net/ipv4/route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              ROUTE - implementation of the IP router.
   8  *
   9  * Authors:     Ross Biro
  10  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  12  *              Linus Torvalds, <Linus.Torvalds@helsinki.fi>
  13  *              Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  14  *
  15  * Fixes:
  16  *              Alan Cox        :       Verify area fixes.
  17  *              Alan Cox        :       cli() protects routing changes
  18  *              Rui Oliveira    :       ICMP routing table updates
  19  *              (rco@di.uminho.pt)      Routing table insertion and update
  20  *              Linus Torvalds  :       Rewrote bits to be sensible
  21  *              Alan Cox        :       Added BSD route gw semantics
  22  *              Alan Cox        :       Super /proc >4K
  23  *              Alan Cox        :       MTU in route table
  24  *              Alan Cox        :       MSS actually. Also added the window
  25  *                                      clamper.
  26  *              Sam Lantinga    :       Fixed route matching in rt_del()
  27  *              Alan Cox        :       Routing cache support.
  28  *              Alan Cox        :       Removed compatibility cruft.
  29  *              Alan Cox        :       RTF_REJECT support.
  30  *              Alan Cox        :       TCP irtt support.
  31  *              Jonathan Naylor :       Added Metric support.
  32  *      Miquel van Smoorenburg  :       BSD API fixes.
  33  *      Miquel van Smoorenburg  :       Metrics.
  34  *              Alan Cox        :       Use __u32 properly
  35  *              Alan Cox        :       Aligned routing errors more closely with BSD
  36  *                                      our system is still very different.
  37  *              Alan Cox        :       Faster /proc handling
  38  *      Alexey Kuznetsov        :       Massive rework to support tree based routing,
  39  *                                      routing caches and better behaviour.
  40  *
  41  *              Olaf Erb        :       irtt wasn't being copied right.
  42  *              Bjorn Ekwall    :       Kerneld route support.
  43  *              Alan Cox        :       Multicast fixed (I hope)
  44  *              Pavel Krauz     :       Limited broadcast fixed
  45  *              Mike McLagan    :       Routing by source
  46  *      Alexey Kuznetsov        :       End of old history. Split to fib.c and
  47  *                                      route.c and rewritten from scratch.
  48  *              Andi Kleen      :       Load-limit warning messages.
  49  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  50  *      Vitaly E. Lavrov        :       Race condition in ip_route_input_slow.
  51  *      Tobias Ringstrom        :       Uninitialized res.type in ip_route_output_slow.
  52  *      Vladimir V. Ivanov      :       IP rule info (flowid) is really useful.
  53  *              Marc Boucher    :       routing by fwmark
  54  *      Robert Olsson           :       Added rt_cache statistics
  55  *      Arnaldo C. Melo         :       Convert proc stuff to seq_file
  56  *      Eric Dumazet            :       hashed spinlocks and rt_check_expire() fixes.
  57  *      Ilia Sotnikov           :       Ignore TOS on PMTUD and Redirect
  58  *      Ilia Sotnikov           :       Removed TOS from hash calculations
  59  */
  60
  61 #define pr_fmt(fmt) "IPv4: " fmt
  62
  63 #include <linux/module.h>
  64 #include <linux/bitops.h>
  65 #include <linux/kernel.h>
  66 #include <linux/mm.h>
  67 #include <linux/memblock.h>
  68 #include <linux/socket.h>
  69 #include <linux/errno.h>
  70 #include <linux/in.h>
  71 #include <linux/inet.h>
  72 #include <linux/netdevice.h>
  73 #include <linux/proc_fs.h>
  74 #include <linux/init.h>
  75 #include <linux/skbuff.h>
  76 #include <linux/inetdevice.h>
  77 #include <linux/igmp.h>
  78 #include <linux/pkt_sched.h>
  79 #include <linux/mroute.h>
  80 #include <linux/netfilter_ipv4.h>
  81 #include <linux/random.h>
  82 #include <linux/rcupdate.h>
  83 #include <linux/slab.h>
  84 #include <linux/jhash.h>
  85 #include <net/dst.h>
  86 #include <net/dst_metadata.h>
  87 #include <net/net_namespace.h>
  88 #include <net/ip.h>
  89 #include <net/route.h>
  90 #include <net/inetpeer.h>
  91 #include <net/sock.h>
  92 #include <net/ip_fib.h>
  93 #include <net/nexthop.h>
  94 #include <net/tcp.h>
  95 #include <net/icmp.h>
  96 #include <net/xfrm.h>
  97 #include <net/lwtunnel.h>
  98 #include <net/netevent.h>
  99 #include <net/rtnetlink.h>
 100 #ifdef CONFIG_SYSCTL
 101 #include <linux/sysctl.h>
 102 #endif
 103 #include <net/secure_seq.h>
 104 #include <net/ip_tunnels.h>
 105
 106 #include "fib_lookup.h"
 107
 108 #define RT_FL_TOS(oldflp4) \
 109         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
 110
 111 #define RT_GC_TIMEOUT (300*HZ)
 112
 113 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
 114 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
 115
 116 static int ip_rt_max_size;
 117 static int ip_rt_redirect_number __read_mostly  = 9;
 118 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 119 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 120 static int ip_rt_error_cost __read_mostly       = HZ;
 121 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 122 static int ip_rt_min_advmss __read_mostly       = 256;
 123
 124 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
 125
 126 /*
 127  *      Interface to generic destination cache.
 128  */
 129
 130 INDIRECT_CALLABLE_SCOPE
 131 struct dst_entry        *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 132 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 133 INDIRECT_CALLABLE_SCOPE
 134 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                                            bool confirm_neigh);
 140 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 141                                         struct sk_buff *skb);
 142 static void             ipv4_dst_destroy(struct dst_entry *dst);
 143
 144 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 145 {
 146         WARN_ON(1);
 147         return NULL;
 148 }
 149
 150 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 151                                            struct sk_buff *skb,
 152                                            const void *daddr);
 153 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
 154
 155 static struct dst_ops ipv4_dst_ops = {
 156         .family =               AF_INET,
 157         .check =                ipv4_dst_check,
 158         .default_advmss =       ipv4_default_advmss,
 159         .mtu =                  ipv4_mtu,
 160         .cow_metrics =          ipv4_cow_metrics,
 161         .destroy =              ipv4_dst_destroy,
 162         .negative_advice =      ipv4_negative_advice,
 163         .link_failure =         ipv4_link_failure,
 164         .update_pmtu =          ip_rt_update_pmtu,
 165         .redirect =             ip_do_redirect,
 166         .local_out =            __ip_local_out,
 167         .neigh_lookup =         ipv4_neigh_lookup,
 168         .confirm_neigh =        ipv4_confirm_neigh,
 169 };
 170
 171 #define ECN_OR_COST(class)      TC_PRIO_##class
 172
 173 const __u8 ip_tos2prio[16] = {
 174         TC_PRIO_BESTEFFORT,
 175         ECN_OR_COST(BESTEFFORT),
 176         TC_PRIO_BESTEFFORT,
 177         ECN_OR_COST(BESTEFFORT),
 178         TC_PRIO_BULK,
 179         ECN_OR_COST(BULK),
 180         TC_PRIO_BULK,
 181         ECN_OR_COST(BULK),
 182         TC_PRIO_INTERACTIVE,
 183         ECN_OR_COST(INTERACTIVE),
 184         TC_PRIO_INTERACTIVE,
 185         ECN_OR_COST(INTERACTIVE),
 186         TC_PRIO_INTERACTIVE_BULK,
 187         ECN_OR_COST(INTERACTIVE_BULK),
 188         TC_PRIO_INTERACTIVE_BULK,
 189         ECN_OR_COST(INTERACTIVE_BULK)
 190 };
 191 EXPORT_SYMBOL(ip_tos2prio);
 192
 193 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 194 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
 195
 196 #ifdef CONFIG_PROC_FS
 197 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 198 {
 199         if (*pos)
 200                 return NULL;
 201         return SEQ_START_TOKEN;
 202 }
 203
 204 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 205 {
 206         ++*pos;
 207         return NULL;
 208 }
 209
 210 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 211 {
 212 }
 213
 214 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 215 {
 216         if (v == SEQ_START_TOKEN)
 217                 seq_printf(seq, "%-127s\n",
 218                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 219                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 220                            "HHUptod\tSpecDst");
 221         return 0;
 222 }
 223
 224 static const struct seq_operations rt_cache_seq_ops = {
 225         .start  = rt_cache_seq_start,
 226         .next   = rt_cache_seq_next,
 227         .stop   = rt_cache_seq_stop,
 228         .show   = rt_cache_seq_show,
 229 };
 230
 231 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 232 {
 233         int cpu;
 234
 235         if (*pos == 0)
 236                 return SEQ_START_TOKEN;
 237
 238         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 239                 if (!cpu_possible(cpu))
 240                         continue;
 241                 *pos = cpu+1;
 242                 return &per_cpu(rt_cache_stat, cpu);
 243         }
 244         return NULL;
 245 }
 246
 247 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 248 {
 249         int cpu;
 250
 251         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 252                 if (!cpu_possible(cpu))
 253                         continue;
 254                 *pos = cpu+1;
 255                 return &per_cpu(rt_cache_stat, cpu);
 256         }
 257         (*pos)++;
 258         return NULL;
 259
 260 }
 261
 262 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 263 {
 264
 265 }
 266
 267 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 268 {
 269         struct rt_cache_stat *st = v;
 270
 271         if (v == SEQ_START_TOKEN) {
 272                 seq_puts(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");
 273                 return 0;
 274         }
 275
 276         seq_printf(seq, "%08x %08x %08x    %08x   %08x    %08x %08x       "
 277                         "%08x       %08x %08x     %08x    %08x %08x   "
 278                         "%08x     %08x        %08x        %08x\n",
 279                    dst_entries_get_slow(&ipv4_dst_ops),
 280                    0, /* st->in_hit */
 281                    st->in_slow_tot,
 282                    st->in_slow_mc,
 283                    st->in_no_route,
 284                    st->in_brd,
 285                    st->in_martian_dst,
 286                    st->in_martian_src,
 287
 288                    0, /* st->out_hit */
 289                    st->out_slow_tot,
 290                    st->out_slow_mc,
 291
 292                    0, /* st->gc_total */
 293                    0, /* st->gc_ignored */
 294                    0, /* st->gc_goal_miss */
 295                    0, /* st->gc_dst_overflow */
 296                    0, /* st->in_hlist_search */
 297                    0  /* st->out_hlist_search */
 298                 );
 299         return 0;
 300 }
 301
 302 static const struct seq_operations rt_cpu_seq_ops = {
 303         .start  = rt_cpu_seq_start,
 304         .next   = rt_cpu_seq_next,
 305         .stop   = rt_cpu_seq_stop,
 306         .show   = rt_cpu_seq_show,
 307 };
 308
 309 #ifdef CONFIG_IP_ROUTE_CLASSID
 310 static int rt_acct_proc_show(struct seq_file *m, void *v)
 311 {
 312         struct ip_rt_acct *dst, *src;
 313         unsigned int i, j;
 314
 315         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 316         if (!dst)
 317                 return -ENOMEM;
 318
 319         for_each_possible_cpu(i) {
 320                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 321                 for (j = 0; j < 256; j++) {
 322                         dst[j].o_bytes   += src[j].o_bytes;
 323                         dst[j].o_packets += src[j].o_packets;
 324                         dst[j].i_bytes   += src[j].i_bytes;
 325                         dst[j].i_packets += src[j].i_packets;
 326                 }
 327         }
 328
 329         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 330         kfree(dst);
 331         return 0;
 332 }
 333 #endif
 334
 335 static int __net_init ip_rt_do_proc_init(struct net *net)
 336 {
 337         struct proc_dir_entry *pde;
 338
 339         pde = proc_create_seq("rt_cache", 0444, net->proc_net,
 340                               &rt_cache_seq_ops);
 341         if (!pde)
 342                 goto err1;
 343
 344         pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
 345                               &rt_cpu_seq_ops);
 346         if (!pde)
 347                 goto err2;
 348
 349 #ifdef CONFIG_IP_ROUTE_CLASSID
 350         pde = proc_create_single("rt_acct", 0, net->proc_net,
 351                         rt_acct_proc_show);
 352         if (!pde)
 353                 goto err3;
 354 #endif
 355         return 0;
 356
 357 #ifdef CONFIG_IP_ROUTE_CLASSID
 358 err3:
 359         remove_proc_entry("rt_cache", net->proc_net_stat);
 360 #endif
 361 err2:
 362         remove_proc_entry("rt_cache", net->proc_net);
 363 err1:
 364         return -ENOMEM;
 365 }
 366
 367 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 368 {
 369         remove_proc_entry("rt_cache", net->proc_net_stat);
 370         remove_proc_entry("rt_cache", net->proc_net);
 371 #ifdef CONFIG_IP_ROUTE_CLASSID
 372         remove_proc_entry("rt_acct", net->proc_net);
 373 #endif
 374 }
 375
 376 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 377         .init = ip_rt_do_proc_init,
 378         .exit = ip_rt_do_proc_exit,
 379 };
 380
 381 static int __init ip_rt_proc_init(void)
 382 {
 383         return register_pernet_subsys(&ip_rt_proc_ops);
 384 }
 385
 386 #else
 387 static inline int ip_rt_proc_init(void)
 388 {
 389         return 0;
 390 }
 391 #endif /* CONFIG_PROC_FS */
 392
 393 static inline bool rt_is_expired(const struct rtable *rth)
 394 {
 395         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 396 }
 397
 398 void rt_cache_flush(struct net *net)
 399 {
 400         rt_genid_bump_ipv4(net);
 401 }
 402
 403 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 404                                            struct sk_buff *skb,
 405                                            const void *daddr)
 406 {
 407         const struct rtable *rt = container_of(dst, struct rtable, dst);
 408         struct net_device *dev = dst->dev;
 409         struct neighbour *n;
 410
 411         rcu_read_lock_bh();
 412
 413         if (likely(rt->rt_gw_family == AF_INET)) {
 414                 n = ip_neigh_gw4(dev, rt->rt_gw4);
 415         } else if (rt->rt_gw_family == AF_INET6) {
 416                 n = ip_neigh_gw6(dev, &rt->rt_gw6);
 417         } else {
 418                 __be32 pkey;
 419
 420                 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
 421                 n = ip_neigh_gw4(dev, pkey);
 422         }
 423
 424         if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
 425                 n = NULL;
 426
 427         rcu_read_unlock_bh();
 428
 429         return n;
 430 }
 431
 432 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 433 {
 434         const struct rtable *rt = container_of(dst, struct rtable, dst);
 435         struct net_device *dev = dst->dev;
 436         const __be32 *pkey = daddr;
 437
 438         if (rt->rt_gw_family == AF_INET) {
 439                 pkey = (const __be32 *)&rt->rt_gw4;
 440         } else if (rt->rt_gw_family == AF_INET6) {
 441                 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
 442         } else if (!daddr ||
 443                  (rt->rt_flags &
 444                   (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
 445                 return;
 446         }
 447         __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
 448 }
 449
 450 /* Hash tables of size 2048..262144 depending on RAM size.
 451  * Each bucket uses 8 bytes.
 452  */
 453 static u32 ip_idents_mask __read_mostly;
 454 static atomic_t *ip_idents __read_mostly;
 455 static u32 *ip_tstamps __read_mostly;
 456
 457 /* In order to protect privacy, we add a perturbation to identifiers
 458  * if one generator is seldom used. This makes hard for an attacker
 459  * to infer how many packets were sent between two points in time.
 460  */
 461 u32 ip_idents_reserve(u32 hash, int segs)
 462 {
 463         u32 bucket, old, now = (u32)jiffies;
 464         atomic_t *p_id;
 465         u32 *p_tstamp;
 466         u32 delta = 0;
 467
 468         bucket = hash & ip_idents_mask;
 469         p_tstamp = ip_tstamps + bucket;
 470         p_id = ip_idents + bucket;
 471         old = READ_ONCE(*p_tstamp);
 472
 473         if (old != now && cmpxchg(p_tstamp, old, now) == old)
 474                 delta = prandom_u32_max(now - old);
 475
 476         /* If UBSAN reports an error there, please make sure your compiler
 477          * supports -fno-strict-overflow before reporting it that was a bug
 478          * in UBSAN, and it has been fixed in GCC-8.
 479          */
 480         return atomic_add_return(segs + delta, p_id) - segs;
 481 }
 482 EXPORT_SYMBOL(ip_idents_reserve);
 483
 484 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
 485 {
 486         u32 hash, id;
 487
 488         /* Note the following code is not safe, but this is okay. */
 489         if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
 490                 get_random_bytes(&net->ipv4.ip_id_key,
 491                                  sizeof(net->ipv4.ip_id_key));
 492
 493         hash = siphash_3u32((__force u32)iph->daddr,
 494                             (__force u32)iph->saddr,
 495                             iph->protocol,
 496                             &net->ipv4.ip_id_key);
 497         id = ip_idents_reserve(hash, segs);
 498         iph->id = htons(id);
 499 }
 500 EXPORT_SYMBOL(__ip_select_ident);
 501
 502 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
 503                              const struct sock *sk,
 504                              const struct iphdr *iph,
 505                              int oif, u8 tos,
 506                              u8 prot, u32 mark, int flow_flags)
 507 {
 508         if (sk) {
 509                 const struct inet_sock *inet = inet_sk(sk);
 510
 511                 oif = sk->sk_bound_dev_if;
 512                 mark = sk->sk_mark;
 513                 tos = RT_CONN_FLAGS(sk);
 514                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 515         }
 516         flowi4_init_output(fl4, oif, mark, tos,
 517                            RT_SCOPE_UNIVERSE, prot,
 518                            flow_flags,
 519                            iph->daddr, iph->saddr, 0, 0,
 520                            sock_net_uid(net, sk));
 521 }
 522
 523 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 524                                const struct sock *sk)
 525 {
 526         const struct net *net = dev_net(skb->dev);
 527         const struct iphdr *iph = ip_hdr(skb);
 528         int oif = skb->dev->ifindex;
 529         u8 tos = RT_TOS(iph->tos);
 530         u8 prot = iph->protocol;
 531         u32 mark = skb->mark;
 532
 533         __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
 534 }
 535
 536 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 537 {
 538         const struct inet_sock *inet = inet_sk(sk);
 539         const struct ip_options_rcu *inet_opt;
 540         __be32 daddr = inet->inet_daddr;
 541
 542         rcu_read_lock();
 543         inet_opt = rcu_dereference(inet->inet_opt);
 544         if (inet_opt && inet_opt->opt.srr)
 545                 daddr = inet_opt->opt.faddr;
 546         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 547                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 548                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 549                            inet_sk_flowi_flags(sk),
 550                            daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
 551         rcu_read_unlock();
 552 }
 553
 554 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 555                                  const struct sk_buff *skb)
 556 {
 557         if (skb)
 558                 build_skb_flow_key(fl4, skb, sk);
 559         else
 560                 build_sk_flow_key(fl4, sk);
 561 }
 562
 563 static DEFINE_SPINLOCK(fnhe_lock);
 564
 565 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 566 {
 567         struct rtable *rt;
 568
 569         rt = rcu_dereference(fnhe->fnhe_rth_input);
 570         if (rt) {
 571                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 572                 dst_dev_put(&rt->dst);
 573                 dst_release(&rt->dst);
 574         }
 575         rt = rcu_dereference(fnhe->fnhe_rth_output);
 576         if (rt) {
 577                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 578                 dst_dev_put(&rt->dst);
 579                 dst_release(&rt->dst);
 580         }
 581 }
 582
 583 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
 584 {
 585         struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
 586         struct fib_nh_exception *fnhe, *oldest = NULL;
 587
 588         for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
 589                 fnhe = rcu_dereference_protected(*fnhe_p,
 590                                                  lockdep_is_held(&fnhe_lock));
 591                 if (!fnhe)
 592                         break;
 593                 if (!oldest ||
 594                     time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
 595                         oldest = fnhe;
 596                         oldest_p = fnhe_p;
 597                 }
 598         }
 599         fnhe_flush_routes(oldest);
 600         *oldest_p = oldest->fnhe_next;
 601         kfree_rcu(oldest, rcu);
 602 }
 603
 604 static u32 fnhe_hashfun(__be32 daddr)
 605 {
 606         static siphash_aligned_key_t fnhe_hash_key;
 607         u64 hval;
 608
 609         net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
 610         hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
 611         return hash_64(hval, FNHE_HASH_SHIFT);
 612 }
 613
 614 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 615 {
 616         rt->rt_pmtu = fnhe->fnhe_pmtu;
 617         rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
 618         rt->dst.expires = fnhe->fnhe_expires;
 619
 620         if (fnhe->fnhe_gw) {
 621                 rt->rt_flags |= RTCF_REDIRECTED;
 622                 rt->rt_uses_gateway = 1;
 623                 rt->rt_gw_family = AF_INET;
 624                 rt->rt_gw4 = fnhe->fnhe_gw;
 625         }
 626 }
 627
 628 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
 629                                   __be32 gw, u32 pmtu, bool lock,
 630                                   unsigned long expires)
 631 {
 632         struct fnhe_hash_bucket *hash;
 633         struct fib_nh_exception *fnhe;
 634         struct rtable *rt;
 635         u32 genid, hval;
 636         unsigned int i;
 637         int depth;
 638
 639         genid = fnhe_genid(dev_net(nhc->nhc_dev));
 640         hval = fnhe_hashfun(daddr);
 641
 642         spin_lock_bh(&fnhe_lock);
 643
 644         hash = rcu_dereference(nhc->nhc_exceptions);
 645         if (!hash) {
 646                 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
 647                 if (!hash)
 648                         goto out_unlock;
 649                 rcu_assign_pointer(nhc->nhc_exceptions, hash);
 650         }
 651
 652         hash += hval;
 653
 654         depth = 0;
 655         for (fnhe = rcu_dereference(hash->chain); fnhe;
 656              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 657                 if (fnhe->fnhe_daddr == daddr)
 658                         break;
 659                 depth++;
 660         }
 661
 662         if (fnhe) {
 663                 if (fnhe->fnhe_genid != genid)
 664                         fnhe->fnhe_genid = genid;
 665                 if (gw)
 666                         fnhe->fnhe_gw = gw;
 667                 if (pmtu) {
 668                         fnhe->fnhe_pmtu = pmtu;
 669                         fnhe->fnhe_mtu_locked = lock;
 670                 }
 671                 fnhe->fnhe_expires = max(1UL, expires);
 672                 /* Update all cached dsts too */
 673                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 674                 if (rt)
 675                         fill_route_from_fnhe(rt, fnhe);
 676                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 677                 if (rt)
 678                         fill_route_from_fnhe(rt, fnhe);
 679         } else {
 680                 /* Randomize max depth to avoid some side channels attacks. */
 681                 int max_depth = FNHE_RECLAIM_DEPTH +
 682                                 prandom_u32_max(FNHE_RECLAIM_DEPTH);
 683
 684                 while (depth > max_depth) {
 685                         fnhe_remove_oldest(hash);
 686                         depth--;
 687                 }
 688
 689                 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 690                 if (!fnhe)
 691                         goto out_unlock;
 692
 693                 fnhe->fnhe_next = hash->chain;
 694
 695                 fnhe->fnhe_genid = genid;
 696                 fnhe->fnhe_daddr = daddr;
 697                 fnhe->fnhe_gw = gw;
 698                 fnhe->fnhe_pmtu = pmtu;
 699                 fnhe->fnhe_mtu_locked = lock;
 700                 fnhe->fnhe_expires = max(1UL, expires);
 701
 702                 rcu_assign_pointer(hash->chain, fnhe);
 703
 704                 /* Exception created; mark the cached routes for the nexthop
 705                  * stale, so anyone caching it rechecks if this exception
 706                  * applies to them.
 707                  */
 708                 rt = rcu_dereference(nhc->nhc_rth_input);
 709                 if (rt)
 710                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 711
 712                 for_each_possible_cpu(i) {
 713                         struct rtable __rcu **prt;
 714
 715                         prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
 716                         rt = rcu_dereference(*prt);
 717                         if (rt)
 718                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 719                 }
 720         }
 721
 722         fnhe->fnhe_stamp = jiffies;
 723
 724 out_unlock:
 725         spin_unlock_bh(&fnhe_lock);
 726 }
 727
 728 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 729                              bool kill_route)
 730 {
 731         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 732         __be32 old_gw = ip_hdr(skb)->saddr;
 733         struct net_device *dev = skb->dev;
 734         struct in_device *in_dev;
 735         struct fib_result res;
 736         struct neighbour *n;
 737         struct net *net;
 738
 739         switch (icmp_hdr(skb)->code & 7) {
 740         case ICMP_REDIR_NET:
 741         case ICMP_REDIR_NETTOS:
 742         case ICMP_REDIR_HOST:
 743         case ICMP_REDIR_HOSTTOS:
 744                 break;
 745
 746         default:
 747                 return;
 748         }
 749
 750         if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
 751                 return;
 752
 753         in_dev = __in_dev_get_rcu(dev);
 754         if (!in_dev)
 755                 return;
 756
 757         net = dev_net(dev);
 758         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 759             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 760             ipv4_is_zeronet(new_gw))
 761                 goto reject_redirect;
 762
 763         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 764                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 765                         goto reject_redirect;
 766                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 767                         goto reject_redirect;
 768         } else {
 769                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 770                         goto reject_redirect;
 771         }
 772
 773         n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
 774         if (!n)
 775                 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
 776         if (!IS_ERR(n)) {
 777                 if (!(n->nud_state & NUD_VALID)) {
 778                         neigh_event_send(n, NULL);
 779                 } else {
 780                         if (fib_lookup(net, fl4, &res, 0) == 0) {
 781                                 struct fib_nh_common *nhc;
 782
 783                                 fib_select_path(net, &res, fl4, skb);
 784                                 nhc = FIB_RES_NHC(res);
 785                                 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
 786                                                 0, false,
 787                                                 jiffies + ip_rt_gc_timeout);
 788                         }
 789                         if (kill_route)
 790                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 791                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 792                 }
 793                 neigh_release(n);
 794         }
 795         return;
 796
 797 reject_redirect:
 798 #ifdef CONFIG_IP_ROUTE_VERBOSE
 799         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 800                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 801                 __be32 daddr = iph->daddr;
 802                 __be32 saddr = iph->saddr;
 803
 804                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 805                                      "  Advised path = %pI4 -> %pI4\n",
 806                                      &old_gw, dev->name, &new_gw,
 807                                      &saddr, &daddr);
 808         }
 809 #endif
 810         ;
 811 }
 812
 813 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 814 {
 815         struct rtable *rt;
 816         struct flowi4 fl4;
 817         const struct iphdr *iph = (const struct iphdr *) skb->data;
 818         struct net *net = dev_net(skb->dev);
 819         int oif = skb->dev->ifindex;
 820         u8 tos = RT_TOS(iph->tos);
 821         u8 prot = iph->protocol;
 822         u32 mark = skb->mark;
 823
 824         rt = (struct rtable *) dst;
 825
 826         __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
 827         __ip_do_redirect(rt, skb, &fl4, true);
 828 }
 829
 830 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 831 {
 832         struct rtable *rt = (struct rtable *)dst;
 833         struct dst_entry *ret = dst;
 834
 835         if (rt) {
 836                 if (dst->obsolete > 0) {
 837                         ip_rt_put(rt);
 838                         ret = NULL;
 839                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 840                            rt->dst.expires) {
 841                         ip_rt_put(rt);
 842                         ret = NULL;
 843                 }
 844         }
 845         return ret;
 846 }
 847
 848 /*
 849  * Algorithm:
 850  *      1. The first ip_rt_redirect_number redirects are sent
 851  *         with exponential backoff, then we stop sending them at all,
 852  *         assuming that the host ignores our redirects.
 853  *      2. If we did not see packets requiring redirects
 854  *         during ip_rt_redirect_silence, we assume that the host
 855  *         forgot redirected route and start to send redirects again.
 856  *
 857  * This algorithm is much cheaper and more intelligent than dumb load limiting
 858  * in icmp.c.
 859  *
 860  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 861  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 862  */
 863
 864 void ip_rt_send_redirect(struct sk_buff *skb)
 865 {
 866         struct rtable *rt = skb_rtable(skb);
 867         struct in_device *in_dev;
 868         struct inet_peer *peer;
 869         struct net *net;
 870         int log_martians;
 871         int vif;
 872
 873         rcu_read_lock();
 874         in_dev = __in_dev_get_rcu(rt->dst.dev);
 875         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 876                 rcu_read_unlock();
 877                 return;
 878         }
 879         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 880         vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
 881         rcu_read_unlock();
 882
 883         net = dev_net(rt->dst.dev);
 884         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
 885         if (!peer) {
 886                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 887                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 888                 return;
 889         }
 890
 891         /* No redirected packets during ip_rt_redirect_silence;
 892          * reset the algorithm.
 893          */
 894         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
 895                 peer->rate_tokens = 0;
 896                 peer->n_redirects = 0;
 897         }
 898
 899         /* Too many ignored redirects; do not send anything
 900          * set dst.rate_last to the last seen redirected packet.
 901          */
 902         if (peer->n_redirects >= ip_rt_redirect_number) {
 903                 peer->rate_last = jiffies;
 904                 goto out_put_peer;
 905         }
 906
 907         /* Check for load limit; set rate_last to the latest sent
 908          * redirect.
 909          */
 910         if (peer->n_redirects == 0 ||
 911             time_after(jiffies,
 912                        (peer->rate_last +
 913                         (ip_rt_redirect_load << peer->n_redirects)))) {
 914                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 915
 916                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 917                 peer->rate_last = jiffies;
 918                 ++peer->n_redirects;
 919 #ifdef CONFIG_IP_ROUTE_VERBOSE
 920                 if (log_martians &&
 921                     peer->n_redirects == ip_rt_redirect_number)
 922                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 923                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 924                                              &ip_hdr(skb)->daddr, &gw);
 925 #endif
 926         }
 927 out_put_peer:
 928         inet_putpeer(peer);
 929 }
 930
 931 static int ip_error(struct sk_buff *skb)
 932 {
 933         struct rtable *rt = skb_rtable(skb);
 934         struct net_device *dev = skb->dev;
 935         struct in_device *in_dev;
 936         struct inet_peer *peer;
 937         unsigned long now;
 938         struct net *net;
 939         bool send;
 940         int code;
 941
 942         if (netif_is_l3_master(skb->dev)) {
 943                 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
 944                 if (!dev)
 945                         goto out;
 946         }
 947
 948         in_dev = __in_dev_get_rcu(dev);
 949
 950         /* IP on this device is disabled. */
 951         if (!in_dev)
 952                 goto out;
 953
 954         net = dev_net(rt->dst.dev);
 955         if (!IN_DEV_FORWARD(in_dev)) {
 956                 switch (rt->dst.error) {
 957                 case EHOSTUNREACH:
 958                         __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
 959                         break;
 960
 961                 case ENETUNREACH:
 962                         __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 963                         break;
 964                 }
 965                 goto out;
 966         }
 967
 968         switch (rt->dst.error) {
 969         case EINVAL:
 970         default:
 971                 goto out;
 972         case EHOSTUNREACH:
 973                 code = ICMP_HOST_UNREACH;
 974                 break;
 975         case ENETUNREACH:
 976                 code = ICMP_NET_UNREACH;
 977                 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 978                 break;
 979         case EACCES:
 980                 code = ICMP_PKT_FILTERED;
 981                 break;
 982         }
 983
 984         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
 985                                l3mdev_master_ifindex(skb->dev), 1);
 986
 987         send = true;
 988         if (peer) {
 989                 now = jiffies;
 990                 peer->rate_tokens += now - peer->rate_last;
 991                 if (peer->rate_tokens > ip_rt_error_burst)
 992                         peer->rate_tokens = ip_rt_error_burst;
 993                 peer->rate_last = now;
 994                 if (peer->rate_tokens >= ip_rt_error_cost)
 995                         peer->rate_tokens -= ip_rt_error_cost;
 996                 else
 997                         send = false;
 998                 inet_putpeer(peer);
 999         }
1000         if (send)
1001                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1002
1003 out:    kfree_skb(skb);
1004         return 0;
1005 }
1006
1007 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1008 {
1009         struct dst_entry *dst = &rt->dst;
1010         struct net *net = dev_net(dst->dev);
1011         struct fib_result res;
1012         bool lock = false;
1013         u32 old_mtu;
1014
1015         if (ip_mtu_locked(dst))
1016                 return;
1017
1018         old_mtu = ipv4_mtu(dst);
1019         if (old_mtu < mtu)
1020                 return;
1021
1022         if (mtu < net->ipv4.ip_rt_min_pmtu) {
1023                 lock = true;
1024                 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1025         }
1026
1027         if (rt->rt_pmtu == mtu && !lock &&
1028             time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1029                 return;
1030
1031         rcu_read_lock();
1032         if (fib_lookup(net, fl4, &res, 0) == 0) {
1033                 struct fib_nh_common *nhc;
1034
1035                 fib_select_path(net, &res, fl4, NULL);
1036                 nhc = FIB_RES_NHC(res);
1037                 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1038                                       jiffies + net->ipv4.ip_rt_mtu_expires);
1039         }
1040         rcu_read_unlock();
1041 }
1042
1043 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1044                               struct sk_buff *skb, u32 mtu,
1045                               bool confirm_neigh)
1046 {
1047         struct rtable *rt = (struct rtable *) dst;
1048         struct flowi4 fl4;
1049
1050         ip_rt_build_flow_key(&fl4, sk, skb);
1051
1052         /* Don't make lookup fail for bridged encapsulations */
1053         if (skb && netif_is_any_bridge_port(skb->dev))
1054                 fl4.flowi4_oif = 0;
1055
1056         __ip_rt_update_pmtu(rt, &fl4, mtu);
1057 }
1058
1059 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1060                       int oif, u8 protocol)
1061 {
1062         const struct iphdr *iph = (const struct iphdr *)skb->data;
1063         struct flowi4 fl4;
1064         struct rtable *rt;
1065         u32 mark = IP4_REPLY_MARK(net, skb->mark);
1066
1067         __build_flow_key(net, &fl4, NULL, iph, oif,
1068                          RT_TOS(iph->tos), protocol, mark, 0);
1069         rt = __ip_route_output_key(net, &fl4);
1070         if (!IS_ERR(rt)) {
1071                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1072                 ip_rt_put(rt);
1073         }
1074 }
1075 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1076
1077 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1078 {
1079         const struct iphdr *iph = (const struct iphdr *)skb->data;
1080         struct flowi4 fl4;
1081         struct rtable *rt;
1082
1083         __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1084
1085         if (!fl4.flowi4_mark)
1086                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1087
1088         rt = __ip_route_output_key(sock_net(sk), &fl4);
1089         if (!IS_ERR(rt)) {
1090                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1091                 ip_rt_put(rt);
1092         }
1093 }
1094
1095 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1096 {
1097         const struct iphdr *iph = (const struct iphdr *)skb->data;
1098         struct flowi4 fl4;
1099         struct rtable *rt;
1100         struct dst_entry *odst = NULL;
1101         bool new = false;
1102         struct net *net = sock_net(sk);
1103
1104         bh_lock_sock(sk);
1105
1106         if (!ip_sk_accept_pmtu(sk))
1107                 goto out;
1108
1109         odst = sk_dst_get(sk);
1110
1111         if (sock_owned_by_user(sk) || !odst) {
1112                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1113                 goto out;
1114         }
1115
1116         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1117
1118         rt = (struct rtable *)odst;
1119         if (odst->obsolete && !odst->ops->check(odst, 0)) {
1120                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1121                 if (IS_ERR(rt))
1122                         goto out;
1123
1124                 new = true;
1125         }
1126
1127         __ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu);
1128
1129         if (!dst_check(&rt->dst, 0)) {
1130                 if (new)
1131                         dst_release(&rt->dst);
1132
1133                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1134                 if (IS_ERR(rt))
1135                         goto out;
1136
1137                 new = true;
1138         }
1139
1140         if (new)
1141                 sk_dst_set(sk, &rt->dst);
1142
1143 out:
1144         bh_unlock_sock(sk);
1145         dst_release(odst);
1146 }
1147 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1148
1149 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1150                    int oif, u8 protocol)
1151 {
1152         const struct iphdr *iph = (const struct iphdr *)skb->data;
1153         struct flowi4 fl4;
1154         struct rtable *rt;
1155
1156         __build_flow_key(net, &fl4, NULL, iph, oif,
1157                          RT_TOS(iph->tos), protocol, 0, 0);
1158         rt = __ip_route_output_key(net, &fl4);
1159         if (!IS_ERR(rt)) {
1160                 __ip_do_redirect(rt, skb, &fl4, false);
1161                 ip_rt_put(rt);
1162         }
1163 }
1164 EXPORT_SYMBOL_GPL(ipv4_redirect);
1165
1166 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1167 {
1168         const struct iphdr *iph = (const struct iphdr *)skb->data;
1169         struct flowi4 fl4;
1170         struct rtable *rt;
1171         struct net *net = sock_net(sk);
1172
1173         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1174         rt = __ip_route_output_key(net, &fl4);
1175         if (!IS_ERR(rt)) {
1176                 __ip_do_redirect(rt, skb, &fl4, false);
1177                 ip_rt_put(rt);
1178         }
1179 }
1180 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1181
1182 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1183                                                          u32 cookie)
1184 {
1185         struct rtable *rt = (struct rtable *) dst;
1186
1187         /* All IPV4 dsts are created with ->obsolete set to the value
1188          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1189          * into this function always.
1190          *
1191          * When a PMTU/redirect information update invalidates a route,
1192          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1193          * DST_OBSOLETE_DEAD.
1194          */
1195         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1196                 return NULL;
1197         return dst;
1198 }
1199 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1200
1201 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1202 {
1203         struct ip_options opt;
1204         int res;
1205
1206         /* Recompile ip options since IPCB may not be valid anymore.
1207          * Also check we have a reasonable ipv4 header.
1208          */
1209         if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1210             ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1211                 return;
1212
1213         memset(&opt, 0, sizeof(opt));
1214         if (ip_hdr(skb)->ihl > 5) {
1215                 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1216                         return;
1217                 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1218
1219                 rcu_read_lock();
1220                 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1221                 rcu_read_unlock();
1222
1223                 if (res)
1224                         return;
1225         }
1226         __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1227 }
1228
1229 static void ipv4_link_failure(struct sk_buff *skb)
1230 {
1231         struct rtable *rt;
1232
1233         ipv4_send_dest_unreach(skb);
1234
1235         rt = skb_rtable(skb);
1236         if (rt)
1237                 dst_set_expires(&rt->dst, 0);
1238 }
1239
1240 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1241 {
1242         pr_debug("%s: %pI4 -> %pI4, %s\n",
1243                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1244                  skb->dev ? skb->dev->name : "?");
1245         kfree_skb(skb);
1246         WARN_ON(1);
1247         return 0;
1248 }
1249
1250 /*
1251  * We do not cache source address of outgoing interface,
1252  * because it is used only by IP RR, TS and SRR options,
1253  * so that it out of fast path.
1254  *
1255  * BTW remember: "addr" is allowed to be not aligned
1256  * in IP options!
1257  */
1258
1259 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1260 {
1261         __be32 src;
1262
1263         if (rt_is_output_route(rt))
1264                 src = ip_hdr(skb)->saddr;
1265         else {
1266                 struct fib_result res;
1267                 struct iphdr *iph = ip_hdr(skb);
1268                 struct flowi4 fl4 = {
1269                         .daddr = iph->daddr,
1270                         .saddr = iph->saddr,
1271                         .flowi4_tos = RT_TOS(iph->tos),
1272                         .flowi4_oif = rt->dst.dev->ifindex,
1273                         .flowi4_iif = skb->dev->ifindex,
1274                         .flowi4_mark = skb->mark,
1275                 };
1276
1277                 rcu_read_lock();
1278                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1279                         src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1280                 else
1281                         src = inet_select_addr(rt->dst.dev,
1282                                                rt_nexthop(rt, iph->daddr),
1283                                                RT_SCOPE_UNIVERSE);
1284                 rcu_read_unlock();
1285         }
1286         memcpy(addr, &src, 4);
1287 }
1288
1289 #ifdef CONFIG_IP_ROUTE_CLASSID
1290 static void set_class_tag(struct rtable *rt, u32 tag)
1291 {
1292         if (!(rt->dst.tclassid & 0xFFFF))
1293                 rt->dst.tclassid |= tag & 0xFFFF;
1294         if (!(rt->dst.tclassid & 0xFFFF0000))
1295                 rt->dst.tclassid |= tag & 0xFFFF0000;
1296 }
1297 #endif
1298
1299 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1300 {
1301         unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1302         unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1303                                     ip_rt_min_advmss);
1304
1305         return min(advmss, IPV4_MAX_PMTU - header_size);
1306 }
1307
1308 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1309 {
1310         return ip_dst_mtu_maybe_forward(dst, false);
1311 }
1312 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1313
1314 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1315 {
1316         struct fnhe_hash_bucket *hash;
1317         struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1318         u32 hval = fnhe_hashfun(daddr);
1319
1320         spin_lock_bh(&fnhe_lock);
1321
1322         hash = rcu_dereference_protected(nhc->nhc_exceptions,
1323                                          lockdep_is_held(&fnhe_lock));
1324         hash += hval;
1325
1326         fnhe_p = &hash->chain;
1327         fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1328         while (fnhe) {
1329                 if (fnhe->fnhe_daddr == daddr) {
1330                         rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1331                                 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1332                         /* set fnhe_daddr to 0 to ensure it won't bind with
1333                          * new dsts in rt_bind_exception().
1334                          */
1335                         fnhe->fnhe_daddr = 0;
1336                         fnhe_flush_routes(fnhe);
1337                         kfree_rcu(fnhe, rcu);
1338                         break;
1339                 }
1340                 fnhe_p = &fnhe->fnhe_next;
1341                 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1342                                                  lockdep_is_held(&fnhe_lock));
1343         }
1344
1345         spin_unlock_bh(&fnhe_lock);
1346 }
1347
1348 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1349                                                __be32 daddr)
1350 {
1351         struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1352         struct fib_nh_exception *fnhe;
1353         u32 hval;
1354
1355         if (!hash)
1356                 return NULL;
1357
1358         hval = fnhe_hashfun(daddr);
1359
1360         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1361              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1362                 if (fnhe->fnhe_daddr == daddr) {
1363                         if (fnhe->fnhe_expires &&
1364                             time_after(jiffies, fnhe->fnhe_expires)) {
1365                                 ip_del_fnhe(nhc, daddr);
1366                                 break;
1367                         }
1368                         return fnhe;
1369                 }
1370         }
1371         return NULL;
1372 }
1373
1374 /* MTU selection:
1375  * 1. mtu on route is locked - use it
1376  * 2. mtu from nexthop exception
1377  * 3. mtu from egress device
1378  */
1379
1380 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1381 {
1382         struct fib_nh_common *nhc = res->nhc;
1383         struct net_device *dev = nhc->nhc_dev;
1384         struct fib_info *fi = res->fi;
1385         u32 mtu = 0;
1386
1387         if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1388             fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1389                 mtu = fi->fib_mtu;
1390
1391         if (likely(!mtu)) {
1392                 struct fib_nh_exception *fnhe;
1393
1394                 fnhe = find_exception(nhc, daddr);
1395                 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1396                         mtu = fnhe->fnhe_pmtu;
1397         }
1398
1399         if (likely(!mtu))
1400                 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1401
1402         return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1403 }
1404
1405 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1406                               __be32 daddr, const bool do_cache)
1407 {
1408         bool ret = false;
1409
1410         spin_lock_bh(&fnhe_lock);
1411
1412         if (daddr == fnhe->fnhe_daddr) {
1413                 struct rtable __rcu **porig;
1414                 struct rtable *orig;
1415                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1416
1417                 if (rt_is_input_route(rt))
1418                         porig = &fnhe->fnhe_rth_input;
1419                 else
1420                         porig = &fnhe->fnhe_rth_output;
1421                 orig = rcu_dereference(*porig);
1422
1423                 if (fnhe->fnhe_genid != genid) {
1424                         fnhe->fnhe_genid = genid;
1425                         fnhe->fnhe_gw = 0;
1426                         fnhe->fnhe_pmtu = 0;
1427                         fnhe->fnhe_expires = 0;
1428                         fnhe->fnhe_mtu_locked = false;
1429                         fnhe_flush_routes(fnhe);
1430                         orig = NULL;
1431                 }
1432                 fill_route_from_fnhe(rt, fnhe);
1433                 if (!rt->rt_gw4) {
1434                         rt->rt_gw4 = daddr;
1435                         rt->rt_gw_family = AF_INET;
1436                 }
1437
1438                 if (do_cache) {
1439                         dst_hold(&rt->dst);
1440                         rcu_assign_pointer(*porig, rt);
1441                         if (orig) {
1442                                 dst_dev_put(&orig->dst);
1443                                 dst_release(&orig->dst);
1444                         }
1445                         ret = true;
1446                 }
1447
1448                 fnhe->fnhe_stamp = jiffies;
1449         }
1450         spin_unlock_bh(&fnhe_lock);
1451
1452         return ret;
1453 }
1454
1455 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1456 {
1457         struct rtable *orig, *prev, **p;
1458         bool ret = true;
1459
1460         if (rt_is_input_route(rt)) {
1461                 p = (struct rtable **)&nhc->nhc_rth_input;
1462         } else {
1463                 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1464         }
1465         orig = *p;
1466
1467         /* hold dst before doing cmpxchg() to avoid race condition
1468          * on this dst
1469          */
1470         dst_hold(&rt->dst);
1471         prev = cmpxchg(p, orig, rt);
1472         if (prev == orig) {
1473                 if (orig) {
1474                         rt_add_uncached_list(orig);
1475                         dst_release(&orig->dst);
1476                 }
1477         } else {
1478                 dst_release(&rt->dst);
1479                 ret = false;
1480         }
1481
1482         return ret;
1483 }
1484
1485 struct uncached_list {
1486         spinlock_t              lock;
1487         struct list_head        head;
1488 };
1489
1490 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1491
1492 void rt_add_uncached_list(struct rtable *rt)
1493 {
1494         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1495
1496         rt->rt_uncached_list = ul;
1497
1498         spin_lock_bh(&ul->lock);
1499         list_add_tail(&rt->rt_uncached, &ul->head);
1500         spin_unlock_bh(&ul->lock);
1501 }
1502
1503 void rt_del_uncached_list(struct rtable *rt)
1504 {
1505         if (!list_empty(&rt->rt_uncached)) {
1506                 struct uncached_list *ul = rt->rt_uncached_list;
1507
1508                 spin_lock_bh(&ul->lock);
1509                 list_del(&rt->rt_uncached);
1510                 spin_unlock_bh(&ul->lock);
1511         }
1512 }
1513
1514 static void ipv4_dst_destroy(struct dst_entry *dst)
1515 {
1516         struct rtable *rt = (struct rtable *)dst;
1517
1518         ip_dst_metrics_put(dst);
1519         rt_del_uncached_list(rt);
1520 }
1521
1522 void rt_flush_dev(struct net_device *dev)
1523 {
1524         struct rtable *rt;
1525         int cpu;
1526
1527         for_each_possible_cpu(cpu) {
1528                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1529
1530                 spin_lock_bh(&ul->lock);
1531                 list_for_each_entry(rt, &ul->head, rt_uncached) {
1532                         if (rt->dst.dev != dev)
1533                                 continue;
1534                         rt->dst.dev = blackhole_netdev;
1535                         dev_replace_track(dev, blackhole_netdev,
1536                                           &rt->dst.dev_tracker,
1537                                           GFP_ATOMIC);
1538                 }
1539                 spin_unlock_bh(&ul->lock);
1540         }
1541 }
1542
1543 static bool rt_cache_valid(const struct rtable *rt)
1544 {
1545         return  rt &&
1546                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1547                 !rt_is_expired(rt);
1548 }
1549
1550 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1551                            const struct fib_result *res,
1552                            struct fib_nh_exception *fnhe,
1553                            struct fib_info *fi, u16 type, u32 itag,
1554                            const bool do_cache)
1555 {
1556         bool cached = false;
1557
1558         if (fi) {
1559                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1560
1561                 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1562                         rt->rt_uses_gateway = 1;
1563                         rt->rt_gw_family = nhc->nhc_gw_family;
1564                         /* only INET and INET6 are supported */
1565                         if (likely(nhc->nhc_gw_family == AF_INET))
1566                                 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1567                         else
1568                                 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1569                 }
1570
1571                 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1572
1573 #ifdef CONFIG_IP_ROUTE_CLASSID
1574                 if (nhc->nhc_family == AF_INET) {
1575                         struct fib_nh *nh;
1576
1577                         nh = container_of(nhc, struct fib_nh, nh_common);
1578                         rt->dst.tclassid = nh->nh_tclassid;
1579                 }
1580 #endif
1581                 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1582                 if (unlikely(fnhe))
1583                         cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1584                 else if (do_cache)
1585                         cached = rt_cache_route(nhc, rt);
1586                 if (unlikely(!cached)) {
1587                         /* Routes we intend to cache in nexthop exception or
1588                          * FIB nexthop have the DST_NOCACHE bit clear.
1589                          * However, if we are unsuccessful at storing this
1590                          * route into the cache we really need to set it.
1591                          */
1592                         if (!rt->rt_gw4) {
1593                                 rt->rt_gw_family = AF_INET;
1594                                 rt->rt_gw4 = daddr;
1595                         }
1596                         rt_add_uncached_list(rt);
1597                 }
1598         } else
1599                 rt_add_uncached_list(rt);
1600
1601 #ifdef CONFIG_IP_ROUTE_CLASSID
1602 #ifdef CONFIG_IP_MULTIPLE_TABLES
1603         set_class_tag(rt, res->tclassid);
1604 #endif
1605         set_class_tag(rt, itag);
1606 #endif
1607 }
1608
1609 struct rtable *rt_dst_alloc(struct net_device *dev,
1610                             unsigned int flags, u16 type,
1611                             bool nopolicy, bool noxfrm)
1612 {
1613         struct rtable *rt;
1614
1615         rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1616                        (nopolicy ? DST_NOPOLICY : 0) |
1617                        (noxfrm ? DST_NOXFRM : 0));
1618
1619         if (rt) {
1620                 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1621                 rt->rt_flags = flags;
1622                 rt->rt_type = type;
1623                 rt->rt_is_input = 0;
1624                 rt->rt_iif = 0;
1625                 rt->rt_pmtu = 0;
1626                 rt->rt_mtu_locked = 0;
1627                 rt->rt_uses_gateway = 0;
1628                 rt->rt_gw_family = 0;
1629                 rt->rt_gw4 = 0;
1630                 INIT_LIST_HEAD(&rt->rt_uncached);
1631
1632                 rt->dst.output = ip_output;
1633                 if (flags & RTCF_LOCAL)
1634                         rt->dst.input = ip_local_deliver;
1635         }
1636
1637         return rt;
1638 }
1639 EXPORT_SYMBOL(rt_dst_alloc);
1640
1641 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1642 {
1643         struct rtable *new_rt;
1644
1645         new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1646                            rt->dst.flags);
1647
1648         if (new_rt) {
1649                 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1650                 new_rt->rt_flags = rt->rt_flags;
1651                 new_rt->rt_type = rt->rt_type;
1652                 new_rt->rt_is_input = rt->rt_is_input;
1653                 new_rt->rt_iif = rt->rt_iif;
1654                 new_rt->rt_pmtu = rt->rt_pmtu;
1655                 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1656                 new_rt->rt_gw_family = rt->rt_gw_family;
1657                 if (rt->rt_gw_family == AF_INET)
1658                         new_rt->rt_gw4 = rt->rt_gw4;
1659                 else if (rt->rt_gw_family == AF_INET6)
1660                         new_rt->rt_gw6 = rt->rt_gw6;
1661                 INIT_LIST_HEAD(&new_rt->rt_uncached);
1662
1663                 new_rt->dst.input = rt->dst.input;
1664                 new_rt->dst.output = rt->dst.output;
1665                 new_rt->dst.error = rt->dst.error;
1666                 new_rt->dst.lastuse = jiffies;
1667                 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1668         }
1669         return new_rt;
1670 }
1671 EXPORT_SYMBOL(rt_dst_clone);
1672
1673 /* called in rcu_read_lock() section */
1674 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1675                           u8 tos, struct net_device *dev,
1676                           struct in_device *in_dev, u32 *itag)
1677 {
1678         int err;
1679
1680         /* Primary sanity checks. */
1681         if (!in_dev)
1682                 return -EINVAL;
1683
1684         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1685             skb->protocol != htons(ETH_P_IP))
1686                 return -EINVAL;
1687
1688         if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1689                 return -EINVAL;
1690
1691         if (ipv4_is_zeronet(saddr)) {
1692                 if (!ipv4_is_local_multicast(daddr) &&
1693                     ip_hdr(skb)->protocol != IPPROTO_IGMP)
1694                         return -EINVAL;
1695         } else {
1696                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1697                                           in_dev, itag);
1698                 if (err < 0)
1699                         return err;
1700         }
1701         return 0;
1702 }
1703
1704 /* called in rcu_read_lock() section */
1705 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1706                              u8 tos, struct net_device *dev, int our)
1707 {
1708         struct in_device *in_dev = __in_dev_get_rcu(dev);
1709         unsigned int flags = RTCF_MULTICAST;
1710         struct rtable *rth;
1711         u32 itag = 0;
1712         int err;
1713
1714         err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1715         if (err)
1716                 return err;
1717
1718         if (our)
1719                 flags |= RTCF_LOCAL;
1720
1721         rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1722                            IN_DEV_ORCONF(in_dev, NOPOLICY), false);
1723         if (!rth)
1724                 return -ENOBUFS;
1725
1726 #ifdef CONFIG_IP_ROUTE_CLASSID
1727         rth->dst.tclassid = itag;
1728 #endif
1729         rth->dst.output = ip_rt_bug;
1730         rth->rt_is_input= 1;
1731
1732 #ifdef CONFIG_IP_MROUTE
1733         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1734                 rth->dst.input = ip_mr_input;
1735 #endif
1736         RT_CACHE_STAT_INC(in_slow_mc);
1737
1738         skb_dst_set(skb, &rth->dst);
1739         return 0;
1740 }
1741
1742
1743 static void ip_handle_martian_source(struct net_device *dev,
1744                                      struct in_device *in_dev,
1745                                      struct sk_buff *skb,
1746                                      __be32 daddr,
1747                                      __be32 saddr)
1748 {
1749         RT_CACHE_STAT_INC(in_martian_src);
1750 #ifdef CONFIG_IP_ROUTE_VERBOSE
1751         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1752                 /*
1753                  *      RFC1812 recommendation, if source is martian,
1754                  *      the only hint is MAC header.
1755                  */
1756                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1757                         &daddr, &saddr, dev->name);
1758                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1759                         print_hex_dump(KERN_WARNING, "ll header: ",
1760                                        DUMP_PREFIX_OFFSET, 16, 1,
1761                                        skb_mac_header(skb),
1762                                        dev->hard_header_len, false);
1763                 }
1764         }
1765 #endif
1766 }
1767
1768 /* called in rcu_read_lock() section */
1769 static int __mkroute_input(struct sk_buff *skb,
1770                            const struct fib_result *res,
1771                            struct in_device *in_dev,
1772                            __be32 daddr, __be32 saddr, u32 tos)
1773 {
1774         struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1775         struct net_device *dev = nhc->nhc_dev;
1776         struct fib_nh_exception *fnhe;
1777         struct rtable *rth;
1778         int err;
1779         struct in_device *out_dev;
1780         bool do_cache;
1781         u32 itag = 0;
1782
1783         /* get a working reference to the output device */
1784         out_dev = __in_dev_get_rcu(dev);
1785         if (!out_dev) {
1786                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1787                 return -EINVAL;
1788         }
1789
1790         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1791                                   in_dev->dev, in_dev, &itag);
1792         if (err < 0) {
1793                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1794                                          saddr);
1795
1796                 goto cleanup;
1797         }
1798
1799         do_cache = res->fi && !itag;
1800         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1801             skb->protocol == htons(ETH_P_IP)) {
1802                 __be32 gw;
1803
1804                 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1805                 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1806                     inet_addr_onlink(out_dev, saddr, gw))
1807                         IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1808         }
1809
1810         if (skb->protocol != htons(ETH_P_IP)) {
1811                 /* Not IP (i.e. ARP). Do not create route, if it is
1812                  * invalid for proxy arp. DNAT routes are always valid.
1813                  *
1814                  * Proxy arp feature have been extended to allow, ARP
1815                  * replies back to the same interface, to support
1816                  * Private VLAN switch technologies. See arp.c.
1817                  */
1818                 if (out_dev == in_dev &&
1819                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1820                         err = -EINVAL;
1821                         goto cleanup;
1822                 }
1823         }
1824
1825         fnhe = find_exception(nhc, daddr);
1826         if (do_cache) {
1827                 if (fnhe)
1828                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1829                 else
1830                         rth = rcu_dereference(nhc->nhc_rth_input);
1831                 if (rt_cache_valid(rth)) {
1832                         skb_dst_set_noref(skb, &rth->dst);
1833                         goto out;
1834                 }
1835         }
1836
1837         rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1838                            IN_DEV_ORCONF(in_dev, NOPOLICY),
1839                            IN_DEV_ORCONF(out_dev, NOXFRM));
1840         if (!rth) {
1841                 err = -ENOBUFS;
1842                 goto cleanup;
1843         }
1844
1845         rth->rt_is_input = 1;
1846         RT_CACHE_STAT_INC(in_slow_tot);
1847
1848         rth->dst.input = ip_forward;
1849
1850         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1851                        do_cache);
1852         lwtunnel_set_redirect(&rth->dst);
1853         skb_dst_set(skb, &rth->dst);
1854 out:
1855         err = 0;
1856  cleanup:
1857         return err;
1858 }
1859
1860 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1861 /* To make ICMP packets follow the right flow, the multipath hash is
1862  * calculated from the inner IP addresses.
1863  */
1864 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1865                                  struct flow_keys *hash_keys)
1866 {
1867         const struct iphdr *outer_iph = ip_hdr(skb);
1868         const struct iphdr *key_iph = outer_iph;
1869         const struct iphdr *inner_iph;
1870         const struct icmphdr *icmph;
1871         struct iphdr _inner_iph;
1872         struct icmphdr _icmph;
1873
1874         if (likely(outer_iph->protocol != IPPROTO_ICMP))
1875                 goto out;
1876
1877         if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1878                 goto out;
1879
1880         icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1881                                    &_icmph);
1882         if (!icmph)
1883                 goto out;
1884
1885         if (!icmp_is_err(icmph->type))
1886                 goto out;
1887
1888         inner_iph = skb_header_pointer(skb,
1889                                        outer_iph->ihl * 4 + sizeof(_icmph),
1890                                        sizeof(_inner_iph), &_inner_iph);
1891         if (!inner_iph)
1892                 goto out;
1893
1894         key_iph = inner_iph;
1895 out:
1896         hash_keys->addrs.v4addrs.src = key_iph->saddr;
1897         hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1898 }
1899
1900 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1901                                            const struct sk_buff *skb,
1902                                            bool *p_has_inner)
1903 {
1904         u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
1905         struct flow_keys keys, hash_keys;
1906
1907         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1908                 return 0;
1909
1910         memset(&hash_keys, 0, sizeof(hash_keys));
1911         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1912
1913         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1914         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1915                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1916         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1917                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1918         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1919                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1920         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1921                 hash_keys.ports.src = keys.ports.src;
1922         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1923                 hash_keys.ports.dst = keys.ports.dst;
1924
1925         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1926         return flow_hash_from_keys(&hash_keys);
1927 }
1928
1929 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1930                                            const struct sk_buff *skb,
1931                                            bool has_inner)
1932 {
1933         u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
1934         struct flow_keys keys, hash_keys;
1935
1936         /* We assume the packet carries an encapsulation, but if none was
1937          * encountered during dissection of the outer flow, then there is no
1938          * point in calling the flow dissector again.
1939          */
1940         if (!has_inner)
1941                 return 0;
1942
1943         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1944                 return 0;
1945
1946         memset(&hash_keys, 0, sizeof(hash_keys));
1947         skb_flow_dissect_flow_keys(skb, &keys, 0);
1948
1949         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1950                 return 0;
1951
1952         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1953                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1954                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1955                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1956                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1957                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1958         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1959                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1960                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1961                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1962                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1963                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1964                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1965                         hash_keys.tags.flow_label = keys.tags.flow_label;
1966         }
1967
1968         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1969                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1970         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1971                 hash_keys.ports.src = keys.ports.src;
1972         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1973                 hash_keys.ports.dst = keys.ports.dst;
1974
1975         return flow_hash_from_keys(&hash_keys);
1976 }
1977
1978 static u32 fib_multipath_custom_hash_skb(const struct net *net,
1979                                          const struct sk_buff *skb)
1980 {
1981         u32 mhash, mhash_inner;
1982         bool has_inner = true;
1983
1984         mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
1985         mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
1986
1987         return jhash_2words(mhash, mhash_inner, 0);
1988 }
1989
1990 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
1991                                          const struct flowi4 *fl4)
1992 {
1993         u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
1994         struct flow_keys hash_keys;
1995
1996         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1997                 return 0;
1998
1999         memset(&hash_keys, 0, sizeof(hash_keys));
2000         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2001         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2002                 hash_keys.addrs.v4addrs.src = fl4->saddr;
2003         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2004                 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2005         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2006                 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2007         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2008                 hash_keys.ports.src = fl4->fl4_sport;
2009         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2010                 hash_keys.ports.dst = fl4->fl4_dport;
2011
2012         return flow_hash_from_keys(&hash_keys);
2013 }
2014
2015 /* if skb is set it will be used and fl4 can be NULL */
2016 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2017                        const struct sk_buff *skb, struct flow_keys *flkeys)
2018 {
2019         u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2020         struct flow_keys hash_keys;
2021         u32 mhash = 0;
2022
2023         switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
2024         case 0:
2025                 memset(&hash_keys, 0, sizeof(hash_keys));
2026                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2027                 if (skb) {
2028                         ip_multipath_l3_keys(skb, &hash_keys);
2029                 } else {
2030                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2031                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2032                 }
2033                 mhash = flow_hash_from_keys(&hash_keys);
2034                 break;
2035         case 1:
2036                 /* skb is currently provided only when forwarding */
2037                 if (skb) {
2038                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2039                         struct flow_keys keys;
2040
2041                         /* short-circuit if we already have L4 hash present */
2042                         if (skb->l4_hash)
2043                                 return skb_get_hash_raw(skb) >> 1;
2044
2045                         memset(&hash_keys, 0, sizeof(hash_keys));
2046
2047                         if (!flkeys) {
2048                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2049                                 flkeys = &keys;
2050                         }
2051
2052                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2053                         hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2054                         hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2055                         hash_keys.ports.src = flkeys->ports.src;
2056                         hash_keys.ports.dst = flkeys->ports.dst;
2057                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2058                 } else {
2059                         memset(&hash_keys, 0, sizeof(hash_keys));
2060                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2061                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2062                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2063                         hash_keys.ports.src = fl4->fl4_sport;
2064                         hash_keys.ports.dst = fl4->fl4_dport;
2065                         hash_keys.basic.ip_proto = fl4->flowi4_proto;
2066                 }
2067                 mhash = flow_hash_from_keys(&hash_keys);
2068                 break;
2069         case 2:
2070                 memset(&hash_keys, 0, sizeof(hash_keys));
2071                 /* skb is currently provided only when forwarding */
2072                 if (skb) {
2073                         struct flow_keys keys;
2074
2075                         skb_flow_dissect_flow_keys(skb, &keys, 0);
2076                         /* Inner can be v4 or v6 */
2077                         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2078                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2079                                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2080                                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2081                         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2082                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2083                                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2084                                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2085                                 hash_keys.tags.flow_label = keys.tags.flow_label;
2086                                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2087                         } else {
2088                                 /* Same as case 0 */
2089                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2090                                 ip_multipath_l3_keys(skb, &hash_keys);
2091                         }
2092                 } else {
2093                         /* Same as case 0 */
2094                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2095                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2096                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2097                 }
2098                 mhash = flow_hash_from_keys(&hash_keys);
2099                 break;
2100         case 3:
2101                 if (skb)
2102                         mhash = fib_multipath_custom_hash_skb(net, skb);
2103                 else
2104                         mhash = fib_multipath_custom_hash_fl4(net, fl4);
2105                 break;
2106         }
2107
2108         if (multipath_hash)
2109                 mhash = jhash_2words(mhash, multipath_hash, 0);
2110
2111         return mhash >> 1;
2112 }
2113 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2114
2115 static int ip_mkroute_input(struct sk_buff *skb,
2116                             struct fib_result *res,
2117                             struct in_device *in_dev,
2118                             __be32 daddr, __be32 saddr, u32 tos,
2119                             struct flow_keys *hkeys)
2120 {
2121 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2122         if (res->fi && fib_info_num_path(res->fi) > 1) {
2123                 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2124
2125                 fib_select_multipath(res, h);
2126         }
2127 #endif
2128
2129         /* create a routing cache entry */
2130         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2131 }
2132
2133 /* Implements all the saddr-related checks as ip_route_input_slow(),
2134  * assuming daddr is valid and the destination is not a local broadcast one.
2135  * Uses the provided hint instead of performing a route lookup.
2136  */
2137 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2138                       u8 tos, struct net_device *dev,
2139                       const struct sk_buff *hint)
2140 {
2141         struct in_device *in_dev = __in_dev_get_rcu(dev);
2142         struct rtable *rt = skb_rtable(hint);
2143         struct net *net = dev_net(dev);
2144         int err = -EINVAL;
2145         u32 tag = 0;
2146
2147         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2148                 goto martian_source;
2149
2150         if (ipv4_is_zeronet(saddr))
2151                 goto martian_source;
2152
2153         if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2154                 goto martian_source;
2155
2156         if (rt->rt_type != RTN_LOCAL)
2157                 goto skip_validate_source;
2158
2159         tos &= IPTOS_RT_MASK;
2160         err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2161         if (err < 0)
2162                 goto martian_source;
2163
2164 skip_validate_source:
2165         skb_dst_copy(skb, hint);
2166         return 0;
2167
2168 martian_source:
2169         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2170         return err;
2171 }
2172
2173 /* get device for dst_alloc with local routes */
2174 static struct net_device *ip_rt_get_dev(struct net *net,
2175                                         const struct fib_result *res)
2176 {
2177         struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2178         struct net_device *dev = NULL;
2179
2180         if (nhc)
2181                 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2182
2183         return dev ? : net->loopback_dev;
2184 }
2185
2186 /*
2187  *      NOTE. We drop all the packets that has local source
2188  *      addresses, because every properly looped back packet
2189  *      must have correct destination already attached by output routine.
2190  *      Changes in the enforced policies must be applied also to
2191  *      ip_route_use_hint().
2192  *
2193  *      Such approach solves two big problems:
2194  *      1. Not simplex devices are handled properly.
2195  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2196  *      called with rcu_read_lock()
2197  */
2198
2199 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2200                                u8 tos, struct net_device *dev,
2201                                struct fib_result *res)
2202 {
2203         struct in_device *in_dev = __in_dev_get_rcu(dev);
2204         struct flow_keys *flkeys = NULL, _flkeys;
2205         struct net    *net = dev_net(dev);
2206         struct ip_tunnel_info *tun_info;
2207         int             err = -EINVAL;
2208         unsigned int    flags = 0;
2209         u32             itag = 0;
2210         struct rtable   *rth;
2211         struct flowi4   fl4;
2212         bool do_cache = true;
2213
2214         /* IP on this device is disabled. */
2215
2216         if (!in_dev)
2217                 goto out;
2218
2219         /* Check for the most weird martians, which can be not detected
2220          * by fib_lookup.
2221          */
2222
2223         tun_info = skb_tunnel_info(skb);
2224         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2225                 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2226         else
2227                 fl4.flowi4_tun_key.tun_id = 0;
2228         skb_dst_drop(skb);
2229
2230         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2231                 goto martian_source;
2232
2233         res->fi = NULL;
2234         res->table = NULL;
2235         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2236                 goto brd_input;
2237
2238         /* Accept zero addresses only to limited broadcast;
2239          * I even do not know to fix it or not. Waiting for complains :-)
2240          */
2241         if (ipv4_is_zeronet(saddr))
2242                 goto martian_source;
2243
2244         if (ipv4_is_zeronet(daddr))
2245                 goto martian_destination;
2246
2247         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2248          * and call it once if daddr or/and saddr are loopback addresses
2249          */
2250         if (ipv4_is_loopback(daddr)) {
2251                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2252                         goto martian_destination;
2253         } else if (ipv4_is_loopback(saddr)) {
2254                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2255                         goto martian_source;
2256         }
2257
2258         /*
2259          *      Now we are ready to route packet.
2260          */
2261         fl4.flowi4_oif = 0;
2262         fl4.flowi4_iif = dev->ifindex;
2263         fl4.flowi4_mark = skb->mark;
2264         fl4.flowi4_tos = tos;
2265         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2266         fl4.flowi4_flags = 0;
2267         fl4.daddr = daddr;
2268         fl4.saddr = saddr;
2269         fl4.flowi4_uid = sock_net_uid(net, NULL);
2270         fl4.flowi4_multipath_hash = 0;
2271
2272         if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2273                 flkeys = &_flkeys;
2274         } else {
2275                 fl4.flowi4_proto = 0;
2276                 fl4.fl4_sport = 0;
2277                 fl4.fl4_dport = 0;
2278         }
2279
2280         err = fib_lookup(net, &fl4, res, 0);
2281         if (err != 0) {
2282                 if (!IN_DEV_FORWARD(in_dev))
2283                         err = -EHOSTUNREACH;
2284                 goto no_route;
2285         }
2286
2287         if (res->type == RTN_BROADCAST) {
2288                 if (IN_DEV_BFORWARD(in_dev))
2289                         goto make_route;
2290                 /* not do cache if bc_forwarding is enabled */
2291                 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2292                         do_cache = false;
2293                 goto brd_input;
2294         }
2295
2296         if (res->type == RTN_LOCAL) {
2297                 err = fib_validate_source(skb, saddr, daddr, tos,
2298                                           0, dev, in_dev, &itag);
2299                 if (err < 0)
2300                         goto martian_source;
2301                 goto local_input;
2302         }
2303
2304         if (!IN_DEV_FORWARD(in_dev)) {
2305                 err = -EHOSTUNREACH;
2306                 goto no_route;
2307         }
2308         if (res->type != RTN_UNICAST)
2309                 goto martian_destination;
2310
2311 make_route:
2312         err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2313 out:    return err;
2314
2315 brd_input:
2316         if (skb->protocol != htons(ETH_P_IP))
2317                 goto e_inval;
2318
2319         if (!ipv4_is_zeronet(saddr)) {
2320                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2321                                           in_dev, &itag);
2322                 if (err < 0)
2323                         goto martian_source;
2324         }
2325         flags |= RTCF_BROADCAST;
2326         res->type = RTN_BROADCAST;
2327         RT_CACHE_STAT_INC(in_brd);
2328
2329 local_input:
2330         do_cache &= res->fi && !itag;
2331         if (do_cache) {
2332                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2333
2334                 rth = rcu_dereference(nhc->nhc_rth_input);
2335                 if (rt_cache_valid(rth)) {
2336                         skb_dst_set_noref(skb, &rth->dst);
2337                         err = 0;
2338                         goto out;
2339                 }
2340         }
2341
2342         rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2343                            flags | RTCF_LOCAL, res->type,
2344                            IN_DEV_ORCONF(in_dev, NOPOLICY), false);
2345         if (!rth)
2346                 goto e_nobufs;
2347
2348         rth->dst.output= ip_rt_bug;
2349 #ifdef CONFIG_IP_ROUTE_CLASSID
2350         rth->dst.tclassid = itag;
2351 #endif
2352         rth->rt_is_input = 1;
2353
2354         RT_CACHE_STAT_INC(in_slow_tot);
2355         if (res->type == RTN_UNREACHABLE) {
2356                 rth->dst.input= ip_error;
2357                 rth->dst.error= -err;
2358                 rth->rt_flags   &= ~RTCF_LOCAL;
2359         }
2360
2361         if (do_cache) {
2362                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2363
2364                 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2365                 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2366                         WARN_ON(rth->dst.input == lwtunnel_input);
2367                         rth->dst.lwtstate->orig_input = rth->dst.input;
2368                         rth->dst.input = lwtunnel_input;
2369                 }
2370
2371                 if (unlikely(!rt_cache_route(nhc, rth)))
2372                         rt_add_uncached_list(rth);
2373         }
2374         skb_dst_set(skb, &rth->dst);
2375         err = 0;
2376         goto out;
2377
2378 no_route:
2379         RT_CACHE_STAT_INC(in_no_route);
2380         res->type = RTN_UNREACHABLE;
2381         res->fi = NULL;
2382         res->table = NULL;
2383         goto local_input;
2384
2385         /*
2386          *      Do not cache martian addresses: they should be logged (RFC1812)
2387          */
2388 martian_destination:
2389         RT_CACHE_STAT_INC(in_martian_dst);
2390 #ifdef CONFIG_IP_ROUTE_VERBOSE
2391         if (IN_DEV_LOG_MARTIANS(in_dev))
2392                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2393                                      &daddr, &saddr, dev->name);
2394 #endif
2395
2396 e_inval:
2397         err = -EINVAL;
2398         goto out;
2399
2400 e_nobufs:
2401         err = -ENOBUFS;
2402         goto out;
2403
2404 martian_source:
2405         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2406         goto out;
2407 }
2408
2409 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2410                          u8 tos, struct net_device *dev)
2411 {
2412         struct fib_result res;
2413         int err;
2414
2415         tos &= IPTOS_RT_MASK;
2416         rcu_read_lock();
2417         err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2418         rcu_read_unlock();
2419
2420         return err;
2421 }
2422 EXPORT_SYMBOL(ip_route_input_noref);
2423
2424 /* called with rcu_read_lock held */
2425 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2426                        u8 tos, struct net_device *dev, struct fib_result *res)
2427 {
2428         /* Multicast recognition logic is moved from route cache to here.
2429          * The problem was that too many Ethernet cards have broken/missing
2430          * hardware multicast filters :-( As result the host on multicasting
2431          * network acquires a lot of useless route cache entries, sort of
2432          * SDR messages from all the world. Now we try to get rid of them.
2433          * Really, provided software IP multicast filter is organized
2434          * reasonably (at least, hashed), it does not result in a slowdown
2435          * comparing with route cache reject entries.
2436          * Note, that multicast routers are not affected, because
2437          * route cache entry is created eventually.
2438          */
2439         if (ipv4_is_multicast(daddr)) {
2440                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2441                 int our = 0;
2442                 int err = -EINVAL;
2443
2444                 if (!in_dev)
2445                         return err;
2446                 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2447                                       ip_hdr(skb)->protocol);
2448
2449                 /* check l3 master if no match yet */
2450                 if (!our && netif_is_l3_slave(dev)) {
2451                         struct in_device *l3_in_dev;
2452
2453                         l3_in_dev = __in_dev_get_rcu(skb->dev);
2454                         if (l3_in_dev)
2455                                 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2456                                                       ip_hdr(skb)->protocol);
2457                 }
2458
2459                 if (our
2460 #ifdef CONFIG_IP_MROUTE
2461                         ||
2462                     (!ipv4_is_local_multicast(daddr) &&
2463                      IN_DEV_MFORWARD(in_dev))
2464 #endif
2465                    ) {
2466                         err = ip_route_input_mc(skb, daddr, saddr,
2467                                                 tos, dev, our);
2468                 }
2469                 return err;
2470         }
2471
2472         return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2473 }
2474
2475 /* called with rcu_read_lock() */
2476 static struct rtable *__mkroute_output(const struct fib_result *res,
2477                                        const struct flowi4 *fl4, int orig_oif,
2478                                        struct net_device *dev_out,
2479                                        unsigned int flags)
2480 {
2481         struct fib_info *fi = res->fi;
2482         struct fib_nh_exception *fnhe;
2483         struct in_device *in_dev;
2484         u16 type = res->type;
2485         struct rtable *rth;
2486         bool do_cache;
2487
2488         in_dev = __in_dev_get_rcu(dev_out);
2489         if (!in_dev)
2490                 return ERR_PTR(-EINVAL);
2491
2492         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2493                 if (ipv4_is_loopback(fl4->saddr) &&
2494                     !(dev_out->flags & IFF_LOOPBACK) &&
2495                     !netif_is_l3_master(dev_out))
2496                         return ERR_PTR(-EINVAL);
2497
2498         if (ipv4_is_lbcast(fl4->daddr))
2499                 type = RTN_BROADCAST;
2500         else if (ipv4_is_multicast(fl4->daddr))
2501                 type = RTN_MULTICAST;
2502         else if (ipv4_is_zeronet(fl4->daddr))
2503                 return ERR_PTR(-EINVAL);
2504
2505         if (dev_out->flags & IFF_LOOPBACK)
2506                 flags |= RTCF_LOCAL;
2507
2508         do_cache = true;
2509         if (type == RTN_BROADCAST) {
2510                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2511                 fi = NULL;
2512         } else if (type == RTN_MULTICAST) {
2513                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2514                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2515                                      fl4->flowi4_proto))
2516                         flags &= ~RTCF_LOCAL;
2517                 else
2518                         do_cache = false;
2519                 /* If multicast route do not exist use
2520                  * default one, but do not gateway in this case.
2521                  * Yes, it is hack.
2522                  */
2523                 if (fi && res->prefixlen < 4)
2524                         fi = NULL;
2525         } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2526                    (orig_oif != dev_out->ifindex)) {
2527                 /* For local routes that require a particular output interface
2528                  * we do not want to cache the result.  Caching the result
2529                  * causes incorrect behaviour when there are multiple source
2530                  * addresses on the interface, the end result being that if the
2531                  * intended recipient is waiting on that interface for the
2532                  * packet he won't receive it because it will be delivered on
2533                  * the loopback interface and the IP_PKTINFO ipi_ifindex will
2534                  * be set to the loopback interface as well.
2535                  */
2536                 do_cache = false;
2537         }
2538
2539         fnhe = NULL;
2540         do_cache &= fi != NULL;
2541         if (fi) {
2542                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2543                 struct rtable __rcu **prth;
2544
2545                 fnhe = find_exception(nhc, fl4->daddr);
2546                 if (!do_cache)
2547                         goto add;
2548                 if (fnhe) {
2549                         prth = &fnhe->fnhe_rth_output;
2550                 } else {
2551                         if (unlikely(fl4->flowi4_flags &
2552                                      FLOWI_FLAG_KNOWN_NH &&
2553                                      !(nhc->nhc_gw_family &&
2554                                        nhc->nhc_scope == RT_SCOPE_LINK))) {
2555                                 do_cache = false;
2556                                 goto add;
2557                         }
2558                         prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2559                 }
2560                 rth = rcu_dereference(*prth);
2561                 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2562                         return rth;
2563         }
2564
2565 add:
2566         rth = rt_dst_alloc(dev_out, flags, type,
2567                            IN_DEV_ORCONF(in_dev, NOPOLICY),
2568                            IN_DEV_ORCONF(in_dev, NOXFRM));
2569         if (!rth)
2570                 return ERR_PTR(-ENOBUFS);
2571
2572         rth->rt_iif = orig_oif;
2573
2574         RT_CACHE_STAT_INC(out_slow_tot);
2575
2576         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2577                 if (flags & RTCF_LOCAL &&
2578                     !(dev_out->flags & IFF_LOOPBACK)) {
2579                         rth->dst.output = ip_mc_output;
2580                         RT_CACHE_STAT_INC(out_slow_mc);
2581                 }
2582 #ifdef CONFIG_IP_MROUTE
2583                 if (type == RTN_MULTICAST) {
2584                         if (IN_DEV_MFORWARD(in_dev) &&
2585                             !ipv4_is_local_multicast(fl4->daddr)) {
2586                                 rth->dst.input = ip_mr_input;
2587                                 rth->dst.output = ip_mc_output;
2588                         }
2589                 }
2590 #endif
2591         }
2592
2593         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2594         lwtunnel_set_redirect(&rth->dst);
2595
2596         return rth;
2597 }
2598
2599 /*
2600  * Major route resolver routine.
2601  */
2602
2603 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2604                                         const struct sk_buff *skb)
2605 {
2606         __u8 tos = RT_FL_TOS(fl4);
2607         struct fib_result res = {
2608                 .type           = RTN_UNSPEC,
2609                 .fi             = NULL,
2610                 .table          = NULL,
2611                 .tclassid       = 0,
2612         };
2613         struct rtable *rth;
2614
2615         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2616         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2617         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2618                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2619
2620         rcu_read_lock();
2621         rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2622         rcu_read_unlock();
2623
2624         return rth;
2625 }
2626 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2627
2628 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2629                                             struct fib_result *res,
2630                                             const struct sk_buff *skb)
2631 {
2632         struct net_device *dev_out = NULL;
2633         int orig_oif = fl4->flowi4_oif;
2634         unsigned int flags = 0;
2635         struct rtable *rth;
2636         int err;
2637
2638         if (fl4->saddr) {
2639                 if (ipv4_is_multicast(fl4->saddr) ||
2640                     ipv4_is_lbcast(fl4->saddr) ||
2641                     ipv4_is_zeronet(fl4->saddr)) {
2642                         rth = ERR_PTR(-EINVAL);
2643                         goto out;
2644                 }
2645
2646                 rth = ERR_PTR(-ENETUNREACH);
2647
2648                 /* I removed check for oif == dev_out->oif here.
2649                  * It was wrong for two reasons:
2650                  * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2651                  *    is assigned to multiple interfaces.
2652                  * 2. Moreover, we are allowed to send packets with saddr
2653                  *    of another iface. --ANK
2654                  */
2655
2656                 if (fl4->flowi4_oif == 0 &&
2657                     (ipv4_is_multicast(fl4->daddr) ||
2658                      ipv4_is_lbcast(fl4->daddr))) {
2659                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2660                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2661                         if (!dev_out)
2662                                 goto out;
2663
2664                         /* Special hack: user can direct multicasts
2665                          * and limited broadcast via necessary interface
2666                          * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2667                          * This hack is not just for fun, it allows
2668                          * vic,vat and friends to work.
2669                          * They bind socket to loopback, set ttl to zero
2670                          * and expect that it will work.
2671                          * From the viewpoint of routing cache they are broken,
2672                          * because we are not allowed to build multicast path
2673                          * with loopback source addr (look, routing cache
2674                          * cannot know, that ttl is zero, so that packet
2675                          * will not leave this host and route is valid).
2676                          * Luckily, this hack is good workaround.
2677                          */
2678
2679                         fl4->flowi4_oif = dev_out->ifindex;
2680                         goto make_route;
2681                 }
2682
2683                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2684                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2685                         if (!__ip_dev_find(net, fl4->saddr, false))
2686                                 goto out;
2687                 }
2688         }
2689
2690
2691         if (fl4->flowi4_oif) {
2692                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2693                 rth = ERR_PTR(-ENODEV);
2694                 if (!dev_out)
2695                         goto out;
2696
2697                 /* RACE: Check return value of inet_select_addr instead. */
2698                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2699                         rth = ERR_PTR(-ENETUNREACH);
2700                         goto out;
2701                 }
2702                 if (ipv4_is_local_multicast(fl4->daddr) ||
2703                     ipv4_is_lbcast(fl4->daddr) ||
2704                     fl4->flowi4_proto == IPPROTO_IGMP) {
2705                         if (!fl4->saddr)
2706                                 fl4->saddr = inet_select_addr(dev_out, 0,
2707                                                               RT_SCOPE_LINK);
2708                         goto make_route;
2709                 }
2710                 if (!fl4->saddr) {
2711                         if (ipv4_is_multicast(fl4->daddr))
2712                                 fl4->saddr = inet_select_addr(dev_out, 0,
2713                                                               fl4->flowi4_scope);
2714                         else if (!fl4->daddr)
2715                                 fl4->saddr = inet_select_addr(dev_out, 0,
2716                                                               RT_SCOPE_HOST);
2717                 }
2718         }
2719
2720         if (!fl4->daddr) {
2721                 fl4->daddr = fl4->saddr;
2722                 if (!fl4->daddr)
2723                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2724                 dev_out = net->loopback_dev;
2725                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2726                 res->type = RTN_LOCAL;
2727                 flags |= RTCF_LOCAL;
2728                 goto make_route;
2729         }
2730
2731         err = fib_lookup(net, fl4, res, 0);
2732         if (err) {
2733                 res->fi = NULL;
2734                 res->table = NULL;
2735                 if (fl4->flowi4_oif &&
2736                     (ipv4_is_multicast(fl4->daddr) ||
2737                     !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2738                         /* Apparently, routing tables are wrong. Assume,
2739                          * that the destination is on link.
2740                          *
2741                          * WHY? DW.
2742                          * Because we are allowed to send to iface
2743                          * even if it has NO routes and NO assigned
2744                          * addresses. When oif is specified, routing
2745                          * tables are looked up with only one purpose:
2746                          * to catch if destination is gatewayed, rather than
2747                          * direct. Moreover, if MSG_DONTROUTE is set,
2748                          * we send packet, ignoring both routing tables
2749                          * and ifaddr state. --ANK
2750                          *
2751                          *
2752                          * We could make it even if oif is unknown,
2753                          * likely IPv6, but we do not.
2754                          */
2755
2756                         if (fl4->saddr == 0)
2757                                 fl4->saddr = inet_select_addr(dev_out, 0,
2758                                                               RT_SCOPE_LINK);
2759                         res->type = RTN_UNICAST;
2760                         goto make_route;
2761                 }
2762                 rth = ERR_PTR(err);
2763                 goto out;
2764         }
2765
2766         if (res->type == RTN_LOCAL) {
2767                 if (!fl4->saddr) {
2768                         if (res->fi->fib_prefsrc)
2769                                 fl4->saddr = res->fi->fib_prefsrc;
2770                         else
2771                                 fl4->saddr = fl4->daddr;
2772                 }
2773
2774                 /* L3 master device is the loopback for that domain */
2775                 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2776                         net->loopback_dev;
2777
2778                 /* make sure orig_oif points to fib result device even
2779                  * though packet rx/tx happens over loopback or l3mdev
2780                  */
2781                 orig_oif = FIB_RES_OIF(*res);
2782
2783                 fl4->flowi4_oif = dev_out->ifindex;
2784                 flags |= RTCF_LOCAL;
2785                 goto make_route;
2786         }
2787
2788         fib_select_path(net, res, fl4, skb);
2789
2790         dev_out = FIB_RES_DEV(*res);
2791
2792 make_route:
2793         rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2794
2795 out:
2796         return rth;
2797 }
2798
2799 static struct dst_ops ipv4_dst_blackhole_ops = {
2800         .family                 = AF_INET,
2801         .default_advmss         = ipv4_default_advmss,
2802         .neigh_lookup           = ipv4_neigh_lookup,
2803         .check                  = dst_blackhole_check,
2804         .cow_metrics            = dst_blackhole_cow_metrics,
2805         .update_pmtu            = dst_blackhole_update_pmtu,
2806         .redirect               = dst_blackhole_redirect,
2807         .mtu                    = dst_blackhole_mtu,
2808 };
2809
2810 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2811 {
2812         struct rtable *ort = (struct rtable *) dst_orig;
2813         struct rtable *rt;
2814
2815         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2816         if (rt) {
2817                 struct dst_entry *new = &rt->dst;
2818
2819                 new->__use = 1;
2820                 new->input = dst_discard;
2821                 new->output = dst_discard_out;
2822
2823                 new->dev = net->loopback_dev;
2824                 dev_hold_track(new->dev, &new->dev_tracker, GFP_ATOMIC);
2825
2826                 rt->rt_is_input = ort->rt_is_input;
2827                 rt->rt_iif = ort->rt_iif;
2828                 rt->rt_pmtu = ort->rt_pmtu;
2829                 rt->rt_mtu_locked = ort->rt_mtu_locked;
2830
2831                 rt->rt_genid = rt_genid_ipv4(net);
2832                 rt->rt_flags = ort->rt_flags;
2833                 rt->rt_type = ort->rt_type;
2834                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2835                 rt->rt_gw_family = ort->rt_gw_family;
2836                 if (rt->rt_gw_family == AF_INET)
2837                         rt->rt_gw4 = ort->rt_gw4;
2838                 else if (rt->rt_gw_family == AF_INET6)
2839                         rt->rt_gw6 = ort->rt_gw6;
2840
2841                 INIT_LIST_HEAD(&rt->rt_uncached);
2842         }
2843
2844         dst_release(dst_orig);
2845
2846         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2847 }
2848
2849 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2850                                     const struct sock *sk)
2851 {
2852         struct rtable *rt = __ip_route_output_key(net, flp4);
2853
2854         if (IS_ERR(rt))
2855                 return rt;
2856
2857         if (flp4->flowi4_proto) {
2858                 flp4->flowi4_oif = rt->dst.dev->ifindex;
2859                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2860                                                         flowi4_to_flowi(flp4),
2861                                                         sk, 0);
2862         }
2863
2864         return rt;
2865 }
2866 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2867
2868 struct rtable *ip_route_output_tunnel(struct sk_buff *skb,
2869                                       struct net_device *dev,
2870                                       struct net *net, __be32 *saddr,
2871                                       const struct ip_tunnel_info *info,
2872                                       u8 protocol, bool use_cache)
2873 {
2874 #ifdef CONFIG_DST_CACHE
2875         struct dst_cache *dst_cache;
2876 #endif
2877         struct rtable *rt = NULL;
2878         struct flowi4 fl4;
2879         __u8 tos;
2880
2881 #ifdef CONFIG_DST_CACHE
2882         dst_cache = (struct dst_cache *)&info->dst_cache;
2883         if (use_cache) {
2884                 rt = dst_cache_get_ip4(dst_cache, saddr);
2885                 if (rt)
2886                         return rt;
2887         }
2888 #endif
2889         memset(&fl4, 0, sizeof(fl4));
2890         fl4.flowi4_mark = skb->mark;
2891         fl4.flowi4_proto = protocol;
2892         fl4.daddr = info->key.u.ipv4.dst;
2893         fl4.saddr = info->key.u.ipv4.src;
2894         tos = info->key.tos;
2895         fl4.flowi4_tos = RT_TOS(tos);
2896
2897         rt = ip_route_output_key(net, &fl4);
2898         if (IS_ERR(rt)) {
2899                 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
2900                 return ERR_PTR(-ENETUNREACH);
2901         }
2902         if (rt->dst.dev == dev) { /* is this necessary? */
2903                 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
2904                 ip_rt_put(rt);
2905                 return ERR_PTR(-ELOOP);
2906         }
2907 #ifdef CONFIG_DST_CACHE
2908         if (use_cache)
2909                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2910 #endif
2911         *saddr = fl4.saddr;
2912         return rt;
2913 }
2914 EXPORT_SYMBOL_GPL(ip_route_output_tunnel);
2915
2916 /* called with rcu_read_lock held */
2917 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2918                         struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2919                         struct sk_buff *skb, u32 portid, u32 seq,
2920                         unsigned int flags)
2921 {
2922         struct rtmsg *r;
2923         struct nlmsghdr *nlh;
2924         unsigned long expires = 0;
2925         u32 error;
2926         u32 metrics[RTAX_MAX];
2927
2928         nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2929         if (!nlh)
2930                 return -EMSGSIZE;
2931
2932         r = nlmsg_data(nlh);
2933         r->rtm_family    = AF_INET;
2934         r->rtm_dst_len  = 32;
2935         r->rtm_src_len  = 0;
2936         r->rtm_tos      = fl4 ? fl4->flowi4_tos : 0;
2937         r->rtm_table    = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2938         if (nla_put_u32(skb, RTA_TABLE, table_id))
2939                 goto nla_put_failure;
2940         r->rtm_type     = rt->rt_type;
2941         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2942         r->rtm_protocol = RTPROT_UNSPEC;
2943         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2944         if (rt->rt_flags & RTCF_NOTIFY)
2945                 r->rtm_flags |= RTM_F_NOTIFY;
2946         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2947                 r->rtm_flags |= RTCF_DOREDIRECT;
2948
2949         if (nla_put_in_addr(skb, RTA_DST, dst))
2950                 goto nla_put_failure;
2951         if (src) {
2952                 r->rtm_src_len = 32;
2953                 if (nla_put_in_addr(skb, RTA_SRC, src))
2954                         goto nla_put_failure;
2955         }
2956         if (rt->dst.dev &&
2957             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2958                 goto nla_put_failure;
2959         if (rt->dst.lwtstate &&
2960             lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2961                 goto nla_put_failure;
2962 #ifdef CONFIG_IP_ROUTE_CLASSID
2963         if (rt->dst.tclassid &&
2964             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2965                 goto nla_put_failure;
2966 #endif
2967         if (fl4 && !rt_is_input_route(rt) &&
2968             fl4->saddr != src) {
2969                 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2970                         goto nla_put_failure;
2971         }
2972         if (rt->rt_uses_gateway) {
2973                 if (rt->rt_gw_family == AF_INET &&
2974                     nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2975                         goto nla_put_failure;
2976                 } else if (rt->rt_gw_family == AF_INET6) {
2977                         int alen = sizeof(struct in6_addr);
2978                         struct nlattr *nla;
2979                         struct rtvia *via;
2980
2981                         nla = nla_reserve(skb, RTA_VIA, alen + 2);
2982                         if (!nla)
2983                                 goto nla_put_failure;
2984
2985                         via = nla_data(nla);
2986                         via->rtvia_family = AF_INET6;
2987                         memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2988                 }
2989         }
2990
2991         expires = rt->dst.expires;
2992         if (expires) {
2993                 unsigned long now = jiffies;
2994
2995                 if (time_before(now, expires))
2996                         expires -= now;
2997                 else
2998                         expires = 0;
2999         }
3000
3001         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3002         if (rt->rt_pmtu && expires)
3003                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3004         if (rt->rt_mtu_locked && expires)
3005                 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3006         if (rtnetlink_put_metrics(skb, metrics) < 0)
3007                 goto nla_put_failure;
3008
3009         if (fl4) {
3010                 if (fl4->flowi4_mark &&
3011                     nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3012                         goto nla_put_failure;
3013
3014                 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3015                     nla_put_u32(skb, RTA_UID,
3016                                 from_kuid_munged(current_user_ns(),
3017                                                  fl4->flowi4_uid)))
3018                         goto nla_put_failure;
3019
3020                 if (rt_is_input_route(rt)) {
3021 #ifdef CONFIG_IP_MROUTE
3022                         if (ipv4_is_multicast(dst) &&
3023                             !ipv4_is_local_multicast(dst) &&
3024                             IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
3025                                 int err = ipmr_get_route(net, skb,
3026                                                          fl4->saddr, fl4->daddr,
3027                                                          r, portid);
3028
3029                                 if (err <= 0) {
3030                                         if (err == 0)
3031                                                 return 0;
3032                                         goto nla_put_failure;
3033                                 }
3034                         } else
3035 #endif
3036                                 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3037                                         goto nla_put_failure;
3038                 }
3039         }
3040
3041         error = rt->dst.error;
3042
3043         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3044                 goto nla_put_failure;
3045
3046         nlmsg_end(skb, nlh);
3047         return 0;
3048
3049 nla_put_failure:
3050         nlmsg_cancel(skb, nlh);
3051         return -EMSGSIZE;
3052 }
3053
3054 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3055                             struct netlink_callback *cb, u32 table_id,
3056                             struct fnhe_hash_bucket *bucket, int genid,
3057                             int *fa_index, int fa_start, unsigned int flags)
3058 {
3059         int i;
3060
3061         for (i = 0; i < FNHE_HASH_SIZE; i++) {
3062                 struct fib_nh_exception *fnhe;
3063
3064                 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3065                      fnhe = rcu_dereference(fnhe->fnhe_next)) {
3066                         struct rtable *rt;
3067                         int err;
3068
3069                         if (*fa_index < fa_start)
3070                                 goto next;
3071
3072                         if (fnhe->fnhe_genid != genid)
3073                                 goto next;
3074
3075                         if (fnhe->fnhe_expires &&
3076                             time_after(jiffies, fnhe->fnhe_expires))
3077                                 goto next;
3078
3079                         rt = rcu_dereference(fnhe->fnhe_rth_input);
3080                         if (!rt)
3081                                 rt = rcu_dereference(fnhe->fnhe_rth_output);
3082                         if (!rt)
3083                                 goto next;
3084
3085                         err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3086                                            table_id, NULL, skb,
3087                                            NETLINK_CB(cb->skb).portid,
3088                                            cb->nlh->nlmsg_seq, flags);
3089                         if (err)
3090                                 return err;
3091 next:
3092                         (*fa_index)++;
3093                 }
3094         }
3095
3096         return 0;
3097 }
3098
3099 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3100                        u32 table_id, struct fib_info *fi,
3101                        int *fa_index, int fa_start, unsigned int flags)
3102 {
3103         struct net *net = sock_net(cb->skb->sk);
3104         int nhsel, genid = fnhe_genid(net);
3105
3106         for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3107                 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3108                 struct fnhe_hash_bucket *bucket;
3109                 int err;
3110
3111                 if (nhc->nhc_flags & RTNH_F_DEAD)
3112                         continue;
3113
3114                 rcu_read_lock();
3115                 bucket = rcu_dereference(nhc->nhc_exceptions);
3116                 err = 0;
3117                 if (bucket)
3118                         err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3119                                                genid, fa_index, fa_start,
3120                                                flags);
3121                 rcu_read_unlock();
3122                 if (err)
3123                         return err;
3124         }
3125
3126         return 0;
3127 }
3128
3129 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3130                                                    u8 ip_proto, __be16 sport,
3131                                                    __be16 dport)
3132 {
3133         struct sk_buff *skb;
3134         struct iphdr *iph;
3135
3136         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3137         if (!skb)
3138                 return NULL;
3139
3140         /* Reserve room for dummy headers, this skb can pass
3141          * through good chunk of routing engine.
3142          */
3143         skb_reset_mac_header(skb);
3144         skb_reset_network_header(skb);
3145         skb->protocol = htons(ETH_P_IP);
3146         iph = skb_put(skb, sizeof(struct iphdr));
3147         iph->protocol = ip_proto;
3148         iph->saddr = src;
3149         iph->daddr = dst;
3150         iph->version = 0x4;
3151         iph->frag_off = 0;
3152         iph->ihl = 0x5;
3153         skb_set_transport_header(skb, skb->len);
3154
3155         switch (iph->protocol) {
3156         case IPPROTO_UDP: {
3157                 struct udphdr *udph;
3158
3159                 udph = skb_put_zero(skb, sizeof(struct udphdr));
3160                 udph->source = sport;
3161                 udph->dest = dport;
3162                 udph->len = htons(sizeof(struct udphdr));
3163                 udph->check = 0;
3164                 break;
3165         }
3166         case IPPROTO_TCP: {
3167                 struct tcphdr *tcph;
3168
3169                 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3170                 tcph->source    = sport;
3171                 tcph->dest      = dport;
3172                 tcph->doff      = sizeof(struct tcphdr) / 4;
3173                 tcph->rst = 1;
3174                 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3175                                             src, dst, 0);
3176                 break;
3177         }
3178         case IPPROTO_ICMP: {
3179                 struct icmphdr *icmph;
3180
3181                 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3182                 icmph->type = ICMP_ECHO;
3183                 icmph->code = 0;
3184         }
3185         }
3186
3187         return skb;
3188 }
3189
3190 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3191                                        const struct nlmsghdr *nlh,
3192                                        struct nlattr **tb,
3193                                        struct netlink_ext_ack *extack)
3194 {
3195         struct rtmsg *rtm;
3196         int i, err;
3197
3198         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3199                 NL_SET_ERR_MSG(extack,
3200                                "ipv4: Invalid header for route get request");
3201                 return -EINVAL;
3202         }
3203
3204         if (!netlink_strict_get_check(skb))
3205                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3206                                               rtm_ipv4_policy, extack);
3207
3208         rtm = nlmsg_data(nlh);
3209         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3210             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3211             rtm->rtm_table || rtm->rtm_protocol ||
3212             rtm->rtm_scope || rtm->rtm_type) {
3213                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3214                 return -EINVAL;
3215         }
3216
3217         if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3218                                RTM_F_LOOKUP_TABLE |
3219                                RTM_F_FIB_MATCH)) {
3220                 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3221                 return -EINVAL;
3222         }
3223
3224         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3225                                             rtm_ipv4_policy, extack);
3226         if (err)
3227                 return err;
3228
3229         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3230             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3231                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3232                 return -EINVAL;
3233         }
3234
3235         for (i = 0; i <= RTA_MAX; i++) {
3236                 if (!tb[i])
3237                         continue;
3238
3239                 switch (i) {
3240                 case RTA_IIF:
3241                 case RTA_OIF:
3242                 case RTA_SRC:
3243                 case RTA_DST:
3244                 case RTA_IP_PROTO:
3245                 case RTA_SPORT:
3246                 case RTA_DPORT:
3247                 case RTA_MARK:
3248                 case RTA_UID:
3249                         break;
3250                 default:
3251                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3252                         return -EINVAL;
3253                 }
3254         }
3255
3256         return 0;
3257 }
3258
3259 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3260                              struct netlink_ext_ack *extack)
3261 {
3262         struct net *net = sock_net(in_skb->sk);
3263         struct nlattr *tb[RTA_MAX+1];
3264         u32 table_id = RT_TABLE_MAIN;
3265         __be16 sport = 0, dport = 0;
3266         struct fib_result res = {};
3267         u8 ip_proto = IPPROTO_UDP;
3268         struct rtable *rt = NULL;
3269         struct sk_buff *skb;
3270         struct rtmsg *rtm;
3271         struct flowi4 fl4 = {};
3272         __be32 dst = 0;
3273         __be32 src = 0;
3274         kuid_t uid;
3275         u32 iif;
3276         int err;
3277         int mark;
3278
3279         err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3280         if (err < 0)
3281                 return err;
3282
3283         rtm = nlmsg_data(nlh);
3284         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3285         dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3286         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3287         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3288         if (tb[RTA_UID])
3289                 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3290         else
3291                 uid = (iif ? INVALID_UID : current_uid());
3292
3293         if (tb[RTA_IP_PROTO]) {
3294                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3295                                                   &ip_proto, AF_INET, extack);
3296                 if (err)
3297                         return err;
3298         }
3299
3300         if (tb[RTA_SPORT])
3301                 sport = nla_get_be16(tb[RTA_SPORT]);
3302
3303         if (tb[RTA_DPORT])
3304                 dport = nla_get_be16(tb[RTA_DPORT]);
3305
3306         skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3307         if (!skb)
3308                 return -ENOBUFS;
3309
3310         fl4.daddr = dst;
3311         fl4.saddr = src;
3312         fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3313         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3314         fl4.flowi4_mark = mark;
3315         fl4.flowi4_uid = uid;
3316         if (sport)
3317                 fl4.fl4_sport = sport;
3318         if (dport)
3319                 fl4.fl4_dport = dport;
3320         fl4.flowi4_proto = ip_proto;
3321
3322         rcu_read_lock();
3323
3324         if (iif) {
3325                 struct net_device *dev;
3326
3327                 dev = dev_get_by_index_rcu(net, iif);
3328                 if (!dev) {
3329                         err = -ENODEV;
3330                         goto errout_rcu;
3331                 }
3332
3333                 fl4.flowi4_iif = iif; /* for rt_fill_info */
3334                 skb->dev        = dev;
3335                 skb->mark       = mark;
3336                 err = ip_route_input_rcu(skb, dst, src,
3337                                          rtm->rtm_tos & IPTOS_RT_MASK, dev,
3338                                          &res);
3339
3340                 rt = skb_rtable(skb);
3341                 if (err == 0 && rt->dst.error)
3342                         err = -rt->dst.error;
3343         } else {
3344                 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3345                 skb->dev = net->loopback_dev;
3346                 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3347                 err = 0;
3348                 if (IS_ERR(rt))
3349                         err = PTR_ERR(rt);
3350                 else
3351                         skb_dst_set(skb, &rt->dst);
3352         }
3353
3354         if (err)
3355                 goto errout_rcu;
3356
3357         if (rtm->rtm_flags & RTM_F_NOTIFY)
3358                 rt->rt_flags |= RTCF_NOTIFY;
3359
3360         if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3361                 table_id = res.table ? res.table->tb_id : 0;
3362
3363         /* reset skb for netlink reply msg */
3364         skb_trim(skb, 0);
3365         skb_reset_network_header(skb);
3366         skb_reset_transport_header(skb);
3367         skb_reset_mac_header(skb);
3368
3369         if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3370                 struct fib_rt_info fri;
3371
3372                 if (!res.fi) {
3373                         err = fib_props[res.type].error;
3374                         if (!err)
3375                                 err = -EHOSTUNREACH;
3376                         goto errout_rcu;
3377                 }
3378                 fri.fi = res.fi;
3379                 fri.tb_id = table_id;
3380                 fri.dst = res.prefix;
3381                 fri.dst_len = res.prefixlen;
3382                 fri.tos = fl4.flowi4_tos;
3383                 fri.type = rt->rt_type;
3384                 fri.offload = 0;
3385                 fri.trap = 0;
3386                 fri.offload_failed = 0;
3387                 if (res.fa_head) {
3388                         struct fib_alias *fa;
3389
3390                         hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3391                                 u8 slen = 32 - fri.dst_len;
3392
3393                                 if (fa->fa_slen == slen &&
3394                                     fa->tb_id == fri.tb_id &&
3395                                     fa->fa_tos == fri.tos &&
3396                                     fa->fa_info == res.fi &&
3397                                     fa->fa_type == fri.type) {
3398                                         fri.offload = READ_ONCE(fa->offload);
3399                                         fri.trap = READ_ONCE(fa->trap);
3400                                         break;
3401                                 }
3402                         }
3403                 }
3404                 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3405                                     nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3406         } else {
3407                 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3408                                    NETLINK_CB(in_skb).portid,
3409                                    nlh->nlmsg_seq, 0);
3410         }
3411         if (err < 0)
3412                 goto errout_rcu;
3413
3414         rcu_read_unlock();
3415
3416         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3417
3418 errout_free:
3419         return err;
3420 errout_rcu:
3421         rcu_read_unlock();
3422         kfree_skb(skb);
3423         goto errout_free;
3424 }
3425
3426 void ip_rt_multicast_event(struct in_device *in_dev)
3427 {
3428         rt_cache_flush(dev_net(in_dev->dev));
3429 }
3430
3431 #ifdef CONFIG_SYSCTL
3432 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3433 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
3434 static int ip_rt_gc_elasticity __read_mostly    = 8;
3435 static int ip_min_valid_pmtu __read_mostly      = IPV4_MIN_MTU;
3436
3437 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3438                 void *buffer, size_t *lenp, loff_t *ppos)
3439 {
3440         struct net *net = (struct net *)__ctl->extra1;
3441
3442         if (write) {
3443                 rt_cache_flush(net);
3444                 fnhe_genid_bump(net);
3445                 return 0;
3446         }
3447
3448         return -EINVAL;
3449 }
3450
3451 static struct ctl_table ipv4_route_table[] = {
3452         {
3453                 .procname       = "gc_thresh",
3454                 .data           = &ipv4_dst_ops.gc_thresh,
3455                 .maxlen         = sizeof(int),
3456                 .mode           = 0644,
3457                 .proc_handler   = proc_dointvec,
3458         },
3459         {
3460                 .procname       = "max_size",
3461                 .data           = &ip_rt_max_size,
3462                 .maxlen         = sizeof(int),
3463                 .mode           = 0644,
3464                 .proc_handler   = proc_dointvec,
3465         },
3466         {
3467                 /*  Deprecated. Use gc_min_interval_ms */
3468
3469                 .procname       = "gc_min_interval",
3470                 .data           = &ip_rt_gc_min_interval,
3471                 .maxlen         = sizeof(int),
3472                 .mode           = 0644,
3473                 .proc_handler   = proc_dointvec_jiffies,
3474         },
3475         {
3476                 .procname       = "gc_min_interval_ms",
3477                 .data           = &ip_rt_gc_min_interval,
3478                 .maxlen         = sizeof(int),
3479                 .mode           = 0644,
3480                 .proc_handler   = proc_dointvec_ms_jiffies,
3481         },
3482         {
3483                 .procname       = "gc_timeout",
3484                 .data           = &ip_rt_gc_timeout,
3485                 .maxlen         = sizeof(int),
3486                 .mode           = 0644,
3487                 .proc_handler   = proc_dointvec_jiffies,
3488         },
3489         {
3490                 .procname       = "gc_interval",
3491                 .data           = &ip_rt_gc_interval,
3492                 .maxlen         = sizeof(int),
3493                 .mode           = 0644,
3494                 .proc_handler   = proc_dointvec_jiffies,
3495         },
3496         {
3497                 .procname       = "redirect_load",
3498                 .data           = &ip_rt_redirect_load,
3499                 .maxlen         = sizeof(int),
3500                 .mode           = 0644,
3501                 .proc_handler   = proc_dointvec,
3502         },
3503         {
3504                 .procname       = "redirect_number",
3505                 .data           = &ip_rt_redirect_number,
3506                 .maxlen         = sizeof(int),
3507                 .mode           = 0644,
3508                 .proc_handler   = proc_dointvec,
3509         },
3510         {
3511                 .procname       = "redirect_silence",
3512                 .data           = &ip_rt_redirect_silence,
3513                 .maxlen         = sizeof(int),
3514                 .mode           = 0644,
3515                 .proc_handler   = proc_dointvec,
3516         },
3517         {
3518                 .procname       = "error_cost",
3519                 .data           = &ip_rt_error_cost,
3520                 .maxlen         = sizeof(int),
3521                 .mode           = 0644,
3522                 .proc_handler   = proc_dointvec,
3523         },
3524         {
3525                 .procname       = "error_burst",
3526                 .data           = &ip_rt_error_burst,
3527                 .maxlen         = sizeof(int),
3528                 .mode           = 0644,
3529                 .proc_handler   = proc_dointvec,
3530         },
3531         {
3532                 .procname       = "gc_elasticity",
3533                 .data           = &ip_rt_gc_elasticity,
3534                 .maxlen         = sizeof(int),
3535                 .mode           = 0644,
3536                 .proc_handler   = proc_dointvec,
3537         },
3538         {
3539                 .procname       = "min_adv_mss",
3540                 .data           = &ip_rt_min_advmss,
3541                 .maxlen         = sizeof(int),
3542                 .mode           = 0644,
3543                 .proc_handler   = proc_dointvec,
3544         },
3545         { }
3546 };
3547
3548 static const char ipv4_route_flush_procname[] = "flush";
3549
3550 static struct ctl_table ipv4_route_netns_table[] = {
3551         {
3552                 .procname       = ipv4_route_flush_procname,
3553                 .maxlen         = sizeof(int),
3554                 .mode           = 0200,
3555                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3556         },
3557         {
3558                 .procname       = "min_pmtu",
3559                 .data           = &init_net.ipv4.ip_rt_min_pmtu,
3560                 .maxlen         = sizeof(int),
3561                 .mode           = 0644,
3562                 .proc_handler   = proc_dointvec_minmax,
3563                 .extra1         = &ip_min_valid_pmtu,
3564         },
3565         {
3566                 .procname       = "mtu_expires",
3567                 .data           = &init_net.ipv4.ip_rt_mtu_expires,
3568                 .maxlen         = sizeof(int),
3569                 .mode           = 0644,
3570                 .proc_handler   = proc_dointvec_jiffies,
3571         },
3572         { },
3573 };
3574
3575 static __net_init int sysctl_route_net_init(struct net *net)
3576 {
3577         struct ctl_table *tbl;
3578
3579         tbl = ipv4_route_netns_table;
3580         if (!net_eq(net, &init_net)) {
3581                 int i;
3582
3583                 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3584                 if (!tbl)
3585                         goto err_dup;
3586
3587                 /* Don't export non-whitelisted sysctls to unprivileged users */
3588                 if (net->user_ns != &init_user_ns) {
3589                         if (tbl[0].procname != ipv4_route_flush_procname)
3590                                 tbl[0].procname = NULL;
3591                 }
3592
3593                 /* Update the variables to point into the current struct net
3594                  * except for the first element flush
3595                  */
3596                 for (i = 1; i < ARRAY_SIZE(ipv4_route_netns_table) - 1; i++)
3597                         tbl[i].data += (void *)net - (void *)&init_net;
3598         }
3599         tbl[0].extra1 = net;
3600
3601         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3602         if (!net->ipv4.route_hdr)
3603                 goto err_reg;
3604         return 0;
3605
3606 err_reg:
3607         if (tbl != ipv4_route_netns_table)
3608                 kfree(tbl);
3609 err_dup:
3610         return -ENOMEM;
3611 }
3612
3613 static __net_exit void sysctl_route_net_exit(struct net *net)
3614 {
3615         struct ctl_table *tbl;
3616
3617         tbl = net->ipv4.route_hdr->ctl_table_arg;
3618         unregister_net_sysctl_table(net->ipv4.route_hdr);
3619         BUG_ON(tbl == ipv4_route_netns_table);
3620         kfree(tbl);
3621 }
3622
3623 static __net_initdata struct pernet_operations sysctl_route_ops = {
3624         .init = sysctl_route_net_init,
3625         .exit = sysctl_route_net_exit,
3626 };
3627 #endif
3628
3629 static __net_init int netns_ip_rt_init(struct net *net)
3630 {
3631         /* Set default value for namespaceified sysctls */
3632         net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3633         net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3634         return 0;
3635 }
3636
3637 static struct pernet_operations __net_initdata ip_rt_ops = {
3638         .init = netns_ip_rt_init,
3639 };
3640
3641 static __net_init int rt_genid_init(struct net *net)
3642 {
3643         atomic_set(&net->ipv4.rt_genid, 0);
3644         atomic_set(&net->fnhe_genid, 0);
3645         atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3646         return 0;
3647 }
3648
3649 static __net_initdata struct pernet_operations rt_genid_ops = {
3650         .init = rt_genid_init,
3651 };
3652
3653 static int __net_init ipv4_inetpeer_init(struct net *net)
3654 {
3655         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3656
3657         if (!bp)
3658                 return -ENOMEM;
3659         inet_peer_base_init(bp);
3660         net->ipv4.peers = bp;
3661         return 0;
3662 }
3663
3664 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3665 {
3666         struct inet_peer_base *bp = net->ipv4.peers;
3667
3668         net->ipv4.peers = NULL;
3669         inetpeer_invalidate_tree(bp);
3670         kfree(bp);
3671 }
3672
3673 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3674         .init   =       ipv4_inetpeer_init,
3675         .exit   =       ipv4_inetpeer_exit,
3676 };
3677
3678 #ifdef CONFIG_IP_ROUTE_CLASSID
3679 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3680 #endif /* CONFIG_IP_ROUTE_CLASSID */
3681
3682 int __init ip_rt_init(void)
3683 {
3684         void *idents_hash;
3685         int cpu;
3686
3687         /* For modern hosts, this will use 2 MB of memory */
3688         idents_hash = alloc_large_system_hash("IP idents",
3689                                               sizeof(*ip_idents) + sizeof(*ip_tstamps),
3690                                               0,
3691                                               16, /* one bucket per 64 KB */
3692                                               HASH_ZERO,
3693                                               NULL,
3694                                               &ip_idents_mask,
3695                                               2048,
3696                                               256*1024);
3697
3698         ip_idents = idents_hash;
3699
3700         prandom_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3701
3702         ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3703
3704         for_each_possible_cpu(cpu) {
3705                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3706
3707                 INIT_LIST_HEAD(&ul->head);
3708                 spin_lock_init(&ul->lock);
3709         }
3710 #ifdef CONFIG_IP_ROUTE_CLASSID
3711         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3712         if (!ip_rt_acct)
3713                 panic("IP: failed to allocate ip_rt_acct\n");
3714 #endif
3715
3716         ipv4_dst_ops.kmem_cachep =
3717                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3718                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3719
3720         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3721
3722         if (dst_entries_init(&ipv4_dst_ops) < 0)
3723                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3724
3725         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3726                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3727
3728         ipv4_dst_ops.gc_thresh = ~0;
3729         ip_rt_max_size = INT_MAX;
3730
3731         devinet_init();
3732         ip_fib_init();
3733
3734         if (ip_rt_proc_init())
3735                 pr_err("Unable to create route proc files\n");
3736 #ifdef CONFIG_XFRM
3737         xfrm_init();
3738         xfrm4_init();
3739 #endif
3740         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3741                       RTNL_FLAG_DOIT_UNLOCKED);
3742
3743 #ifdef CONFIG_SYSCTL
3744         register_pernet_subsys(&sysctl_route_ops);
3745 #endif
3746         register_pernet_subsys(&ip_rt_ops);
3747         register_pernet_subsys(&rt_genid_ops);
3748         register_pernet_subsys(&ipv4_inetpeer_ops);
3749         return 0;
3750 }
3751
3752 #ifdef CONFIG_SYSCTL
3753 /*
3754  * We really need to sanitize the damn ipv4 init order, then all
3755  * this nonsense will go away.
3756  */
3757 void __init ip_static_sysctl_init(void)
3758 {
3759         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3760 }
3761 #endif