast(fl4->daddr))
2108                                 fl4->saddr = inet_select_addr(dev_out, 0,
2109                                                               fl4->flowi4_scope);
2110                         else if (!fl4->daddr)
2111                                 fl4->saddr = inet_select_addr(dev_out, 0,
2112                                                               RT_SCOPE_HOST);
2113                 }
2114         }
2115
2116         if (!fl4->daddr) {
2117                 fl4->daddr = fl4->saddr;
2118                 if (!fl4->daddr)
2119                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2120                 dev_out = net->loopback_dev;
2121                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2122                 res.type = RTN_LOCAL;
2123                 flags |= RTCF_LOCAL;
2124                 goto make_route;
2125         }
2126
2127         if (fib_lookup(net, fl4, &res)) {
2128                 res.fi = NULL;
2129                 res.table = NULL;
2130                 if (fl4->flowi4_oif) {
2131                         /* Apparently, routing tables are wrong. Assume,
2132                            that the destination is on link.
2133
2134                            WHY? DW.
2135                            Because we are allowed to send to iface
2136                            even if it has NO routes and NO assigned
2137                            addresses. When oif is specified, routing
2138                            tables are looked up with only one purpose:
2139                            to catch if destination is gatewayed, rather than
2140                            direct. Moreover, if MSG_DONTROUTE is set,
2141                            we send packet, ignoring both routing tables
2142                            and ifaddr state. --ANK
2143
2144
2145                            We could make it even if oif is unknown,
2146                            likely IPv6, but we do not.
2147                          */
2148
2149                         if (fl4->saddr == 0)
2150                                 fl4->saddr = inet_select_addr(dev_out, 0,
2151                                                               RT_SCOPE_LINK);
2152                         res.type = RTN_UNICAST;
2153                         goto make_route;
2154                 }
2155                 rth = ERR_PTR(-ENETUNREACH);
2156                 goto out;
2157         }
2158
2159         if (res.type == RTN_LOCAL) {
2160                 if (!fl4->saddr) {
2161                         if (res.fi->fib_prefsrc)
2162                                 fl4->saddr = res.fi->fib_prefsrc;
2163                         else
2164                                 fl4->saddr = fl4->daddr;
2165                 }
2166                 dev_out = net->loopback_dev;
2167                 fl4->flowi4_oif = dev_out->ifindex;
2168                 flags |= RTCF_LOCAL;
2169                 goto make_route;
2170         }
2171
2172 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2173         if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2174                 fib_select_multipath(&res);
2175         else
2176 #endif
2177         if (!res.prefixlen &&
2178             res.table->tb_num_default > 1 &&
2179             res.type == RTN_UNICAST && !fl4->flowi4_oif)
2180                 fib_select_default(&res);
2181
2182         if (!fl4->saddr)
2183                 fl4->saddr = FIB_RES_PREFSRC(net, res);
2184
2185         dev_out = FIB_RES_DEV(res);
2186         fl4->flowi4_oif = dev_out->ifindex;
2187
2188
2189 make_route:
2190         rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2191
2192 out:
2193         rcu_read_unlock();
2194         return rth;
2195 }
2196 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2197
2198 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2199 {
2200         return NULL;
2201 }
2202
2203 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2204 {
2205         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2206
2207         return mtu ? : dst->dev->mtu;
2208 }
2209
2210 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2211                                           struct sk_buff *skb, u32 mtu)
2212 {
2213 }
2214
2215 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2216                                        struct sk_buff *skb)
2217 {
2218 }
2219
2220 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2221                                           unsigned long old)
2222 {
2223         return NULL;
2224 }
2225
2226 static struct dst_ops ipv4_dst_blackhole_ops = {
2227         .family                 =       AF_INET,
2228         .check                  =       ipv4_blackhole_dst_check,
2229         .mtu                    =       ipv4_blackhole_mtu,
2230         .default_advmss         =       ipv4_default_advmss,
2231         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2232         .redirect               =       ipv4_rt_blackhole_redirect,
2233         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
2234         .neigh_lookup           =       ipv4_neigh_lookup,
2235 };
2236
2237 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2238 {
2239         struct rtable *ort = (struct rtable *) dst_orig;
2240         struct rtable *rt;
2241
2242         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2243         if (rt) {
2244                 struct dst_entry *new = &rt->dst;
2245
2246                 new->__use = 1;
2247                 new->input = dst_discard;
2248                 new->output = dst_discard_sk;
2249
2250                 new->dev = ort->dst.dev;
2251                 if (new->dev)
2252                         dev_hold(new->dev);
2253
2254                 rt->rt_is_input = ort->rt_is_input;
2255                 rt->rt_iif = ort->rt_iif;
2256                 rt->rt_pmtu = ort->rt_pmtu;
2257
2258                 rt->rt_genid = rt_genid_ipv4(net);
2259                 rt->rt_flags = ort->rt_flags;
2260                 rt->rt_type = ort->rt_type;
2261                 rt->rt_gateway = ort->rt_gateway;
2262                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2263
2264                 INIT_LIST_HEAD(&rt->rt_uncached);
2265
2266                 dst_free(new);
2267         }
2268
2269         dst_release(dst_orig);
2270
2271         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2272 }
2273
2274 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2275                                     struct sock *sk)
2276 {
2277         struct rtable *rt = __ip_route_output_key(net, flp4);
2278
2279         if (IS_ERR(rt))
2280                 return rt;
2281
2282         if (flp4->flowi4_proto)
2283                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2284                                                         flowi4_to_flowi(flp4),
2285                                                         sk, 0);
2286
2287         return rt;
2288 }
2289 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2290
2291 static int rt_fill_info(struct net *net,  __be32 dst, __be32 src,
2292                         struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2293                         u32 seq, int event, int nowait, unsigned int flags)
2294 {
2295         struct rtable *rt = skb_rtable(skb);
2296         struct rtmsg *r;
2297         struct nlmsghdr *nlh;
2298         unsigned long expires = 0;
2299         u32 error;
2300         u32 metrics[RTAX_MAX];
2301
2302         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
2303         if (!nlh)
2304                 return -EMSGSIZE;
2305
2306         r = nlmsg_data(nlh);
2307         r->rtm_family    = AF_INET;
2308         r->rtm_dst_len  = 32;
2309         r->rtm_src_len  = 0;
2310         r->rtm_tos      = fl4->flowi4_tos;
2311         r->rtm_table    = RT_TABLE_MAIN;
2312         if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2313                 goto nla_put_failure;
2314         r->rtm_type     = rt->rt_type;
2315         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2316         r->rtm_protocol = RTPROT_UNSPEC;
2317         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2318         if (rt->rt_flags & RTCF_NOTIFY)
2319                 r->rtm_flags |= RTM_F_NOTIFY;
2320         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2321                 r->rtm_flags |= RTCF_DOREDIRECT;
2322
2323         if (nla_put_in_addr(skb, RTA_DST, dst))
2324                 goto nla_put_failure;
2325         if (src) {
2326                 r->rtm_src_len = 32;
2327                 if (nla_put_in_addr(skb, RTA_SRC, src))
2328                         goto nla_put_failure;
2329         }
2330         if (rt->dst.dev &&
2331             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2332                 goto nla_put_failure;
2333 #ifdef CONFIG_IP_ROUTE_CLASSID
2334         if (rt->dst.tclassid &&
2335             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2336                 goto nla_put_failure;
2337 #endif
2338         if (!rt_is_input_route(rt) &&
2339             fl4->saddr != src) {
2340                 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2341                         goto nla_put_failure;
2342         }
2343         if (rt->rt_uses_gateway &&
2344             nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2345                 goto nla_put_failure;
2346
2347         expires = rt->dst.expires;
2348         if (expires) {
2349                 unsigned long now = jiffies;
2350
2351                 if (time_before(now, expires))
2352                         expires -= now;
2353                 else
2354                         expires = 0;
2355         }
2356
2357         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2358         if (rt->rt_pmtu && expires)
2359                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2360         if (rtnetlink_put_metrics(skb, metrics) < 0)
2361                 goto nla_put_failure;
2362
2363         if (fl4->flowi4_mark &&
2364             nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2365                 goto nla_put_failure;
2366
2367         error = rt->dst.error;
2368
2369         if (rt_is_input_route(rt)) {
2370 #ifdef CONFIG_IP_MROUTE
2371                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2372                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2373                         int err = ipmr_get_route(net, skb,
2374                                                  fl4->saddr, fl4->daddr,
2375                                                  r, nowait);
2376                         if (err <= 0) {
2377                                 if (!nowait) {
2378                                         if (err == 0)
2379                                                 return 0;
2380                                         goto nla_put_failure;
2381                                 } else {
2382                                         if (err == -EMSGSIZE)
2383                                                 goto nla_put_failure;
2384                                         error = err;
2385                                 }
2386                         }
2387                 } else
2388 #endif
2389                         if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2390                                 goto nla_put_failure;
2391         }
2392
2393         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2394                 goto nla_put_failure;
2395
2396         nlmsg_end(skb, nlh);
2397         return 0;
2398
2399 nla_put_failure:
2400         nlmsg_cancel(skb, nlh);
2401         return -EMSGSIZE;
2402 }
2403
2404 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2405 {
2406         struct net *net = sock_net(in_skb->sk);
2407         struct rtmsg *rtm;
2408         struct nlattr *tb[RTA_MAX+1];
2409         struct rtable *rt = NULL;
2410         struct flowi4 fl4;
2411         __be32 dst = 0;
2412         __be32 src = 0;
2413         u32 iif;
2414         int err;
2415         int mark;
2416         struct sk_buff *skb;
2417
2418         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2419         if (err < 0)
2420                 goto errout;
2421
2422         rtm = nlmsg_data(nlh);
2423
2424         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2425         if (!skb) {
2426                 err = -ENOBUFS;
2427                 goto errout;
2428         }
2429
2430         /* Reserve room for dummy headers, this skb can pass
2431            through good chunk of routing engine.
2432          */
2433         skb_reset_mac_header(skb);
2434         skb_reset_network_header(skb);
2435
2436         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2437         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2438         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2439
2440         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2441         dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2442         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2443         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2444
2445         memset(&fl4, 0, sizeof(fl4));
2446         fl4.daddr = dst;
2447         fl4.saddr = src;
2448         fl4.flowi4_tos = rtm->rtm_tos;
2449         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2450         fl4.flowi4_mark = mark;
2451
2452         if (iif) {
2453                 struct net_device *dev;
2454
2455                 dev = __dev_get_by_index(net, iif);
2456                 if (!dev) {
2457                         err = -ENODEV;
2458                         goto errout_free;
2459                 }
2460
2461                 skb->protocol   = htons(ETH_P_IP);
2462                 skb->dev        = dev;
2463                 skb->mark       = mark;
2464                 local_bh_disable();
2465                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2466                 local_bh_enable();
2467
2468                 rt = skb_rtable(skb);
2469                 if (err == 0 && rt->dst.error)
2470                         err = -rt->dst.error;
2471         } else {
2472                 rt = ip_route_output_key(net, &fl4);
2473
2474                 err = 0;
2475                 if (IS_ERR(rt))
2476                         err = PTR_ERR(rt);
2477         }
2478
2479         if (err)
2480                 goto errout_free;
2481
2482         skb_dst_set(skb, &rt->dst);
2483         if (rtm->rtm_flags & RTM_F_NOTIFY)
2484                 rt->rt_flags |= RTCF_NOTIFY;
2485
2486         err = rt_fill_info(net, dst, src, &fl4, skb,
2487                            NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2488                            RTM_NEWROUTE, 0, 0);
2489         if (err < 0)
2490                 goto errout_free;
2491
2492         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2493 errout:
2494         return err;
2495
2496 errout_free:
2497         kfree_skb(skb);
2498         goto errout;
2499 }
2500
2501 void ip_rt_multicast_event(struct in_device *in_dev)
2502 {
2503         rt_cache_flush(dev_net(in_dev->dev));
2504 }
2505
2506 #ifdef CONFIG_SYSCTL
2507 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
2508 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
2509 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
2510 static int ip_rt_gc_elasticity __read_mostly    = 8;
2511
2512 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2513                                         void __user *buffer,
2514                                         size_t *lenp, loff_t *ppos)
2515 {
2516         struct net *net = (struct net *)__ctl->extra1;
2517
2518         if (write) {
2519                 rt_cache_flush(net);
2520                 fnhe_genid_bump(net);
2521                 return 0;
2522         }
2523
2524         return -EINVAL;
2525 }
2526
2527 static struct ctl_table ipv4_route_table[] = {
2528         {
2529                 .procname       = "gc_thresh",
2530                 .data           = &ipv4_dst_ops.gc_thresh,
2531                 .maxlen         = sizeof(int),
2532                 .mode           = 0644,
2533                 .proc_handler   = proc_dointvec,
2534         },
2535         {
2536                 .procname       = "max_size",
2537                 .data           = &ip_rt_max_size,
2538                 .maxlen         = sizeof(int),
2539                 .mode           = 0644,
2540                 .proc_handler   = proc_dointvec,
2541         },
2542         {
2543                 /*  Deprecated. Use gc_min_interval_ms */
2544
2545                 .procname       = "gc_min_interval",
2546                 .data           = &ip_rt_gc_min_interval,
2547                 .maxlen         = sizeof(int),
2548                 .mode           = 0644,
2549                 .proc_handler   = proc_dointvec_jiffies,
2550         },
2551         {
2552                 .procname       = "gc_min_interval_ms",
2553                 .data           = &ip_rt_gc_min_interval,
2554                 .maxlen         = sizeof(int),
2555                 .mode           = 0644,
2556                 .proc_handler   = proc_dointvec_ms_jiffies,
2557         },
2558         {
2559                 .procname       = "gc_timeout",
2560                 .data           = &ip_rt_gc_timeout,
2561                 .maxlen         = sizeof(int),
2562                 .mode           = 0644,
2563                 .proc_handler   = proc_dointvec_jiffies,
2564         },
2565         {
2566                 .procname       = "gc_interval",
2567                 .data           = &ip_rt_gc_interval,
2568                 .maxlen         = sizeof(int),
2569                 .mode           = 0644,
2570                 .proc_handler   = proc_dointvec_jiffies,
2571         },
2572         {
2573                 .procname       = "redirect_load",
2574                 .data           = &ip_rt_redirect_load,
2575                 .maxlen         = sizeof(int),
2576                 .mode           = 0644,
2577                 .proc_handler   = proc_dointvec,
2578         },
2579         {
2580                 .procname       = "redirect_number",
2581                 .data           = &ip_rt_redirect_number,
2582                 .maxlen         = sizeof(int),
2583                 .mode           = 0644,
2584                 .proc_handler   = proc_dointvec,
2585         },
2586         {
2587                 .procname       = "redirect_silence",
2588                 .data           = &ip_rt_redirect_silence,
2589                 .maxlen         = sizeof(int),
2590                 .mode           = 0644,
2591                 .proc_handler   = proc_dointvec,
2592         },
2593         {
2594                 .procname       = "error_cost",
2595                 .data           = &ip_rt_error_cost,
2596                 .maxlen         = sizeof(int),
2597                 .mode           = 0644,
2598                 .proc_handler   = proc_dointvec,
2599         },
2600         {
2601                 .procname       = "error_burst",
2602                 .data           = &ip_rt_error_burst,
2603                 .maxlen         = sizeof(int),
2604                 .mode           = 0644,
2605                 .proc_handler   = proc_dointvec,
2606         },
2607         {
2608                 .procname       = "gc_elasticity",
2609                 .data           = &ip_rt_gc_elasticity,
2610                 .maxlen         = sizeof(int),
2611                 .mode           = 0644,
2612                 .proc_handler   = proc_dointvec,
2613         },
2614         {
2615                 .procname       = "mtu_expires",
2616                 .data           = &ip_rt_mtu_expires,
2617                 .maxlen         = sizeof(int),
2618                 .mode           = 0644,
2619                 .proc_handler   = proc_dointvec_jiffies,
2620         },
2621         {
2622                 .procname       = "min_pmtu",
2623                 .data           = &ip_rt_min_pmtu,
2624                 .maxlen         = sizeof(int),
2625                 .mode           = 0644,
2626                 .proc_handler   = proc_dointvec,
2627         },
2628         {
2629                 .procname       = "min_adv_mss",
2630                 .data           = &ip_rt_min_advmss,
2631                 .maxlen         = sizeof(int),
2632                 .mode           = 0644,
2633                 .proc_handler   = proc_dointvec,
2634         },
2635         { }
2636 };
2637
2638 static struct ctl_table ipv4_route_flush_table[] = {
2639         {
2640                 .procname       = "flush",
2641                 .maxlen         = sizeof(int),
2642                 .mode           = 0200,
2643                 .proc_handler   = ipv4_sysctl_rtcache_flush,
2644         },
2645         { },
2646 };
2647
2648 static __net_init int sysctl_route_net_init(struct net *net)
2649 {
2650         struct ctl_table *tbl;
2651
2652         tbl = ipv4_route_flush_table;
2653         if (!net_eq(net, &init_net)) {
2654                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2655                 if (!tbl)
2656                         goto err_dup;
2657
2658                 /* Don't export sysctls to unprivileged users */
2659                 if (net->user_ns != &init_user_ns)
2660                         tbl[0].procname = NULL;
2661         }
2662         tbl[0].extra1 = net;
2663
2664         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2665         if (!net->ipv4.route_hdr)
2666                 goto err_reg;
2667         return 0;
2668
2669 err_reg:
2670         if (tbl != ipv4_route_flush_table)
2671                 kfree(tbl);
2672 err_dup:
2673         return -ENOMEM;
2674 }
2675
2676 static __net_exit void sysctl_route_net_exit(struct net *net)
2677 {
2678         struct ctl_table *tbl;
2679
2680         tbl = net->ipv4.route_hdr->ctl_table_arg;
2681         unregister_net_sysctl_table(net->ipv4.route_hdr);
2682         BUG_ON(tbl == ipv4_route_flush_table);
2683         kfree(tbl);
2684 }
2685
2686 static __net_initdata struct pernet_operations sysctl_route_ops = {
2687         .init = sysctl_route_net_init,
2688         .exit = sysctl_route_net_exit,
2689 };
2690 #endif
2691
2692 static __net_init int rt_genid_init(struct net *net)
2693 {
2694         atomic_set(&net->ipv4.rt_genid, 0);
2695         atomic_set(&net->fnhe_genid, 0);
2696         get_random_bytes(&net->ipv4.dev_addr_genid,
2697                          sizeof(net->ipv4.dev_addr_genid));
2698         return 0;
2699 }
2700
2701 static __net_initdata struct pernet_operations rt_genid_ops = {
2702         .init = rt_genid_init,
2703 };
2704
2705 static int __net_init ipv4_inetpeer_init(struct net *net)
2706 {
2707         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2708
2709         if (!bp)
2710                 return -ENOMEM;
2711         inet_peer_base_init(bp);
2712         net->ipv4.peers = bp;
2713         return 0;
2714 }
2715
2716 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2717 {
2718         struct inet_peer_base *bp = net->ipv4.peers;
2719
2720         net->ipv4.peers = NULL;
2721         inetpeer_invalidate_tree(bp);
2722         kfree(bp);
2723 }
2724
2725 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2726         .init   =       ipv4_inetpeer_init,
2727         .exit   =       ipv4_inetpeer_exit,
2728 };
2729
2730 #ifdef CONFIG_IP_ROUTE_CLASSID
2731 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2732 #endif /* CONFIG_IP_ROUTE_CLASSID */
2733
2734 int __init ip_rt_init(void)
2735 {
2736         int rc = 0;
2737         int cpu;
2738
2739         ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2740         if (!ip_idents)
2741                 panic("IP: failed to allocate ip_idents\n");
2742
2743         prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2744
2745         for_each_possible_cpu(cpu) {
2746                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
2747
2748                 INIT_LIST_HEAD(&ul->head);
2749                 spin_lock_init(&ul->lock);
2750         }
2751 #ifdef CONFIG_IP_ROUTE_CLASSID
2752         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2753         if (!ip_rt_acct)
2754                 panic("IP: failed to allocate ip_rt_acct\n");
2755 #endif
2756
2757         ipv4_dst_ops.kmem_cachep =
2758                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2759                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2760
2761         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2762
2763         if (dst_entries_init(&ipv4_dst_ops) < 0)
2764                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2765
2766         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2767                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2768
2769         ipv4_dst_ops.gc_thresh = ~0;
2770         ip_rt_max_size = INT_MAX;
2771
2772         devinet_init();
2773         ip_fib_init();
2774
2775         if (ip_rt_proc_init())
2776                 pr_err("Unable to create route proc files\n");
2777 #ifdef CONFIG_XFRM
2778         xfrm_init();
2779         xfrm4_init();
2780 #endif
2781         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2782
2783 #ifdef CONFIG_SYSCTL
2784         register_pernet_subsys(&sysctl_route_ops);
2785 #endif
2786         register_pernet_subsys(&rt_genid_ops);
2787         register_pernet_subsys(&ipv4_inetpeer_ops);
2788         return rc;
2789 }
2790
2791 #ifdef CONFIG_SYSCTL
2792 /*
2793  * We really need to sanitize the damn ipv4 init order, then all
2794  * this nonsense will go away.
2795  */
2796 void __init ip_static_sysctl_init(void)
2797 {
2798         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
2799 }
2800 #endif