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/inet_dscp.h>
  88 #include <net/net_namespace.h>
  89 #include <net/ip.h>
  90 #include <net/route.h>
  91 #include <net/inetpeer.h>
  92 #include <net/sock.h>
  93 #include <net/ip_fib.h>
  94 #include <net/nexthop.h>
  95 #include <net/tcp.h>
  96 #include <net/icmp.h>
  97 #include <net/xfrm.h>
  98 #include <net/lwtunnel.h>
  99 #include <net/netevent.h>
 100 #include <net/rtnetlink.h>
 101 #ifdef CONFIG_SYSCTL
 102 #include <linux/sysctl.h>
 103 #endif
 104 #include <net/secure_seq.h>
 105 #include <net/ip_tunnels.h>
 106
 107 #include "fib_lookup.h"
 108
 109 #define RT_FL_TOS(oldflp4) \
 110         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
 111
 112 #define RT_GC_TIMEOUT (300*HZ)
 113
 114 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
 115 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
 116 #define DEFAULT_MIN_ADVMSS 256
 117 static int ip_rt_max_size;
 118 static int ip_rt_redirect_number __read_mostly  = 9;
 119 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 120 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 121 static int ip_rt_error_cost __read_mostly       = HZ;
 122 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 123
 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();
 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();
 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 static 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 = get_random_u32_below(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
 483 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
 484 {
 485         u32 hash, id;
 486
 487         /* Note the following code is not safe, but this is okay. */
 488         if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
 489                 get_random_bytes(&net->ipv4.ip_id_key,
 490                                  sizeof(net->ipv4.ip_id_key));
 491
 492         hash = siphash_3u32((__force u32)iph->daddr,
 493                             (__force u32)iph->saddr,
 494                             iph->protocol,
 495                             &net->ipv4.ip_id_key);
 496         id = ip_idents_reserve(hash, segs);
 497         iph->id = htons(id);
 498 }
 499 EXPORT_SYMBOL(__ip_select_ident);
 500
 501 static void ip_rt_fix_tos(struct flowi4 *fl4)
 502 {
 503         __u8 tos = RT_FL_TOS(fl4);
 504
 505         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
 506         if (tos & RTO_ONLINK)
 507                 fl4->flowi4_scope = RT_SCOPE_LINK;
 508 }
 509
 510 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
 511                              const struct sock *sk, const struct iphdr *iph,
 512                              int oif, __u8 tos, u8 prot, u32 mark,
 513                              int flow_flags)
 514 {
 515         __u8 scope = RT_SCOPE_UNIVERSE;
 516
 517         if (sk) {
 518                 oif = sk->sk_bound_dev_if;
 519                 mark = READ_ONCE(sk->sk_mark);
 520                 tos = ip_sock_rt_tos(sk);
 521                 scope = ip_sock_rt_scope(sk);
 522                 prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
 523                                                     sk->sk_protocol;
 524         }
 525
 526         flowi4_init_output(fl4, oif, mark, tos & IPTOS_RT_MASK, scope,
 527                            prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
 528                            sock_net_uid(net, sk));
 529 }
 530
 531 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 532                                const struct sock *sk)
 533 {
 534         const struct net *net = dev_net(skb->dev);
 535         const struct iphdr *iph = ip_hdr(skb);
 536         int oif = skb->dev->ifindex;
 537         u8 prot = iph->protocol;
 538         u32 mark = skb->mark;
 539         __u8 tos = iph->tos;
 540
 541         __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
 542 }
 543
 544 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 545 {
 546         const struct inet_sock *inet = inet_sk(sk);
 547         const struct ip_options_rcu *inet_opt;
 548         __be32 daddr = inet->inet_daddr;
 549
 550         rcu_read_lock();
 551         inet_opt = rcu_dereference(inet->inet_opt);
 552         if (inet_opt && inet_opt->opt.srr)
 553                 daddr = inet_opt->opt.faddr;
 554         flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
 555                            ip_sock_rt_tos(sk) & IPTOS_RT_MASK,
 556                            ip_sock_rt_scope(sk),
 557                            inet_test_bit(HDRINCL, sk) ?
 558                                 IPPROTO_RAW : sk->sk_protocol,
 559                            inet_sk_flowi_flags(sk),
 560                            daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
 561         rcu_read_unlock();
 562 }
 563
 564 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 565                                  const struct sk_buff *skb)
 566 {
 567         if (skb)
 568                 build_skb_flow_key(fl4, skb, sk);
 569         else
 570                 build_sk_flow_key(fl4, sk);
 571 }
 572
 573 static DEFINE_SPINLOCK(fnhe_lock);
 574
 575 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 576 {
 577         struct rtable *rt;
 578
 579         rt = rcu_dereference(fnhe->fnhe_rth_input);
 580         if (rt) {
 581                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 582                 dst_dev_put(&rt->dst);
 583                 dst_release(&rt->dst);
 584         }
 585         rt = rcu_dereference(fnhe->fnhe_rth_output);
 586         if (rt) {
 587                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 588                 dst_dev_put(&rt->dst);
 589                 dst_release(&rt->dst);
 590         }
 591 }
 592
 593 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
 594 {
 595         struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
 596         struct fib_nh_exception *fnhe, *oldest = NULL;
 597
 598         for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
 599                 fnhe = rcu_dereference_protected(*fnhe_p,
 600                                                  lockdep_is_held(&fnhe_lock));
 601                 if (!fnhe)
 602                         break;
 603                 if (!oldest ||
 604                     time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
 605                         oldest = fnhe;
 606                         oldest_p = fnhe_p;
 607                 }
 608         }
 609         fnhe_flush_routes(oldest);
 610         *oldest_p = oldest->fnhe_next;
 611         kfree_rcu(oldest, rcu);
 612 }
 613
 614 static u32 fnhe_hashfun(__be32 daddr)
 615 {
 616         static siphash_aligned_key_t fnhe_hash_key;
 617         u64 hval;
 618
 619         net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
 620         hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
 621         return hash_64(hval, FNHE_HASH_SHIFT);
 622 }
 623
 624 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 625 {
 626         rt->rt_pmtu = fnhe->fnhe_pmtu;
 627         rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
 628         rt->dst.expires = fnhe->fnhe_expires;
 629
 630         if (fnhe->fnhe_gw) {
 631                 rt->rt_flags |= RTCF_REDIRECTED;
 632                 rt->rt_uses_gateway = 1;
 633                 rt->rt_gw_family = AF_INET;
 634                 rt->rt_gw4 = fnhe->fnhe_gw;
 635         }
 636 }
 637
 638 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
 639                                   __be32 gw, u32 pmtu, bool lock,
 640                                   unsigned long expires)
 641 {
 642         struct fnhe_hash_bucket *hash;
 643         struct fib_nh_exception *fnhe;
 644         struct rtable *rt;
 645         u32 genid, hval;
 646         unsigned int i;
 647         int depth;
 648
 649         genid = fnhe_genid(dev_net(nhc->nhc_dev));
 650         hval = fnhe_hashfun(daddr);
 651
 652         spin_lock_bh(&fnhe_lock);
 653
 654         hash = rcu_dereference(nhc->nhc_exceptions);
 655         if (!hash) {
 656                 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
 657                 if (!hash)
 658                         goto out_unlock;
 659                 rcu_assign_pointer(nhc->nhc_exceptions, hash);
 660         }
 661
 662         hash += hval;
 663
 664         depth = 0;
 665         for (fnhe = rcu_dereference(hash->chain); fnhe;
 666              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 667                 if (fnhe->fnhe_daddr == daddr)
 668                         break;
 669                 depth++;
 670         }
 671
 672         if (fnhe) {
 673                 if (fnhe->fnhe_genid != genid)
 674                         fnhe->fnhe_genid = genid;
 675                 if (gw)
 676                         fnhe->fnhe_gw = gw;
 677                 if (pmtu) {
 678                         fnhe->fnhe_pmtu = pmtu;
 679                         fnhe->fnhe_mtu_locked = lock;
 680                 }
 681                 fnhe->fnhe_expires = max(1UL, expires);
 682                 /* Update all cached dsts too */
 683                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 684                 if (rt)
 685                         fill_route_from_fnhe(rt, fnhe);
 686                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 687                 if (rt)
 688                         fill_route_from_fnhe(rt, fnhe);
 689         } else {
 690                 /* Randomize max depth to avoid some side channels attacks. */
 691                 int max_depth = FNHE_RECLAIM_DEPTH +
 692                                 get_random_u32_below(FNHE_RECLAIM_DEPTH);
 693
 694                 while (depth > max_depth) {
 695                         fnhe_remove_oldest(hash);
 696                         depth--;
 697                 }
 698
 699                 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 700                 if (!fnhe)
 701                         goto out_unlock;
 702
 703                 fnhe->fnhe_next = hash->chain;
 704
 705                 fnhe->fnhe_genid = genid;
 706                 fnhe->fnhe_daddr = daddr;
 707                 fnhe->fnhe_gw = gw;
 708                 fnhe->fnhe_pmtu = pmtu;
 709                 fnhe->fnhe_mtu_locked = lock;
 710                 fnhe->fnhe_expires = max(1UL, expires);
 711
 712                 rcu_assign_pointer(hash->chain, fnhe);
 713
 714                 /* Exception created; mark the cached routes for the nexthop
 715                  * stale, so anyone caching it rechecks if this exception
 716                  * applies to them.
 717                  */
 718                 rt = rcu_dereference(nhc->nhc_rth_input);
 719                 if (rt)
 720                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 721
 722                 for_each_possible_cpu(i) {
 723                         struct rtable __rcu **prt;
 724
 725                         prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
 726                         rt = rcu_dereference(*prt);
 727                         if (rt)
 728                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 729                 }
 730         }
 731
 732         fnhe->fnhe_stamp = jiffies;
 733
 734 out_unlock:
 735         spin_unlock_bh(&fnhe_lock);
 736 }
 737
 738 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 739                              bool kill_route)
 740 {
 741         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 742         __be32 old_gw = ip_hdr(skb)->saddr;
 743         struct net_device *dev = skb->dev;
 744         struct in_device *in_dev;
 745         struct fib_result res;
 746         struct neighbour *n;
 747         struct net *net;
 748
 749         switch (icmp_hdr(skb)->code & 7) {
 750         case ICMP_REDIR_NET:
 751         case ICMP_REDIR_NETTOS:
 752         case ICMP_REDIR_HOST:
 753         case ICMP_REDIR_HOSTTOS:
 754                 break;
 755
 756         default:
 757                 return;
 758         }
 759
 760         if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
 761                 return;
 762
 763         in_dev = __in_dev_get_rcu(dev);
 764         if (!in_dev)
 765                 return;
 766
 767         net = dev_net(dev);
 768         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 769             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 770             ipv4_is_zeronet(new_gw))
 771                 goto reject_redirect;
 772
 773         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 774                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 775                         goto reject_redirect;
 776                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 777                         goto reject_redirect;
 778         } else {
 779                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 780                         goto reject_redirect;
 781         }
 782
 783         n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
 784         if (!n)
 785                 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
 786         if (!IS_ERR(n)) {
 787                 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
 788                         neigh_event_send(n, NULL);
 789                 } else {
 790                         if (fib_lookup(net, fl4, &res, 0) == 0) {
 791                                 struct fib_nh_common *nhc;
 792
 793                                 fib_select_path(net, &res, fl4, skb);
 794                                 nhc = FIB_RES_NHC(res);
 795                                 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
 796                                                 0, false,
 797                                                 jiffies + ip_rt_gc_timeout);
 798                         }
 799                         if (kill_route)
 800                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 801                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 802                 }
 803                 neigh_release(n);
 804         }
 805         return;
 806
 807 reject_redirect:
 808 #ifdef CONFIG_IP_ROUTE_VERBOSE
 809         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 810                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 811                 __be32 daddr = iph->daddr;
 812                 __be32 saddr = iph->saddr;
 813
 814                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 815                                      "  Advised path = %pI4 -> %pI4\n",
 816                                      &old_gw, dev->name, &new_gw,
 817                                      &saddr, &daddr);
 818         }
 819 #endif
 820         ;
 821 }
 822
 823 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 824 {
 825         struct rtable *rt;
 826         struct flowi4 fl4;
 827         const struct iphdr *iph = (const struct iphdr *) skb->data;
 828         struct net *net = dev_net(skb->dev);
 829         int oif = skb->dev->ifindex;
 830         u8 prot = iph->protocol;
 831         u32 mark = skb->mark;
 832         __u8 tos = iph->tos;
 833
 834         rt = (struct rtable *) dst;
 835
 836         __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
 837         __ip_do_redirect(rt, skb, &fl4, true);
 838 }
 839
 840 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 841 {
 842         struct rtable *rt = (struct rtable *)dst;
 843         struct dst_entry *ret = dst;
 844
 845         if (rt) {
 846                 if (dst->obsolete > 0) {
 847                         ip_rt_put(rt);
 848                         ret = NULL;
 849                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 850                            rt->dst.expires) {
 851                         ip_rt_put(rt);
 852                         ret = NULL;
 853                 }
 854         }
 855         return ret;
 856 }
 857
 858 /*
 859  * Algorithm:
 860  *      1. The first ip_rt_redirect_number redirects are sent
 861  *         with exponential backoff, then we stop sending them at all,
 862  *         assuming that the host ignores our redirects.
 863  *      2. If we did not see packets requiring redirects
 864  *         during ip_rt_redirect_silence, we assume that the host
 865  *         forgot redirected route and start to send redirects again.
 866  *
 867  * This algorithm is much cheaper and more intelligent than dumb load limiting
 868  * in icmp.c.
 869  *
 870  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 871  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 872  */
 873
 874 void ip_rt_send_redirect(struct sk_buff *skb)
 875 {
 876         struct rtable *rt = skb_rtable(skb);
 877         struct in_device *in_dev;
 878         struct inet_peer *peer;
 879         struct net *net;
 880         int log_martians;
 881         int vif;
 882
 883         rcu_read_lock();
 884         in_dev = __in_dev_get_rcu(rt->dst.dev);
 885         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 886                 rcu_read_unlock();
 887                 return;
 888         }
 889         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 890         vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
 891         rcu_read_unlock();
 892
 893         net = dev_net(rt->dst.dev);
 894         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
 895         if (!peer) {
 896                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 897                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 898                 return;
 899         }
 900
 901         /* No redirected packets during ip_rt_redirect_silence;
 902          * reset the algorithm.
 903          */
 904         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
 905                 peer->rate_tokens = 0;
 906                 peer->n_redirects = 0;
 907         }
 908
 909         /* Too many ignored redirects; do not send anything
 910          * set dst.rate_last to the last seen redirected packet.
 911          */
 912         if (peer->n_redirects >= ip_rt_redirect_number) {
 913                 peer->rate_last = jiffies;
 914                 goto out_put_peer;
 915         }
 916
 917         /* Check for load limit; set rate_last to the latest sent
 918          * redirect.
 919          */
 920         if (peer->n_redirects == 0 ||
 921             time_after(jiffies,
 922                        (peer->rate_last +
 923                         (ip_rt_redirect_load << peer->n_redirects)))) {
 924                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 925
 926                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 927                 peer->rate_last = jiffies;
 928                 ++peer->n_redirects;
 929 #ifdef CONFIG_IP_ROUTE_VERBOSE
 930                 if (log_martians &&
 931                     peer->n_redirects == ip_rt_redirect_number)
 932                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 933                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 934                                              &ip_hdr(skb)->daddr, &gw);
 935 #endif
 936         }
 937 out_put_peer:
 938         inet_putpeer(peer);
 939 }
 940
 941 static int ip_error(struct sk_buff *skb)
 942 {
 943         struct rtable *rt = skb_rtable(skb);
 944         struct net_device *dev = skb->dev;
 945         struct in_device *in_dev;
 946         struct inet_peer *peer;
 947         unsigned long now;
 948         struct net *net;
 949         SKB_DR(reason);
 950         bool send;
 951         int code;
 952
 953         if (netif_is_l3_master(skb->dev)) {
 954                 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
 955                 if (!dev)
 956                         goto out;
 957         }
 958
 959         in_dev = __in_dev_get_rcu(dev);
 960
 961         /* IP on this device is disabled. */
 962         if (!in_dev)
 963                 goto out;
 964
 965         net = dev_net(rt->dst.dev);
 966         if (!IN_DEV_FORWARD(in_dev)) {
 967                 switch (rt->dst.error) {
 968                 case EHOSTUNREACH:
 969                         SKB_DR_SET(reason, IP_INADDRERRORS);
 970                         __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
 971                         break;
 972
 973                 case ENETUNREACH:
 974                         SKB_DR_SET(reason, IP_INNOROUTES);
 975                         __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 976                         break;
 977                 }
 978                 goto out;
 979         }
 980
 981         switch (rt->dst.error) {
 982         case EINVAL:
 983         default:
 984                 goto out;
 985         case EHOSTUNREACH:
 986                 code = ICMP_HOST_UNREACH;
 987                 break;
 988         case ENETUNREACH:
 989                 code = ICMP_NET_UNREACH;
 990                 SKB_DR_SET(reason, IP_INNOROUTES);
 991                 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 992                 break;
 993         case EACCES:
 994                 code = ICMP_PKT_FILTERED;
 995                 break;
 996         }
 997
 998         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
 999                                l3mdev_master_ifindex(skb->dev), 1);
1000
1001         send = true;
1002         if (peer) {
1003                 now = jiffies;
1004                 peer->rate_tokens += now - peer->rate_last;
1005                 if (peer->rate_tokens > ip_rt_error_burst)
1006                         peer->rate_tokens = ip_rt_error_burst;
1007                 peer->rate_last = now;
1008                 if (peer->rate_tokens >= ip_rt_error_cost)
1009                         peer->rate_tokens -= ip_rt_error_cost;
1010                 else
1011                         send = false;
1012                 inet_putpeer(peer);
1013         }
1014         if (send)
1015                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1016
1017 out:    kfree_skb_reason(skb, reason);
1018         return 0;
1019 }
1020
1021 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1022 {
1023         struct dst_entry *dst = &rt->dst;
1024         struct net *net = dev_net(dst->dev);
1025         struct fib_result res;
1026         bool lock = false;
1027         u32 old_mtu;
1028
1029         if (ip_mtu_locked(dst))
1030                 return;
1031
1032         old_mtu = ipv4_mtu(dst);
1033         if (old_mtu < mtu)
1034                 return;
1035
1036         if (mtu < net->ipv4.ip_rt_min_pmtu) {
1037                 lock = true;
1038                 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1039         }
1040
1041         if (rt->rt_pmtu == mtu && !lock &&
1042             time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1043                 return;
1044
1045         rcu_read_lock();
1046         if (fib_lookup(net, fl4, &res, 0) == 0) {
1047                 struct fib_nh_common *nhc;
1048
1049                 fib_select_path(net, &res, fl4, NULL);
1050                 nhc = FIB_RES_NHC(res);
1051                 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1052                                       jiffies + net->ipv4.ip_rt_mtu_expires);
1053         }
1054         rcu_read_unlock();
1055 }
1056
1057 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1058                               struct sk_buff *skb, u32 mtu,
1059                               bool confirm_neigh)
1060 {
1061         struct rtable *rt = (struct rtable *) dst;
1062         struct flowi4 fl4;
1063
1064         ip_rt_build_flow_key(&fl4, sk, skb);
1065
1066         /* Don't make lookup fail for bridged encapsulations */
1067         if (skb && netif_is_any_bridge_port(skb->dev))
1068                 fl4.flowi4_oif = 0;
1069
1070         __ip_rt_update_pmtu(rt, &fl4, mtu);
1071 }
1072
1073 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1074                       int oif, u8 protocol)
1075 {
1076         const struct iphdr *iph = (const struct iphdr *)skb->data;
1077         struct flowi4 fl4;
1078         struct rtable *rt;
1079         u32 mark = IP4_REPLY_MARK(net, skb->mark);
1080
1081         __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1082                          0);
1083         rt = __ip_route_output_key(net, &fl4);
1084         if (!IS_ERR(rt)) {
1085                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1086                 ip_rt_put(rt);
1087         }
1088 }
1089 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1090
1091 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1092 {
1093         const struct iphdr *iph = (const struct iphdr *)skb->data;
1094         struct flowi4 fl4;
1095         struct rtable *rt;
1096
1097         __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1098
1099         if (!fl4.flowi4_mark)
1100                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1101
1102         rt = __ip_route_output_key(sock_net(sk), &fl4);
1103         if (!IS_ERR(rt)) {
1104                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1105                 ip_rt_put(rt);
1106         }
1107 }
1108
1109 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1110 {
1111         const struct iphdr *iph = (const struct iphdr *)skb->data;
1112         struct flowi4 fl4;
1113         struct rtable *rt;
1114         struct dst_entry *odst = NULL;
1115         bool new = false;
1116         struct net *net = sock_net(sk);
1117
1118         bh_lock_sock(sk);
1119
1120         if (!ip_sk_accept_pmtu(sk))
1121                 goto out;
1122
1123         odst = sk_dst_get(sk);
1124
1125         if (sock_owned_by_user(sk) || !odst) {
1126                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1127                 goto out;
1128         }
1129
1130         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1131
1132         rt = (struct rtable *)odst;
1133         if (odst->obsolete && !odst->ops->check(odst, 0)) {
1134                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1135                 if (IS_ERR(rt))
1136                         goto out;
1137
1138                 new = true;
1139         }
1140
1141         __ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu);
1142
1143         if (!dst_check(&rt->dst, 0)) {
1144                 if (new)
1145                         dst_release(&rt->dst);
1146
1147                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1148                 if (IS_ERR(rt))
1149                         goto out;
1150
1151                 new = true;
1152         }
1153
1154         if (new)
1155                 sk_dst_set(sk, &rt->dst);
1156
1157 out:
1158         bh_unlock_sock(sk);
1159         dst_release(odst);
1160 }
1161 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1162
1163 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1164                    int oif, u8 protocol)
1165 {
1166         const struct iphdr *iph = (const struct iphdr *)skb->data;
1167         struct flowi4 fl4;
1168         struct rtable *rt;
1169
1170         __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1171         rt = __ip_route_output_key(net, &fl4);
1172         if (!IS_ERR(rt)) {
1173                 __ip_do_redirect(rt, skb, &fl4, false);
1174                 ip_rt_put(rt);
1175         }
1176 }
1177 EXPORT_SYMBOL_GPL(ipv4_redirect);
1178
1179 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1180 {
1181         const struct iphdr *iph = (const struct iphdr *)skb->data;
1182         struct flowi4 fl4;
1183         struct rtable *rt;
1184         struct net *net = sock_net(sk);
1185
1186         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1187         rt = __ip_route_output_key(net, &fl4);
1188         if (!IS_ERR(rt)) {
1189                 __ip_do_redirect(rt, skb, &fl4, false);
1190                 ip_rt_put(rt);
1191         }
1192 }
1193 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1194
1195 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1196                                                          u32 cookie)
1197 {
1198         struct rtable *rt = (struct rtable *) dst;
1199
1200         /* All IPV4 dsts are created with ->obsolete set to the value
1201          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1202          * into this function always.
1203          *
1204          * When a PMTU/redirect information update invalidates a route,
1205          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1206          * DST_OBSOLETE_DEAD.
1207          */
1208         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1209                 return NULL;
1210         return dst;
1211 }
1212 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1213
1214 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1215 {
1216         struct ip_options opt;
1217         int res;
1218
1219         /* Recompile ip options since IPCB may not be valid anymore.
1220          * Also check we have a reasonable ipv4 header.
1221          */
1222         if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1223             ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1224                 return;
1225
1226         memset(&opt, 0, sizeof(opt));
1227         if (ip_hdr(skb)->ihl > 5) {
1228                 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1229                         return;
1230                 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1231
1232                 rcu_read_lock();
1233                 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1234                 rcu_read_unlock();
1235
1236                 if (res)
1237                         return;
1238         }
1239         __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1240 }
1241
1242 static void ipv4_link_failure(struct sk_buff *skb)
1243 {
1244         struct rtable *rt;
1245
1246         ipv4_send_dest_unreach(skb);
1247
1248         rt = skb_rtable(skb);
1249         if (rt)
1250                 dst_set_expires(&rt->dst, 0);
1251 }
1252
1253 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1254 {
1255         pr_debug("%s: %pI4 -> %pI4, %s\n",
1256                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1257                  skb->dev ? skb->dev->name : "?");
1258         kfree_skb(skb);
1259         WARN_ON(1);
1260         return 0;
1261 }
1262
1263 /*
1264  * We do not cache source address of outgoing interface,
1265  * because it is used only by IP RR, TS and SRR options,
1266  * so that it out of fast path.
1267  *
1268  * BTW remember: "addr" is allowed to be not aligned
1269  * in IP options!
1270  */
1271
1272 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1273 {
1274         __be32 src;
1275
1276         if (rt_is_output_route(rt))
1277                 src = ip_hdr(skb)->saddr;
1278         else {
1279                 struct fib_result res;
1280                 struct iphdr *iph = ip_hdr(skb);
1281                 struct flowi4 fl4 = {
1282                         .daddr = iph->daddr,
1283                         .saddr = iph->saddr,
1284                         .flowi4_tos = RT_TOS(iph->tos),
1285                         .flowi4_oif = rt->dst.dev->ifindex,
1286                         .flowi4_iif = skb->dev->ifindex,
1287                         .flowi4_mark = skb->mark,
1288                 };
1289
1290                 rcu_read_lock();
1291                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1292                         src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1293                 else
1294                         src = inet_select_addr(rt->dst.dev,
1295                                                rt_nexthop(rt, iph->daddr),
1296                                                RT_SCOPE_UNIVERSE);
1297                 rcu_read_unlock();
1298         }
1299         memcpy(addr, &src, 4);
1300 }
1301
1302 #ifdef CONFIG_IP_ROUTE_CLASSID
1303 static void set_class_tag(struct rtable *rt, u32 tag)
1304 {
1305         if (!(rt->dst.tclassid & 0xFFFF))
1306                 rt->dst.tclassid |= tag & 0xFFFF;
1307         if (!(rt->dst.tclassid & 0xFFFF0000))
1308                 rt->dst.tclassid |= tag & 0xFFFF0000;
1309 }
1310 #endif
1311
1312 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1313 {
1314         struct net *net = dev_net(dst->dev);
1315         unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1316         unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1317                                     net->ipv4.ip_rt_min_advmss);
1318
1319         return min(advmss, IPV4_MAX_PMTU - header_size);
1320 }
1321
1322 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1323 {
1324         return ip_dst_mtu_maybe_forward(dst, false);
1325 }
1326 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1327
1328 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1329 {
1330         struct fnhe_hash_bucket *hash;
1331         struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1332         u32 hval = fnhe_hashfun(daddr);
1333
1334         spin_lock_bh(&fnhe_lock);
1335
1336         hash = rcu_dereference_protected(nhc->nhc_exceptions,
1337                                          lockdep_is_held(&fnhe_lock));
1338         hash += hval;
1339
1340         fnhe_p = &hash->chain;
1341         fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1342         while (fnhe) {
1343                 if (fnhe->fnhe_daddr == daddr) {
1344                         rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1345                                 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1346                         /* set fnhe_daddr to 0 to ensure it won't bind with
1347                          * new dsts in rt_bind_exception().
1348                          */
1349                         fnhe->fnhe_daddr = 0;
1350                         fnhe_flush_routes(fnhe);
1351                         kfree_rcu(fnhe, rcu);
1352                         break;
1353                 }
1354                 fnhe_p = &fnhe->fnhe_next;
1355                 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1356                                                  lockdep_is_held(&fnhe_lock));
1357         }
1358
1359         spin_unlock_bh(&fnhe_lock);
1360 }
1361
1362 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1363                                                __be32 daddr)
1364 {
1365         struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1366         struct fib_nh_exception *fnhe;
1367         u32 hval;
1368
1369         if (!hash)
1370                 return NULL;
1371
1372         hval = fnhe_hashfun(daddr);
1373
1374         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1375              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1376                 if (fnhe->fnhe_daddr == daddr) {
1377                         if (fnhe->fnhe_expires &&
1378                             time_after(jiffies, fnhe->fnhe_expires)) {
1379                                 ip_del_fnhe(nhc, daddr);
1380                                 break;
1381                         }
1382                         return fnhe;
1383                 }
1384         }
1385         return NULL;
1386 }
1387
1388 /* MTU selection:
1389  * 1. mtu on route is locked - use it
1390  * 2. mtu from nexthop exception
1391  * 3. mtu from egress device
1392  */
1393
1394 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1395 {
1396         struct fib_nh_common *nhc = res->nhc;
1397         struct net_device *dev = nhc->nhc_dev;
1398         struct fib_info *fi = res->fi;
1399         u32 mtu = 0;
1400
1401         if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1402             fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1403                 mtu = fi->fib_mtu;
1404
1405         if (likely(!mtu)) {
1406                 struct fib_nh_exception *fnhe;
1407
1408                 fnhe = find_exception(nhc, daddr);
1409                 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1410                         mtu = fnhe->fnhe_pmtu;
1411         }
1412
1413         if (likely(!mtu))
1414                 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1415
1416         return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1417 }
1418
1419 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1420                               __be32 daddr, const bool do_cache)
1421 {
1422         bool ret = false;
1423
1424         spin_lock_bh(&fnhe_lock);
1425
1426         if (daddr == fnhe->fnhe_daddr) {
1427                 struct rtable __rcu **porig;
1428                 struct rtable *orig;
1429                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1430
1431                 if (rt_is_input_route(rt))
1432                         porig = &fnhe->fnhe_rth_input;
1433                 else
1434                         porig = &fnhe->fnhe_rth_output;
1435                 orig = rcu_dereference(*porig);
1436
1437                 if (fnhe->fnhe_genid != genid) {
1438                         fnhe->fnhe_genid = genid;
1439                         fnhe->fnhe_gw = 0;
1440                         fnhe->fnhe_pmtu = 0;
1441                         fnhe->fnhe_expires = 0;
1442                         fnhe->fnhe_mtu_locked = false;
1443                         fnhe_flush_routes(fnhe);
1444                         orig = NULL;
1445                 }
1446                 fill_route_from_fnhe(rt, fnhe);
1447                 if (!rt->rt_gw4) {
1448                         rt->rt_gw4 = daddr;
1449                         rt->rt_gw_family = AF_INET;
1450                 }
1451
1452                 if (do_cache) {
1453                         dst_hold(&rt->dst);
1454                         rcu_assign_pointer(*porig, rt);
1455                         if (orig) {
1456                                 dst_dev_put(&orig->dst);
1457                                 dst_release(&orig->dst);
1458                         }
1459                         ret = true;
1460                 }
1461
1462                 fnhe->fnhe_stamp = jiffies;
1463         }
1464         spin_unlock_bh(&fnhe_lock);
1465
1466         return ret;
1467 }
1468
1469 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1470 {
1471         struct rtable *orig, *prev, **p;
1472         bool ret = true;
1473
1474         if (rt_is_input_route(rt)) {
1475                 p = (struct rtable **)&nhc->nhc_rth_input;
1476         } else {
1477                 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1478         }
1479         orig = *p;
1480
1481         /* hold dst before doing cmpxchg() to avoid race condition
1482          * on this dst
1483          */
1484         dst_hold(&rt->dst);
1485         prev = cmpxchg(p, orig, rt);
1486         if (prev == orig) {
1487                 if (orig) {
1488                         rt_add_uncached_list(orig);
1489                         dst_release(&orig->dst);
1490                 }
1491         } else {
1492                 dst_release(&rt->dst);
1493                 ret = false;
1494         }
1495
1496         return ret;
1497 }
1498
1499 struct uncached_list {
1500         spinlock_t              lock;
1501         struct list_head        head;
1502         struct list_head        quarantine;
1503 };
1504
1505 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1506
1507 void rt_add_uncached_list(struct rtable *rt)
1508 {
1509         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1510
1511         rt->dst.rt_uncached_list = ul;
1512
1513         spin_lock_bh(&ul->lock);
1514         list_add_tail(&rt->dst.rt_uncached, &ul->head);
1515         spin_unlock_bh(&ul->lock);
1516 }
1517
1518 void rt_del_uncached_list(struct rtable *rt)
1519 {
1520         if (!list_empty(&rt->dst.rt_uncached)) {
1521                 struct uncached_list *ul = rt->dst.rt_uncached_list;
1522
1523                 spin_lock_bh(&ul->lock);
1524                 list_del_init(&rt->dst.rt_uncached);
1525                 spin_unlock_bh(&ul->lock);
1526         }
1527 }
1528
1529 static void ipv4_dst_destroy(struct dst_entry *dst)
1530 {
1531         struct rtable *rt = (struct rtable *)dst;
1532
1533         ip_dst_metrics_put(dst);
1534         rt_del_uncached_list(rt);
1535 }
1536
1537 void rt_flush_dev(struct net_device *dev)
1538 {
1539         struct rtable *rt, *safe;
1540         int cpu;
1541
1542         for_each_possible_cpu(cpu) {
1543                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1544
1545                 if (list_empty(&ul->head))
1546                         continue;
1547
1548                 spin_lock_bh(&ul->lock);
1549                 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1550                         if (rt->dst.dev != dev)
1551                                 continue;
1552                         rt->dst.dev = blackhole_netdev;
1553                         netdev_ref_replace(dev, blackhole_netdev,
1554                                            &rt->dst.dev_tracker, GFP_ATOMIC);
1555                         list_move(&rt->dst.rt_uncached, &ul->quarantine);
1556                 }
1557                 spin_unlock_bh(&ul->lock);
1558         }
1559 }
1560
1561 static bool rt_cache_valid(const struct rtable *rt)
1562 {
1563         return  rt &&
1564                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1565                 !rt_is_expired(rt);
1566 }
1567
1568 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1569                            const struct fib_result *res,
1570                            struct fib_nh_exception *fnhe,
1571                            struct fib_info *fi, u16 type, u32 itag,
1572                            const bool do_cache)
1573 {
1574         bool cached = false;
1575
1576         if (fi) {
1577                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1578
1579                 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1580                         rt->rt_uses_gateway = 1;
1581                         rt->rt_gw_family = nhc->nhc_gw_family;
1582                         /* only INET and INET6 are supported */
1583                         if (likely(nhc->nhc_gw_family == AF_INET))
1584                                 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1585                         else
1586                                 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1587                 }
1588
1589                 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1590
1591 #ifdef CONFIG_IP_ROUTE_CLASSID
1592                 if (nhc->nhc_family == AF_INET) {
1593                         struct fib_nh *nh;
1594
1595                         nh = container_of(nhc, struct fib_nh, nh_common);
1596                         rt->dst.tclassid = nh->nh_tclassid;
1597                 }
1598 #endif
1599                 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1600                 if (unlikely(fnhe))
1601                         cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1602                 else if (do_cache)
1603                         cached = rt_cache_route(nhc, rt);
1604                 if (unlikely(!cached)) {
1605                         /* Routes we intend to cache in nexthop exception or
1606                          * FIB nexthop have the DST_NOCACHE bit clear.
1607                          * However, if we are unsuccessful at storing this
1608                          * route into the cache we really need to set it.
1609                          */
1610                         if (!rt->rt_gw4) {
1611                                 rt->rt_gw_family = AF_INET;
1612                                 rt->rt_gw4 = daddr;
1613                         }
1614                         rt_add_uncached_list(rt);
1615                 }
1616         } else
1617                 rt_add_uncached_list(rt);
1618
1619 #ifdef CONFIG_IP_ROUTE_CLASSID
1620 #ifdef CONFIG_IP_MULTIPLE_TABLES
1621         set_class_tag(rt, res->tclassid);
1622 #endif
1623         set_class_tag(rt, itag);
1624 #endif
1625 }
1626
1627 struct rtable *rt_dst_alloc(struct net_device *dev,
1628                             unsigned int flags, u16 type,
1629                             bool noxfrm)
1630 {
1631         struct rtable *rt;
1632
1633         rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1634                        (noxfrm ? DST_NOXFRM : 0));
1635
1636         if (rt) {
1637                 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1638                 rt->rt_flags = flags;
1639                 rt->rt_type = type;
1640                 rt->rt_is_input = 0;
1641                 rt->rt_iif = 0;
1642                 rt->rt_pmtu = 0;
1643                 rt->rt_mtu_locked = 0;
1644                 rt->rt_uses_gateway = 0;
1645                 rt->rt_gw_family = 0;
1646                 rt->rt_gw4 = 0;
1647
1648                 rt->dst.output = ip_output;
1649                 if (flags & RTCF_LOCAL)
1650                         rt->dst.input = ip_local_deliver;
1651         }
1652
1653         return rt;
1654 }
1655 EXPORT_SYMBOL(rt_dst_alloc);
1656
1657 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1658 {
1659         struct rtable *new_rt;
1660
1661         new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1662                            rt->dst.flags);
1663
1664         if (new_rt) {
1665                 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1666                 new_rt->rt_flags = rt->rt_flags;
1667                 new_rt->rt_type = rt->rt_type;
1668                 new_rt->rt_is_input = rt->rt_is_input;
1669                 new_rt->rt_iif = rt->rt_iif;
1670                 new_rt->rt_pmtu = rt->rt_pmtu;
1671                 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1672                 new_rt->rt_gw_family = rt->rt_gw_family;
1673                 if (rt->rt_gw_family == AF_INET)
1674                         new_rt->rt_gw4 = rt->rt_gw4;
1675                 else if (rt->rt_gw_family == AF_INET6)
1676                         new_rt->rt_gw6 = rt->rt_gw6;
1677
1678                 new_rt->dst.input = rt->dst.input;
1679                 new_rt->dst.output = rt->dst.output;
1680                 new_rt->dst.error = rt->dst.error;
1681                 new_rt->dst.lastuse = jiffies;
1682                 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1683         }
1684         return new_rt;
1685 }
1686 EXPORT_SYMBOL(rt_dst_clone);
1687
1688 /* called in rcu_read_lock() section */
1689 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1690                           u8 tos, struct net_device *dev,
1691                           struct in_device *in_dev, u32 *itag)
1692 {
1693         int err;
1694
1695         /* Primary sanity checks. */
1696         if (!in_dev)
1697                 return -EINVAL;
1698
1699         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1700             skb->protocol != htons(ETH_P_IP))
1701                 return -EINVAL;
1702
1703         if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1704                 return -EINVAL;
1705
1706         if (ipv4_is_zeronet(saddr)) {
1707                 if (!ipv4_is_local_multicast(daddr) &&
1708                     ip_hdr(skb)->protocol != IPPROTO_IGMP)
1709                         return -EINVAL;
1710         } else {
1711                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1712                                           in_dev, itag);
1713                 if (err < 0)
1714                         return err;
1715         }
1716         return 0;
1717 }
1718
1719 /* called in rcu_read_lock() section */
1720 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1721                              u8 tos, struct net_device *dev, int our)
1722 {
1723         struct in_device *in_dev = __in_dev_get_rcu(dev);
1724         unsigned int flags = RTCF_MULTICAST;
1725         struct rtable *rth;
1726         u32 itag = 0;
1727         int err;
1728
1729         err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1730         if (err)
1731                 return err;
1732
1733         if (our)
1734                 flags |= RTCF_LOCAL;
1735
1736         if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1737                 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1738
1739         rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1740                            false);
1741         if (!rth)
1742                 return -ENOBUFS;
1743
1744 #ifdef CONFIG_IP_ROUTE_CLASSID
1745         rth->dst.tclassid = itag;
1746 #endif
1747         rth->dst.output = ip_rt_bug;
1748         rth->rt_is_input= 1;
1749
1750 #ifdef CONFIG_IP_MROUTE
1751         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1752                 rth->dst.input = ip_mr_input;
1753 #endif
1754         RT_CACHE_STAT_INC(in_slow_mc);
1755
1756         skb_dst_drop(skb);
1757         skb_dst_set(skb, &rth->dst);
1758         return 0;
1759 }
1760
1761
1762 static void ip_handle_martian_source(struct net_device *dev,
1763                                      struct in_device *in_dev,
1764                                      struct sk_buff *skb,
1765                                      __be32 daddr,
1766                                      __be32 saddr)
1767 {
1768         RT_CACHE_STAT_INC(in_martian_src);
1769 #ifdef CONFIG_IP_ROUTE_VERBOSE
1770         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1771                 /*
1772                  *      RFC1812 recommendation, if source is martian,
1773                  *      the only hint is MAC header.
1774                  */
1775                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1776                         &daddr, &saddr, dev->name);
1777                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1778                         print_hex_dump(KERN_WARNING, "ll header: ",
1779                                        DUMP_PREFIX_OFFSET, 16, 1,
1780                                        skb_mac_header(skb),
1781                                        dev->hard_header_len, false);
1782                 }
1783         }
1784 #endif
1785 }
1786
1787 /* called in rcu_read_lock() section */
1788 static int __mkroute_input(struct sk_buff *skb,
1789                            const struct fib_result *res,
1790                            struct in_device *in_dev,
1791                            __be32 daddr, __be32 saddr, u32 tos)
1792 {
1793         struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1794         struct net_device *dev = nhc->nhc_dev;
1795         struct fib_nh_exception *fnhe;
1796         struct rtable *rth;
1797         int err;
1798         struct in_device *out_dev;
1799         bool do_cache;
1800         u32 itag = 0;
1801
1802         /* get a working reference to the output device */
1803         out_dev = __in_dev_get_rcu(dev);
1804         if (!out_dev) {
1805                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1806                 return -EINVAL;
1807         }
1808
1809         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1810                                   in_dev->dev, in_dev, &itag);
1811         if (err < 0) {
1812                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1813                                          saddr);
1814
1815                 goto cleanup;
1816         }
1817
1818         do_cache = res->fi && !itag;
1819         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1820             skb->protocol == htons(ETH_P_IP)) {
1821                 __be32 gw;
1822
1823                 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1824                 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1825                     inet_addr_onlink(out_dev, saddr, gw))
1826                         IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1827         }
1828
1829         if (skb->protocol != htons(ETH_P_IP)) {
1830                 /* Not IP (i.e. ARP). Do not create route, if it is
1831                  * invalid for proxy arp. DNAT routes are always valid.
1832                  *
1833                  * Proxy arp feature have been extended to allow, ARP
1834                  * replies back to the same interface, to support
1835                  * Private VLAN switch technologies. See arp.c.
1836                  */
1837                 if (out_dev == in_dev &&
1838                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1839                         err = -EINVAL;
1840                         goto cleanup;
1841                 }
1842         }
1843
1844         if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1845                 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1846
1847         fnhe = find_exception(nhc, daddr);
1848         if (do_cache) {
1849                 if (fnhe)
1850                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1851                 else
1852                         rth = rcu_dereference(nhc->nhc_rth_input);
1853                 if (rt_cache_valid(rth)) {
1854                         skb_dst_set_noref(skb, &rth->dst);
1855                         goto out;
1856                 }
1857         }
1858
1859         rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1860                            IN_DEV_ORCONF(out_dev, NOXFRM));
1861         if (!rth) {
1862                 err = -ENOBUFS;
1863                 goto cleanup;
1864         }
1865
1866         rth->rt_is_input = 1;
1867         RT_CACHE_STAT_INC(in_slow_tot);
1868
1869         rth->dst.input = ip_forward;
1870
1871         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1872                        do_cache);
1873         lwtunnel_set_redirect(&rth->dst);
1874         skb_dst_set(skb, &rth->dst);
1875 out:
1876         err = 0;
1877  cleanup:
1878         return err;
1879 }
1880
1881 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1882 /* To make ICMP packets follow the right flow, the multipath hash is
1883  * calculated from the inner IP addresses.
1884  */
1885 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1886                                  struct flow_keys *hash_keys)
1887 {
1888         const struct iphdr *outer_iph = ip_hdr(skb);
1889         const struct iphdr *key_iph = outer_iph;
1890         const struct iphdr *inner_iph;
1891         const struct icmphdr *icmph;
1892         struct iphdr _inner_iph;
1893         struct icmphdr _icmph;
1894
1895         if (likely(outer_iph->protocol != IPPROTO_ICMP))
1896                 goto out;
1897
1898         if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1899                 goto out;
1900
1901         icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1902                                    &_icmph);
1903         if (!icmph)
1904                 goto out;
1905
1906         if (!icmp_is_err(icmph->type))
1907                 goto out;
1908
1909         inner_iph = skb_header_pointer(skb,
1910                                        outer_iph->ihl * 4 + sizeof(_icmph),
1911                                        sizeof(_inner_iph), &_inner_iph);
1912         if (!inner_iph)
1913                 goto out;
1914
1915         key_iph = inner_iph;
1916 out:
1917         hash_keys->addrs.v4addrs.src = key_iph->saddr;
1918         hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1919 }
1920
1921 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1922                                            const struct sk_buff *skb,
1923                                            bool *p_has_inner)
1924 {
1925         u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1926         struct flow_keys keys, hash_keys;
1927
1928         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1929                 return 0;
1930
1931         memset(&hash_keys, 0, sizeof(hash_keys));
1932         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1933
1934         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1935         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1936                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1937         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1938                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1939         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1940                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1941         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1942                 hash_keys.ports.src = keys.ports.src;
1943         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1944                 hash_keys.ports.dst = keys.ports.dst;
1945
1946         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1947         return flow_hash_from_keys(&hash_keys);
1948 }
1949
1950 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1951                                            const struct sk_buff *skb,
1952                                            bool has_inner)
1953 {
1954         u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1955         struct flow_keys keys, hash_keys;
1956
1957         /* We assume the packet carries an encapsulation, but if none was
1958          * encountered during dissection of the outer flow, then there is no
1959          * point in calling the flow dissector again.
1960          */
1961         if (!has_inner)
1962                 return 0;
1963
1964         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1965                 return 0;
1966
1967         memset(&hash_keys, 0, sizeof(hash_keys));
1968         skb_flow_dissect_flow_keys(skb, &keys, 0);
1969
1970         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1971                 return 0;
1972
1973         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1974                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1975                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1976                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1977                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1978                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1979         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1980                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1981                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1982                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1983                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1984                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1985                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1986                         hash_keys.tags.flow_label = keys.tags.flow_label;
1987         }
1988
1989         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1990                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1991         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1992                 hash_keys.ports.src = keys.ports.src;
1993         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1994                 hash_keys.ports.dst = keys.ports.dst;
1995
1996         return flow_hash_from_keys(&hash_keys);
1997 }
1998
1999 static u32 fib_multipath_custom_hash_skb(const struct net *net,
2000                                          const struct sk_buff *skb)
2001 {
2002         u32 mhash, mhash_inner;
2003         bool has_inner = true;
2004
2005         mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2006         mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2007
2008         return jhash_2words(mhash, mhash_inner, 0);
2009 }
2010
2011 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2012                                          const struct flowi4 *fl4)
2013 {
2014         u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2015         struct flow_keys hash_keys;
2016
2017         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2018                 return 0;
2019
2020         memset(&hash_keys, 0, sizeof(hash_keys));
2021         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2022         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2023                 hash_keys.addrs.v4addrs.src = fl4->saddr;
2024         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2025                 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2026         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2027                 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2028         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2029                 hash_keys.ports.src = fl4->fl4_sport;
2030         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2031                 hash_keys.ports.dst = fl4->fl4_dport;
2032
2033         return flow_hash_from_keys(&hash_keys);
2034 }
2035
2036 /* if skb is set it will be used and fl4 can be NULL */
2037 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2038                        const struct sk_buff *skb, struct flow_keys *flkeys)
2039 {
2040         u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2041         struct flow_keys hash_keys;
2042         u32 mhash = 0;
2043
2044         switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2045         case 0:
2046                 memset(&hash_keys, 0, sizeof(hash_keys));
2047                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2048                 if (skb) {
2049                         ip_multipath_l3_keys(skb, &hash_keys);
2050                 } else {
2051                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2052                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2053                 }
2054                 mhash = flow_hash_from_keys(&hash_keys);
2055                 break;
2056         case 1:
2057                 /* skb is currently provided only when forwarding */
2058                 if (skb) {
2059                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2060                         struct flow_keys keys;
2061
2062                         /* short-circuit if we already have L4 hash present */
2063                         if (skb->l4_hash)
2064                                 return skb_get_hash_raw(skb) >> 1;
2065
2066                         memset(&hash_keys, 0, sizeof(hash_keys));
2067
2068                         if (!flkeys) {
2069                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2070                                 flkeys = &keys;
2071                         }
2072
2073                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2074                         hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2075                         hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2076                         hash_keys.ports.src = flkeys->ports.src;
2077                         hash_keys.ports.dst = flkeys->ports.dst;
2078                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2079                 } else {
2080                         memset(&hash_keys, 0, sizeof(hash_keys));
2081                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2082                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2083                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2084                         hash_keys.ports.src = fl4->fl4_sport;
2085                         hash_keys.ports.dst = fl4->fl4_dport;
2086                         hash_keys.basic.ip_proto = fl4->flowi4_proto;
2087                 }
2088                 mhash = flow_hash_from_keys(&hash_keys);
2089                 break;
2090         case 2:
2091                 memset(&hash_keys, 0, sizeof(hash_keys));
2092                 /* skb is currently provided only when forwarding */
2093                 if (skb) {
2094                         struct flow_keys keys;
2095
2096                         skb_flow_dissect_flow_keys(skb, &keys, 0);
2097                         /* Inner can be v4 or v6 */
2098                         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2099                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2100                                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2101                                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2102                         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2103                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2104                                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2105                                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2106                                 hash_keys.tags.flow_label = keys.tags.flow_label;
2107                                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2108                         } else {
2109                                 /* Same as case 0 */
2110                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2111                                 ip_multipath_l3_keys(skb, &hash_keys);
2112                         }
2113                 } else {
2114                         /* Same as case 0 */
2115                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2116                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2117                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2118                 }
2119                 mhash = flow_hash_from_keys(&hash_keys);
2120                 break;
2121         case 3:
2122                 if (skb)
2123                         mhash = fib_multipath_custom_hash_skb(net, skb);
2124                 else
2125                         mhash = fib_multipath_custom_hash_fl4(net, fl4);
2126                 break;
2127         }
2128
2129         if (multipath_hash)
2130                 mhash = jhash_2words(mhash, multipath_hash, 0);
2131
2132         return mhash >> 1;
2133 }
2134 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2135
2136 static int ip_mkroute_input(struct sk_buff *skb,
2137                             struct fib_result *res,
2138                             struct in_device *in_dev,
2139                             __be32 daddr, __be32 saddr, u32 tos,
2140                             struct flow_keys *hkeys)
2141 {
2142 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2143         if (res->fi && fib_info_num_path(res->fi) > 1) {
2144                 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2145
2146                 fib_select_multipath(res, h);
2147         }
2148 #endif
2149
2150         /* create a routing cache entry */
2151         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2152 }
2153
2154 /* Implements all the saddr-related checks as ip_route_input_slow(),
2155  * assuming daddr is valid and the destination is not a local broadcast one.
2156  * Uses the provided hint instead of performing a route lookup.
2157  */
2158 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2159                       u8 tos, struct net_device *dev,
2160                       const struct sk_buff *hint)
2161 {
2162         struct in_device *in_dev = __in_dev_get_rcu(dev);
2163         struct rtable *rt = skb_rtable(hint);
2164         struct net *net = dev_net(dev);
2165         int err = -EINVAL;
2166         u32 tag = 0;
2167
2168         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2169                 goto martian_source;
2170
2171         if (ipv4_is_zeronet(saddr))
2172                 goto martian_source;
2173
2174         if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2175                 goto martian_source;
2176
2177         if (rt->rt_type != RTN_LOCAL)
2178                 goto skip_validate_source;
2179
2180         tos &= IPTOS_RT_MASK;
2181         err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2182         if (err < 0)
2183                 goto martian_source;
2184
2185 skip_validate_source:
2186         skb_dst_copy(skb, hint);
2187         return 0;
2188
2189 martian_source:
2190         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2191         return err;
2192 }
2193
2194 /* get device for dst_alloc with local routes */
2195 static struct net_device *ip_rt_get_dev(struct net *net,
2196                                         const struct fib_result *res)
2197 {
2198         struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2199         struct net_device *dev = NULL;
2200
2201         if (nhc)
2202                 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2203
2204         return dev ? : net->loopback_dev;
2205 }
2206
2207 /*
2208  *      NOTE. We drop all the packets that has local source
2209  *      addresses, because every properly looped back packet
2210  *      must have correct destination already attached by output routine.
2211  *      Changes in the enforced policies must be applied also to
2212  *      ip_route_use_hint().
2213  *
2214  *      Such approach solves two big problems:
2215  *      1. Not simplex devices are handled properly.
2216  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2217  *      called with rcu_read_lock()
2218  */
2219
2220 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2221                                u8 tos, struct net_device *dev,
2222                                struct fib_result *res)
2223 {
2224         struct in_device *in_dev = __in_dev_get_rcu(dev);
2225         struct flow_keys *flkeys = NULL, _flkeys;
2226         struct net    *net = dev_net(dev);
2227         struct ip_tunnel_info *tun_info;
2228         int             err = -EINVAL;
2229         unsigned int    flags = 0;
2230         u32             itag = 0;
2231         struct rtable   *rth;
2232         struct flowi4   fl4;
2233         bool do_cache = true;
2234
2235         /* IP on this device is disabled. */
2236
2237         if (!in_dev)
2238                 goto out;
2239
2240         /* Check for the most weird martians, which can be not detected
2241          * by fib_lookup.
2242          */
2243
2244         tun_info = skb_tunnel_info(skb);
2245         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2246                 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2247         else
2248                 fl4.flowi4_tun_key.tun_id = 0;
2249         skb_dst_drop(skb);
2250
2251         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2252                 goto martian_source;
2253
2254         res->fi = NULL;
2255         res->table = NULL;
2256         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2257                 goto brd_input;
2258
2259         /* Accept zero addresses only to limited broadcast;
2260          * I even do not know to fix it or not. Waiting for complains :-)
2261          */
2262         if (ipv4_is_zeronet(saddr))
2263                 goto martian_source;
2264
2265         if (ipv4_is_zeronet(daddr))
2266                 goto martian_destination;
2267
2268         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2269          * and call it once if daddr or/and saddr are loopback addresses
2270          */
2271         if (ipv4_is_loopback(daddr)) {
2272                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2273                         goto martian_destination;
2274         } else if (ipv4_is_loopback(saddr)) {
2275                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2276                         goto martian_source;
2277         }
2278
2279         /*
2280          *      Now we are ready to route packet.
2281          */
2282         fl4.flowi4_l3mdev = 0;
2283         fl4.flowi4_oif = 0;
2284         fl4.flowi4_iif = dev->ifindex;
2285         fl4.flowi4_mark = skb->mark;
2286         fl4.flowi4_tos = tos;
2287         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2288         fl4.flowi4_flags = 0;
2289         fl4.daddr = daddr;
2290         fl4.saddr = saddr;
2291         fl4.flowi4_uid = sock_net_uid(net, NULL);
2292         fl4.flowi4_multipath_hash = 0;
2293
2294         if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2295                 flkeys = &_flkeys;
2296         } else {
2297                 fl4.flowi4_proto = 0;
2298                 fl4.fl4_sport = 0;
2299                 fl4.fl4_dport = 0;
2300         }
2301
2302         err = fib_lookup(net, &fl4, res, 0);
2303         if (err != 0) {
2304                 if (!IN_DEV_FORWARD(in_dev))
2305                         err = -EHOSTUNREACH;
2306                 goto no_route;
2307         }
2308
2309         if (res->type == RTN_BROADCAST) {
2310                 if (IN_DEV_BFORWARD(in_dev))
2311                         goto make_route;
2312                 /* not do cache if bc_forwarding is enabled */
2313                 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2314                         do_cache = false;
2315                 goto brd_input;
2316         }
2317
2318         if (res->type == RTN_LOCAL) {
2319                 err = fib_validate_source(skb, saddr, daddr, tos,
2320                                           0, dev, in_dev, &itag);
2321                 if (err < 0)
2322                         goto martian_source;
2323                 goto local_input;
2324         }
2325
2326         if (!IN_DEV_FORWARD(in_dev)) {
2327                 err = -EHOSTUNREACH;
2328                 goto no_route;
2329         }
2330         if (res->type != RTN_UNICAST)
2331                 goto martian_destination;
2332
2333 make_route:
2334         err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2335 out:    return err;
2336
2337 brd_input:
2338         if (skb->protocol != htons(ETH_P_IP))
2339                 goto e_inval;
2340
2341         if (!ipv4_is_zeronet(saddr)) {
2342                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2343                                           in_dev, &itag);
2344                 if (err < 0)
2345                         goto martian_source;
2346         }
2347         flags |= RTCF_BROADCAST;
2348         res->type = RTN_BROADCAST;
2349         RT_CACHE_STAT_INC(in_brd);
2350
2351 local_input:
2352         if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2353                 IPCB(skb)->flags |= IPSKB_NOPOLICY;
2354
2355         do_cache &= res->fi && !itag;
2356         if (do_cache) {
2357                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2358
2359                 rth = rcu_dereference(nhc->nhc_rth_input);
2360                 if (rt_cache_valid(rth)) {
2361                         skb_dst_set_noref(skb, &rth->dst);
2362                         err = 0;
2363                         goto out;
2364                 }
2365         }
2366
2367         rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2368                            flags | RTCF_LOCAL, res->type, false);
2369         if (!rth)
2370                 goto e_nobufs;
2371
2372         rth->dst.output= ip_rt_bug;
2373 #ifdef CONFIG_IP_ROUTE_CLASSID
2374         rth->dst.tclassid = itag;
2375 #endif
2376         rth->rt_is_input = 1;
2377
2378         RT_CACHE_STAT_INC(in_slow_tot);
2379         if (res->type == RTN_UNREACHABLE) {
2380                 rth->dst.input= ip_error;
2381                 rth->dst.error= -err;
2382                 rth->rt_flags   &= ~RTCF_LOCAL;
2383         }
2384
2385         if (do_cache) {
2386                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2387
2388                 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2389                 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2390                         WARN_ON(rth->dst.input == lwtunnel_input);
2391                         rth->dst.lwtstate->orig_input = rth->dst.input;
2392                         rth->dst.input = lwtunnel_input;
2393                 }
2394
2395                 if (unlikely(!rt_cache_route(nhc, rth)))
2396                         rt_add_uncached_list(rth);
2397         }
2398         skb_dst_set(skb, &rth->dst);
2399         err = 0;
2400         goto out;
2401
2402 no_route:
2403         RT_CACHE_STAT_INC(in_no_route);
2404         res->type = RTN_UNREACHABLE;
2405         res->fi = NULL;
2406         res->table = NULL;
2407         goto local_input;
2408
2409         /*
2410          *      Do not cache martian addresses: they should be logged (RFC1812)
2411          */
2412 martian_destination:
2413         RT_CACHE_STAT_INC(in_martian_dst);
2414 #ifdef CONFIG_IP_ROUTE_VERBOSE
2415         if (IN_DEV_LOG_MARTIANS(in_dev))
2416                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2417                                      &daddr, &saddr, dev->name);
2418 #endif
2419
2420 e_inval:
2421         err = -EINVAL;
2422         goto out;
2423
2424 e_nobufs:
2425         err = -ENOBUFS;
2426         goto out;
2427
2428 martian_source:
2429         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2430         goto out;
2431 }
2432
2433 /* called with rcu_read_lock held */
2434 static int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2435                               u8 tos, struct net_device *dev, struct fib_result *res)
2436 {
2437         /* Multicast recognition logic is moved from route cache to here.
2438          * The problem was that too many Ethernet cards have broken/missing
2439          * hardware multicast filters :-( As result the host on multicasting
2440          * network acquires a lot of useless route cache entries, sort of
2441          * SDR messages from all the world. Now we try to get rid of them.
2442          * Really, provided software IP multicast filter is organized
2443          * reasonably (at least, hashed), it does not result in a slowdown
2444          * comparing with route cache reject entries.
2445          * Note, that multicast routers are not affected, because
2446          * route cache entry is created eventually.
2447          */
2448         if (ipv4_is_multicast(daddr)) {
2449                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2450                 int our = 0;
2451                 int err = -EINVAL;
2452
2453                 if (!in_dev)
2454                         return err;
2455                 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2456                                       ip_hdr(skb)->protocol);
2457
2458                 /* check l3 master if no match yet */
2459                 if (!our && netif_is_l3_slave(dev)) {
2460                         struct in_device *l3_in_dev;
2461
2462                         l3_in_dev = __in_dev_get_rcu(skb->dev);
2463                         if (l3_in_dev)
2464                                 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2465                                                       ip_hdr(skb)->protocol);
2466                 }
2467
2468                 if (our
2469 #ifdef CONFIG_IP_MROUTE
2470                         ||
2471                     (!ipv4_is_local_multicast(daddr) &&
2472                      IN_DEV_MFORWARD(in_dev))
2473 #endif
2474                    ) {
2475                         err = ip_route_input_mc(skb, daddr, saddr,
2476                                                 tos, dev, our);
2477                 }
2478                 return err;
2479         }
2480
2481         return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2482 }
2483
2484 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2485                          u8 tos, struct net_device *dev)
2486 {
2487         struct fib_result res;
2488         int err;
2489
2490         tos &= IPTOS_RT_MASK;
2491         rcu_read_lock();
2492         err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2493         rcu_read_unlock();
2494
2495         return err;
2496 }
2497 EXPORT_SYMBOL(ip_route_input_noref);
2498
2499 /* called with rcu_read_lock() */
2500 static struct rtable *__mkroute_output(const struct fib_result *res,
2501                                        const struct flowi4 *fl4, int orig_oif,
2502                                        struct net_device *dev_out,
2503                                        unsigned int flags)
2504 {
2505         struct fib_info *fi = res->fi;
2506         struct fib_nh_exception *fnhe;
2507         struct in_device *in_dev;
2508         u16 type = res->type;
2509         struct rtable *rth;
2510         bool do_cache;
2511
2512         in_dev = __in_dev_get_rcu(dev_out);
2513         if (!in_dev)
2514                 return ERR_PTR(-EINVAL);
2515
2516         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2517                 if (ipv4_is_loopback(fl4->saddr) &&
2518                     !(dev_out->flags & IFF_LOOPBACK) &&
2519                     !netif_is_l3_master(dev_out))
2520                         return ERR_PTR(-EINVAL);
2521
2522         if (ipv4_is_lbcast(fl4->daddr))
2523                 type = RTN_BROADCAST;
2524         else if (ipv4_is_multicast(fl4->daddr))
2525                 type = RTN_MULTICAST;
2526         else if (ipv4_is_zeronet(fl4->daddr))
2527                 return ERR_PTR(-EINVAL);
2528
2529         if (dev_out->flags & IFF_LOOPBACK)
2530                 flags |= RTCF_LOCAL;
2531
2532         do_cache = true;
2533         if (type == RTN_BROADCAST) {
2534                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2535                 fi = NULL;
2536         } else if (type == RTN_MULTICAST) {
2537                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2538                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2539                                      fl4->flowi4_proto))
2540                         flags &= ~RTCF_LOCAL;
2541                 else
2542                         do_cache = false;
2543                 /* If multicast route do not exist use
2544                  * default one, but do not gateway in this case.
2545                  * Yes, it is hack.
2546                  */
2547                 if (fi && res->prefixlen < 4)
2548                         fi = NULL;
2549         } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2550                    (orig_oif != dev_out->ifindex)) {
2551                 /* For local routes that require a particular output interface
2552                  * we do not want to cache the result.  Caching the result
2553                  * causes incorrect behaviour when there are multiple source
2554                  * addresses on the interface, the end result being that if the
2555                  * intended recipient is waiting on that interface for the
2556                  * packet he won't receive it because it will be delivered on
2557                  * the loopback interface and the IP_PKTINFO ipi_ifindex will
2558                  * be set to the loopback interface as well.
2559                  */
2560                 do_cache = false;
2561         }
2562
2563         fnhe = NULL;
2564         do_cache &= fi != NULL;
2565         if (fi) {
2566                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2567                 struct rtable __rcu **prth;
2568
2569                 fnhe = find_exception(nhc, fl4->daddr);
2570                 if (!do_cache)
2571                         goto add;
2572                 if (fnhe) {
2573                         prth = &fnhe->fnhe_rth_output;
2574                 } else {
2575                         if (unlikely(fl4->flowi4_flags &
2576                                      FLOWI_FLAG_KNOWN_NH &&
2577                                      !(nhc->nhc_gw_family &&
2578                                        nhc->nhc_scope == RT_SCOPE_LINK))) {
2579                                 do_cache = false;
2580                                 goto add;
2581                         }
2582                         prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2583                 }
2584                 rth = rcu_dereference(*prth);
2585                 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2586                         return rth;
2587         }
2588
2589 add:
2590         rth = rt_dst_alloc(dev_out, flags, type,
2591                            IN_DEV_ORCONF(in_dev, NOXFRM));
2592         if (!rth)
2593                 return ERR_PTR(-ENOBUFS);
2594
2595         rth->rt_iif = orig_oif;
2596
2597         RT_CACHE_STAT_INC(out_slow_tot);
2598
2599         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2600                 if (flags & RTCF_LOCAL &&
2601                     !(dev_out->flags & IFF_LOOPBACK)) {
2602                         rth->dst.output = ip_mc_output;
2603                         RT_CACHE_STAT_INC(out_slow_mc);
2604                 }
2605 #ifdef CONFIG_IP_MROUTE
2606                 if (type == RTN_MULTICAST) {
2607                         if (IN_DEV_MFORWARD(in_dev) &&
2608                             !ipv4_is_local_multicast(fl4->daddr)) {
2609                                 rth->dst.input = ip_mr_input;
2610                                 rth->dst.output = ip_mc_output;
2611                         }
2612                 }
2613 #endif
2614         }
2615
2616         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2617         lwtunnel_set_redirect(&rth->dst);
2618
2619         return rth;
2620 }
2621
2622 /*
2623  * Major route resolver routine.
2624  */
2625
2626 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2627                                         const struct sk_buff *skb)
2628 {
2629         struct fib_result res = {
2630                 .type           = RTN_UNSPEC,
2631                 .fi             = NULL,
2632                 .table          = NULL,
2633                 .tclassid       = 0,
2634         };
2635         struct rtable *rth;
2636
2637         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2638         ip_rt_fix_tos(fl4);
2639
2640         rcu_read_lock();
2641         rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2642         rcu_read_unlock();
2643
2644         return rth;
2645 }
2646 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2647
2648 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2649                                             struct fib_result *res,
2650                                             const struct sk_buff *skb)
2651 {
2652         struct net_device *dev_out = NULL;
2653         int orig_oif = fl4->flowi4_oif;
2654         unsigned int flags = 0;
2655         struct rtable *rth;
2656         int err;
2657
2658         if (fl4->saddr) {
2659                 if (ipv4_is_multicast(fl4->saddr) ||
2660                     ipv4_is_lbcast(fl4->saddr) ||
2661                     ipv4_is_zeronet(fl4->saddr)) {
2662                         rth = ERR_PTR(-EINVAL);
2663                         goto out;
2664                 }
2665
2666                 rth = ERR_PTR(-ENETUNREACH);
2667
2668                 /* I removed check for oif == dev_out->oif here.
2669                  * It was wrong for two reasons:
2670                  * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2671                  *    is assigned to multiple interfaces.
2672                  * 2. Moreover, we are allowed to send packets with saddr
2673                  *    of another iface. --ANK
2674                  */
2675
2676                 if (fl4->flowi4_oif == 0 &&
2677                     (ipv4_is_multicast(fl4->daddr) ||
2678                      ipv4_is_lbcast(fl4->daddr))) {
2679                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2680                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2681                         if (!dev_out)
2682                                 goto out;
2683
2684                         /* Special hack: user can direct multicasts
2685                          * and limited broadcast via necessary interface
2686                          * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2687                          * This hack is not just for fun, it allows
2688                          * vic,vat and friends to work.
2689                          * They bind socket to loopback, set ttl to zero
2690                          * and expect that it will work.
2691                          * From the viewpoint of routing cache they are broken,
2692                          * because we are not allowed to build multicast path
2693                          * with loopback source addr (look, routing cache
2694                          * cannot know, that ttl is zero, so that packet
2695                          * will not leave this host and route is valid).
2696                          * Luckily, this hack is good workaround.
2697                          */
2698
2699                         fl4->flowi4_oif = dev_out->ifindex;
2700                         goto make_route;
2701                 }
2702
2703                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2704                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2705                         if (!__ip_dev_find(net, fl4->saddr, false))
2706                                 goto out;
2707                 }
2708         }
2709
2710
2711         if (fl4->flowi4_oif) {
2712                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2713                 rth = ERR_PTR(-ENODEV);
2714                 if (!dev_out)
2715                         goto out;
2716
2717                 /* RACE: Check return value of inet_select_addr instead. */
2718                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2719                         rth = ERR_PTR(-ENETUNREACH);
2720                         goto out;
2721                 }
2722                 if (ipv4_is_local_multicast(fl4->daddr) ||
2723                     ipv4_is_lbcast(fl4->daddr) ||
2724                     fl4->flowi4_proto == IPPROTO_IGMP) {
2725                         if (!fl4->saddr)
2726                                 fl4->saddr = inet_select_addr(dev_out, 0,
2727                                                               RT_SCOPE_LINK);
2728                         goto make_route;
2729                 }
2730                 if (!fl4->saddr) {
2731                         if (ipv4_is_multicast(fl4->daddr))
2732                                 fl4->saddr = inet_select_addr(dev_out, 0,
2733                                                               fl4->flowi4_scope);
2734                         else if (!fl4->daddr)
2735                                 fl4->saddr = inet_select_addr(dev_out, 0,
2736                                                               RT_SCOPE_HOST);
2737                 }
2738         }
2739
2740         if (!fl4->daddr) {
2741                 fl4->daddr = fl4->saddr;
2742                 if (!fl4->daddr)
2743                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2744                 dev_out = net->loopback_dev;
2745                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2746                 res->type = RTN_LOCAL;
2747                 flags |= RTCF_LOCAL;
2748                 goto make_route;
2749         }
2750
2751         err = fib_lookup(net, fl4, res, 0);
2752         if (err) {
2753                 res->fi = NULL;
2754                 res->table = NULL;
2755                 if (fl4->flowi4_oif &&
2756                     (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2757                         /* Apparently, routing tables are wrong. Assume,
2758                          * that the destination is on link.
2759                          *
2760                          * WHY? DW.
2761                          * Because we are allowed to send to iface
2762                          * even if it has NO routes and NO assigned
2763                          * addresses. When oif is specified, routing
2764                          * tables are looked up with only one purpose:
2765                          * to catch if destination is gatewayed, rather than
2766                          * direct. Moreover, if MSG_DONTROUTE is set,
2767                          * we send packet, ignoring both routing tables
2768                          * and ifaddr state. --ANK
2769                          *
2770                          *
2771                          * We could make it even if oif is unknown,
2772                          * likely IPv6, but we do not.
2773                          */
2774
2775                         if (fl4->saddr == 0)
2776                                 fl4->saddr = inet_select_addr(dev_out, 0,
2777                                                               RT_SCOPE_LINK);
2778                         res->type = RTN_UNICAST;
2779                         goto make_route;
2780                 }
2781                 rth = ERR_PTR(err);
2782                 goto out;
2783         }
2784
2785         if (res->type == RTN_LOCAL) {
2786                 if (!fl4->saddr) {
2787                         if (res->fi->fib_prefsrc)
2788                                 fl4->saddr = res->fi->fib_prefsrc;
2789                         else
2790                                 fl4->saddr = fl4->daddr;
2791                 }
2792
2793                 /* L3 master device is the loopback for that domain */
2794                 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2795                         net->loopback_dev;
2796
2797                 /* make sure orig_oif points to fib result device even
2798                  * though packet rx/tx happens over loopback or l3mdev
2799                  */
2800                 orig_oif = FIB_RES_OIF(*res);
2801
2802                 fl4->flowi4_oif = dev_out->ifindex;
2803                 flags |= RTCF_LOCAL;
2804                 goto make_route;
2805         }
2806
2807         fib_select_path(net, res, fl4, skb);
2808
2809         dev_out = FIB_RES_DEV(*res);
2810
2811 make_route:
2812         rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2813
2814 out:
2815         return rth;
2816 }
2817
2818 static struct dst_ops ipv4_dst_blackhole_ops = {
2819         .family                 = AF_INET,
2820         .default_advmss         = ipv4_default_advmss,
2821         .neigh_lookup           = ipv4_neigh_lookup,
2822         .check                  = dst_blackhole_check,
2823         .cow_metrics            = dst_blackhole_cow_metrics,
2824         .update_pmtu            = dst_blackhole_update_pmtu,
2825         .redirect               = dst_blackhole_redirect,
2826         .mtu                    = dst_blackhole_mtu,
2827 };
2828
2829 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2830 {
2831         struct rtable *ort = (struct rtable *) dst_orig;
2832         struct rtable *rt;
2833
2834         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2835         if (rt) {
2836                 struct dst_entry *new = &rt->dst;
2837
2838                 new->__use = 1;
2839                 new->input = dst_discard;
2840                 new->output = dst_discard_out;
2841
2842                 new->dev = net->loopback_dev;
2843                 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2844
2845                 rt->rt_is_input = ort->rt_is_input;
2846                 rt->rt_iif = ort->rt_iif;
2847                 rt->rt_pmtu = ort->rt_pmtu;
2848                 rt->rt_mtu_locked = ort->rt_mtu_locked;
2849
2850                 rt->rt_genid = rt_genid_ipv4(net);
2851                 rt->rt_flags = ort->rt_flags;
2852                 rt->rt_type = ort->rt_type;
2853                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2854                 rt->rt_gw_family = ort->rt_gw_family;
2855                 if (rt->rt_gw_family == AF_INET)
2856                         rt->rt_gw4 = ort->rt_gw4;
2857                 else if (rt->rt_gw_family == AF_INET6)
2858                         rt->rt_gw6 = ort->rt_gw6;
2859         }
2860
2861         dst_release(dst_orig);
2862
2863         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2864 }
2865
2866 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2867                                     const struct sock *sk)
2868 {
2869         struct rtable *rt = __ip_route_output_key(net, flp4);
2870
2871         if (IS_ERR(rt))
2872                 return rt;
2873
2874         if (flp4->flowi4_proto) {
2875                 flp4->flowi4_oif = rt->dst.dev->ifindex;
2876                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2877                                                         flowi4_to_flowi(flp4),
2878                                                         sk, 0);
2879         }
2880
2881         return rt;
2882 }
2883 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2884
2885 struct rtable *ip_route_output_tunnel(struct sk_buff *skb,
2886                                       struct net_device *dev,
2887                                       struct net *net, __be32 *saddr,
2888                                       const struct ip_tunnel_info *info,
2889                                       u8 protocol, bool use_cache)
2890 {
2891 #ifdef CONFIG_DST_CACHE
2892         struct dst_cache *dst_cache;
2893 #endif
2894         struct rtable *rt = NULL;
2895         struct flowi4 fl4;
2896         __u8 tos;
2897
2898 #ifdef CONFIG_DST_CACHE
2899         dst_cache = (struct dst_cache *)&info->dst_cache;
2900         if (use_cache) {
2901                 rt = dst_cache_get_ip4(dst_cache, saddr);
2902                 if (rt)
2903                         return rt;
2904         }
2905 #endif
2906         memset(&fl4, 0, sizeof(fl4));
2907         fl4.flowi4_mark = skb->mark;
2908         fl4.flowi4_proto = protocol;
2909         fl4.daddr = info->key.u.ipv4.dst;
2910         fl4.saddr = info->key.u.ipv4.src;
2911         tos = info->key.tos;
2912         fl4.flowi4_tos = RT_TOS(tos);
2913
2914         rt = ip_route_output_key(net, &fl4);
2915         if (IS_ERR(rt)) {
2916                 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
2917                 return ERR_PTR(-ENETUNREACH);
2918         }
2919         if (rt->dst.dev == dev) { /* is this necessary? */
2920                 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
2921                 ip_rt_put(rt);
2922                 return ERR_PTR(-ELOOP);
2923         }
2924 #ifdef CONFIG_DST_CACHE
2925         if (use_cache)
2926                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2927 #endif
2928         *saddr = fl4.saddr;
2929         return rt;
2930 }
2931 EXPORT_SYMBOL_GPL(ip_route_output_tunnel);
2932
2933 /* called with rcu_read_lock held */
2934 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2935                         struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2936                         struct sk_buff *skb, u32 portid, u32 seq,
2937                         unsigned int flags)
2938 {
2939         struct rtmsg *r;
2940         struct nlmsghdr *nlh;
2941         unsigned long expires = 0;
2942         u32 error;
2943         u32 metrics[RTAX_MAX];
2944
2945         nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2946         if (!nlh)
2947                 return -EMSGSIZE;
2948
2949         r = nlmsg_data(nlh);
2950         r->rtm_family    = AF_INET;
2951         r->rtm_dst_len  = 32;
2952         r->rtm_src_len  = 0;
2953         r->rtm_tos      = fl4 ? fl4->flowi4_tos : 0;
2954         r->rtm_table    = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2955         if (nla_put_u32(skb, RTA_TABLE, table_id))
2956                 goto nla_put_failure;
2957         r->rtm_type     = rt->rt_type;
2958         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2959         r->rtm_protocol = RTPROT_UNSPEC;
2960         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2961         if (rt->rt_flags & RTCF_NOTIFY)
2962                 r->rtm_flags |= RTM_F_NOTIFY;
2963         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2964                 r->rtm_flags |= RTCF_DOREDIRECT;
2965
2966         if (nla_put_in_addr(skb, RTA_DST, dst))
2967                 goto nla_put_failure;
2968         if (src) {
2969                 r->rtm_src_len = 32;
2970                 if (nla_put_in_addr(skb, RTA_SRC, src))
2971                         goto nla_put_failure;
2972         }
2973         if (rt->dst.dev &&
2974             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2975                 goto nla_put_failure;
2976         if (rt->dst.lwtstate &&
2977             lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2978                 goto nla_put_failure;
2979 #ifdef CONFIG_IP_ROUTE_CLASSID
2980         if (rt->dst.tclassid &&
2981             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2982                 goto nla_put_failure;
2983 #endif
2984         if (fl4 && !rt_is_input_route(rt) &&
2985             fl4->saddr != src) {
2986                 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2987                         goto nla_put_failure;
2988         }
2989         if (rt->rt_uses_gateway) {
2990                 if (rt->rt_gw_family == AF_INET &&
2991                     nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2992                         goto nla_put_failure;
2993                 } else if (rt->rt_gw_family == AF_INET6) {
2994                         int alen = sizeof(struct in6_addr);
2995                         struct nlattr *nla;
2996                         struct rtvia *via;
2997
2998                         nla = nla_reserve(skb, RTA_VIA, alen + 2);
2999                         if (!nla)
3000                                 goto nla_put_failure;
3001
3002                         via = nla_data(nla);
3003                         via->rtvia_family = AF_INET6;
3004                         memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3005                 }
3006         }
3007
3008         expires = rt->dst.expires;
3009         if (expires) {
3010                 unsigned long now = jiffies;
3011
3012                 if (time_before(now, expires))
3013                         expires -= now;
3014                 else
3015                         expires = 0;
3016         }
3017
3018         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3019         if (rt->rt_pmtu && expires)
3020                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3021         if (rt->rt_mtu_locked && expires)
3022                 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3023         if (rtnetlink_put_metrics(skb, metrics) < 0)
3024                 goto nla_put_failure;
3025
3026         if (fl4) {
3027                 if (fl4->flowi4_mark &&
3028                     nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3029                         goto nla_put_failure;
3030
3031                 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3032                     nla_put_u32(skb, RTA_UID,
3033                                 from_kuid_munged(current_user_ns(),
3034                                                  fl4->flowi4_uid)))
3035                         goto nla_put_failure;
3036
3037                 if (rt_is_input_route(rt)) {
3038 #ifdef CONFIG_IP_MROUTE
3039                         if (ipv4_is_multicast(dst) &&
3040                             !ipv4_is_local_multicast(dst) &&
3041                             IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
3042                                 int err = ipmr_get_route(net, skb,
3043                                                          fl4->saddr, fl4->daddr,
3044                                                          r, portid);
3045
3046                                 if (err <= 0) {
3047                                         if (err == 0)
3048                                                 return 0;
3049                                         goto nla_put_failure;
3050                                 }
3051                         } else
3052 #endif
3053                                 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3054                                         goto nla_put_failure;
3055                 }
3056         }
3057
3058         error = rt->dst.error;
3059
3060         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3061                 goto nla_put_failure;
3062
3063         nlmsg_end(skb, nlh);
3064         return 0;
3065
3066 nla_put_failure:
3067         nlmsg_cancel(skb, nlh);
3068         return -EMSGSIZE;
3069 }
3070
3071 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3072                             struct netlink_callback *cb, u32 table_id,
3073                             struct fnhe_hash_bucket *bucket, int genid,
3074                             int *fa_index, int fa_start, unsigned int flags)
3075 {
3076         int i;
3077
3078         for (i = 0; i < FNHE_HASH_SIZE; i++) {
3079                 struct fib_nh_exception *fnhe;
3080
3081                 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3082                      fnhe = rcu_dereference(fnhe->fnhe_next)) {
3083                         struct rtable *rt;
3084                         int err;
3085
3086                         if (*fa_index < fa_start)
3087                                 goto next;
3088
3089                         if (fnhe->fnhe_genid != genid)
3090                                 goto next;
3091
3092                         if (fnhe->fnhe_expires &&
3093                             time_after(jiffies, fnhe->fnhe_expires))
3094                                 goto next;
3095
3096                         rt = rcu_dereference(fnhe->fnhe_rth_input);
3097                         if (!rt)
3098                                 rt = rcu_dereference(fnhe->fnhe_rth_output);
3099                         if (!rt)
3100                                 goto next;
3101
3102                         err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3103                                            table_id, NULL, skb,
3104                                            NETLINK_CB(cb->skb).portid,
3105                                            cb->nlh->nlmsg_seq, flags);
3106                         if (err)
3107                                 return err;
3108 next:
3109                         (*fa_index)++;
3110                 }
3111         }
3112
3113         return 0;
3114 }
3115
3116 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3117                        u32 table_id, struct fib_info *fi,
3118                        int *fa_index, int fa_start, unsigned int flags)
3119 {
3120         struct net *net = sock_net(cb->skb->sk);
3121         int nhsel, genid = fnhe_genid(net);
3122
3123         for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3124                 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3125                 struct fnhe_hash_bucket *bucket;
3126                 int err;
3127
3128                 if (nhc->nhc_flags & RTNH_F_DEAD)
3129                         continue;
3130
3131                 rcu_read_lock();
3132                 bucket = rcu_dereference(nhc->nhc_exceptions);
3133                 err = 0;
3134                 if (bucket)
3135                         err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3136                                                genid, fa_index, fa_start,
3137                                                flags);
3138                 rcu_read_unlock();
3139                 if (err)
3140                         return err;
3141         }
3142
3143         return 0;
3144 }
3145
3146 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3147                                                    u8 ip_proto, __be16 sport,
3148                                                    __be16 dport)
3149 {
3150         struct sk_buff *skb;
3151         struct iphdr *iph;
3152
3153         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3154         if (!skb)
3155                 return NULL;
3156
3157         /* Reserve room for dummy headers, this skb can pass
3158          * through good chunk of routing engine.
3159          */
3160         skb_reset_mac_header(skb);
3161         skb_reset_network_header(skb);
3162         skb->protocol = htons(ETH_P_IP);
3163         iph = skb_put(skb, sizeof(struct iphdr));
3164         iph->protocol = ip_proto;
3165         iph->saddr = src;
3166         iph->daddr = dst;
3167         iph->version = 0x4;
3168         iph->frag_off = 0;
3169         iph->ihl = 0x5;
3170         skb_set_transport_header(skb, skb->len);
3171
3172         switch (iph->protocol) {
3173         case IPPROTO_UDP: {
3174                 struct udphdr *udph;
3175
3176                 udph = skb_put_zero(skb, sizeof(struct udphdr));
3177                 udph->source = sport;
3178                 udph->dest = dport;
3179                 udph->len = htons(sizeof(struct udphdr));
3180                 udph->check = 0;
3181                 break;
3182         }
3183         case IPPROTO_TCP: {
3184                 struct tcphdr *tcph;
3185
3186                 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3187                 tcph->source    = sport;
3188                 tcph->dest      = dport;
3189                 tcph->doff      = sizeof(struct tcphdr) / 4;
3190                 tcph->rst = 1;
3191                 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3192                                             src, dst, 0);
3193                 break;
3194         }
3195         case IPPROTO_ICMP: {
3196                 struct icmphdr *icmph;
3197
3198                 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3199                 icmph->type = ICMP_ECHO;
3200                 icmph->code = 0;
3201         }
3202         }
3203
3204         return skb;
3205 }
3206
3207 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3208                                        const struct nlmsghdr *nlh,
3209                                        struct nlattr **tb,
3210                                        struct netlink_ext_ack *extack)
3211 {
3212         struct rtmsg *rtm;
3213         int i, err;
3214
3215         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3216                 NL_SET_ERR_MSG(extack,
3217                                "ipv4: Invalid header for route get request");
3218                 return -EINVAL;
3219         }
3220
3221         if (!netlink_strict_get_check(skb))
3222                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3223                                               rtm_ipv4_policy, extack);
3224
3225         rtm = nlmsg_data(nlh);
3226         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3227             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3228             rtm->rtm_table || rtm->rtm_protocol ||
3229             rtm->rtm_scope || rtm->rtm_type) {
3230                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3231                 return -EINVAL;
3232         }
3233
3234         if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3235                                RTM_F_LOOKUP_TABLE |
3236                                RTM_F_FIB_MATCH)) {
3237                 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3238                 return -EINVAL;
3239         }
3240
3241         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3242                                             rtm_ipv4_policy, extack);
3243         if (err)
3244                 return err;
3245
3246         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3247             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3248                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3249                 return -EINVAL;
3250         }
3251
3252         for (i = 0; i <= RTA_MAX; i++) {
3253                 if (!tb[i])
3254                         continue;
3255
3256                 switch (i) {
3257                 case RTA_IIF:
3258                 case RTA_OIF:
3259                 case RTA_SRC:
3260                 case RTA_DST:
3261                 case RTA_IP_PROTO:
3262                 case RTA_SPORT:
3263                 case RTA_DPORT:
3264                 case RTA_MARK:
3265                 case RTA_UID:
3266                         break;
3267                 default:
3268                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3269                         return -EINVAL;
3270                 }
3271         }
3272
3273         return 0;
3274 }
3275
3276 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3277                              struct netlink_ext_ack *extack)
3278 {
3279         struct net *net = sock_net(in_skb->sk);
3280         struct nlattr *tb[RTA_MAX+1];
3281         u32 table_id = RT_TABLE_MAIN;
3282         __be16 sport = 0, dport = 0;
3283         struct fib_result res = {};
3284         u8 ip_proto = IPPROTO_UDP;
3285         struct rtable *rt = NULL;
3286         struct sk_buff *skb;
3287         struct rtmsg *rtm;
3288         struct flowi4 fl4 = {};
3289         __be32 dst = 0;
3290         __be32 src = 0;
3291         kuid_t uid;
3292         u32 iif;
3293         int err;
3294         int mark;
3295
3296         err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3297         if (err < 0)
3298                 return err;
3299
3300         rtm = nlmsg_data(nlh);
3301         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3302         dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3303         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3304         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3305         if (tb[RTA_UID])
3306                 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3307         else
3308                 uid = (iif ? INVALID_UID : current_uid());
3309
3310         if (tb[RTA_IP_PROTO]) {
3311                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3312                                                   &ip_proto, AF_INET, extack);
3313                 if (err)
3314                         return err;
3315         }
3316
3317         if (tb[RTA_SPORT])
3318                 sport = nla_get_be16(tb[RTA_SPORT]);
3319
3320         if (tb[RTA_DPORT])
3321                 dport = nla_get_be16(tb[RTA_DPORT]);
3322
3323         skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3324         if (!skb)
3325                 return -ENOBUFS;
3326
3327         fl4.daddr = dst;
3328         fl4.saddr = src;
3329         fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3330         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3331         fl4.flowi4_mark = mark;
3332         fl4.flowi4_uid = uid;
3333         if (sport)
3334                 fl4.fl4_sport = sport;
3335         if (dport)
3336                 fl4.fl4_dport = dport;
3337         fl4.flowi4_proto = ip_proto;
3338
3339         rcu_read_lock();
3340
3341         if (iif) {
3342                 struct net_device *dev;
3343
3344                 dev = dev_get_by_index_rcu(net, iif);
3345                 if (!dev) {
3346                         err = -ENODEV;
3347                         goto errout_rcu;
3348                 }
3349
3350                 fl4.flowi4_iif = iif; /* for rt_fill_info */
3351                 skb->dev        = dev;
3352                 skb->mark       = mark;
3353                 err = ip_route_input_rcu(skb, dst, src,
3354                                          rtm->rtm_tos & IPTOS_RT_MASK, dev,
3355                                          &res);
3356
3357                 rt = skb_rtable(skb);
3358                 if (err == 0 && rt->dst.error)
3359                         err = -rt->dst.error;
3360         } else {
3361                 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3362                 skb->dev = net->loopback_dev;
3363                 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3364                 err = 0;
3365                 if (IS_ERR(rt))
3366                         err = PTR_ERR(rt);
3367                 else
3368                         skb_dst_set(skb, &rt->dst);
3369         }
3370
3371         if (err)
3372                 goto errout_rcu;
3373
3374         if (rtm->rtm_flags & RTM_F_NOTIFY)
3375                 rt->rt_flags |= RTCF_NOTIFY;
3376
3377         if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3378                 table_id = res.table ? res.table->tb_id : 0;
3379
3380         /* reset skb for netlink reply msg */
3381         skb_trim(skb, 0);
3382         skb_reset_network_header(skb);
3383         skb_reset_transport_header(skb);
3384         skb_reset_mac_header(skb);
3385
3386         if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3387                 struct fib_rt_info fri;
3388
3389                 if (!res.fi) {
3390                         err = fib_props[res.type].error;
3391                         if (!err)
3392                                 err = -EHOSTUNREACH;
3393                         goto errout_rcu;
3394                 }
3395                 fri.fi = res.fi;
3396                 fri.tb_id = table_id;
3397                 fri.dst = res.prefix;
3398                 fri.dst_len = res.prefixlen;
3399                 fri.dscp = inet_dsfield_to_dscp(fl4.flowi4_tos);
3400                 fri.type = rt->rt_type;
3401                 fri.offload = 0;
3402                 fri.trap = 0;
3403                 fri.offload_failed = 0;
3404                 if (res.fa_head) {
3405                         struct fib_alias *fa;
3406
3407                         hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3408                                 u8 slen = 32 - fri.dst_len;
3409
3410                                 if (fa->fa_slen == slen &&
3411                                     fa->tb_id == fri.tb_id &&
3412                                     fa->fa_dscp == fri.dscp &&
3413                                     fa->fa_info == res.fi &&
3414                                     fa->fa_type == fri.type) {
3415                                         fri.offload = READ_ONCE(fa->offload);
3416                                         fri.trap = READ_ONCE(fa->trap);
3417                                         break;
3418                                 }
3419                         }
3420                 }
3421                 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3422                                     nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3423         } else {
3424                 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3425                                    NETLINK_CB(in_skb).portid,
3426                                    nlh->nlmsg_seq, 0);
3427         }
3428         if (err < 0)
3429                 goto errout_rcu;
3430
3431         rcu_read_unlock();
3432
3433         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3434
3435 errout_free:
3436         return err;
3437 errout_rcu:
3438         rcu_read_unlock();
3439         kfree_skb(skb);
3440         goto errout_free;
3441 }
3442
3443 void ip_rt_multicast_event(struct in_device *in_dev)
3444 {
3445         rt_cache_flush(dev_net(in_dev->dev));
3446 }
3447
3448 #ifdef CONFIG_SYSCTL
3449 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3450 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
3451 static int ip_rt_gc_elasticity __read_mostly    = 8;
3452 static int ip_min_valid_pmtu __read_mostly      = IPV4_MIN_MTU;
3453
3454 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3455                 void *buffer, size_t *lenp, loff_t *ppos)
3456 {
3457         struct net *net = (struct net *)__ctl->extra1;
3458
3459         if (write) {
3460                 rt_cache_flush(net);
3461                 fnhe_genid_bump(net);
3462                 return 0;
3463         }
3464
3465         return -EINVAL;
3466 }
3467
3468 static struct ctl_table ipv4_route_table[] = {
3469         {
3470                 .procname       = "gc_thresh",
3471                 .data           = &ipv4_dst_ops.gc_thresh,
3472                 .maxlen         = sizeof(int),
3473                 .mode           = 0644,
3474                 .proc_handler   = proc_dointvec,
3475         },
3476         {
3477                 .procname       = "max_size",
3478                 .data           = &ip_rt_max_size,
3479                 .maxlen         = sizeof(int),
3480                 .mode           = 0644,
3481                 .proc_handler   = proc_dointvec,
3482         },
3483         {
3484                 /*  Deprecated. Use gc_min_interval_ms */
3485
3486                 .procname       = "gc_min_interval",
3487                 .data           = &ip_rt_gc_min_interval,
3488                 .maxlen         = sizeof(int),
3489                 .mode           = 0644,
3490                 .proc_handler   = proc_dointvec_jiffies,
3491         },
3492         {
3493                 .procname       = "gc_min_interval_ms",
3494                 .data           = &ip_rt_gc_min_interval,
3495                 .maxlen         = sizeof(int),
3496                 .mode           = 0644,
3497                 .proc_handler   = proc_dointvec_ms_jiffies,
3498         },
3499         {
3500                 .procname       = "gc_timeout",
3501                 .data           = &ip_rt_gc_timeout,
3502                 .maxlen         = sizeof(int),
3503                 .mode           = 0644,
3504                 .proc_handler   = proc_dointvec_jiffies,
3505         },
3506         {
3507                 .procname       = "gc_interval",
3508                 .data           = &ip_rt_gc_interval,
3509                 .maxlen         = sizeof(int),
3510                 .mode           = 0644,
3511                 .proc_handler   = proc_dointvec_jiffies,
3512         },
3513         {
3514                 .procname       = "redirect_load",
3515                 .data           = &ip_rt_redirect_load,
3516                 .maxlen         = sizeof(int),
3517                 .mode           = 0644,
3518                 .proc_handler   = proc_dointvec,
3519         },
3520         {
3521                 .procname       = "redirect_number",
3522                 .data           = &ip_rt_redirect_number,
3523                 .maxlen         = sizeof(int),
3524                 .mode           = 0644,
3525                 .proc_handler   = proc_dointvec,
3526         },
3527         {
3528                 .procname       = "redirect_silence",
3529                 .data           = &ip_rt_redirect_silence,
3530                 .maxlen         = sizeof(int),
3531                 .mode           = 0644,
3532                 .proc_handler   = proc_dointvec,
3533         },
3534         {
3535                 .procname       = "error_cost",
3536                 .data           = &ip_rt_error_cost,
3537                 .maxlen         = sizeof(int),
3538                 .mode           = 0644,
3539                 .proc_handler   = proc_dointvec,
3540         },
3541         {
3542                 .procname       = "error_burst",
3543                 .data           = &ip_rt_error_burst,
3544                 .maxlen         = sizeof(int),
3545                 .mode           = 0644,
3546                 .proc_handler   = proc_dointvec,
3547         },
3548         {
3549                 .procname       = "gc_elasticity",
3550                 .data           = &ip_rt_gc_elasticity,
3551                 .maxlen         = sizeof(int),
3552                 .mode           = 0644,
3553                 .proc_handler   = proc_dointvec,
3554         },
3555         { }
3556 };
3557
3558 static const char ipv4_route_flush_procname[] = "flush";
3559
3560 static struct ctl_table ipv4_route_netns_table[] = {
3561         {
3562                 .procname       = ipv4_route_flush_procname,
3563                 .maxlen         = sizeof(int),
3564                 .mode           = 0200,
3565                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3566         },
3567         {
3568                 .procname       = "min_pmtu",
3569                 .data           = &init_net.ipv4.ip_rt_min_pmtu,
3570                 .maxlen         = sizeof(int),
3571                 .mode           = 0644,
3572                 .proc_handler   = proc_dointvec_minmax,
3573                 .extra1         = &ip_min_valid_pmtu,
3574         },
3575         {
3576                 .procname       = "mtu_expires",
3577                 .data           = &init_net.ipv4.ip_rt_mtu_expires,
3578                 .maxlen         = sizeof(int),
3579                 .mode           = 0644,
3580                 .proc_handler   = proc_dointvec_jiffies,
3581         },
3582         {
3583                 .procname   = "min_adv_mss",
3584                 .data       = &init_net.ipv4.ip_rt_min_advmss,
3585                 .maxlen     = sizeof(int),
3586                 .mode       = 0644,
3587                 .proc_handler   = proc_dointvec,
3588         },
3589         { },
3590 };
3591
3592 static __net_init int sysctl_route_net_init(struct net *net)
3593 {
3594         struct ctl_table *tbl;
3595         size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3596
3597         tbl = ipv4_route_netns_table;
3598         if (!net_eq(net, &init_net)) {
3599                 int i;
3600
3601                 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3602                 if (!tbl)
3603                         goto err_dup;
3604
3605                 /* Don't export non-whitelisted sysctls to unprivileged users */
3606                 if (net->user_ns != &init_user_ns) {
3607                         if (tbl[0].procname != ipv4_route_flush_procname) {
3608                                 tbl[0].procname = NULL;
3609                                 table_size = 0;
3610                         }
3611                 }
3612
3613                 /* Update the variables to point into the current struct net
3614                  * except for the first element flush
3615                  */
3616                 for (i = 1; i < ARRAY_SIZE(ipv4_route_netns_table) - 1; i++)
3617                         tbl[i].data += (void *)net - (void *)&init_net;
3618         }
3619         tbl[0].extra1 = net;
3620
3621         net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3622                                                      tbl, table_size);
3623         if (!net->ipv4.route_hdr)
3624                 goto err_reg;
3625         return 0;
3626
3627 err_reg:
3628         if (tbl != ipv4_route_netns_table)
3629                 kfree(tbl);
3630 err_dup:
3631         return -ENOMEM;
3632 }
3633
3634 static __net_exit void sysctl_route_net_exit(struct net *net)
3635 {
3636         struct ctl_table *tbl;
3637
3638         tbl = net->ipv4.route_hdr->ctl_table_arg;
3639         unregister_net_sysctl_table(net->ipv4.route_hdr);
3640         BUG_ON(tbl == ipv4_route_netns_table);
3641         kfree(tbl);
3642 }
3643
3644 static __net_initdata struct pernet_operations sysctl_route_ops = {
3645         .init = sysctl_route_net_init,
3646         .exit = sysctl_route_net_exit,
3647 };
3648 #endif
3649
3650 static __net_init int netns_ip_rt_init(struct net *net)
3651 {
3652         /* Set default value for namespaceified sysctls */
3653         net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3654         net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3655         net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3656         return 0;
3657 }
3658
3659 static struct pernet_operations __net_initdata ip_rt_ops = {
3660         .init = netns_ip_rt_init,
3661 };
3662
3663 static __net_init int rt_genid_init(struct net *net)
3664 {
3665         atomic_set(&net->ipv4.rt_genid, 0);
3666         atomic_set(&net->fnhe_genid, 0);
3667         atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3668         return 0;
3669 }
3670
3671 static __net_initdata struct pernet_operations rt_genid_ops = {
3672         .init = rt_genid_init,
3673 };
3674
3675 static int __net_init ipv4_inetpeer_init(struct net *net)
3676 {
3677         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3678
3679         if (!bp)
3680                 return -ENOMEM;
3681         inet_peer_base_init(bp);
3682         net->ipv4.peers = bp;
3683         return 0;
3684 }
3685
3686 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3687 {
3688         struct inet_peer_base *bp = net->ipv4.peers;
3689
3690         net->ipv4.peers = NULL;
3691         inetpeer_invalidate_tree(bp);
3692         kfree(bp);
3693 }
3694
3695 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3696         .init   =       ipv4_inetpeer_init,
3697         .exit   =       ipv4_inetpeer_exit,
3698 };
3699
3700 #ifdef CONFIG_IP_ROUTE_CLASSID
3701 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3702 #endif /* CONFIG_IP_ROUTE_CLASSID */
3703
3704 int __init ip_rt_init(void)
3705 {
3706         void *idents_hash;
3707         int cpu;
3708
3709         /* For modern hosts, this will use 2 MB of memory */
3710         idents_hash = alloc_large_system_hash("IP idents",
3711                                               sizeof(*ip_idents) + sizeof(*ip_tstamps),
3712                                               0,
3713                                               16, /* one bucket per 64 KB */
3714                                               HASH_ZERO,
3715                                               NULL,
3716                                               &ip_idents_mask,
3717                                               2048,
3718                                               256*1024);
3719
3720         ip_idents = idents_hash;
3721
3722         get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3723
3724         ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3725
3726         for_each_possible_cpu(cpu) {
3727                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3728
3729                 INIT_LIST_HEAD(&ul->head);
3730                 INIT_LIST_HEAD(&ul->quarantine);
3731                 spin_lock_init(&ul->lock);
3732         }
3733 #ifdef CONFIG_IP_ROUTE_CLASSID
3734         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3735         if (!ip_rt_acct)
3736                 panic("IP: failed to allocate ip_rt_acct\n");
3737 #endif
3738
3739         ipv4_dst_ops.kmem_cachep =
3740                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3741                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3742
3743         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3744
3745         if (dst_entries_init(&ipv4_dst_ops) < 0)
3746                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3747
3748         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3749                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3750
3751         ipv4_dst_ops.gc_thresh = ~0;
3752         ip_rt_max_size = INT_MAX;
3753
3754         devinet_init();
3755         ip_fib_init();
3756
3757         if (ip_rt_proc_init())
3758                 pr_err("Unable to create route proc files\n");
3759 #ifdef CONFIG_XFRM
3760         xfrm_init();
3761         xfrm4_init();
3762 #endif
3763         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3764                       RTNL_FLAG_DOIT_UNLOCKED);
3765
3766 #ifdef CONFIG_SYSCTL
3767         register_pernet_subsys(&sysctl_route_ops);
3768 #endif
3769         register_pernet_subsys(&ip_rt_ops);
3770         register_pernet_subsys(&rt_genid_ops);
3771         register_pernet_subsys(&ipv4_inetpeer_ops);
3772         return 0;
3773 }
3774
3775 #ifdef CONFIG_SYSCTL
3776 /*
3777  * We really need to sanitize the damn ipv4 init order, then all
3778  * this nonsense will go away.
3779  */
3780 void __init ip_static_sysctl_init(void)
3781 {
3782         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3783 }
3784 #endif