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, <[email protected]>
  11  *              Alan Cox, <[email protected]>
  12  *              Linus Torvalds, <[email protected]>
  13  *              Alexey Kuznetsov, <[email protected]>
  14  *
  15  * Fixes:
  16  *              Alan Cox        :       Verify area fixes.
  17  *              Alan Cox        :       cli() protects routing changes
  18  *              Rui Oliveira    :       ICMP routing table updates
  19  *              ([email protected])      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_GC_TIMEOUT (300*HZ)
 110
 111 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
 112 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
 113 #define DEFAULT_MIN_ADVMSS 256
 114 static int ip_rt_max_size;
 115 static int ip_rt_redirect_number __read_mostly  = 9;
 116 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 117 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 118 static int ip_rt_error_cost __read_mostly       = HZ;
 119 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 120
 121 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
 122
 123 /*
 124  *      Interface to generic destination cache.
 125  */
 126
 127 INDIRECT_CALLABLE_SCOPE
 128 struct dst_entry        *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 129 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 130 INDIRECT_CALLABLE_SCOPE
 131 unsigned int            ipv4_mtu(const struct dst_entry *dst);
 132 static void             ipv4_negative_advice(struct sock *sk,
 133                                              struct dst_entry *dst);
 134 static void              ipv4_link_failure(struct sk_buff *skb);
 135 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 136                                            struct sk_buff *skb, u32 mtu,
 137                                            bool confirm_neigh);
 138 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 139                                         struct sk_buff *skb);
 140 static void             ipv4_dst_destroy(struct dst_entry *dst);
 141
 142 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 143 {
 144         WARN_ON(1);
 145         return NULL;
 146 }
 147
 148 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 149                                            struct sk_buff *skb,
 150                                            const void *daddr);
 151 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
 152
 153 static struct dst_ops ipv4_dst_ops = {
 154         .family =               AF_INET,
 155         .check =                ipv4_dst_check,
 156         .default_advmss =       ipv4_default_advmss,
 157         .mtu =                  ipv4_mtu,
 158         .cow_metrics =          ipv4_cow_metrics,
 159         .destroy =              ipv4_dst_destroy,
 160         .negative_advice =      ipv4_negative_advice,
 161         .link_failure =         ipv4_link_failure,
 162         .update_pmtu =          ip_rt_update_pmtu,
 163         .redirect =             ip_do_redirect,
 164         .local_out =            __ip_local_out,
 165         .neigh_lookup =         ipv4_neigh_lookup,
 166         .confirm_neigh =        ipv4_confirm_neigh,
 167 };
 168
 169 #define ECN_OR_COST(class)      TC_PRIO_##class
 170
 171 const __u8 ip_tos2prio[16] = {
 172         TC_PRIO_BESTEFFORT,
 173         ECN_OR_COST(BESTEFFORT),
 174         TC_PRIO_BESTEFFORT,
 175         ECN_OR_COST(BESTEFFORT),
 176         TC_PRIO_BULK,
 177         ECN_OR_COST(BULK),
 178         TC_PRIO_BULK,
 179         ECN_OR_COST(BULK),
 180         TC_PRIO_INTERACTIVE,
 181         ECN_OR_COST(INTERACTIVE),
 182         TC_PRIO_INTERACTIVE,
 183         ECN_OR_COST(INTERACTIVE),
 184         TC_PRIO_INTERACTIVE_BULK,
 185         ECN_OR_COST(INTERACTIVE_BULK),
 186         TC_PRIO_INTERACTIVE_BULK,
 187         ECN_OR_COST(INTERACTIVE_BULK)
 188 };
 189 EXPORT_SYMBOL(ip_tos2prio);
 190
 191 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 192 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
 193
 194 #ifdef CONFIG_PROC_FS
 195 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 196 {
 197         if (*pos)
 198                 return NULL;
 199         return SEQ_START_TOKEN;
 200 }
 201
 202 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 203 {
 204         ++*pos;
 205         return NULL;
 206 }
 207
 208 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 209 {
 210 }
 211
 212 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 213 {
 214         if (v == SEQ_START_TOKEN)
 215                 seq_printf(seq, "%-127s\n",
 216                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 217                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 218                            "HHUptod\tSpecDst");
 219         return 0;
 220 }
 221
 222 static const struct seq_operations rt_cache_seq_ops = {
 223         .start  = rt_cache_seq_start,
 224         .next   = rt_cache_seq_next,
 225         .stop   = rt_cache_seq_stop,
 226         .show   = rt_cache_seq_show,
 227 };
 228
 229 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 230 {
 231         int cpu;
 232
 233         if (*pos == 0)
 234                 return SEQ_START_TOKEN;
 235
 236         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 237                 if (!cpu_possible(cpu))
 238                         continue;
 239                 *pos = cpu+1;
 240                 return &per_cpu(rt_cache_stat, cpu);
 241         }
 242         return NULL;
 243 }
 244
 245 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 246 {
 247         int cpu;
 248
 249         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 250                 if (!cpu_possible(cpu))
 251                         continue;
 252                 *pos = cpu+1;
 253                 return &per_cpu(rt_cache_stat, cpu);
 254         }
 255         (*pos)++;
 256         return NULL;
 257
 258 }
 259
 260 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 261 {
 262
 263 }
 264
 265 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 266 {
 267         struct rt_cache_stat *st = v;
 268
 269         if (v == SEQ_START_TOKEN) {
 270                 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");
 271                 return 0;
 272         }
 273
 274         seq_printf(seq, "%08x %08x %08x    %08x   %08x    %08x %08x       "
 275                         "%08x       %08x %08x     %08x    %08x %08x   "
 276                         "%08x     %08x        %08x        %08x\n",
 277                    dst_entries_get_slow(&ipv4_dst_ops),
 278                    0, /* st->in_hit */
 279                    st->in_slow_tot,
 280                    st->in_slow_mc,
 281                    st->in_no_route,
 282                    st->in_brd,
 283                    st->in_martian_dst,
 284                    st->in_martian_src,
 285
 286                    0, /* st->out_hit */
 287                    st->out_slow_tot,
 288                    st->out_slow_mc,
 289
 290                    0, /* st->gc_total */
 291                    0, /* st->gc_ignored */
 292                    0, /* st->gc_goal_miss */
 293                    0, /* st->gc_dst_overflow */
 294                    0, /* st->in_hlist_search */
 295                    0  /* st->out_hlist_search */
 296                 );
 297         return 0;
 298 }
 299
 300 static const struct seq_operations rt_cpu_seq_ops = {
 301         .start  = rt_cpu_seq_start,
 302         .next   = rt_cpu_seq_next,
 303         .stop   = rt_cpu_seq_stop,
 304         .show   = rt_cpu_seq_show,
 305 };
 306
 307 #ifdef CONFIG_IP_ROUTE_CLASSID
 308 static int rt_acct_proc_show(struct seq_file *m, void *v)
 309 {
 310         struct ip_rt_acct *dst, *src;
 311         unsigned int i, j;
 312
 313         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 314         if (!dst)
 315                 return -ENOMEM;
 316
 317         for_each_possible_cpu(i) {
 318                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 319                 for (j = 0; j < 256; j++) {
 320                         dst[j].o_bytes   += src[j].o_bytes;
 321                         dst[j].o_packets += src[j].o_packets;
 322                         dst[j].i_bytes   += src[j].i_bytes;
 323                         dst[j].i_packets += src[j].i_packets;
 324                 }
 325         }
 326
 327         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 328         kfree(dst);
 329         return 0;
 330 }
 331 #endif
 332
 333 static int __net_init ip_rt_do_proc_init(struct net *net)
 334 {
 335         struct proc_dir_entry *pde;
 336
 337         pde = proc_create_seq("rt_cache", 0444, net->proc_net,
 338                               &rt_cache_seq_ops);
 339         if (!pde)
 340                 goto err1;
 341
 342         pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
 343                               &rt_cpu_seq_ops);
 344         if (!pde)
 345                 goto err2;
 346
 347 #ifdef CONFIG_IP_ROUTE_CLASSID
 348         pde = proc_create_single("rt_acct", 0, net->proc_net,
 349                         rt_acct_proc_show);
 350         if (!pde)
 351                 goto err3;
 352 #endif
 353         return 0;
 354
 355 #ifdef CONFIG_IP_ROUTE_CLASSID
 356 err3:
 357         remove_proc_entry("rt_cache", net->proc_net_stat);
 358 #endif
 359 err2:
 360         remove_proc_entry("rt_cache", net->proc_net);
 361 err1:
 362         return -ENOMEM;
 363 }
 364
 365 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 366 {
 367         remove_proc_entry("rt_cache", net->proc_net_stat);
 368         remove_proc_entry("rt_cache", net->proc_net);
 369 #ifdef CONFIG_IP_ROUTE_CLASSID
 370         remove_proc_entry("rt_acct", net->proc_net);
 371 #endif
 372 }
 373
 374 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 375         .init = ip_rt_do_proc_init,
 376         .exit = ip_rt_do_proc_exit,
 377 };
 378
 379 static int __init ip_rt_proc_init(void)
 380 {
 381         return register_pernet_subsys(&ip_rt_proc_ops);
 382 }
 383
 384 #else
 385 static inline int ip_rt_proc_init(void)
 386 {
 387         return 0;
 388 }
 389 #endif /* CONFIG_PROC_FS */
 390
 391 static inline bool rt_is_expired(const struct rtable *rth)
 392 {
 393         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 394 }
 395
 396 void rt_cache_flush(struct net *net)
 397 {
 398         rt_genid_bump_ipv4(net);
 399 }
 400
 401 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 402                                            struct sk_buff *skb,
 403                                            const void *daddr)
 404 {
 405         const struct rtable *rt = container_of(dst, struct rtable, dst);
 406         struct net_device *dev = dst->dev;
 407         struct neighbour *n;
 408
 409         rcu_read_lock();
 410
 411         if (likely(rt->rt_gw_family == AF_INET)) {
 412                 n = ip_neigh_gw4(dev, rt->rt_gw4);
 413         } else if (rt->rt_gw_family == AF_INET6) {
 414                 n = ip_neigh_gw6(dev, &rt->rt_gw6);
 415         } else {
 416                 __be32 pkey;
 417
 418                 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
 419                 n = ip_neigh_gw4(dev, pkey);
 420         }
 421
 422         if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
 423                 n = NULL;
 424
 425         rcu_read_unlock();
 426
 427         return n;
 428 }
 429
 430 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 431 {
 432         const struct rtable *rt = container_of(dst, struct rtable, dst);
 433         struct net_device *dev = dst->dev;
 434         const __be32 *pkey = daddr;
 435
 436         if (rt->rt_gw_family == AF_INET) {
 437                 pkey = (const __be32 *)&rt->rt_gw4;
 438         } else if (rt->rt_gw_family == AF_INET6) {
 439                 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
 440         } else if (!daddr ||
 441                  (rt->rt_flags &
 442                   (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
 443                 return;
 444         }
 445         __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
 446 }
 447
 448 /* Hash tables of size 2048..262144 depending on RAM size.
 449  * Each bucket uses 8 bytes.
 450  */
 451 static u32 ip_idents_mask __read_mostly;
 452 static atomic_t *ip_idents __read_mostly;
 453 static u32 *ip_tstamps __read_mostly;
 454
 455 /* In order to protect privacy, we add a perturbation to identifiers
 456  * if one generator is seldom used. This makes hard for an attacker
 457  * to infer how many packets were sent between two points in time.
 458  */
 459 static u32 ip_idents_reserve(u32 hash, int segs)
 460 {
 461         u32 bucket, old, now = (u32)jiffies;
 462         atomic_t *p_id;
 463         u32 *p_tstamp;
 464         u32 delta = 0;
 465
 466         bucket = hash & ip_idents_mask;
 467         p_tstamp = ip_tstamps + bucket;
 468         p_id = ip_idents + bucket;
 469         old = READ_ONCE(*p_tstamp);
 470
 471         if (old != now && cmpxchg(p_tstamp, old, now) == old)
 472                 delta = get_random_u32_below(now - old);
 473
 474         /* If UBSAN reports an error there, please make sure your compiler
 475          * supports -fno-strict-overflow before reporting it that was a bug
 476          * in UBSAN, and it has been fixed in GCC-8.
 477          */
 478         return atomic_add_return(segs + delta, p_id) - segs;
 479 }
 480
 481 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
 482 {
 483         u32 hash, id;
 484
 485         /* Note the following code is not safe, but this is okay. */
 486         if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
 487                 get_random_bytes(&net->ipv4.ip_id_key,
 488                                  sizeof(net->ipv4.ip_id_key));
 489
 490         hash = siphash_3u32((__force u32)iph->daddr,
 491                             (__force u32)iph->saddr,
 492                             iph->protocol,
 493                             &net->ipv4.ip_id_key);
 494         id = ip_idents_reserve(hash, segs);
 495         iph->id = htons(id);
 496 }
 497 EXPORT_SYMBOL(__ip_select_ident);
 498
 499 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
 500                              const struct sock *sk, const struct iphdr *iph,
 501                              int oif, __u8 tos, u8 prot, u32 mark,
 502                              int flow_flags)
 503 {
 504         __u8 scope = RT_SCOPE_UNIVERSE;
 505
 506         if (sk) {
 507                 oif = sk->sk_bound_dev_if;
 508                 mark = READ_ONCE(sk->sk_mark);
 509                 tos = ip_sock_rt_tos(sk);
 510                 scope = ip_sock_rt_scope(sk);
 511                 prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
 512                                                     sk->sk_protocol;
 513         }
 514
 515         flowi4_init_output(fl4, oif, mark, tos & IPTOS_RT_MASK, scope,
 516                            prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
 517                            sock_net_uid(net, sk));
 518 }
 519
 520 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 521                                const struct sock *sk)
 522 {
 523         const struct net *net = dev_net(skb->dev);
 524         const struct iphdr *iph = ip_hdr(skb);
 525         int oif = skb->dev->ifindex;
 526         u8 prot = iph->protocol;
 527         u32 mark = skb->mark;
 528         __u8 tos = iph->tos;
 529
 530         __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
 531 }
 532
 533 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 534 {
 535         const struct inet_sock *inet = inet_sk(sk);
 536         const struct ip_options_rcu *inet_opt;
 537         __be32 daddr = inet->inet_daddr;
 538
 539         rcu_read_lock();
 540         inet_opt = rcu_dereference(inet->inet_opt);
 541         if (inet_opt && inet_opt->opt.srr)
 542                 daddr = inet_opt->opt.faddr;
 543         flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
 544                            ip_sock_rt_tos(sk) & IPTOS_RT_MASK,
 545                            ip_sock_rt_scope(sk),
 546                            inet_test_bit(HDRINCL, sk) ?
 547                                 IPPROTO_RAW : sk->sk_protocol,
 548                            inet_sk_flowi_flags(sk),
 549                            daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
 550         rcu_read_unlock();
 551 }
 552
 553 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 554                                  const struct sk_buff *skb)
 555 {
 556         if (skb)
 557                 build_skb_flow_key(fl4, skb, sk);
 558         else
 559                 build_sk_flow_key(fl4, sk);
 560 }
 561
 562 static DEFINE_SPINLOCK(fnhe_lock);
 563
 564 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 565 {
 566         struct rtable *rt;
 567
 568         rt = rcu_dereference(fnhe->fnhe_rth_input);
 569         if (rt) {
 570                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 571                 dst_dev_put(&rt->dst);
 572                 dst_release(&rt->dst);
 573         }
 574         rt = rcu_dereference(fnhe->fnhe_rth_output);
 575         if (rt) {
 576                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 577                 dst_dev_put(&rt->dst);
 578                 dst_release(&rt->dst);
 579         }
 580 }
 581
 582 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
 583 {
 584         struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
 585         struct fib_nh_exception *fnhe, *oldest = NULL;
 586
 587         for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
 588                 fnhe = rcu_dereference_protected(*fnhe_p,
 589                                                  lockdep_is_held(&fnhe_lock));
 590                 if (!fnhe)
 591                         break;
 592                 if (!oldest ||
 593                     time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
 594                         oldest = fnhe;
 595                         oldest_p = fnhe_p;
 596                 }
 597         }
 598         fnhe_flush_routes(oldest);
 599         *oldest_p = oldest->fnhe_next;
 600         kfree_rcu(oldest, rcu);
 601 }
 602
 603 static u32 fnhe_hashfun(__be32 daddr)
 604 {
 605         static siphash_aligned_key_t fnhe_hash_key;
 606         u64 hval;
 607
 608         net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
 609         hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
 610         return hash_64(hval, FNHE_HASH_SHIFT);
 611 }
 612
 613 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 614 {
 615         rt->rt_pmtu = fnhe->fnhe_pmtu;
 616         rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
 617         rt->dst.expires = fnhe->fnhe_expires;
 618
 619         if (fnhe->fnhe_gw) {
 620                 rt->rt_flags |= RTCF_REDIRECTED;
 621                 rt->rt_uses_gateway = 1;
 622                 rt->rt_gw_family = AF_INET;
 623                 rt->rt_gw4 = fnhe->fnhe_gw;
 624         }
 625 }
 626
 627 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
 628                                   __be32 gw, u32 pmtu, bool lock,
 629                                   unsigned long expires)
 630 {
 631         struct fnhe_hash_bucket *hash;
 632         struct fib_nh_exception *fnhe;
 633         struct rtable *rt;
 634         u32 genid, hval;
 635         unsigned int i;
 636         int depth;
 637
 638         genid = fnhe_genid(dev_net(nhc->nhc_dev));
 639         hval = fnhe_hashfun(daddr);
 640
 641         spin_lock_bh(&fnhe_lock);
 642
 643         hash = rcu_dereference(nhc->nhc_exceptions);
 644         if (!hash) {
 645                 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
 646                 if (!hash)
 647                         goto out_unlock;
 648                 rcu_assign_pointer(nhc->nhc_exceptions, hash);
 649         }
 650
 651         hash += hval;
 652
 653         depth = 0;
 654         for (fnhe = rcu_dereference(hash->chain); fnhe;
 655              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 656                 if (fnhe->fnhe_daddr == daddr)
 657                         break;
 658                 depth++;
 659         }
 660
 661         if (fnhe) {
 662                 if (fnhe->fnhe_genid != genid)
 663                         fnhe->fnhe_genid = genid;
 664                 if (gw)
 665                         fnhe->fnhe_gw = gw;
 666                 if (pmtu) {
 667                         fnhe->fnhe_pmtu = pmtu;
 668                         fnhe->fnhe_mtu_locked = lock;
 669                 }
 670                 fnhe->fnhe_expires = max(1UL, expires);
 671                 /* Update all cached dsts too */
 672                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 673                 if (rt)
 674                         fill_route_from_fnhe(rt, fnhe);
 675                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 676                 if (rt)
 677                         fill_route_from_fnhe(rt, fnhe);
 678         } else {
 679                 /* Randomize max depth to avoid some side channels attacks. */
 680                 int max_depth = FNHE_RECLAIM_DEPTH +
 681                                 get_random_u32_below(FNHE_RECLAIM_DEPTH);
 682
 683                 while (depth > max_depth) {
 684                         fnhe_remove_oldest(hash);
 685                         depth--;
 686                 }
 687
 688                 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 689                 if (!fnhe)
 690                         goto out_unlock;
 691
 692                 fnhe->fnhe_next = hash->chain;
 693
 694                 fnhe->fnhe_genid = genid;
 695                 fnhe->fnhe_daddr = daddr;
 696                 fnhe->fnhe_gw = gw;
 697                 fnhe->fnhe_pmtu = pmtu;
 698                 fnhe->fnhe_mtu_locked = lock;
 699                 fnhe->fnhe_expires = max(1UL, expires);
 700
 701                 rcu_assign_pointer(hash->chain, fnhe);
 702
 703                 /* Exception created; mark the cached routes for the nexthop
 704                  * stale, so anyone caching it rechecks if this exception
 705                  * applies to them.
 706                  */
 707                 rt = rcu_dereference(nhc->nhc_rth_input);
 708                 if (rt)
 709                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 710
 711                 for_each_possible_cpu(i) {
 712                         struct rtable __rcu **prt;
 713
 714                         prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
 715                         rt = rcu_dereference(*prt);
 716                         if (rt)
 717                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 718                 }
 719         }
 720
 721         fnhe->fnhe_stamp = jiffies;
 722
 723 out_unlock:
 724         spin_unlock_bh(&fnhe_lock);
 725 }
 726
 727 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 728                              bool kill_route)
 729 {
 730         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 731         __be32 old_gw = ip_hdr(skb)->saddr;
 732         struct net_device *dev = skb->dev;
 733         struct in_device *in_dev;
 734         struct fib_result res;
 735         struct neighbour *n;
 736         struct net *net;
 737
 738         switch (icmp_hdr(skb)->code & 7) {
 739         case ICMP_REDIR_NET:
 740         case ICMP_REDIR_NETTOS:
 741         case ICMP_REDIR_HOST:
 742         case ICMP_REDIR_HOSTTOS:
 743                 break;
 744
 745         default:
 746                 return;
 747         }
 748
 749         if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
 750                 return;
 751
 752         in_dev = __in_dev_get_rcu(dev);
 753         if (!in_dev)
 754                 return;
 755
 756         net = dev_net(dev);
 757         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 758             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 759             ipv4_is_zeronet(new_gw))
 760                 goto reject_redirect;
 761
 762         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 763                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 764                         goto reject_redirect;
 765                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 766                         goto reject_redirect;
 767         } else {
 768                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 769                         goto reject_redirect;
 770         }
 771
 772         n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
 773         if (!n)
 774                 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
 775         if (!IS_ERR(n)) {
 776                 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
 777                         neigh_event_send(n, NULL);
 778                 } else {
 779                         if (fib_lookup(net, fl4, &res, 0) == 0) {
 780                                 struct fib_nh_common *nhc;
 781
 782                                 fib_select_path(net, &res, fl4, skb);
 783                                 nhc = FIB_RES_NHC(res);
 784                                 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
 785                                                 0, false,
 786                                                 jiffies + ip_rt_gc_timeout);
 787                         }
 788                         if (kill_route)
 789                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 790                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 791                 }
 792                 neigh_release(n);
 793         }
 794         return;
 795
 796 reject_redirect:
 797 #ifdef CONFIG_IP_ROUTE_VERBOSE
 798         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 799                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 800                 __be32 daddr = iph->daddr;
 801                 __be32 saddr = iph->saddr;
 802
 803                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 804                                      "  Advised path = %pI4 -> %pI4\n",
 805                                      &old_gw, dev->name, &new_gw,
 806                                      &saddr, &daddr);
 807         }
 808 #endif
 809         ;
 810 }
 811
 812 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 813 {
 814         struct rtable *rt;
 815         struct flowi4 fl4;
 816         const struct iphdr *iph = (const struct iphdr *) skb->data;
 817         struct net *net = dev_net(skb->dev);
 818         int oif = skb->dev->ifindex;
 819         u8 prot = iph->protocol;
 820         u32 mark = skb->mark;
 821         __u8 tos = iph->tos;
 822
 823         rt = dst_rtable(dst);
 824
 825         __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
 826         __ip_do_redirect(rt, skb, &fl4, true);
 827 }
 828
 829 static void ipv4_negative_advice(struct sock *sk,
 830                                  struct dst_entry *dst)
 831 {
 832         struct rtable *rt = dst_rtable(dst);
 833
 834         if ((dst->obsolete > 0) ||
 835             (rt->rt_flags & RTCF_REDIRECTED) ||
 836             rt->dst.expires)
 837                 sk_dst_reset(sk);
 838 }
 839
 840 /*
 841  * Algorithm:
 842  *      1. The first ip_rt_redirect_number redirects are sent
 843  *         with exponential backoff, then we stop sending them at all,
 844  *         assuming that the host ignores our redirects.
 845  *      2. If we did not see packets requiring redirects
 846  *         during ip_rt_redirect_silence, we assume that the host
 847  *         forgot redirected route and start to send redirects again.
 848  *
 849  * This algorithm is much cheaper and more intelligent than dumb load limiting
 850  * in icmp.c.
 851  *
 852  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 853  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 854  */
 855
 856 void ip_rt_send_redirect(struct sk_buff *skb)
 857 {
 858         struct rtable *rt = skb_rtable(skb);
 859         struct in_device *in_dev;
 860         struct inet_peer *peer;
 861         struct net *net;
 862         int log_martians;
 863         int vif;
 864
 865         rcu_read_lock();
 866         in_dev = __in_dev_get_rcu(rt->dst.dev);
 867         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 868                 rcu_read_unlock();
 869                 return;
 870         }
 871         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 872         vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
 873         rcu_read_unlock();
 874
 875         net = dev_net(rt->dst.dev);
 876         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
 877         if (!peer) {
 878                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 879                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 880                 return;
 881         }
 882
 883         /* No redirected packets during ip_rt_redirect_silence;
 884          * reset the algorithm.
 885          */
 886         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
 887                 peer->rate_tokens = 0;
 888                 peer->n_redirects = 0;
 889         }
 890
 891         /* Too many ignored redirects; do not send anything
 892          * set dst.rate_last to the last seen redirected packet.
 893          */
 894         if (peer->n_redirects >= ip_rt_redirect_number) {
 895                 peer->rate_last = jiffies;
 896                 goto out_put_peer;
 897         }
 898
 899         /* Check for load limit; set rate_last to the latest sent
 900          * redirect.
 901          */
 902         if (peer->n_redirects == 0 ||
 903             time_after(jiffies,
 904                        (peer->rate_last +
 905                         (ip_rt_redirect_load << peer->n_redirects)))) {
 906                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 907
 908                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 909                 peer->rate_last = jiffies;
 910                 ++peer->n_redirects;
 911                 if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
 912                     peer->n_redirects == ip_rt_redirect_number)
 913                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 914                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 915                                              &ip_hdr(skb)->daddr, &gw);
 916         }
 917 out_put_peer:
 918         inet_putpeer(peer);
 919 }
 920
 921 static int ip_error(struct sk_buff *skb)
 922 {
 923         struct rtable *rt = skb_rtable(skb);
 924         struct net_device *dev = skb->dev;
 925         struct in_device *in_dev;
 926         struct inet_peer *peer;
 927         unsigned long now;
 928         struct net *net;
 929         SKB_DR(reason);
 930         bool send;
 931         int code;
 932
 933         if (netif_is_l3_master(skb->dev)) {
 934                 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
 935                 if (!dev)
 936                         goto out;
 937         }
 938
 939         in_dev = __in_dev_get_rcu(dev);
 940
 941         /* IP on this device is disabled. */
 942         if (!in_dev)
 943                 goto out;
 944
 945         net = dev_net(rt->dst.dev);
 946         if (!IN_DEV_FORWARD(in_dev)) {
 947                 switch (rt->dst.error) {
 948                 case EHOSTUNREACH:
 949                         SKB_DR_SET(reason, IP_INADDRERRORS);
 950                         __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
 951                         break;
 952
 953                 case ENETUNREACH:
 954                         SKB_DR_SET(reason, IP_INNOROUTES);
 955                         __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 956                         break;
 957                 }
 958                 goto out;
 959         }
 960
 961         switch (rt->dst.error) {
 962         case EINVAL:
 963         default:
 964                 goto out;
 965         case EHOSTUNREACH:
 966                 code = ICMP_HOST_UNREACH;
 967                 break;
 968         case ENETUNREACH:
 969                 code = ICMP_NET_UNREACH;
 970                 SKB_DR_SET(reason, IP_INNOROUTES);
 971                 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 972                 break;
 973         case EACCES:
 974                 code = ICMP_PKT_FILTERED;
 975                 break;
 976         }
 977
 978         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
 979                                l3mdev_master_ifindex(skb->dev), 1);
 980
 981         send = true;
 982         if (peer) {
 983                 now = jiffies;
 984                 peer->rate_tokens += now - peer->rate_last;
 985                 if (peer->rate_tokens > ip_rt_error_burst)
 986                         peer->rate_tokens = ip_rt_error_burst;
 987                 peer->rate_last = now;
 988                 if (peer->rate_tokens >= ip_rt_error_cost)
 989                         peer->rate_tokens -= ip_rt_error_cost;
 990                 else
 991                         send = false;
 992                 inet_putpeer(peer);
 993         }
 994         if (send)
 995                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
 996
 997 out:    kfree_skb_reason(skb, reason);
 998         return 0;
 999 }
1000
1001 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1002 {
1003         struct dst_entry *dst = &rt->dst;
1004         struct net *net = dev_net(dst->dev);
1005         struct fib_result res;
1006         bool lock = false;
1007         u32 old_mtu;
1008
1009         if (ip_mtu_locked(dst))
1010                 return;
1011
1012         old_mtu = ipv4_mtu(dst);
1013         if (old_mtu < mtu)
1014                 return;
1015
1016         if (mtu < net->ipv4.ip_rt_min_pmtu) {
1017                 lock = true;
1018                 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1019         }
1020
1021         if (rt->rt_pmtu == mtu && !lock &&
1022             time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1023                 return;
1024
1025         rcu_read_lock();
1026         if (fib_lookup(net, fl4, &res, 0) == 0) {
1027                 struct fib_nh_common *nhc;
1028
1029                 fib_select_path(net, &res, fl4, NULL);
1030                 nhc = FIB_RES_NHC(res);
1031                 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1032                                       jiffies + net->ipv4.ip_rt_mtu_expires);
1033         }
1034         rcu_read_unlock();
1035 }
1036
1037 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1038                               struct sk_buff *skb, u32 mtu,
1039                               bool confirm_neigh)
1040 {
1041         struct rtable *rt = dst_rtable(dst);
1042         struct flowi4 fl4;
1043
1044         ip_rt_build_flow_key(&fl4, sk, skb);
1045
1046         /* Don't make lookup fail for bridged encapsulations */
1047         if (skb && netif_is_any_bridge_port(skb->dev))
1048                 fl4.flowi4_oif = 0;
1049
1050         __ip_rt_update_pmtu(rt, &fl4, mtu);
1051 }
1052
1053 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1054                       int oif, u8 protocol)
1055 {
1056         const struct iphdr *iph = (const struct iphdr *)skb->data;
1057         struct flowi4 fl4;
1058         struct rtable *rt;
1059         u32 mark = IP4_REPLY_MARK(net, skb->mark);
1060
1061         __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1062                          0);
1063         rt = __ip_route_output_key(net, &fl4);
1064         if (!IS_ERR(rt)) {
1065                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1066                 ip_rt_put(rt);
1067         }
1068 }
1069 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1070
1071 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1072 {
1073         const struct iphdr *iph = (const struct iphdr *)skb->data;
1074         struct flowi4 fl4;
1075         struct rtable *rt;
1076
1077         __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1078
1079         if (!fl4.flowi4_mark)
1080                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1081
1082         rt = __ip_route_output_key(sock_net(sk), &fl4);
1083         if (!IS_ERR(rt)) {
1084                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1085                 ip_rt_put(rt);
1086         }
1087 }
1088
1089 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1090 {
1091         const struct iphdr *iph = (const struct iphdr *)skb->data;
1092         struct flowi4 fl4;
1093         struct rtable *rt;
1094         struct dst_entry *odst = NULL;
1095         bool new = false;
1096         struct net *net = sock_net(sk);
1097
1098         bh_lock_sock(sk);
1099
1100         if (!ip_sk_accept_pmtu(sk))
1101                 goto out;
1102
1103         odst = sk_dst_get(sk);
1104
1105         if (sock_owned_by_user(sk) || !odst) {
1106                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1107                 goto out;
1108         }
1109
1110         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1111
1112         rt = dst_rtable(odst);
1113         if (odst->obsolete && !odst->ops->check(odst, 0)) {
1114                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1115                 if (IS_ERR(rt))
1116                         goto out;
1117
1118                 new = true;
1119         }
1120
1121         __ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu);
1122
1123         if (!dst_check(&rt->dst, 0)) {
1124                 if (new)
1125                         dst_release(&rt->dst);
1126
1127                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1128                 if (IS_ERR(rt))
1129                         goto out;
1130
1131                 new = true;
1132         }
1133
1134         if (new)
1135                 sk_dst_set(sk, &rt->dst);
1136
1137 out:
1138         bh_unlock_sock(sk);
1139         dst_release(odst);
1140 }
1141 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1142
1143 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1144                    int oif, u8 protocol)
1145 {
1146         const struct iphdr *iph = (const struct iphdr *)skb->data;
1147         struct flowi4 fl4;
1148         struct rtable *rt;
1149
1150         __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1151         rt = __ip_route_output_key(net, &fl4);
1152         if (!IS_ERR(rt)) {
1153                 __ip_do_redirect(rt, skb, &fl4, false);
1154                 ip_rt_put(rt);
1155         }
1156 }
1157 EXPORT_SYMBOL_GPL(ipv4_redirect);
1158
1159 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1160 {
1161         const struct iphdr *iph = (const struct iphdr *)skb->data;
1162         struct flowi4 fl4;
1163         struct rtable *rt;
1164         struct net *net = sock_net(sk);
1165
1166         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1167         rt = __ip_route_output_key(net, &fl4);
1168         if (!IS_ERR(rt)) {
1169                 __ip_do_redirect(rt, skb, &fl4, false);
1170                 ip_rt_put(rt);
1171         }
1172 }
1173 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1174
1175 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1176                                                          u32 cookie)
1177 {
1178         struct rtable *rt = dst_rtable(dst);
1179
1180         /* All IPV4 dsts are created with ->obsolete set to the value
1181          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1182          * into this function always.
1183          *
1184          * When a PMTU/redirect information update invalidates a route,
1185          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1186          * DST_OBSOLETE_DEAD.
1187          */
1188         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1189                 return NULL;
1190         return dst;
1191 }
1192 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1193
1194 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1195 {
1196         struct net_device *dev;
1197         struct ip_options opt;
1198         int res;
1199
1200         /* Recompile ip options since IPCB may not be valid anymore.
1201          * Also check we have a reasonable ipv4 header.
1202          */
1203         if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1204             ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1205                 return;
1206
1207         memset(&opt, 0, sizeof(opt));
1208         if (ip_hdr(skb)->ihl > 5) {
1209                 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1210                         return;
1211                 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1212
1213                 rcu_read_lock();
1214                 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1215                 res = __ip_options_compile(dev_net(dev), &opt, skb, NULL);
1216                 rcu_read_unlock();
1217
1218                 if (res)
1219                         return;
1220         }
1221         __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1222 }
1223
1224 static void ipv4_link_failure(struct sk_buff *skb)
1225 {
1226         struct rtable *rt;
1227
1228         ipv4_send_dest_unreach(skb);
1229
1230         rt = skb_rtable(skb);
1231         if (rt)
1232                 dst_set_expires(&rt->dst, 0);
1233 }
1234
1235 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1236 {
1237         pr_debug("%s: %pI4 -> %pI4, %s\n",
1238                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1239                  skb->dev ? skb->dev->name : "?");
1240         kfree_skb(skb);
1241         WARN_ON(1);
1242         return 0;
1243 }
1244
1245 /*
1246  * We do not cache source address of outgoing interface,
1247  * because it is used only by IP RR, TS and SRR options,
1248  * so that it out of fast path.
1249  *
1250  * BTW remember: "addr" is allowed to be not aligned
1251  * in IP options!
1252  */
1253
1254 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1255 {
1256         __be32 src;
1257
1258         if (rt_is_output_route(rt))
1259                 src = ip_hdr(skb)->saddr;
1260         else {
1261                 struct fib_result res;
1262                 struct iphdr *iph = ip_hdr(skb);
1263                 struct flowi4 fl4 = {
1264                         .daddr = iph->daddr,
1265                         .saddr = iph->saddr,
1266                         .flowi4_tos = RT_TOS(iph->tos),
1267                         .flowi4_oif = rt->dst.dev->ifindex,
1268                         .flowi4_iif = skb->dev->ifindex,
1269                         .flowi4_mark = skb->mark,
1270                 };
1271
1272                 rcu_read_lock();
1273                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1274                         src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1275                 else
1276                         src = inet_select_addr(rt->dst.dev,
1277                                                rt_nexthop(rt, iph->daddr),
1278                                                RT_SCOPE_UNIVERSE);
1279                 rcu_read_unlock();
1280         }
1281         memcpy(addr, &src, 4);
1282 }
1283
1284 #ifdef CONFIG_IP_ROUTE_CLASSID
1285 static void set_class_tag(struct rtable *rt, u32 tag)
1286 {
1287         if (!(rt->dst.tclassid & 0xFFFF))
1288                 rt->dst.tclassid |= tag & 0xFFFF;
1289         if (!(rt->dst.tclassid & 0xFFFF0000))
1290                 rt->dst.tclassid |= tag & 0xFFFF0000;
1291 }
1292 #endif
1293
1294 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1295 {
1296         struct net *net = dev_net(dst->dev);
1297         unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1298         unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1299                                     net->ipv4.ip_rt_min_advmss);
1300
1301         return min(advmss, IPV4_MAX_PMTU - header_size);
1302 }
1303
1304 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1305 {
1306         return ip_dst_mtu_maybe_forward(dst, false);
1307 }
1308 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1309
1310 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1311 {
1312         struct fnhe_hash_bucket *hash;
1313         struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1314         u32 hval = fnhe_hashfun(daddr);
1315
1316         spin_lock_bh(&fnhe_lock);
1317
1318         hash = rcu_dereference_protected(nhc->nhc_exceptions,
1319                                          lockdep_is_held(&fnhe_lock));
1320         hash += hval;
1321
1322         fnhe_p = &hash->chain;
1323         fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1324         while (fnhe) {
1325                 if (fnhe->fnhe_daddr == daddr) {
1326                         rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1327                                 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1328                         /* set fnhe_daddr to 0 to ensure it won't bind with
1329                          * new dsts in rt_bind_exception().
1330                          */
1331                         fnhe->fnhe_daddr = 0;
1332                         fnhe_flush_routes(fnhe);
1333                         kfree_rcu(fnhe, rcu);
1334                         break;
1335                 }
1336                 fnhe_p = &fnhe->fnhe_next;
1337                 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1338                                                  lockdep_is_held(&fnhe_lock));
1339         }
1340
1341         spin_unlock_bh(&fnhe_lock);
1342 }
1343
1344 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1345                                                __be32 daddr)
1346 {
1347         struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1348         struct fib_nh_exception *fnhe;
1349         u32 hval;
1350
1351         if (!hash)
1352                 return NULL;
1353
1354         hval = fnhe_hashfun(daddr);
1355
1356         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1357              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1358                 if (fnhe->fnhe_daddr == daddr) {
1359                         if (fnhe->fnhe_expires &&
1360                             time_after(jiffies, fnhe->fnhe_expires)) {
1361                                 ip_del_fnhe(nhc, daddr);
1362                                 break;
1363                         }
1364                         return fnhe;
1365                 }
1366         }
1367         return NULL;
1368 }
1369
1370 /* MTU selection:
1371  * 1. mtu on route is locked - use it
1372  * 2. mtu from nexthop exception
1373  * 3. mtu from egress device
1374  */
1375
1376 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1377 {
1378         struct fib_nh_common *nhc = res->nhc;
1379         struct net_device *dev = nhc->nhc_dev;
1380         struct fib_info *fi = res->fi;
1381         u32 mtu = 0;
1382
1383         if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1384             fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1385                 mtu = fi->fib_mtu;
1386
1387         if (likely(!mtu)) {
1388                 struct fib_nh_exception *fnhe;
1389
1390                 fnhe = find_exception(nhc, daddr);
1391                 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1392                         mtu = fnhe->fnhe_pmtu;
1393         }
1394
1395         if (likely(!mtu))
1396                 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1397
1398         return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1399 }
1400
1401 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1402                               __be32 daddr, const bool do_cache)
1403 {
1404         bool ret = false;
1405
1406         spin_lock_bh(&fnhe_lock);
1407
1408         if (daddr == fnhe->fnhe_daddr) {
1409                 struct rtable __rcu **porig;
1410                 struct rtable *orig;
1411                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1412
1413                 if (rt_is_input_route(rt))
1414                         porig = &fnhe->fnhe_rth_input;
1415                 else
1416                         porig = &fnhe->fnhe_rth_output;
1417                 orig = rcu_dereference(*porig);
1418
1419                 if (fnhe->fnhe_genid != genid) {
1420                         fnhe->fnhe_genid = genid;
1421                         fnhe->fnhe_gw = 0;
1422                         fnhe->fnhe_pmtu = 0;
1423                         fnhe->fnhe_expires = 0;
1424                         fnhe->fnhe_mtu_locked = false;
1425                         fnhe_flush_routes(fnhe);
1426                         orig = NULL;
1427                 }
1428                 fill_route_from_fnhe(rt, fnhe);
1429                 if (!rt->rt_gw4) {
1430                         rt->rt_gw4 = daddr;
1431                         rt->rt_gw_family = AF_INET;
1432                 }
1433
1434                 if (do_cache) {
1435                         dst_hold(&rt->dst);
1436                         rcu_assign_pointer(*porig, rt);
1437                         if (orig) {
1438                                 dst_dev_put(&orig->dst);
1439                                 dst_release(&orig->dst);
1440                         }
1441                         ret = true;
1442                 }
1443
1444                 fnhe->fnhe_stamp = jiffies;
1445         }
1446         spin_unlock_bh(&fnhe_lock);
1447
1448         return ret;
1449 }
1450
1451 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1452 {
1453         struct rtable *orig, *prev, **p;
1454         bool ret = true;
1455
1456         if (rt_is_input_route(rt)) {
1457                 p = (struct rtable **)&nhc->nhc_rth_input;
1458         } else {
1459                 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1460         }
1461         orig = *p;
1462
1463         /* hold dst before doing cmpxchg() to avoid race condition
1464          * on this dst
1465          */
1466         dst_hold(&rt->dst);
1467         prev = cmpxchg(p, orig, rt);
1468         if (prev == orig) {
1469                 if (orig) {
1470                         rt_add_uncached_list(orig);
1471                         dst_release(&orig->dst);
1472                 }
1473         } else {
1474                 dst_release(&rt->dst);
1475                 ret = false;
1476         }
1477
1478         return ret;
1479 }
1480
1481 struct uncached_list {
1482         spinlock_t              lock;
1483         struct list_head        head;
1484         struct list_head        quarantine;
1485 };
1486
1487 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1488
1489 void rt_add_uncached_list(struct rtable *rt)
1490 {
1491         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1492
1493         rt->dst.rt_uncached_list = ul;
1494
1495         spin_lock_bh(&ul->lock);
1496         list_add_tail(&rt->dst.rt_uncached, &ul->head);
1497         spin_unlock_bh(&ul->lock);
1498 }
1499
1500 void rt_del_uncached_list(struct rtable *rt)
1501 {
1502         if (!list_empty(&rt->dst.rt_uncached)) {
1503                 struct uncached_list *ul = rt->dst.rt_uncached_list;
1504
1505                 spin_lock_bh(&ul->lock);
1506                 list_del_init(&rt->dst.rt_uncached);
1507                 spin_unlock_bh(&ul->lock);
1508         }
1509 }
1510
1511 static void ipv4_dst_destroy(struct dst_entry *dst)
1512 {
1513         ip_dst_metrics_put(dst);
1514         rt_del_uncached_list(dst_rtable(dst));
1515 }
1516
1517 void rt_flush_dev(struct net_device *dev)
1518 {
1519         struct rtable *rt, *safe;
1520         int cpu;
1521
1522         for_each_possible_cpu(cpu) {
1523                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1524
1525                 if (list_empty(&ul->head))
1526                         continue;
1527
1528                 spin_lock_bh(&ul->lock);
1529                 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1530                         if (rt->dst.dev != dev)
1531                                 continue;
1532                         rt->dst.dev = blackhole_netdev;
1533                         netdev_ref_replace(dev, blackhole_netdev,
1534                                            &rt->dst.dev_tracker, GFP_ATOMIC);
1535                         list_move(&rt->dst.rt_uncached, &ul->quarantine);
1536                 }
1537                 spin_unlock_bh(&ul->lock);
1538         }
1539 }
1540
1541 static bool rt_cache_valid(const struct rtable *rt)
1542 {
1543         return  rt &&
1544                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1545                 !rt_is_expired(rt);
1546 }
1547
1548 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1549                            const struct fib_result *res,
1550                            struct fib_nh_exception *fnhe,
1551                            struct fib_info *fi, u16 type, u32 itag,
1552                            const bool do_cache)
1553 {
1554         bool cached = false;
1555
1556         if (fi) {
1557                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1558
1559                 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1560                         rt->rt_uses_gateway = 1;
1561                         rt->rt_gw_family = nhc->nhc_gw_family;
1562                         /* only INET and INET6 are supported */
1563                         if (likely(nhc->nhc_gw_family == AF_INET))
1564                                 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1565                         else
1566                                 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1567                 }
1568
1569                 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1570
1571 #ifdef CONFIG_IP_ROUTE_CLASSID
1572                 if (nhc->nhc_family == AF_INET) {
1573                         struct fib_nh *nh;
1574
1575                         nh = container_of(nhc, struct fib_nh, nh_common);
1576                         rt->dst.tclassid = nh->nh_tclassid;
1577                 }
1578 #endif
1579                 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1580                 if (unlikely(fnhe))
1581                         cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1582                 else if (do_cache)
1583                         cached = rt_cache_route(nhc, rt);
1584                 if (unlikely(!cached)) {
1585                         /* Routes we intend to cache in nexthop exception or
1586                          * FIB nexthop have the DST_NOCACHE bit clear.
1587                          * However, if we are unsuccessful at storing this
1588                          * route into the cache we really need to set it.
1589                          */
1590                         if (!rt->rt_gw4) {
1591                                 rt->rt_gw_family = AF_INET;
1592                                 rt->rt_gw4 = daddr;
1593                         }
1594                         rt_add_uncached_list(rt);
1595                 }
1596         } else
1597                 rt_add_uncached_list(rt);
1598
1599 #ifdef CONFIG_IP_ROUTE_CLASSID
1600 #ifdef CONFIG_IP_MULTIPLE_TABLES
1601         set_class_tag(rt, res->tclassid);
1602 #endif
1603         set_class_tag(rt, itag);
1604 #endif
1605 }
1606
1607 struct rtable *rt_dst_alloc(struct net_device *dev,
1608                             unsigned int flags, u16 type,
1609                             bool noxfrm)
1610 {
1611         struct rtable *rt;
1612
1613         rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1614                        (noxfrm ? DST_NOXFRM : 0));
1615
1616         if (rt) {
1617                 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1618                 rt->rt_flags = flags;
1619                 rt->rt_type = type;
1620                 rt->rt_is_input = 0;
1621                 rt->rt_iif = 0;
1622                 rt->rt_pmtu = 0;
1623                 rt->rt_mtu_locked = 0;
1624                 rt->rt_uses_gateway = 0;
1625                 rt->rt_gw_family = 0;
1626                 rt->rt_gw4 = 0;
1627
1628                 rt->dst.output = ip_output;
1629                 if (flags & RTCF_LOCAL)
1630                         rt->dst.input = ip_local_deliver;
1631         }
1632
1633         return rt;
1634 }
1635 EXPORT_SYMBOL(rt_dst_alloc);
1636
1637 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1638 {
1639         struct rtable *new_rt;
1640
1641         new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1642                            rt->dst.flags);
1643
1644         if (new_rt) {
1645                 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1646                 new_rt->rt_flags = rt->rt_flags;
1647                 new_rt->rt_type = rt->rt_type;
1648                 new_rt->rt_is_input = rt->rt_is_input;
1649                 new_rt->rt_iif = rt->rt_iif;
1650                 new_rt->rt_pmtu = rt->rt_pmtu;
1651                 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1652                 new_rt->rt_gw_family = rt->rt_gw_family;
1653                 if (rt->rt_gw_family == AF_INET)
1654                         new_rt->rt_gw4 = rt->rt_gw4;
1655                 else if (rt->rt_gw_family == AF_INET6)
1656                         new_rt->rt_gw6 = rt->rt_gw6;
1657
1658                 new_rt->dst.input = rt->dst.input;
1659                 new_rt->dst.output = rt->dst.output;
1660                 new_rt->dst.error = rt->dst.error;
1661                 new_rt->dst.lastuse = jiffies;
1662                 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1663         }
1664         return new_rt;
1665 }
1666 EXPORT_SYMBOL(rt_dst_clone);
1667
1668 /* called in rcu_read_lock() section */
1669 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1670                           u8 tos, struct net_device *dev,
1671                           struct in_device *in_dev, u32 *itag)
1672 {
1673         int err;
1674
1675         /* Primary sanity checks. */
1676         if (!in_dev)
1677                 return -EINVAL;
1678
1679         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1680             skb->protocol != htons(ETH_P_IP))
1681                 return -EINVAL;
1682
1683         if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1684                 return -EINVAL;
1685
1686         if (ipv4_is_zeronet(saddr)) {
1687                 if (!ipv4_is_local_multicast(daddr) &&
1688                     ip_hdr(skb)->protocol != IPPROTO_IGMP)
1689                         return -EINVAL;
1690         } else {
1691                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1692                                           in_dev, itag);
1693                 if (err < 0)
1694                         return err;
1695         }
1696         return 0;
1697 }
1698
1699 /* called in rcu_read_lock() section */
1700 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1701                              u8 tos, struct net_device *dev, int our)
1702 {
1703         struct in_device *in_dev = __in_dev_get_rcu(dev);
1704         unsigned int flags = RTCF_MULTICAST;
1705         struct rtable *rth;
1706         u32 itag = 0;
1707         int err;
1708
1709         err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1710         if (err)
1711                 return err;
1712
1713         if (our)
1714                 flags |= RTCF_LOCAL;
1715
1716         if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1717                 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1718
1719         rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1720                            false);
1721         if (!rth)
1722                 return -ENOBUFS;
1723
1724 #ifdef CONFIG_IP_ROUTE_CLASSID
1725         rth->dst.tclassid = itag;
1726 #endif
1727         rth->dst.output = ip_rt_bug;
1728         rth->rt_is_input= 1;
1729
1730 #ifdef CONFIG_IP_MROUTE
1731         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1732                 rth->dst.input = ip_mr_input;
1733 #endif
1734         RT_CACHE_STAT_INC(in_slow_mc);
1735
1736         skb_dst_drop(skb);
1737         skb_dst_set(skb, &rth->dst);
1738         return 0;
1739 }
1740
1741
1742 static void ip_handle_martian_source(struct net_device *dev,
1743                                      struct in_device *in_dev,
1744                                      struct sk_buff *skb,
1745                                      __be32 daddr,
1746                                      __be32 saddr)
1747 {
1748         RT_CACHE_STAT_INC(in_martian_src);
1749 #ifdef CONFIG_IP_ROUTE_VERBOSE
1750         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1751                 /*
1752                  *      RFC1812 recommendation, if source is martian,
1753                  *      the only hint is MAC header.
1754                  */
1755                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1756                         &daddr, &saddr, dev->name);
1757                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1758                         print_hex_dump(KERN_WARNING, "ll header: ",
1759                                        DUMP_PREFIX_OFFSET, 16, 1,
1760                                        skb_mac_header(skb),
1761                                        dev->hard_header_len, false);
1762                 }
1763         }
1764 #endif
1765 }
1766
1767 /* called in rcu_read_lock() section */
1768 static int __mkroute_input(struct sk_buff *skb,
1769                            const struct fib_result *res,
1770                            struct in_device *in_dev,
1771                            __be32 daddr, __be32 saddr, u32 tos)
1772 {
1773         struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1774         struct net_device *dev = nhc->nhc_dev;
1775         struct fib_nh_exception *fnhe;
1776         struct rtable *rth;
1777         int err;
1778         struct in_device *out_dev;
1779         bool do_cache;
1780         u32 itag = 0;
1781
1782         /* get a working reference to the output device */
1783         out_dev = __in_dev_get_rcu(dev);
1784         if (!out_dev) {
1785                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1786                 return -EINVAL;
1787         }
1788
1789         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1790                                   in_dev->dev, in_dev, &itag);
1791         if (err < 0) {
1792                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1793                                          saddr);
1794
1795                 goto cleanup;
1796         }
1797
1798         do_cache = res->fi && !itag;
1799         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1800             skb->protocol == htons(ETH_P_IP)) {
1801                 __be32 gw;
1802
1803                 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1804                 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1805                     inet_addr_onlink(out_dev, saddr, gw))
1806                         IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1807         }
1808
1809         if (skb->protocol != htons(ETH_P_IP)) {
1810                 /* Not IP (i.e. ARP). Do not create route, if it is
1811                  * invalid for proxy arp. DNAT routes are always valid.
1812                  *
1813                  * Proxy arp feature have been extended to allow, ARP
1814                  * replies back to the same interface, to support
1815                  * Private VLAN switch technologies. See arp.c.
1816                  */
1817                 if (out_dev == in_dev &&
1818                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1819                         err = -EINVAL;
1820                         goto cleanup;
1821                 }
1822         }
1823
1824         if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1825                 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1826
1827         fnhe = find_exception(nhc, daddr);
1828         if (do_cache) {
1829                 if (fnhe)
1830                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1831                 else
1832                         rth = rcu_dereference(nhc->nhc_rth_input);
1833                 if (rt_cache_valid(rth)) {
1834                         skb_dst_set_noref(skb, &rth->dst);
1835                         goto out;
1836                 }
1837         }
1838
1839         rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1840                            IN_DEV_ORCONF(out_dev, NOXFRM));
1841         if (!rth) {
1842                 err = -ENOBUFS;
1843                 goto cleanup;
1844         }
1845
1846         rth->rt_is_input = 1;
1847         RT_CACHE_STAT_INC(in_slow_tot);
1848
1849         rth->dst.input = ip_forward;
1850
1851         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1852                        do_cache);
1853         lwtunnel_set_redirect(&rth->dst);
1854         skb_dst_set(skb, &rth->dst);
1855 out:
1856         err = 0;
1857  cleanup:
1858         return err;
1859 }
1860
1861 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1862 /* To make ICMP packets follow the right flow, the multipath hash is
1863  * calculated from the inner IP addresses.
1864  */
1865 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1866                                  struct flow_keys *hash_keys)
1867 {
1868         const struct iphdr *outer_iph = ip_hdr(skb);
1869         const struct iphdr *key_iph = outer_iph;
1870         const struct iphdr *inner_iph;
1871         const struct icmphdr *icmph;
1872         struct iphdr _inner_iph;
1873         struct icmphdr _icmph;
1874
1875         if (likely(outer_iph->protocol != IPPROTO_ICMP))
1876                 goto out;
1877
1878         if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1879                 goto out;
1880
1881         icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1882                                    &_icmph);
1883         if (!icmph)
1884                 goto out;
1885
1886         if (!icmp_is_err(icmph->type))
1887                 goto out;
1888
1889         inner_iph = skb_header_pointer(skb,
1890                                        outer_iph->ihl * 4 + sizeof(_icmph),
1891                                        sizeof(_inner_iph), &_inner_iph);
1892         if (!inner_iph)
1893                 goto out;
1894
1895         key_iph = inner_iph;
1896 out:
1897         hash_keys->addrs.v4addrs.src = key_iph->saddr;
1898         hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1899 }
1900
1901 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1902                                            const struct sk_buff *skb,
1903                                            bool *p_has_inner)
1904 {
1905         u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1906         struct flow_keys keys, hash_keys;
1907
1908         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1909                 return 0;
1910
1911         memset(&hash_keys, 0, sizeof(hash_keys));
1912         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1913
1914         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1915         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1916                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1917         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1918                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1919         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1920                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1921         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1922                 hash_keys.ports.src = keys.ports.src;
1923         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1924                 hash_keys.ports.dst = keys.ports.dst;
1925
1926         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1927         return flow_hash_from_keys(&hash_keys);
1928 }
1929
1930 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1931                                            const struct sk_buff *skb,
1932                                            bool has_inner)
1933 {
1934         u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1935         struct flow_keys keys, hash_keys;
1936
1937         /* We assume the packet carries an encapsulation, but if none was
1938          * encountered during dissection of the outer flow, then there is no
1939          * point in calling the flow dissector again.
1940          */
1941         if (!has_inner)
1942                 return 0;
1943
1944         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1945                 return 0;
1946
1947         memset(&hash_keys, 0, sizeof(hash_keys));
1948         skb_flow_dissect_flow_keys(skb, &keys, 0);
1949
1950         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1951                 return 0;
1952
1953         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1954                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1955                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1956                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1957                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1958                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1959         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1960                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1961                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1962                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1963                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1964                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1965                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1966                         hash_keys.tags.flow_label = keys.tags.flow_label;
1967         }
1968
1969         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1970                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1971         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1972                 hash_keys.ports.src = keys.ports.src;
1973         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1974                 hash_keys.ports.dst = keys.ports.dst;
1975
1976         return flow_hash_from_keys(&hash_keys);
1977 }
1978
1979 static u32 fib_multipath_custom_hash_skb(const struct net *net,
1980                                          const struct sk_buff *skb)
1981 {
1982         u32 mhash, mhash_inner;
1983         bool has_inner = true;
1984
1985         mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
1986         mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
1987
1988         return jhash_2words(mhash, mhash_inner, 0);
1989 }
1990
1991 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
1992                                          const struct flowi4 *fl4)
1993 {
1994         u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1995         struct flow_keys hash_keys;
1996
1997         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1998                 return 0;
1999
2000         memset(&hash_keys, 0, sizeof(hash_keys));
2001         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2002         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2003                 hash_keys.addrs.v4addrs.src = fl4->saddr;
2004         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2005                 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2006         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2007                 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2008         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2009                 hash_keys.ports.src = fl4->fl4_sport;
2010         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2011                 hash_keys.ports.dst = fl4->fl4_dport;
2012
2013         return flow_hash_from_keys(&hash_keys);
2014 }
2015
2016 /* if skb is set it will be used and fl4 can be NULL */
2017 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2018                        const struct sk_buff *skb, struct flow_keys *flkeys)
2019 {
2020         u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2021         struct flow_keys hash_keys;
2022         u32 mhash = 0;
2023
2024         switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2025         case 0:
2026                 memset(&hash_keys, 0, sizeof(hash_keys));
2027                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2028                 if (skb) {
2029                         ip_multipath_l3_keys(skb, &hash_keys);
2030                 } else {
2031                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2032                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2033                 }
2034                 mhash = flow_hash_from_keys(&hash_keys);
2035                 break;
2036         case 1:
2037                 /* skb is currently provided only when forwarding */
2038                 if (skb) {
2039                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2040                         struct flow_keys keys;
2041
2042                         /* short-circuit if we already have L4 hash present */
2043                         if (skb->l4_hash)
2044                                 return skb_get_hash_raw(skb) >> 1;
2045
2046                         memset(&hash_keys, 0, sizeof(hash_keys));
2047
2048                         if (!flkeys) {
2049                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2050                                 flkeys = &keys;
2051                         }
2052
2053                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2054                         hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2055                         hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2056                         hash_keys.ports.src = flkeys->ports.src;
2057                         hash_keys.ports.dst = flkeys->ports.dst;
2058                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2059                 } else {
2060                         memset(&hash_keys, 0, sizeof(hash_keys));
2061                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2062                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2063                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2064                         hash_keys.ports.src = fl4->fl4_sport;
2065                         hash_keys.ports.dst = fl4->fl4_dport;
2066                         hash_keys.basic.ip_proto = fl4->flowi4_proto;
2067                 }
2068                 mhash = flow_hash_from_keys(&hash_keys);
2069                 break;
2070         case 2:
2071                 memset(&hash_keys, 0, sizeof(hash_keys));
2072                 /* skb is currently provided only when forwarding */
2073                 if (skb) {
2074                         struct flow_keys keys;
2075
2076                         skb_flow_dissect_flow_keys(skb, &keys, 0);
2077                         /* Inner can be v4 or v6 */
2078                         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2079                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2080                                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2081                                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2082                         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2083                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2084                                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2085                                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2086                                 hash_keys.tags.flow_label = keys.tags.flow_label;
2087                                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2088                         } else {
2089                                 /* Same as case 0 */
2090                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2091                                 ip_multipath_l3_keys(skb, &hash_keys);
2092                         }
2093                 } else {
2094                         /* Same as case 0 */
2095                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2096                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2097                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2098                 }
2099                 mhash = flow_hash_from_keys(&hash_keys);
2100                 break;
2101         case 3:
2102                 if (skb)
2103                         mhash = fib_multipath_custom_hash_skb(net, skb);
2104                 else
2105                         mhash = fib_multipath_custom_hash_fl4(net, fl4);
2106                 break;
2107         }
2108
2109         if (multipath_hash)
2110                 mhash = jhash_2words(mhash, multipath_hash, 0);
2111
2112         return mhash >> 1;
2113 }
2114 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2115
2116 static int ip_mkroute_input(struct sk_buff *skb,
2117                             struct fib_result *res,
2118                             struct in_device *in_dev,
2119                             __be32 daddr, __be32 saddr, u32 tos,
2120                             struct flow_keys *hkeys)
2121 {
2122 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2123         if (res->fi && fib_info_num_path(res->fi) > 1) {
2124                 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2125
2126                 fib_select_multipath(res, h);
2127                 IPCB(skb)->flags |= IPSKB_MULTIPATH;
2128         }
2129 #endif
2130
2131         /* create a routing cache entry */
2132         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2133 }
2134
2135 /* Implements all the saddr-related checks as ip_route_input_slow(),
2136  * assuming daddr is valid and the destination is not a local broadcast one.
2137  * Uses the provided hint instead of performing a route lookup.
2138  */
2139 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2140                       u8 tos, struct net_device *dev,
2141                       const struct sk_buff *hint)
2142 {
2143         struct in_device *in_dev = __in_dev_get_rcu(dev);
2144         struct rtable *rt = skb_rtable(hint);
2145         struct net *net = dev_net(dev);
2146         int err = -EINVAL;
2147         u32 tag = 0;
2148
2149         if (!in_dev)
2150                 return -EINVAL;
2151
2152         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2153                 goto martian_source;
2154
2155         if (ipv4_is_zeronet(saddr))
2156                 goto martian_source;
2157
2158         if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2159                 goto martian_source;
2160
2161         if (rt->rt_type != RTN_LOCAL)
2162                 goto skip_validate_source;
2163
2164         tos &= IPTOS_RT_MASK;
2165         err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2166         if (err < 0)
2167                 goto martian_source;
2168
2169 skip_validate_source:
2170         skb_dst_copy(skb, hint);
2171         return 0;
2172
2173 martian_source:
2174         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2175         return err;
2176 }
2177
2178 /* get device for dst_alloc with local routes */
2179 static struct net_device *ip_rt_get_dev(struct net *net,
2180                                         const struct fib_result *res)
2181 {
2182         struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2183         struct net_device *dev = NULL;
2184
2185         if (nhc)
2186                 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2187
2188         return dev ? : net->loopback_dev;
2189 }
2190
2191 /*
2192  *      NOTE. We drop all the packets that has local source
2193  *      addresses, because every properly looped back packet
2194  *      must have correct destination already attached by output routine.
2195  *      Changes in the enforced policies must be applied also to
2196  *      ip_route_use_hint().
2197  *
2198  *      Such approach solves two big problems:
2199  *      1. Not simplex devices are handled properly.
2200  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2201  *      called with rcu_read_lock()
2202  */
2203
2204 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2205                                u8 tos, struct net_device *dev,
2206                                struct fib_result *res)
2207 {
2208         struct in_device *in_dev = __in_dev_get_rcu(dev);
2209         struct flow_keys *flkeys = NULL, _flkeys;
2210         struct net    *net = dev_net(dev);
2211         struct ip_tunnel_info *tun_info;
2212         int             err = -EINVAL;
2213         unsigned int    flags = 0;
2214         u32             itag = 0;
2215         struct rtable   *rth;
2216         struct flowi4   fl4;
2217         bool do_cache = true;
2218
2219         /* IP on this device is disabled. */
2220
2221         if (!in_dev)
2222                 goto out;
2223
2224         /* Check for the most weird martians, which can be not detected
2225          * by fib_lookup.
2226          */
2227
2228         tun_info = skb_tunnel_info(skb);
2229         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2230                 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2231         else
2232                 fl4.flowi4_tun_key.tun_id = 0;
2233         skb_dst_drop(skb);
2234
2235         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2236                 goto martian_source;
2237
2238         res->fi = NULL;
2239         res->table = NULL;
2240         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2241                 goto brd_input;
2242
2243         /* Accept zero addresses only to limited broadcast;
2244          * I even do not know to fix it or not. Waiting for complains :-)
2245          */
2246         if (ipv4_is_zeronet(saddr))
2247                 goto martian_source;
2248
2249         if (ipv4_is_zeronet(daddr))
2250                 goto martian_destination;
2251
2252         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2253          * and call it once if daddr or/and saddr are loopback addresses
2254          */
2255         if (ipv4_is_loopback(daddr)) {
2256                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2257                         goto martian_destination;
2258         } else if (ipv4_is_loopback(saddr)) {
2259                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2260                         goto martian_source;
2261         }
2262
2263         /*
2264          *      Now we are ready to route packet.
2265          */
2266         fl4.flowi4_l3mdev = 0;
2267         fl4.flowi4_oif = 0;
2268         fl4.flowi4_iif = dev->ifindex;
2269         fl4.flowi4_mark = skb->mark;
2270         fl4.flowi4_tos = tos;
2271         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2272         fl4.flowi4_flags = 0;
2273         fl4.daddr = daddr;
2274         fl4.saddr = saddr;
2275         fl4.flowi4_uid = sock_net_uid(net, NULL);
2276         fl4.flowi4_multipath_hash = 0;
2277
2278         if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2279                 flkeys = &_flkeys;
2280         } else {
2281                 fl4.flowi4_proto = 0;
2282                 fl4.fl4_sport = 0;
2283                 fl4.fl4_dport = 0;
2284         }
2285
2286         err = fib_lookup(net, &fl4, res, 0);
2287         if (err != 0) {
2288                 if (!IN_DEV_FORWARD(in_dev))
2289                         err = -EHOSTUNREACH;
2290                 goto no_route;
2291         }
2292
2293         if (res->type == RTN_BROADCAST) {
2294                 if (IN_DEV_BFORWARD(in_dev))
2295                         goto make_route;
2296                 /* not do cache if bc_forwarding is enabled */
2297                 if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2298                         do_cache = false;
2299                 goto brd_input;
2300         }
2301
2302         if (res->type == RTN_LOCAL) {
2303                 err = fib_validate_source(skb, saddr, daddr, tos,
2304                                           0, dev, in_dev, &itag);
2305                 if (err < 0)
2306                         goto martian_source;
2307                 goto local_input;
2308         }
2309
2310         if (!IN_DEV_FORWARD(in_dev)) {
2311                 err = -EHOSTUNREACH;
2312                 goto no_route;
2313         }
2314         if (res->type != RTN_UNICAST)
2315                 goto martian_destination;
2316
2317 make_route:
2318         err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2319 out:    return err;
2320
2321 brd_input:
2322         if (skb->protocol != htons(ETH_P_IP))
2323                 goto e_inval;
2324
2325         if (!ipv4_is_zeronet(saddr)) {
2326                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2327                                           in_dev, &itag);
2328                 if (err < 0)
2329                         goto martian_source;
2330         }
2331         flags |= RTCF_BROADCAST;
2332         res->type = RTN_BROADCAST;
2333         RT_CACHE_STAT_INC(in_brd);
2334
2335 local_input:
2336         if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2337                 IPCB(skb)->flags |= IPSKB_NOPOLICY;
2338
2339         do_cache &= res->fi && !itag;
2340         if (do_cache) {
2341                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2342
2343                 rth = rcu_dereference(nhc->nhc_rth_input);
2344                 if (rt_cache_valid(rth)) {
2345                         skb_dst_set_noref(skb, &rth->dst);
2346                         err = 0;
2347                         goto out;
2348                 }
2349         }
2350
2351         rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2352                            flags | RTCF_LOCAL, res->type, false);
2353         if (!rth)
2354                 goto e_nobufs;
2355
2356         rth->dst.output= ip_rt_bug;
2357 #ifdef CONFIG_IP_ROUTE_CLASSID
2358         rth->dst.tclassid = itag;
2359 #endif
2360         rth->rt_is_input = 1;
2361
2362         RT_CACHE_STAT_INC(in_slow_tot);
2363         if (res->type == RTN_UNREACHABLE) {
2364                 rth->dst.input= ip_error;
2365                 rth->dst.error= -err;
2366                 rth->rt_flags   &= ~RTCF_LOCAL;
2367         }
2368
2369         if (do_cache) {
2370                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2371
2372                 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2373                 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2374                         WARN_ON(rth->dst.input == lwtunnel_input);
2375                         rth->dst.lwtstate->orig_input = rth->dst.input;
2376                         rth->dst.input = lwtunnel_input;
2377                 }
2378
2379                 if (unlikely(!rt_cache_route(nhc, rth)))
2380                         rt_add_uncached_list(rth);
2381         }
2382         skb_dst_set(skb, &rth->dst);
2383         err = 0;
2384         goto out;
2385
2386 no_route:
2387         RT_CACHE_STAT_INC(in_no_route);
2388         res->type = RTN_UNREACHABLE;
2389         res->fi = NULL;
2390         res->table = NULL;
2391         goto local_input;
2392
2393         /*
2394          *      Do not cache martian addresses: they should be logged (RFC1812)
2395          */
2396 martian_destination:
2397         RT_CACHE_STAT_INC(in_martian_dst);
2398 #ifdef CONFIG_IP_ROUTE_VERBOSE
2399         if (IN_DEV_LOG_MARTIANS(in_dev))
2400                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2401                                      &daddr, &saddr, dev->name);
2402 #endif
2403
2404 e_inval:
2405         err = -EINVAL;
2406         goto out;
2407
2408 e_nobufs:
2409         err = -ENOBUFS;
2410         goto out;
2411
2412 martian_source:
2413         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2414         goto out;
2415 }
2416
2417 /* called with rcu_read_lock held */
2418 static int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2419                               u8 tos, struct net_device *dev, struct fib_result *res)
2420 {
2421         /* Multicast recognition logic is moved from route cache to here.
2422          * The problem was that too many Ethernet cards have broken/missing
2423          * hardware multicast filters :-( As result the host on multicasting
2424          * network acquires a lot of useless route cache entries, sort of
2425          * SDR messages from all the world. Now we try to get rid of them.
2426          * Really, provided software IP multicast filter is organized
2427          * reasonably (at least, hashed), it does not result in a slowdown
2428          * comparing with route cache reject entries.
2429          * Note, that multicast routers are not affected, because
2430          * route cache entry is created eventually.
2431          */
2432         if (ipv4_is_multicast(daddr)) {
2433                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2434                 int our = 0;
2435                 int err = -EINVAL;
2436
2437                 if (!in_dev)
2438                         return err;
2439                 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2440                                       ip_hdr(skb)->protocol);
2441
2442                 /* check l3 master if no match yet */
2443                 if (!our && netif_is_l3_slave(dev)) {
2444                         struct in_device *l3_in_dev;
2445
2446                         l3_in_dev = __in_dev_get_rcu(skb->dev);
2447                         if (l3_in_dev)
2448                                 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2449                                                       ip_hdr(skb)->protocol);
2450                 }
2451
2452                 if (our
2453 #ifdef CONFIG_IP_MROUTE
2454                         ||
2455                     (!ipv4_is_local_multicast(daddr) &&
2456                      IN_DEV_MFORWARD(in_dev))
2457 #endif
2458                    ) {
2459                         err = ip_route_input_mc(skb, daddr, saddr,
2460                                                 tos, dev, our);
2461                 }
2462                 return err;
2463         }
2464
2465         return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2466 }
2467
2468 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2469                          u8 tos, struct net_device *dev)
2470 {
2471         struct fib_result res;
2472         int err;
2473
2474         tos &= IPTOS_RT_MASK;
2475         rcu_read_lock();
2476         err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2477         rcu_read_unlock();
2478
2479         return err;
2480 }
2481 EXPORT_SYMBOL(ip_route_input_noref);
2482
2483 /* called with rcu_read_lock() */
2484 static struct rtable *__mkroute_output(const struct fib_result *res,
2485                                        const struct flowi4 *fl4, int orig_oif,
2486                                        struct net_device *dev_out,
2487                                        unsigned int flags)
2488 {
2489         struct fib_info *fi = res->fi;
2490         struct fib_nh_exception *fnhe;
2491         struct in_device *in_dev;
2492         u16 type = res->type;
2493         struct rtable *rth;
2494         bool do_cache;
2495
2496         in_dev = __in_dev_get_rcu(dev_out);
2497         if (!in_dev)
2498                 return ERR_PTR(-EINVAL);
2499
2500         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2501                 if (ipv4_is_loopback(fl4->saddr) &&
2502                     !(dev_out->flags & IFF_LOOPBACK) &&
2503                     !netif_is_l3_master(dev_out))
2504                         return ERR_PTR(-EINVAL);
2505
2506         if (ipv4_is_lbcast(fl4->daddr))
2507                 type = RTN_BROADCAST;
2508         else if (ipv4_is_multicast(fl4->daddr))
2509                 type = RTN_MULTICAST;
2510         else if (ipv4_is_zeronet(fl4->daddr))
2511                 return ERR_PTR(-EINVAL);
2512
2513         if (dev_out->flags & IFF_LOOPBACK)
2514                 flags |= RTCF_LOCAL;
2515
2516         do_cache = true;
2517         if (type == RTN_BROADCAST) {
2518                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2519                 fi = NULL;
2520         } else if (type == RTN_MULTICAST) {
2521                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2522                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2523                                      fl4->flowi4_proto))
2524                         flags &= ~RTCF_LOCAL;
2525                 else
2526                         do_cache = false;
2527                 /* If multicast route do not exist use
2528                  * default one, but do not gateway in this case.
2529                  * Yes, it is hack.
2530                  */
2531                 if (fi && res->prefixlen < 4)
2532                         fi = NULL;
2533         } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2534                    (orig_oif != dev_out->ifindex)) {
2535                 /* For local routes that require a particular output interface
2536                  * we do not want to cache the result.  Caching the result
2537                  * causes incorrect behaviour when there are multiple source
2538                  * addresses on the interface, the end result being that if the
2539                  * intended recipient is waiting on that interface for the
2540                  * packet he won't receive it because it will be delivered on
2541                  * the loopback interface and the IP_PKTINFO ipi_ifindex will
2542                  * be set to the loopback interface as well.
2543                  */
2544                 do_cache = false;
2545         }
2546
2547         fnhe = NULL;
2548         do_cache &= fi != NULL;
2549         if (fi) {
2550                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2551                 struct rtable __rcu **prth;
2552
2553                 fnhe = find_exception(nhc, fl4->daddr);
2554                 if (!do_cache)
2555                         goto add;
2556                 if (fnhe) {
2557                         prth = &fnhe->fnhe_rth_output;
2558                 } else {
2559                         if (unlikely(fl4->flowi4_flags &
2560                                      FLOWI_FLAG_KNOWN_NH &&
2561                                      !(nhc->nhc_gw_family &&
2562                                        nhc->nhc_scope == RT_SCOPE_LINK))) {
2563                                 do_cache = false;
2564                                 goto add;
2565                         }
2566                         prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2567                 }
2568                 rth = rcu_dereference(*prth);
2569                 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2570                         return rth;
2571         }
2572
2573 add:
2574         rth = rt_dst_alloc(dev_out, flags, type,
2575                            IN_DEV_ORCONF(in_dev, NOXFRM));
2576         if (!rth)
2577                 return ERR_PTR(-ENOBUFS);
2578
2579         rth->rt_iif = orig_oif;
2580
2581         RT_CACHE_STAT_INC(out_slow_tot);
2582
2583         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2584                 if (flags & RTCF_LOCAL &&
2585                     !(dev_out->flags & IFF_LOOPBACK)) {
2586                         rth->dst.output = ip_mc_output;
2587                         RT_CACHE_STAT_INC(out_slow_mc);
2588                 }
2589 #ifdef CONFIG_IP_MROUTE
2590                 if (type == RTN_MULTICAST) {
2591                         if (IN_DEV_MFORWARD(in_dev) &&
2592                             !ipv4_is_local_multicast(fl4->daddr)) {
2593                                 rth->dst.input = ip_mr_input;
2594                                 rth->dst.output = ip_mc_output;
2595                         }
2596                 }
2597 #endif
2598         }
2599
2600         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2601         lwtunnel_set_redirect(&rth->dst);
2602
2603         return rth;
2604 }
2605
2606 /*
2607  * Major route resolver routine.
2608  */
2609
2610 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2611                                         const struct sk_buff *skb)
2612 {
2613         struct fib_result res = {
2614                 .type           = RTN_UNSPEC,
2615                 .fi             = NULL,
2616                 .table          = NULL,
2617                 .tclassid       = 0,
2618         };
2619         struct rtable *rth;
2620
2621         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2622         fl4->flowi4_tos &= IPTOS_RT_MASK;
2623
2624         rcu_read_lock();
2625         rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2626         rcu_read_unlock();
2627
2628         return rth;
2629 }
2630 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2631
2632 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2633                                             struct fib_result *res,
2634                                             const struct sk_buff *skb)
2635 {
2636         struct net_device *dev_out = NULL;
2637         int orig_oif = fl4->flowi4_oif;
2638         unsigned int flags = 0;
2639         struct rtable *rth;
2640         int err;
2641
2642         if (fl4->saddr) {
2643                 if (ipv4_is_multicast(fl4->saddr) ||
2644                     ipv4_is_lbcast(fl4->saddr) ||
2645                     ipv4_is_zeronet(fl4->saddr)) {
2646                         rth = ERR_PTR(-EINVAL);
2647                         goto out;
2648                 }
2649
2650                 rth = ERR_PTR(-ENETUNREACH);
2651
2652                 /* I removed check for oif == dev_out->oif here.
2653                  * It was wrong for two reasons:
2654                  * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2655                  *    is assigned to multiple interfaces.
2656                  * 2. Moreover, we are allowed to send packets with saddr
2657                  *    of another iface. --ANK
2658                  */
2659
2660                 if (fl4->flowi4_oif == 0 &&
2661                     (ipv4_is_multicast(fl4->daddr) ||
2662                      ipv4_is_lbcast(fl4->daddr))) {
2663                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2664                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2665                         if (!dev_out)
2666                                 goto out;
2667
2668                         /* Special hack: user can direct multicasts
2669                          * and limited broadcast via necessary interface
2670                          * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2671                          * This hack is not just for fun, it allows
2672                          * vic,vat and friends to work.
2673                          * They bind socket to loopback, set ttl to zero
2674                          * and expect that it will work.
2675                          * From the viewpoint of routing cache they are broken,
2676                          * because we are not allowed to build multicast path
2677                          * with loopback source addr (look, routing cache
2678                          * cannot know, that ttl is zero, so that packet
2679                          * will not leave this host and route is valid).
2680                          * Luckily, this hack is good workaround.
2681                          */
2682
2683                         fl4->flowi4_oif = dev_out->ifindex;
2684                         goto make_route;
2685                 }
2686
2687                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2688                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2689                         if (!__ip_dev_find(net, fl4->saddr, false))
2690                                 goto out;
2691                 }
2692         }
2693
2694
2695         if (fl4->flowi4_oif) {
2696                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2697                 rth = ERR_PTR(-ENODEV);
2698                 if (!dev_out)
2699                         goto out;
2700
2701                 /* RACE: Check return value of inet_select_addr instead. */
2702                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2703                         rth = ERR_PTR(-ENETUNREACH);
2704                         goto out;
2705                 }
2706                 if (ipv4_is_local_multicast(fl4->daddr) ||
2707                     ipv4_is_lbcast(fl4->daddr) ||
2708                     fl4->flowi4_proto == IPPROTO_IGMP) {
2709                         if (!fl4->saddr)
2710                                 fl4->saddr = inet_select_addr(dev_out, 0,
2711                                                               RT_SCOPE_LINK);
2712                         goto make_route;
2713                 }
2714                 if (!fl4->saddr) {
2715                         if (ipv4_is_multicast(fl4->daddr))
2716                                 fl4->saddr = inet_select_addr(dev_out, 0,
2717                                                               fl4->flowi4_scope);
2718                         else if (!fl4->daddr)
2719                                 fl4->saddr = inet_select_addr(dev_out, 0,
2720                                                               RT_SCOPE_HOST);
2721                 }
2722         }
2723
2724         if (!fl4->daddr) {
2725                 fl4->daddr = fl4->saddr;
2726                 if (!fl4->daddr)
2727                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2728                 dev_out = net->loopback_dev;
2729                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2730                 res->type = RTN_LOCAL;
2731                 flags |= RTCF_LOCAL;
2732                 goto make_route;
2733         }
2734
2735         err = fib_lookup(net, fl4, res, 0);
2736         if (err) {
2737                 res->fi = NULL;
2738                 res->table = NULL;
2739                 if (fl4->flowi4_oif &&
2740                     (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2741                         /* Apparently, routing tables are wrong. Assume,
2742                          * that the destination is on link.
2743                          *
2744                          * WHY? DW.
2745                          * Because we are allowed to send to iface
2746                          * even if it has NO routes and NO assigned
2747                          * addresses. When oif is specified, routing
2748                          * tables are looked up with only one purpose:
2749                          * to catch if destination is gatewayed, rather than
2750                          * direct. Moreover, if MSG_DONTROUTE is set,
2751                          * we send packet, ignoring both routing tables
2752                          * and ifaddr state. --ANK
2753                          *
2754                          *
2755                          * We could make it even if oif is unknown,
2756                          * likely IPv6, but we do not.
2757                          */
2758
2759                         if (fl4->saddr == 0)
2760                                 fl4->saddr = inet_select_addr(dev_out, 0,
2761                                                               RT_SCOPE_LINK);
2762                         res->type = RTN_UNICAST;
2763                         goto make_route;
2764                 }
2765                 rth = ERR_PTR(err);
2766                 goto out;
2767         }
2768
2769         if (res->type == RTN_LOCAL) {
2770                 if (!fl4->saddr) {
2771                         if (res->fi->fib_prefsrc)
2772                                 fl4->saddr = res->fi->fib_prefsrc;
2773                         else
2774                                 fl4->saddr = fl4->daddr;
2775                 }
2776
2777                 /* L3 master device is the loopback for that domain */
2778                 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2779                         net->loopback_dev;
2780
2781                 /* make sure orig_oif points to fib result device even
2782                  * though packet rx/tx happens over loopback or l3mdev
2783                  */
2784                 orig_oif = FIB_RES_OIF(*res);
2785
2786                 fl4->flowi4_oif = dev_out->ifindex;
2787                 flags |= RTCF_LOCAL;
2788                 goto make_route;
2789         }
2790
2791         fib_select_path(net, res, fl4, skb);
2792
2793         dev_out = FIB_RES_DEV(*res);
2794
2795 make_route:
2796         rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2797
2798 out:
2799         return rth;
2800 }
2801
2802 static struct dst_ops ipv4_dst_blackhole_ops = {
2803         .family                 = AF_INET,
2804         .default_advmss         = ipv4_default_advmss,
2805         .neigh_lookup           = ipv4_neigh_lookup,
2806         .check                  = dst_blackhole_check,
2807         .cow_metrics            = dst_blackhole_cow_metrics,
2808         .update_pmtu            = dst_blackhole_update_pmtu,
2809         .redirect               = dst_blackhole_redirect,
2810         .mtu                    = dst_blackhole_mtu,
2811 };
2812
2813 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2814 {
2815         struct rtable *ort = dst_rtable(dst_orig);
2816         struct rtable *rt;
2817
2818         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2819         if (rt) {
2820                 struct dst_entry *new = &rt->dst;
2821
2822                 new->__use = 1;
2823                 new->input = dst_discard;
2824                 new->output = dst_discard_out;
2825
2826                 new->dev = net->loopback_dev;
2827                 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2828
2829                 rt->rt_is_input = ort->rt_is_input;
2830                 rt->rt_iif = ort->rt_iif;
2831                 rt->rt_pmtu = ort->rt_pmtu;
2832                 rt->rt_mtu_locked = ort->rt_mtu_locked;
2833
2834                 rt->rt_genid = rt_genid_ipv4(net);
2835                 rt->rt_flags = ort->rt_flags;
2836                 rt->rt_type = ort->rt_type;
2837                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2838                 rt->rt_gw_family = ort->rt_gw_family;
2839                 if (rt->rt_gw_family == AF_INET)
2840                         rt->rt_gw4 = ort->rt_gw4;
2841                 else if (rt->rt_gw_family == AF_INET6)
2842                         rt->rt_gw6 = ort->rt_gw6;
2843         }
2844
2845         dst_release(dst_orig);
2846
2847         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2848 }
2849
2850 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2851                                     const struct sock *sk)
2852 {
2853         struct rtable *rt = __ip_route_output_key(net, flp4);
2854
2855         if (IS_ERR(rt))
2856                 return rt;
2857
2858         if (flp4->flowi4_proto) {
2859                 flp4->flowi4_oif = rt->dst.dev->ifindex;
2860                 rt = dst_rtable(xfrm_lookup_route(net, &rt->dst,
2861                                                   flowi4_to_flowi(flp4),
2862                                                   sk, 0));
2863         }
2864
2865         return rt;
2866 }
2867 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2868
2869 /* called with rcu_read_lock held */
2870 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2871                         struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2872                         struct sk_buff *skb, u32 portid, u32 seq,
2873                         unsigned int flags)
2874 {
2875         struct rtmsg *r;
2876         struct nlmsghdr *nlh;
2877         unsigned long expires = 0;
2878         u32 error;
2879         u32 metrics[RTAX_MAX];
2880
2881         nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2882         if (!nlh)
2883                 return -EMSGSIZE;
2884
2885         r = nlmsg_data(nlh);
2886         r->rtm_family    = AF_INET;
2887         r->rtm_dst_len  = 32;
2888         r->rtm_src_len  = 0;
2889         r->rtm_tos      = fl4 ? fl4->flowi4_tos : 0;
2890         r->rtm_table    = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2891         if (nla_put_u32(skb, RTA_TABLE, table_id))
2892                 goto nla_put_failure;
2893         r->rtm_type     = rt->rt_type;
2894         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2895         r->rtm_protocol = RTPROT_UNSPEC;
2896         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2897         if (rt->rt_flags & RTCF_NOTIFY)
2898                 r->rtm_flags |= RTM_F_NOTIFY;
2899         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2900                 r->rtm_flags |= RTCF_DOREDIRECT;
2901
2902         if (nla_put_in_addr(skb, RTA_DST, dst))
2903                 goto nla_put_failure;
2904         if (src) {
2905                 r->rtm_src_len = 32;
2906                 if (nla_put_in_addr(skb, RTA_SRC, src))
2907                         goto nla_put_failure;
2908         }
2909         if (rt->dst.dev &&
2910             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2911                 goto nla_put_failure;
2912         if (rt->dst.lwtstate &&
2913             lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2914                 goto nla_put_failure;
2915 #ifdef CONFIG_IP_ROUTE_CLASSID
2916         if (rt->dst.tclassid &&
2917             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2918                 goto nla_put_failure;
2919 #endif
2920         if (fl4 && !rt_is_input_route(rt) &&
2921             fl4->saddr != src) {
2922                 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2923                         goto nla_put_failure;
2924         }
2925         if (rt->rt_uses_gateway) {
2926                 if (rt->rt_gw_family == AF_INET &&
2927                     nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2928                         goto nla_put_failure;
2929                 } else if (rt->rt_gw_family == AF_INET6) {
2930                         int alen = sizeof(struct in6_addr);
2931                         struct nlattr *nla;
2932                         struct rtvia *via;
2933
2934                         nla = nla_reserve(skb, RTA_VIA, alen + 2);
2935                         if (!nla)
2936                                 goto nla_put_failure;
2937
2938                         via = nla_data(nla);
2939                         via->rtvia_family = AF_INET6;
2940                         memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2941                 }
2942         }
2943
2944         expires = rt->dst.expires;
2945         if (expires) {
2946                 unsigned long now = jiffies;
2947
2948                 if (time_before(now, expires))
2949                         expires -= now;
2950                 else
2951                         expires = 0;
2952         }
2953
2954         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2955         if (rt->rt_pmtu && expires)
2956                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2957         if (rt->rt_mtu_locked && expires)
2958                 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2959         if (rtnetlink_put_metrics(skb, metrics) < 0)
2960                 goto nla_put_failure;
2961
2962         if (fl4) {
2963                 if (fl4->flowi4_mark &&
2964                     nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2965                         goto nla_put_failure;
2966
2967                 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2968                     nla_put_u32(skb, RTA_UID,
2969                                 from_kuid_munged(current_user_ns(),
2970                                                  fl4->flowi4_uid)))
2971                         goto nla_put_failure;
2972
2973                 if (rt_is_input_route(rt)) {
2974 #ifdef CONFIG_IP_MROUTE
2975                         if (ipv4_is_multicast(dst) &&
2976                             !ipv4_is_local_multicast(dst) &&
2977                             IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
2978                                 int err = ipmr_get_route(net, skb,
2979                                                          fl4->saddr, fl4->daddr,
2980                                                          r, portid);
2981
2982                                 if (err <= 0) {
2983                                         if (err == 0)
2984                                                 return 0;
2985                                         goto nla_put_failure;
2986                                 }
2987                         } else
2988 #endif
2989                                 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
2990                                         goto nla_put_failure;
2991                 }
2992         }
2993
2994         error = rt->dst.error;
2995
2996         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2997                 goto nla_put_failure;
2998
2999         nlmsg_end(skb, nlh);
3000         return 0;
3001
3002 nla_put_failure:
3003         nlmsg_cancel(skb, nlh);
3004         return -EMSGSIZE;
3005 }
3006
3007 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3008                             struct netlink_callback *cb, u32 table_id,
3009                             struct fnhe_hash_bucket *bucket, int genid,
3010                             int *fa_index, int fa_start, unsigned int flags)
3011 {
3012         int i;
3013
3014         for (i = 0; i < FNHE_HASH_SIZE; i++) {
3015                 struct fib_nh_exception *fnhe;
3016
3017                 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3018                      fnhe = rcu_dereference(fnhe->fnhe_next)) {
3019                         struct rtable *rt;
3020                         int err;
3021
3022                         if (*fa_index < fa_start)
3023                                 goto next;
3024
3025                         if (fnhe->fnhe_genid != genid)
3026                                 goto next;
3027
3028                         if (fnhe->fnhe_expires &&
3029                             time_after(jiffies, fnhe->fnhe_expires))
3030                                 goto next;
3031
3032                         rt = rcu_dereference(fnhe->fnhe_rth_input);
3033                         if (!rt)
3034                                 rt = rcu_dereference(fnhe->fnhe_rth_output);
3035                         if (!rt)
3036                                 goto next;
3037
3038                         err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3039                                            table_id, NULL, skb,
3040                                            NETLINK_CB(cb->skb).portid,
3041                                            cb->nlh->nlmsg_seq, flags);
3042                         if (err)
3043                                 return err;
3044 next:
3045                         (*fa_index)++;
3046                 }
3047         }
3048
3049         return 0;
3050 }
3051
3052 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3053                        u32 table_id, struct fib_info *fi,
3054                        int *fa_index, int fa_start, unsigned int flags)
3055 {
3056         struct net *net = sock_net(cb->skb->sk);
3057         int nhsel, genid = fnhe_genid(net);
3058
3059         for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3060                 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3061                 struct fnhe_hash_bucket *bucket;
3062                 int err;
3063
3064                 if (nhc->nhc_flags & RTNH_F_DEAD)
3065                         continue;
3066
3067                 rcu_read_lock();
3068                 bucket = rcu_dereference(nhc->nhc_exceptions);
3069                 err = 0;
3070                 if (bucket)
3071                         err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3072                                                genid, fa_index, fa_start,
3073                                                flags);
3074                 rcu_read_unlock();
3075                 if (err)
3076                         return err;
3077         }
3078
3079         return 0;
3080 }
3081
3082 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3083                                                    u8 ip_proto, __be16 sport,
3084                                                    __be16 dport)
3085 {
3086         struct sk_buff *skb;
3087         struct iphdr *iph;
3088
3089         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3090         if (!skb)
3091                 return NULL;
3092
3093         /* Reserve room for dummy headers, this skb can pass
3094          * through good chunk of routing engine.
3095          */
3096         skb_reset_mac_header(skb);
3097         skb_reset_network_header(skb);
3098         skb->protocol = htons(ETH_P_IP);
3099         iph = skb_put(skb, sizeof(struct iphdr));
3100         iph->protocol = ip_proto;
3101         iph->saddr = src;
3102         iph->daddr = dst;
3103         iph->version = 0x4;
3104         iph->frag_off = 0;
3105         iph->ihl = 0x5;
3106         skb_set_transport_header(skb, skb->len);
3107
3108         switch (iph->protocol) {
3109         case IPPROTO_UDP: {
3110                 struct udphdr *udph;
3111
3112                 udph = skb_put_zero(skb, sizeof(struct udphdr));
3113                 udph->source = sport;
3114                 udph->dest = dport;
3115                 udph->len = htons(sizeof(struct udphdr));
3116                 udph->check = 0;
3117                 break;
3118         }
3119         case IPPROTO_TCP: {
3120                 struct tcphdr *tcph;
3121
3122                 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3123                 tcph->source    = sport;
3124                 tcph->dest      = dport;
3125                 tcph->doff      = sizeof(struct tcphdr) / 4;
3126                 tcph->rst = 1;
3127                 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3128                                             src, dst, 0);
3129                 break;
3130         }
3131         case IPPROTO_ICMP: {
3132                 struct icmphdr *icmph;
3133
3134                 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3135                 icmph->type = ICMP_ECHO;
3136                 icmph->code = 0;
3137         }
3138         }
3139
3140         return skb;
3141 }
3142
3143 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3144                                        const struct nlmsghdr *nlh,
3145                                        struct nlattr **tb,
3146                                        struct netlink_ext_ack *extack)
3147 {
3148         struct rtmsg *rtm;
3149         int i, err;
3150
3151         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3152                 NL_SET_ERR_MSG(extack,
3153                                "ipv4: Invalid header for route get request");
3154                 return -EINVAL;
3155         }
3156
3157         if (!netlink_strict_get_check(skb))
3158                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3159                                               rtm_ipv4_policy, extack);
3160
3161         rtm = nlmsg_data(nlh);
3162         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3163             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3164             rtm->rtm_table || rtm->rtm_protocol ||
3165             rtm->rtm_scope || rtm->rtm_type) {
3166                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3167                 return -EINVAL;
3168         }
3169
3170         if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3171                                RTM_F_LOOKUP_TABLE |
3172                                RTM_F_FIB_MATCH)) {
3173                 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3174                 return -EINVAL;
3175         }
3176
3177         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3178                                             rtm_ipv4_policy, extack);
3179         if (err)
3180                 return err;
3181
3182         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3183             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3184                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3185                 return -EINVAL;
3186         }
3187
3188         for (i = 0; i <= RTA_MAX; i++) {
3189                 if (!tb[i])
3190                         continue;
3191
3192                 switch (i) {
3193                 case RTA_IIF:
3194                 case RTA_OIF:
3195                 case RTA_SRC:
3196                 case RTA_DST:
3197                 case RTA_IP_PROTO:
3198                 case RTA_SPORT:
3199                 case RTA_DPORT:
3200                 case RTA_MARK:
3201                 case RTA_UID:
3202                         break;
3203                 default:
3204                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3205                         return -EINVAL;
3206                 }
3207         }
3208
3209         return 0;
3210 }
3211
3212 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3213                              struct netlink_ext_ack *extack)
3214 {
3215         struct net *net = sock_net(in_skb->sk);
3216         struct nlattr *tb[RTA_MAX+1];
3217         u32 table_id = RT_TABLE_MAIN;
3218         __be16 sport = 0, dport = 0;
3219         struct fib_result res = {};
3220         u8 ip_proto = IPPROTO_UDP;
3221         struct rtable *rt = NULL;
3222         struct sk_buff *skb;
3223         struct rtmsg *rtm;
3224         struct flowi4 fl4 = {};
3225         __be32 dst = 0;
3226         __be32 src = 0;
3227         kuid_t uid;
3228         u32 iif;
3229         int err;
3230         int mark;
3231
3232         err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3233         if (err < 0)
3234                 return err;
3235
3236         rtm = nlmsg_data(nlh);
3237         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3238         dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3239         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3240         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3241         if (tb[RTA_UID])
3242                 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3243         else
3244                 uid = (iif ? INVALID_UID : current_uid());
3245
3246         if (tb[RTA_IP_PROTO]) {
3247                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3248                                                   &ip_proto, AF_INET, extack);
3249                 if (err)
3250                         return err;
3251         }
3252
3253         if (tb[RTA_SPORT])
3254                 sport = nla_get_be16(tb[RTA_SPORT]);
3255
3256         if (tb[RTA_DPORT])
3257                 dport = nla_get_be16(tb[RTA_DPORT]);
3258
3259         skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3260         if (!skb)
3261                 return -ENOBUFS;
3262
3263         fl4.daddr = dst;
3264         fl4.saddr = src;
3265         fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3266         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3267         fl4.flowi4_mark = mark;
3268         fl4.flowi4_uid = uid;
3269         if (sport)
3270                 fl4.fl4_sport = sport;
3271         if (dport)
3272                 fl4.fl4_dport = dport;
3273         fl4.flowi4_proto = ip_proto;
3274
3275         rcu_read_lock();
3276
3277         if (iif) {
3278                 struct net_device *dev;
3279
3280                 dev = dev_get_by_index_rcu(net, iif);
3281                 if (!dev) {
3282                         err = -ENODEV;
3283                         goto errout_rcu;
3284                 }
3285
3286                 fl4.flowi4_iif = iif; /* for rt_fill_info */
3287                 skb->dev        = dev;
3288                 skb->mark       = mark;
3289                 err = ip_route_input_rcu(skb, dst, src,
3290                                          rtm->rtm_tos & IPTOS_RT_MASK, dev,
3291                                          &res);
3292
3293                 rt = skb_rtable(skb);
3294                 if (err == 0 && rt->dst.error)
3295                         err = -rt->dst.error;
3296         } else {
3297                 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3298                 skb->dev = net->loopback_dev;
3299                 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3300                 err = 0;
3301                 if (IS_ERR(rt))
3302                         err = PTR_ERR(rt);
3303                 else
3304                         skb_dst_set(skb, &rt->dst);
3305         }
3306
3307         if (err)
3308                 goto errout_rcu;
3309
3310         if (rtm->rtm_flags & RTM_F_NOTIFY)
3311                 rt->rt_flags |= RTCF_NOTIFY;
3312
3313         if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3314                 table_id = res.table ? res.table->tb_id : 0;
3315
3316         /* reset skb for netlink reply msg */
3317         skb_trim(skb, 0);
3318         skb_reset_network_header(skb);
3319         skb_reset_transport_header(skb);
3320         skb_reset_mac_header(skb);
3321
3322         if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3323                 struct fib_rt_info fri;
3324
3325                 if (!res.fi) {
3326                         err = fib_props[res.type].error;
3327                         if (!err)
3328                                 err = -EHOSTUNREACH;
3329                         goto errout_rcu;
3330                 }
3331                 fri.fi = res.fi;
3332                 fri.tb_id = table_id;
3333                 fri.dst = res.prefix;
3334                 fri.dst_len = res.prefixlen;
3335                 fri.dscp = inet_dsfield_to_dscp(fl4.flowi4_tos);
3336                 fri.type = rt->rt_type;
3337                 fri.offload = 0;
3338                 fri.trap = 0;
3339                 fri.offload_failed = 0;
3340                 if (res.fa_head) {
3341                         struct fib_alias *fa;
3342
3343                         hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3344                                 u8 slen = 32 - fri.dst_len;
3345
3346                                 if (fa->fa_slen == slen &&
3347                                     fa->tb_id == fri.tb_id &&
3348                                     fa->fa_dscp == fri.dscp &&
3349                                     fa->fa_info == res.fi &&
3350                                     fa->fa_type == fri.type) {
3351                                         fri.offload = READ_ONCE(fa->offload);
3352                                         fri.trap = READ_ONCE(fa->trap);
3353                                         fri.offload_failed =
3354                                                 READ_ONCE(fa->offload_failed);
3355                                         break;
3356                                 }
3357                         }
3358                 }
3359                 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3360                                     nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3361         } else {
3362                 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3363                                    NETLINK_CB(in_skb).portid,
3364                                    nlh->nlmsg_seq, 0);
3365         }
3366         if (err < 0)
3367                 goto errout_rcu;
3368
3369         rcu_read_unlock();
3370
3371         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3372
3373 errout_free:
3374         return err;
3375 errout_rcu:
3376         rcu_read_unlock();
3377         kfree_skb(skb);
3378         goto errout_free;
3379 }
3380
3381 void ip_rt_multicast_event(struct in_device *in_dev)
3382 {
3383         rt_cache_flush(dev_net(in_dev->dev));
3384 }
3385
3386 #ifdef CONFIG_SYSCTL
3387 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3388 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
3389 static int ip_rt_gc_elasticity __read_mostly    = 8;
3390 static int ip_min_valid_pmtu __read_mostly      = IPV4_MIN_MTU;
3391
3392 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3393                 void *buffer, size_t *lenp, loff_t *ppos)
3394 {
3395         struct net *net = (struct net *)__ctl->extra1;
3396
3397         if (write) {
3398                 rt_cache_flush(net);
3399                 fnhe_genid_bump(net);
3400                 return 0;
3401         }
3402
3403         return -EINVAL;
3404 }
3405
3406 static struct ctl_table ipv4_route_table[] = {
3407         {
3408                 .procname       = "gc_thresh",
3409                 .data           = &ipv4_dst_ops.gc_thresh,
3410                 .maxlen         = sizeof(int),
3411                 .mode           = 0644,
3412                 .proc_handler   = proc_dointvec,
3413         },
3414         {
3415                 .procname       = "max_size",
3416                 .data           = &ip_rt_max_size,
3417                 .maxlen         = sizeof(int),
3418                 .mode           = 0644,
3419                 .proc_handler   = proc_dointvec,
3420         },
3421         {
3422                 /*  Deprecated. Use gc_min_interval_ms */
3423
3424                 .procname       = "gc_min_interval",
3425                 .data           = &ip_rt_gc_min_interval,
3426                 .maxlen         = sizeof(int),
3427                 .mode           = 0644,
3428                 .proc_handler   = proc_dointvec_jiffies,
3429         },
3430         {
3431                 .procname       = "gc_min_interval_ms",
3432                 .data           = &ip_rt_gc_min_interval,
3433                 .maxlen         = sizeof(int),
3434                 .mode           = 0644,
3435                 .proc_handler   = proc_dointvec_ms_jiffies,
3436         },
3437         {
3438                 .procname       = "gc_timeout",
3439                 .data           = &ip_rt_gc_timeout,
3440                 .maxlen         = sizeof(int),
3441                 .mode           = 0644,
3442                 .proc_handler   = proc_dointvec_jiffies,
3443         },
3444         {
3445                 .procname       = "gc_interval",
3446                 .data           = &ip_rt_gc_interval,
3447                 .maxlen         = sizeof(int),
3448                 .mode           = 0644,
3449                 .proc_handler   = proc_dointvec_jiffies,
3450         },
3451         {
3452                 .procname       = "redirect_load",
3453                 .data           = &ip_rt_redirect_load,
3454                 .maxlen         = sizeof(int),
3455                 .mode           = 0644,
3456                 .proc_handler   = proc_dointvec,
3457         },
3458         {
3459                 .procname       = "redirect_number",
3460                 .data           = &ip_rt_redirect_number,
3461                 .maxlen         = sizeof(int),
3462                 .mode           = 0644,
3463                 .proc_handler   = proc_dointvec,
3464         },
3465         {
3466                 .procname       = "redirect_silence",
3467                 .data           = &ip_rt_redirect_silence,
3468                 .maxlen         = sizeof(int),
3469                 .mode           = 0644,
3470                 .proc_handler   = proc_dointvec,
3471         },
3472         {
3473                 .procname       = "error_cost",
3474                 .data           = &ip_rt_error_cost,
3475                 .maxlen         = sizeof(int),
3476                 .mode           = 0644,
3477                 .proc_handler   = proc_dointvec,
3478         },
3479         {
3480                 .procname       = "error_burst",
3481                 .data           = &ip_rt_error_burst,
3482                 .maxlen         = sizeof(int),
3483                 .mode           = 0644,
3484                 .proc_handler   = proc_dointvec,
3485         },
3486         {
3487                 .procname       = "gc_elasticity",
3488                 .data           = &ip_rt_gc_elasticity,
3489                 .maxlen         = sizeof(int),
3490                 .mode           = 0644,
3491                 .proc_handler   = proc_dointvec,
3492         },
3493 };
3494
3495 static const char ipv4_route_flush_procname[] = "flush";
3496
3497 static struct ctl_table ipv4_route_netns_table[] = {
3498         {
3499                 .procname       = ipv4_route_flush_procname,
3500                 .maxlen         = sizeof(int),
3501                 .mode           = 0200,
3502                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3503         },
3504         {
3505                 .procname       = "min_pmtu",
3506                 .data           = &init_net.ipv4.ip_rt_min_pmtu,
3507                 .maxlen         = sizeof(int),
3508                 .mode           = 0644,
3509                 .proc_handler   = proc_dointvec_minmax,
3510                 .extra1         = &ip_min_valid_pmtu,
3511         },
3512         {
3513                 .procname       = "mtu_expires",
3514                 .data           = &init_net.ipv4.ip_rt_mtu_expires,
3515                 .maxlen         = sizeof(int),
3516                 .mode           = 0644,
3517                 .proc_handler   = proc_dointvec_jiffies,
3518         },
3519         {
3520                 .procname   = "min_adv_mss",
3521                 .data       = &init_net.ipv4.ip_rt_min_advmss,
3522                 .maxlen     = sizeof(int),
3523                 .mode       = 0644,
3524                 .proc_handler   = proc_dointvec,
3525         },
3526 };
3527
3528 static __net_init int sysctl_route_net_init(struct net *net)
3529 {
3530         struct ctl_table *tbl;
3531         size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3532
3533         tbl = ipv4_route_netns_table;
3534         if (!net_eq(net, &init_net)) {
3535                 int i;
3536
3537                 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3538                 if (!tbl)
3539                         goto err_dup;
3540
3541                 /* Don't export non-whitelisted sysctls to unprivileged users */
3542                 if (net->user_ns != &init_user_ns) {
3543                         if (tbl[0].procname != ipv4_route_flush_procname)
3544                                 table_size = 0;
3545                 }
3546
3547                 /* Update the variables to point into the current struct net
3548                  * except for the first element flush
3549                  */
3550                 for (i = 1; i < table_size; i++)
3551                         tbl[i].data += (void *)net - (void *)&init_net;
3552         }
3553         tbl[0].extra1 = net;
3554
3555         net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3556                                                      tbl, table_size);
3557         if (!net->ipv4.route_hdr)
3558                 goto err_reg;
3559         return 0;
3560
3561 err_reg:
3562         if (tbl != ipv4_route_netns_table)
3563                 kfree(tbl);
3564 err_dup:
3565         return -ENOMEM;
3566 }
3567
3568 static __net_exit void sysctl_route_net_exit(struct net *net)
3569 {
3570         const struct ctl_table *tbl;
3571
3572         tbl = net->ipv4.route_hdr->ctl_table_arg;
3573         unregister_net_sysctl_table(net->ipv4.route_hdr);
3574         BUG_ON(tbl == ipv4_route_netns_table);
3575         kfree(tbl);
3576 }
3577
3578 static __net_initdata struct pernet_operations sysctl_route_ops = {
3579         .init = sysctl_route_net_init,
3580         .exit = sysctl_route_net_exit,
3581 };
3582 #endif
3583
3584 static __net_init int netns_ip_rt_init(struct net *net)
3585 {
3586         /* Set default value for namespaceified sysctls */
3587         net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3588         net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3589         net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3590         return 0;
3591 }
3592
3593 static struct pernet_operations __net_initdata ip_rt_ops = {
3594         .init = netns_ip_rt_init,
3595 };
3596
3597 static __net_init int rt_genid_init(struct net *net)
3598 {
3599         atomic_set(&net->ipv4.rt_genid, 0);
3600         atomic_set(&net->fnhe_genid, 0);
3601         atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3602         return 0;
3603 }
3604
3605 static __net_initdata struct pernet_operations rt_genid_ops = {
3606         .init = rt_genid_init,
3607 };
3608
3609 static int __net_init ipv4_inetpeer_init(struct net *net)
3610 {
3611         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3612
3613         if (!bp)
3614                 return -ENOMEM;
3615         inet_peer_base_init(bp);
3616         net->ipv4.peers = bp;
3617         return 0;
3618 }
3619
3620 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3621 {
3622         struct inet_peer_base *bp = net->ipv4.peers;
3623
3624         net->ipv4.peers = NULL;
3625         inetpeer_invalidate_tree(bp);
3626         kfree(bp);
3627 }
3628
3629 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3630         .init   =       ipv4_inetpeer_init,
3631         .exit   =       ipv4_inetpeer_exit,
3632 };
3633
3634 #ifdef CONFIG_IP_ROUTE_CLASSID
3635 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3636 #endif /* CONFIG_IP_ROUTE_CLASSID */
3637
3638 int __init ip_rt_init(void)
3639 {
3640         void *idents_hash;
3641         int cpu;
3642
3643         /* For modern hosts, this will use 2 MB of memory */
3644         idents_hash = alloc_large_system_hash("IP idents",
3645                                               sizeof(*ip_idents) + sizeof(*ip_tstamps),
3646                                               0,
3647                                               16, /* one bucket per 64 KB */
3648                                               HASH_ZERO,
3649                                               NULL,
3650                                               &ip_idents_mask,
3651                                               2048,
3652                                               256*1024);
3653
3654         ip_idents = idents_hash;
3655
3656         get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3657
3658         ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3659
3660         for_each_possible_cpu(cpu) {
3661                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3662
3663                 INIT_LIST_HEAD(&ul->head);
3664                 INIT_LIST_HEAD(&ul->quarantine);
3665                 spin_lock_init(&ul->lock);
3666         }
3667 #ifdef CONFIG_IP_ROUTE_CLASSID
3668         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3669         if (!ip_rt_acct)
3670                 panic("IP: failed to allocate ip_rt_acct\n");
3671 #endif
3672
3673         ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3674                                               SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3675
3676         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3677
3678         if (dst_entries_init(&ipv4_dst_ops) < 0)
3679                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3680
3681         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3682                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3683
3684         ipv4_dst_ops.gc_thresh = ~0;
3685         ip_rt_max_size = INT_MAX;
3686
3687         devinet_init();
3688         ip_fib_init();
3689
3690         if (ip_rt_proc_init())
3691                 pr_err("Unable to create route proc files\n");
3692 #ifdef CONFIG_XFRM
3693         xfrm_init();
3694         xfrm4_init();
3695 #endif
3696         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3697                       RTNL_FLAG_DOIT_UNLOCKED);
3698
3699 #ifdef CONFIG_SYSCTL
3700         register_pernet_subsys(&sysctl_route_ops);
3701 #endif
3702         register_pernet_subsys(&ip_rt_ops);
3703         register_pernet_subsys(&rt_genid_ops);
3704         register_pernet_subsys(&ipv4_inetpeer_ops);
3705         return 0;
3706 }
3707
3708 #ifdef CONFIG_SYSCTL
3709 /*
3710  * We really need to sanitize the damn ipv4 init order, then all
3711  * this nonsense will go away.
3712  */
3713 void __init ip_static_sysctl_init(void)
3714 {
3715         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3716 }
3717 #endif