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 struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 133 static void              ipv4_link_failure(struct sk_buff *skb);
 134 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 135                                            struct sk_buff *skb, u32 mtu,
 136                                            bool confirm_neigh);
 137 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 138                                         struct sk_buff *skb);
 139 static void             ipv4_dst_destroy(struct dst_entry *dst);
 140
 141 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 142 {
 143         WARN_ON(1);
 144         return NULL;
 145 }
 146
 147 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 148                                            struct sk_buff *skb,
 149                                            const void *daddr);
 150 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
 151
 152 static struct dst_ops ipv4_dst_ops = {
 153         .family =               AF_INET,
 154         .check =                ipv4_dst_check,
 155         .default_advmss =       ipv4_default_advmss,
 156         .mtu =                  ipv4_mtu,
 157         .cow_metrics =          ipv4_cow_metrics,
 158         .destroy =              ipv4_dst_destroy,
 159         .negative_advice =      ipv4_negative_advice,
 160         .link_failure =         ipv4_link_failure,
 161         .update_pmtu =          ip_rt_update_pmtu,
 162         .redirect =             ip_do_redirect,
 163         .local_out =            __ip_local_out,
 164         .neigh_lookup =         ipv4_neigh_lookup,
 165         .confirm_neigh =        ipv4_confirm_neigh,
 166 };
 167
 168 #define ECN_OR_COST(class)      TC_PRIO_##class
 169
 170 const __u8 ip_tos2prio[16] = {
 171         TC_PRIO_BESTEFFORT,
 172         ECN_OR_COST(BESTEFFORT),
 173         TC_PRIO_BESTEFFORT,
 174         ECN_OR_COST(BESTEFFORT),
 175         TC_PRIO_BULK,
 176         ECN_OR_COST(BULK),
 177         TC_PRIO_BULK,
 178         ECN_OR_COST(BULK),
 179         TC_PRIO_INTERACTIVE,
 180         ECN_OR_COST(INTERACTIVE),
 181         TC_PRIO_INTERACTIVE,
 182         ECN_OR_COST(INTERACTIVE),
 183         TC_PRIO_INTERACTIVE_BULK,
 184         ECN_OR_COST(INTERACTIVE_BULK),
 185         TC_PRIO_INTERACTIVE_BULK,
 186         ECN_OR_COST(INTERACTIVE_BULK)
 187 };
 188 EXPORT_SYMBOL(ip_tos2prio);
 189
 190 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 191 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
 192
 193 #ifdef CONFIG_PROC_FS
 194 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 195 {
 196         if (*pos)
 197                 return NULL;
 198         return SEQ_START_TOKEN;
 199 }
 200
 201 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 202 {
 203         ++*pos;
 204         return NULL;
 205 }
 206
 207 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 208 {
 209 }
 210
 211 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 212 {
 213         if (v == SEQ_START_TOKEN)
 214                 seq_printf(seq, "%-127s\n",
 215                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 216                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 217                            "HHUptod\tSpecDst");
 218         return 0;
 219 }
 220
 221 static const struct seq_operations rt_cache_seq_ops = {
 222         .start  = rt_cache_seq_start,
 223         .next   = rt_cache_seq_next,
 224         .stop   = rt_cache_seq_stop,
 225         .show   = rt_cache_seq_show,
 226 };
 227
 228 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 229 {
 230         int cpu;
 231
 232         if (*pos == 0)
 233                 return SEQ_START_TOKEN;
 234
 235         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 236                 if (!cpu_possible(cpu))
 237                         continue;
 238                 *pos = cpu+1;
 239                 return &per_cpu(rt_cache_stat, cpu);
 240         }
 241         return NULL;
 242 }
 243
 244 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 245 {
 246         int cpu;
 247
 248         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 249                 if (!cpu_possible(cpu))
 250                         continue;
 251                 *pos = cpu+1;
 252                 return &per_cpu(rt_cache_stat, cpu);
 253         }
 254         (*pos)++;
 255         return NULL;
 256
 257 }
 258
 259 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 260 {
 261
 262 }
 263
 264 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 265 {
 266         struct rt_cache_stat *st = v;
 267
 268         if (v == SEQ_START_TOKEN) {
 269                 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");
 270                 return 0;
 271         }
 272
 273         seq_printf(seq, "%08x %08x %08x    %08x   %08x    %08x %08x       "
 274                         "%08x       %08x %08x     %08x    %08x %08x   "
 275                         "%08x     %08x        %08x        %08x\n",
 276                    dst_entries_get_slow(&ipv4_dst_ops),
 277                    0, /* st->in_hit */
 278                    st->in_slow_tot,
 279                    st->in_slow_mc,
 280                    st->in_no_route,
 281                    st->in_brd,
 282                    st->in_martian_dst,
 283                    st->in_martian_src,
 284
 285                    0, /* st->out_hit */
 286                    st->out_slow_tot,
 287                    st->out_slow_mc,
 288
 289                    0, /* st->gc_total */
 290                    0, /* st->gc_ignored */
 291                    0, /* st->gc_goal_miss */
 292                    0, /* st->gc_dst_overflow */
 293                    0, /* st->in_hlist_search */
 294                    0  /* st->out_hlist_search */
 295                 );
 296         return 0;
 297 }
 298
 299 static const struct seq_operations rt_cpu_seq_ops = {
 300         .start  = rt_cpu_seq_start,
 301         .next   = rt_cpu_seq_next,
 302         .stop   = rt_cpu_seq_stop,
 303         .show   = rt_cpu_seq_show,
 304 };
 305
 306 #ifdef CONFIG_IP_ROUTE_CLASSID
 307 static int rt_acct_proc_show(struct seq_file *m, void *v)
 308 {
 309         struct ip_rt_acct *dst, *src;
 310         unsigned int i, j;
 311
 312         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 313         if (!dst)
 314                 return -ENOMEM;
 315
 316         for_each_possible_cpu(i) {
 317                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 318                 for (j = 0; j < 256; j++) {
 319                         dst[j].o_bytes   += src[j].o_bytes;
 320                         dst[j].o_packets += src[j].o_packets;
 321                         dst[j].i_bytes   += src[j].i_bytes;
 322                         dst[j].i_packets += src[j].i_packets;
 323                 }
 324         }
 325
 326         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 327         kfree(dst);
 328         return 0;
 329 }
 330 #endif
 331
 332 static int __net_init ip_rt_do_proc_init(struct net *net)
 333 {
 334         struct proc_dir_entry *pde;
 335
 336         pde = proc_create_seq("rt_cache", 0444, net->proc_net,
 337                               &rt_cache_seq_ops);
 338         if (!pde)
 339                 goto err1;
 340
 341         pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
 342                               &rt_cpu_seq_ops);
 343         if (!pde)
 344                 goto err2;
 345
 346 #ifdef CONFIG_IP_ROUTE_CLASSID
 347         pde = proc_create_single("rt_acct", 0, net->proc_net,
 348                         rt_acct_proc_show);
 349         if (!pde)
 350                 goto err3;
 351 #endif
 352         return 0;
 353
 354 #ifdef CONFIG_IP_ROUTE_CLASSID
 355 err3:
 356         remove_proc_entry("rt_cache", net->proc_net_stat);
 357 #endif
 358 err2:
 359         remove_proc_entry("rt_cache", net->proc_net);
 360 err1:
 361         return -ENOMEM;
 362 }
 363
 364 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 365 {
 366         remove_proc_entry("rt_cache", net->proc_net_stat);
 367         remove_proc_entry("rt_cache", net->proc_net);
 368 #ifdef CONFIG_IP_ROUTE_CLASSID
 369         remove_proc_entry("rt_acct", net->proc_net);
 370 #endif
 371 }
 372
 373 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 374         .init = ip_rt_do_proc_init,
 375         .exit = ip_rt_do_proc_exit,
 376 };
 377
 378 static int __init ip_rt_proc_init(void)
 379 {
 380         return register_pernet_subsys(&ip_rt_proc_ops);
 381 }
 382
 383 #else
 384 static inline int ip_rt_proc_init(void)
 385 {
 386         return 0;
 387 }
 388 #endif /* CONFIG_PROC_FS */
 389
 390 static inline bool rt_is_expired(const struct rtable *rth)
 391 {
 392         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 393 }
 394
 395 void rt_cache_flush(struct net *net)
 396 {
 397         rt_genid_bump_ipv4(net);
 398 }
 399
 400 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 401                                            struct sk_buff *skb,
 402                                            const void *daddr)
 403 {
 404         const struct rtable *rt = container_of(dst, struct rtable, dst);
 405         struct net_device *dev = dst->dev;
 406         struct neighbour *n;
 407
 408         rcu_read_lock();
 409
 410         if (likely(rt->rt_gw_family == AF_INET)) {
 411                 n = ip_neigh_gw4(dev, rt->rt_gw4);
 412         } else if (rt->rt_gw_family == AF_INET6) {
 413                 n = ip_neigh_gw6(dev, &rt->rt_gw6);
 414         } else {
 415                 __be32 pkey;
 416
 417                 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
 418                 n = ip_neigh_gw4(dev, pkey);
 419         }
 420
 421         if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
 422                 n = NULL;
 423
 424         rcu_read_unlock();
 425
 426         return n;
 427 }
 428
 429 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 430 {
 431         const struct rtable *rt = container_of(dst, struct rtable, dst);
 432         struct net_device *dev = dst->dev;
 433         const __be32 *pkey = daddr;
 434
 435         if (rt->rt_gw_family == AF_INET) {
 436                 pkey = (const __be32 *)&rt->rt_gw4;
 437         } else if (rt->rt_gw_family == AF_INET6) {
 438                 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
 439         } else if (!daddr ||
 440                  (rt->rt_flags &
 441                   (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
 442                 return;
 443         }
 444         __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
 445 }
 446
 447 /* Hash tables of size 2048..262144 depending on RAM size.
 448  * Each bucket uses 8 bytes.
 449  */
 450 static u32 ip_idents_mask __read_mostly;
 451 static atomic_t *ip_idents __read_mostly;
 452 static u32 *ip_tstamps __read_mostly;
 453
 454 /* In order to protect privacy, we add a perturbation to identifiers
 455  * if one generator is seldom used. This makes hard for an attacker
 456  * to infer how many packets were sent between two points in time.
 457  */
 458 static u32 ip_idents_reserve(u32 hash, int segs)
 459 {
 460         u32 bucket, old, now = (u32)jiffies;
 461         atomic_t *p_id;
 462         u32 *p_tstamp;
 463         u32 delta = 0;
 464
 465         bucket = hash & ip_idents_mask;
 466         p_tstamp = ip_tstamps + bucket;
 467         p_id = ip_idents + bucket;
 468         old = READ_ONCE(*p_tstamp);
 469
 470         if (old != now && cmpxchg(p_tstamp, old, now) == old)
 471                 delta = get_random_u32_below(now - old);
 472
 473         /* If UBSAN reports an error there, please make sure your compiler
 474          * supports -fno-strict-overflow before reporting it that was a bug
 475          * in UBSAN, and it has been fixed in GCC-8.
 476          */
 477         return atomic_add_return(segs + delta, p_id) - segs;
 478 }
 479
 480 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
 481 {
 482         u32 hash, id;
 483
 484         /* Note the following code is not safe, but this is okay. */
 485         if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
 486                 get_random_bytes(&net->ipv4.ip_id_key,
 487                                  sizeof(net->ipv4.ip_id_key));
 488
 489         hash = siphash_3u32((__force u32)iph->daddr,
 490                             (__force u32)iph->saddr,
 491                             iph->protocol,
 492                             &net->ipv4.ip_id_key);
 493         id = ip_idents_reserve(hash, segs);
 494         iph->id = htons(id);
 495 }
 496 EXPORT_SYMBOL(__ip_select_ident);
 497
 498 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
 499                              const struct sock *sk, const struct iphdr *iph,
 500                              int oif, __u8 tos, u8 prot, u32 mark,
 501                              int flow_flags)
 502 {
 503         __u8 scope = RT_SCOPE_UNIVERSE;
 504
 505         if (sk) {
 506                 oif = sk->sk_bound_dev_if;
 507                 mark = READ_ONCE(sk->sk_mark);
 508                 tos = ip_sock_rt_tos(sk);
 509                 scope = ip_sock_rt_scope(sk);
 510                 prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
 511                                                     sk->sk_protocol;
 512         }
 513
 514         flowi4_init_output(fl4, oif, mark, tos & IPTOS_RT_MASK, scope,
 515                            prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
 516                            sock_net_uid(net, sk));
 517 }
 518
 519 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 520                                const struct sock *sk)
 521 {
 522         const struct net *net = dev_net(skb->dev);
 523         const struct iphdr *iph = ip_hdr(skb);
 524         int oif = skb->dev->ifindex;
 525         u8 prot = iph->protocol;
 526         u32 mark = skb->mark;
 527         __u8 tos = iph->tos;
 528
 529         __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
 530 }
 531
 532 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 533 {
 534         const struct inet_sock *inet = inet_sk(sk);
 535         const struct ip_options_rcu *inet_opt;
 536         __be32 daddr = inet->inet_daddr;
 537
 538         rcu_read_lock();
 539         inet_opt = rcu_dereference(inet->inet_opt);
 540         if (inet_opt && inet_opt->opt.srr)
 541                 daddr = inet_opt->opt.faddr;
 542         flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
 543                            ip_sock_rt_tos(sk) & IPTOS_RT_MASK,
 544                            ip_sock_rt_scope(sk),
 545                            inet_test_bit(HDRINCL, sk) ?
 546                                 IPPROTO_RAW : sk->sk_protocol,
 547                            inet_sk_flowi_flags(sk),
 548                            daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
 549         rcu_read_unlock();
 550 }
 551
 552 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 553                                  const struct sk_buff *skb)
 554 {
 555         if (skb)
 556                 build_skb_flow_key(fl4, skb, sk);
 557         else
 558                 build_sk_flow_key(fl4, sk);
 559 }
 560
 561 static DEFINE_SPINLOCK(fnhe_lock);
 562
 563 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 564 {
 565         struct rtable *rt;
 566
 567         rt = rcu_dereference(fnhe->fnhe_rth_input);
 568         if (rt) {
 569                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 570                 dst_dev_put(&rt->dst);
 571                 dst_release(&rt->dst);
 572         }
 573         rt = rcu_dereference(fnhe->fnhe_rth_output);
 574         if (rt) {
 575                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 576                 dst_dev_put(&rt->dst);
 577                 dst_release(&rt->dst);
 578         }
 579 }
 580
 581 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
 582 {
 583         struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
 584         struct fib_nh_exception *fnhe, *oldest = NULL;
 585
 586         for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
 587                 fnhe = rcu_dereference_protected(*fnhe_p,
 588                                                  lockdep_is_held(&fnhe_lock));
 589                 if (!fnhe)
 590                         break;
 591                 if (!oldest ||
 592                     time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
 593                         oldest = fnhe;
 594                         oldest_p = fnhe_p;
 595                 }
 596         }
 597         fnhe_flush_routes(oldest);
 598         *oldest_p = oldest->fnhe_next;
 599         kfree_rcu(oldest, rcu);
 600 }
 601
 602 static u32 fnhe_hashfun(__be32 daddr)
 603 {
 604         static siphash_aligned_key_t fnhe_hash_key;
 605         u64 hval;
 606
 607         net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
 608         hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
 609         return hash_64(hval, FNHE_HASH_SHIFT);
 610 }
 611
 612 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 613 {
 614         rt->rt_pmtu = fnhe->fnhe_pmtu;
 615         rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
 616         rt->dst.expires = fnhe->fnhe_expires;
 617
 618         if (fnhe->fnhe_gw) {
 619                 rt->rt_flags |= RTCF_REDIRECTED;
 620                 rt->rt_uses_gateway = 1;
 621                 rt->rt_gw_family = AF_INET;
 622                 rt->rt_gw4 = fnhe->fnhe_gw;
 623         }
 624 }
 625
 626 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
 627                                   __be32 gw, u32 pmtu, bool lock,
 628                                   unsigned long expires)
 629 {
 630         struct fnhe_hash_bucket *hash;
 631         struct fib_nh_exception *fnhe;
 632         struct rtable *rt;
 633         u32 genid, hval;
 634         unsigned int i;
 635         int depth;
 636
 637         genid = fnhe_genid(dev_net(nhc->nhc_dev));
 638         hval = fnhe_hashfun(daddr);
 639
 640         spin_lock_bh(&fnhe_lock);
 641
 642         hash = rcu_dereference(nhc->nhc_exceptions);
 643         if (!hash) {
 644                 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
 645                 if (!hash)
 646                         goto out_unlock;
 647                 rcu_assign_pointer(nhc->nhc_exceptions, hash);
 648         }
 649
 650         hash += hval;
 651
 652         depth = 0;
 653         for (fnhe = rcu_dereference(hash->chain); fnhe;
 654              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 655                 if (fnhe->fnhe_daddr == daddr)
 656                         break;
 657                 depth++;
 658         }
 659
 660         if (fnhe) {
 661                 if (fnhe->fnhe_genid != genid)
 662                         fnhe->fnhe_genid = genid;
 663                 if (gw)
 664                         fnhe->fnhe_gw = gw;
 665                 if (pmtu) {
 666                         fnhe->fnhe_pmtu = pmtu;
 667                         fnhe->fnhe_mtu_locked = lock;
 668                 }
 669                 fnhe->fnhe_expires = max(1UL, expires);
 670                 /* Update all cached dsts too */
 671                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 672                 if (rt)
 673                         fill_route_from_fnhe(rt, fnhe);
 674                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 675                 if (rt)
 676                         fill_route_from_fnhe(rt, fnhe);
 677         } else {
 678                 /* Randomize max depth to avoid some side channels attacks. */
 679                 int max_depth = FNHE_RECLAIM_DEPTH +
 680                                 get_random_u32_below(FNHE_RECLAIM_DEPTH);
 681
 682                 while (depth > max_depth) {
 683                         fnhe_remove_oldest(hash);
 684                         depth--;
 685                 }
 686
 687                 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 688                 if (!fnhe)
 689                         goto out_unlock;
 690
 691                 fnhe->fnhe_next = hash->chain;
 692
 693                 fnhe->fnhe_genid = genid;
 694                 fnhe->fnhe_daddr = daddr;
 695                 fnhe->fnhe_gw = gw;
 696                 fnhe->fnhe_pmtu = pmtu;
 697                 fnhe->fnhe_mtu_locked = lock;
 698                 fnhe->fnhe_expires = max(1UL, expires);
 699
 700                 rcu_assign_pointer(hash->chain, fnhe);
 701
 702                 /* Exception created; mark the cached routes for the nexthop
 703                  * stale, so anyone caching it rechecks if this exception
 704                  * applies to them.
 705                  */
 706                 rt = rcu_dereference(nhc->nhc_rth_input);
 707                 if (rt)
 708                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 709
 710                 for_each_possible_cpu(i) {
 711                         struct rtable __rcu **prt;
 712
 713                         prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
 714                         rt = rcu_dereference(*prt);
 715                         if (rt)
 716                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 717                 }
 718         }
 719
 720         fnhe->fnhe_stamp = jiffies;
 721
 722 out_unlock:
 723         spin_unlock_bh(&fnhe_lock);
 724 }
 725
 726 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 727                              bool kill_route)
 728 {
 729         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 730         __be32 old_gw = ip_hdr(skb)->saddr;
 731         struct net_device *dev = skb->dev;
 732         struct in_device *in_dev;
 733         struct fib_result res;
 734         struct neighbour *n;
 735         struct net *net;
 736
 737         switch (icmp_hdr(skb)->code & 7) {
 738         case ICMP_REDIR_NET:
 739         case ICMP_REDIR_NETTOS:
 740         case ICMP_REDIR_HOST:
 741         case ICMP_REDIR_HOSTTOS:
 742                 break;
 743
 744         default:
 745                 return;
 746         }
 747
 748         if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
 749                 return;
 750
 751         in_dev = __in_dev_get_rcu(dev);
 752         if (!in_dev)
 753                 return;
 754
 755         net = dev_net(dev);
 756         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 757             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 758             ipv4_is_zeronet(new_gw))
 759                 goto reject_redirect;
 760
 761         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 762                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 763                         goto reject_redirect;
 764                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 765                         goto reject_redirect;
 766         } else {
 767                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 768                         goto reject_redirect;
 769         }
 770
 771         n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
 772         if (!n)
 773                 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
 774         if (!IS_ERR(n)) {
 775                 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
 776                         neigh_event_send(n, NULL);
 777                 } else {
 778                         if (fib_lookup(net, fl4, &res, 0) == 0) {
 779                                 struct fib_nh_common *nhc;
 780
 781                                 fib_select_path(net, &res, fl4, skb);
 782                                 nhc = FIB_RES_NHC(res);
 783                                 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
 784                                                 0, false,
 785                                                 jiffies + ip_rt_gc_timeout);
 786                         }
 787                         if (kill_route)
 788                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 789                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 790                 }
 791                 neigh_release(n);
 792         }
 793         return;
 794
 795 reject_redirect:
 796 #ifdef CONFIG_IP_ROUTE_VERBOSE
 797         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 798                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 799                 __be32 daddr = iph->daddr;
 800                 __be32 saddr = iph->saddr;
 801
 802                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 803                                      "  Advised path = %pI4 -> %pI4\n",
 804                                      &old_gw, dev->name, &new_gw,
 805                                      &saddr, &daddr);
 806         }
 807 #endif
 808         ;
 809 }
 810
 811 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 812 {
 813         struct rtable *rt;
 814         struct flowi4 fl4;
 815         const struct iphdr *iph = (const struct iphdr *) skb->data;
 816         struct net *net = dev_net(skb->dev);
 817         int oif = skb->dev->ifindex;
 818         u8 prot = iph->protocol;
 819         u32 mark = skb->mark;
 820         __u8 tos = iph->tos;
 821
 822         rt = dst_rtable(dst);
 823
 824         __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
 825         __ip_do_redirect(rt, skb, &fl4, true);
 826 }
 827
 828 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 829 {
 830         struct rtable *rt = dst_rtable(dst);
 831         struct dst_entry *ret = dst;
 832
 833         if (rt) {
 834                 if (dst->obsolete > 0) {
 835                         ip_rt_put(rt);
 836                         ret = NULL;
 837                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 838                            rt->dst.expires) {
 839                         ip_rt_put(rt);
 840                         ret = NULL;
 841                 }
 842         }
 843         return ret;
 844 }
 845
 846 /*
 847  * Algorithm:
 848  *      1. The first ip_rt_redirect_number redirects are sent
 849  *         with exponential backoff, then we stop sending them at all,
 850  *         assuming that the host ignores our redirects.
 851  *      2. If we did not see packets requiring redirects
 852  *         during ip_rt_redirect_silence, we assume that the host
 853  *         forgot redirected route and start to send redirects again.
 854  *
 855  * This algorithm is much cheaper and more intelligent than dumb load limiting
 856  * in icmp.c.
 857  *
 858  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 859  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 860  */
 861
 862 void ip_rt_send_redirect(struct sk_buff *skb)
 863 {
 864         struct rtable *rt = skb_rtable(skb);
 865         struct in_device *in_dev;
 866         struct inet_peer *peer;
 867         struct net *net;
 868         int log_martians;
 869         int vif;
 870
 871         rcu_read_lock();
 872         in_dev = __in_dev_get_rcu(rt->dst.dev);
 873         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 874                 rcu_read_unlock();
 875                 return;
 876         }
 877         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 878         vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
 879         rcu_read_unlock();
 880
 881         net = dev_net(rt->dst.dev);
 882         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
 883         if (!peer) {
 884                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 885                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 886                 return;
 887         }
 888
 889         /* No redirected packets during ip_rt_redirect_silence;
 890          * reset the algorithm.
 891          */
 892         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
 893                 peer->rate_tokens = 0;
 894                 peer->n_redirects = 0;
 895         }
 896
 897         /* Too many ignored redirects; do not send anything
 898          * set dst.rate_last to the last seen redirected packet.
 899          */
 900         if (peer->n_redirects >= ip_rt_redirect_number) {
 901                 peer->rate_last = jiffies;
 902                 goto out_put_peer;
 903         }
 904
 905         /* Check for load limit; set rate_last to the latest sent
 906          * redirect.
 907          */
 908         if (peer->n_redirects == 0 ||
 909             time_after(jiffies,
 910                        (peer->rate_last +
 911                         (ip_rt_redirect_load << peer->n_redirects)))) {
 912                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 913
 914                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 915                 peer->rate_last = jiffies;
 916                 ++peer->n_redirects;
 917                 if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians &&
 918                     peer->n_redirects == ip_rt_redirect_number)
 919                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 920                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 921                                              &ip_hdr(skb)->daddr, &gw);
 922         }
 923 out_put_peer:
 924         inet_putpeer(peer);
 925 }
 926
 927 static int ip_error(struct sk_buff *skb)
 928 {
 929         struct rtable *rt = skb_rtable(skb);
 930         struct net_device *dev = skb->dev;
 931         struct in_device *in_dev;
 932         struct inet_peer *peer;
 933         unsigned long now;
 934         struct net *net;
 935         SKB_DR(reason);
 936         bool send;
 937         int code;
 938
 939         if (netif_is_l3_master(skb->dev)) {
 940                 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
 941                 if (!dev)
 942                         goto out;
 943         }
 944
 945         in_dev = __in_dev_get_rcu(dev);
 946
 947         /* IP on this device is disabled. */
 948         if (!in_dev)
 949                 goto out;
 950
 951         net = dev_net(rt->dst.dev);
 952         if (!IN_DEV_FORWARD(in_dev)) {
 953                 switch (rt->dst.error) {
 954                 case EHOSTUNREACH:
 955                         SKB_DR_SET(reason, IP_INADDRERRORS);
 956                         __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
 957                         break;
 958
 959                 case ENETUNREACH:
 960                         SKB_DR_SET(reason, IP_INNOROUTES);
 961                         __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 962                         break;
 963                 }
 964                 goto out;
 965         }
 966
 967         switch (rt->dst.error) {
 968         case EINVAL:
 969         default:
 970                 goto out;
 971         case EHOSTUNREACH:
 972                 code = ICMP_HOST_UNREACH;
 973                 break;
 974         case ENETUNREACH:
 975                 code = ICMP_NET_UNREACH;
 976                 SKB_DR_SET(reason, IP_INNOROUTES);
 977                 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 978                 break;
 979         case EACCES:
 980                 code = ICMP_PKT_FILTERED;
 981                 break;
 982         }
 983
 984         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
 985                                l3mdev_master_ifindex(skb->dev), 1);
 986
 987         send = true;
 988         if (peer) {
 989                 now = jiffies;
 990                 peer->rate_tokens += now - peer->rate_last;
 991                 if (peer->rate_tokens > ip_rt_error_burst)
 992                         peer->rate_tokens = ip_rt_error_burst;
 993                 peer->rate_last = now;
 994                 if (peer->rate_tokens >= ip_rt_error_cost)
 995                         peer->rate_tokens -= ip_rt_error_cost;
 996                 else
 997                         send = false;
 998                 inet_putpeer(peer);
 999         }
1000         if (send)
1001                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1002
1003 out:    kfree_skb_reason(skb, reason);
1004         return 0;
1005 }
1006
1007 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1008 {
1009         struct dst_entry *dst = &rt->dst;
1010         struct net *net = dev_net(dst->dev);
1011         struct fib_result res;
1012         bool lock = false;
1013         u32 old_mtu;
1014
1015         if (ip_mtu_locked(dst))
1016                 return;
1017
1018         old_mtu = ipv4_mtu(dst);
1019         if (old_mtu < mtu)
1020                 return;
1021
1022         if (mtu < net->ipv4.ip_rt_min_pmtu) {
1023                 lock = true;
1024                 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1025         }
1026
1027         if (rt->rt_pmtu == mtu && !lock &&
1028             time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1029                 return;
1030
1031         rcu_read_lock();
1032         if (fib_lookup(net, fl4, &res, 0) == 0) {
1033                 struct fib_nh_common *nhc;
1034
1035                 fib_select_path(net, &res, fl4, NULL);
1036                 nhc = FIB_RES_NHC(res);
1037                 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1038                                       jiffies + net->ipv4.ip_rt_mtu_expires);
1039         }
1040         rcu_read_unlock();
1041 }
1042
1043 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1044                               struct sk_buff *skb, u32 mtu,
1045                               bool confirm_neigh)
1046 {
1047         struct rtable *rt = dst_rtable(dst);
1048         struct flowi4 fl4;
1049
1050         ip_rt_build_flow_key(&fl4, sk, skb);
1051
1052         /* Don't make lookup fail for bridged encapsulations */
1053         if (skb && netif_is_any_bridge_port(skb->dev))
1054                 fl4.flowi4_oif = 0;
1055
1056         __ip_rt_update_pmtu(rt, &fl4, mtu);
1057 }
1058
1059 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1060                       int oif, u8 protocol)
1061 {
1062         const struct iphdr *iph = (const struct iphdr *)skb->data;
1063         struct flowi4 fl4;
1064         struct rtable *rt;
1065         u32 mark = IP4_REPLY_MARK(net, skb->mark);
1066
1067         __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1068                          0);
1069         rt = __ip_route_output_key(net, &fl4);
1070         if (!IS_ERR(rt)) {
1071                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1072                 ip_rt_put(rt);
1073         }
1074 }
1075 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1076
1077 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1078 {
1079         const struct iphdr *iph = (const struct iphdr *)skb->data;
1080         struct flowi4 fl4;
1081         struct rtable *rt;
1082
1083         __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1084
1085         if (!fl4.flowi4_mark)
1086                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1087
1088         rt = __ip_route_output_key(sock_net(sk), &fl4);
1089         if (!IS_ERR(rt)) {
1090                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1091                 ip_rt_put(rt);
1092         }
1093 }
1094
1095 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1096 {
1097         const struct iphdr *iph = (const struct iphdr *)skb->data;
1098         struct flowi4 fl4;
1099         struct rtable *rt;
1100         struct dst_entry *odst = NULL;
1101         bool new = false;
1102         struct net *net = sock_net(sk);
1103
1104         bh_lock_sock(sk);
1105
1106         if (!ip_sk_accept_pmtu(sk))
1107                 goto out;
1108
1109         odst = sk_dst_get(sk);
1110
1111         if (sock_owned_by_user(sk) || !odst) {
1112                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1113                 goto out;
1114         }
1115
1116         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1117
1118         rt = dst_rtable(odst);
1119         if (odst->obsolete && !odst->ops->check(odst, 0)) {
1120                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1121                 if (IS_ERR(rt))
1122                         goto out;
1123
1124                 new = true;
1125         }
1126
1127         __ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu);
1128
1129         if (!dst_check(&rt->dst, 0)) {
1130                 if (new)
1131                         dst_release(&rt->dst);
1132
1133                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1134                 if (IS_ERR(rt))
1135                         goto out;
1136
1137                 new = true;
1138         }
1139
1140         if (new)
1141                 sk_dst_set(sk, &rt->dst);
1142
1143 out:
1144         bh_unlock_sock(sk);
1145         dst_release(odst);
1146 }
1147 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1148
1149 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1150                    int oif, u8 protocol)
1151 {
1152         const struct iphdr *iph = (const struct iphdr *)skb->data;
1153         struct flowi4 fl4;
1154         struct rtable *rt;
1155
1156         __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1157         rt = __ip_route_output_key(net, &fl4);
1158         if (!IS_ERR(rt)) {
1159                 __ip_do_redirect(rt, skb, &fl4, false);
1160                 ip_rt_put(rt);
1161         }
1162 }
1163 EXPORT_SYMBOL_GPL(ipv4_redirect);
1164
1165 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1166 {
1167         const struct iphdr *iph = (const struct iphdr *)skb->data;
1168         struct flowi4 fl4;
1169         struct rtable *rt;
1170         struct net *net = sock_net(sk);
1171
1172         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1173         rt = __ip_route_output_key(net, &fl4);
1174         if (!IS_ERR(rt)) {
1175                 __ip_do_redirect(rt, skb, &fl4, false);
1176                 ip_rt_put(rt);
1177         }
1178 }
1179 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1180
1181 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1182                                                          u32 cookie)
1183 {
1184         struct rtable *rt = dst_rtable(dst);
1185
1186         /* All IPV4 dsts are created with ->obsolete set to the value
1187          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1188          * into this function always.
1189          *
1190          * When a PMTU/redirect information update invalidates a route,
1191          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1192          * DST_OBSOLETE_DEAD.
1193          */
1194         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1195                 return NULL;
1196         return dst;
1197 }
1198 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1199
1200 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1201 {
1202         struct net_device *dev;
1203         struct ip_options opt;
1204         int res;
1205
1206         /* Recompile ip options since IPCB may not be valid anymore.
1207          * Also check we have a reasonable ipv4 header.
1208          */
1209         if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1210             ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1211                 return;
1212
1213         memset(&opt, 0, sizeof(opt));
1214         if (ip_hdr(skb)->ihl > 5) {
1215                 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1216                         return;
1217                 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1218
1219                 rcu_read_lock();
1220                 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1221                 res = __ip_options_compile(dev_net(dev), &opt, skb, NULL);
1222                 rcu_read_unlock();
1223
1224                 if (res)
1225                         return;
1226         }
1227         __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1228 }
1229
1230 static void ipv4_link_failure(struct sk_buff *skb)
1231 {
1232         struct rtable *rt;
1233
1234         ipv4_send_dest_unreach(skb);
1235
1236         rt = skb_rtable(skb);
1237         if (rt)
1238                 dst_set_expires(&rt->dst, 0);
1239 }
1240
1241 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1242 {
1243         pr_debug("%s: %pI4 -> %pI4, %s\n",
1244                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1245                  skb->dev ? skb->dev->name : "?");
1246         kfree_skb(skb);
1247         WARN_ON(1);
1248         return 0;
1249 }
1250
1251 /*
1252  * We do not cache source address of outgoing interface,
1253  * because it is used only by IP RR, TS and SRR options,
1254  * so that it out of fast path.
1255  *
1256  * BTW remember: "addr" is allowed to be not aligned
1257  * in IP options!
1258  */
1259
1260 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1261 {
1262         __be32 src;
1263
1264         if (rt_is_output_route(rt))
1265                 src = ip_hdr(skb)->saddr;
1266         else {
1267                 struct fib_result res;
1268                 struct iphdr *iph = ip_hdr(skb);
1269                 struct flowi4 fl4 = {
1270                         .daddr = iph->daddr,
1271                         .saddr = iph->saddr,
1272                         .flowi4_tos = RT_TOS(iph->tos),
1273                         .flowi4_oif = rt->dst.dev->ifindex,
1274                         .flowi4_iif = skb->dev->ifindex,
1275                         .flowi4_mark = skb->mark,
1276                 };
1277
1278                 rcu_read_lock();
1279                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1280                         src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1281                 else
1282                         src = inet_select_addr(rt->dst.dev,
1283                                                rt_nexthop(rt, iph->daddr),
1284                                                RT_SCOPE_UNIVERSE);
1285                 rcu_read_unlock();
1286         }
1287         memcpy(addr, &src, 4);
1288 }
1289
1290 #ifdef CONFIG_IP_ROUTE_CLASSID
1291 static void set_class_tag(struct rtable *rt, u32 tag)
1292 {
1293         if (!(rt->dst.tclassid & 0xFFFF))
1294                 rt->dst.tclassid |= tag & 0xFFFF;
1295         if (!(rt->dst.tclassid & 0xFFFF0000))
1296                 rt->dst.tclassid |= tag & 0xFFFF0000;
1297 }
1298 #endif
1299
1300 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1301 {
1302         struct net *net = dev_net(dst->dev);
1303         unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1304         unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1305                                     net->ipv4.ip_rt_min_advmss);
1306
1307         return min(advmss, IPV4_MAX_PMTU - header_size);
1308 }
1309
1310 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1311 {
1312         return ip_dst_mtu_maybe_forward(dst, false);
1313 }
1314 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1315
1316 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1317 {
1318         struct fnhe_hash_bucket *hash;
1319         struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1320         u32 hval = fnhe_hashfun(daddr);
1321
1322         spin_lock_bh(&fnhe_lock);
1323
1324         hash = rcu_dereference_protected(nhc->nhc_exceptions,
1325                                          lockdep_is_held(&fnhe_lock));
1326         hash += hval;
1327
1328         fnhe_p = &hash->chain;
1329         fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1330         while (fnhe) {
1331                 if (fnhe->fnhe_daddr == daddr) {
1332                         rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1333                                 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1334                         /* set fnhe_daddr to 0 to ensure it won't bind with
1335                          * new dsts in rt_bind_exception().
1336                          */
1337                         fnhe->fnhe_daddr = 0;
1338                         fnhe_flush_routes(fnhe);
1339                         kfree_rcu(fnhe, rcu);
1340                         break;
1341                 }
1342                 fnhe_p = &fnhe->fnhe_next;
1343                 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1344                                                  lockdep_is_held(&fnhe_lock));
1345         }
1346
1347         spin_unlock_bh(&fnhe_lock);
1348 }
1349
1350 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1351                                                __be32 daddr)
1352 {
1353         struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1354         struct fib_nh_exception *fnhe;
1355         u32 hval;
1356
1357         if (!hash)
1358                 return NULL;
1359
1360         hval = fnhe_hashfun(daddr);
1361
1362         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1363              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1364                 if (fnhe->fnhe_daddr == daddr) {
1365                         if (fnhe->fnhe_expires &&
1366                             time_after(jiffies, fnhe->fnhe_expires)) {
1367                                 ip_del_fnhe(nhc, daddr);
1368                                 break;
1369                         }
1370                         return fnhe;
1371                 }
1372         }
1373         return NULL;
1374 }
1375
1376 /* MTU selection:
1377  * 1. mtu on route is locked - use it
1378  * 2. mtu from nexthop exception
1379  * 3. mtu from egress device
1380  */
1381
1382 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1383 {
1384         struct fib_nh_common *nhc = res->nhc;
1385         struct net_device *dev = nhc->nhc_dev;
1386         struct fib_info *fi = res->fi;
1387         u32 mtu = 0;
1388
1389         if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1390             fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1391                 mtu = fi->fib_mtu;
1392
1393         if (likely(!mtu)) {
1394                 struct fib_nh_exception *fnhe;
1395
1396                 fnhe = find_exception(nhc, daddr);
1397                 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1398                         mtu = fnhe->fnhe_pmtu;
1399         }
1400
1401         if (likely(!mtu))
1402                 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1403
1404         return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1405 }
1406
1407 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1408                               __be32 daddr, const bool do_cache)
1409 {
1410         bool ret = false;
1411
1412         spin_lock_bh(&fnhe_lock);
1413
1414         if (daddr == fnhe->fnhe_daddr) {
1415                 struct rtable __rcu **porig;
1416                 struct rtable *orig;
1417                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1418
1419                 if (rt_is_input_route(rt))
1420                         porig = &fnhe->fnhe_rth_input;
1421                 else
1422                         porig = &fnhe->fnhe_rth_output;
1423                 orig = rcu_dereference(*porig);
1424
1425                 if (fnhe->fnhe_genid != genid) {
1426                         fnhe->fnhe_genid = genid;
1427                         fnhe->fnhe_gw = 0;
1428                         fnhe->fnhe_pmtu = 0;
1429                         fnhe->fnhe_expires = 0;
1430                         fnhe->fnhe_mtu_locked = false;
1431                         fnhe_flush_routes(fnhe);
1432                         orig = NULL;
1433                 }
1434                 fill_route_from_fnhe(rt, fnhe);
1435                 if (!rt->rt_gw4) {
1436                         rt->rt_gw4 = daddr;
1437                         rt->rt_gw_family = AF_INET;
1438                 }
1439
1440                 if (do_cache) {
1441                         dst_hold(&rt->dst);
1442                         rcu_assign_pointer(*porig, rt);
1443                         if (orig) {
1444                                 dst_dev_put(&orig->dst);
1445                                 dst_release(&orig->dst);
1446                         }
1447                         ret = true;
1448                 }
1449
1450                 fnhe->fnhe_stamp = jiffies;
1451         }
1452         spin_unlock_bh(&fnhe_lock);
1453
1454         return ret;
1455 }
1456
1457 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1458 {
1459         struct rtable *orig, *prev, **p;
1460         bool ret = true;
1461
1462         if (rt_is_input_route(rt)) {
1463                 p = (struct rtable **)&nhc->nhc_rth_input;
1464         } else {
1465                 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1466         }
1467         orig = *p;
1468
1469         /* hold dst before doing cmpxchg() to avoid race condition
1470          * on this dst
1471          */
1472         dst_hold(&rt->dst);
1473         prev = cmpxchg(p, orig, rt);
1474         if (prev == orig) {
1475                 if (orig) {
1476                         rt_add_uncached_list(orig);
1477                         dst_release(&orig->dst);
1478                 }
1479         } else {
1480                 dst_release(&rt->dst);
1481                 ret = false;
1482         }
1483
1484         return ret;
1485 }
1486
1487 struct uncached_list {
1488         spinlock_t              lock;
1489         struct list_head        head;
1490         struct list_head        quarantine;
1491 };
1492
1493 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1494
1495 void rt_add_uncached_list(struct rtable *rt)
1496 {
1497         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1498
1499         rt->dst.rt_uncached_list = ul;
1500
1501         spin_lock_bh(&ul->lock);
1502         list_add_tail(&rt->dst.rt_uncached, &ul->head);
1503         spin_unlock_bh(&ul->lock);
1504 }
1505
1506 void rt_del_uncached_list(struct rtable *rt)
1507 {
1508         if (!list_empty(&rt->dst.rt_uncached)) {
1509                 struct uncached_list *ul = rt->dst.rt_uncached_list;
1510
1511                 spin_lock_bh(&ul->lock);
1512                 list_del_init(&rt->dst.rt_uncached);
1513                 spin_unlock_bh(&ul->lock);
1514         }
1515 }
1516
1517 static void ipv4_dst_destroy(struct dst_entry *dst)
1518 {
1519         ip_dst_metrics_put(dst);
1520         rt_del_uncached_list(dst_rtable(dst));
1521 }
1522
1523 void rt_flush_dev(struct net_device *dev)
1524 {
1525         struct rtable *rt, *safe;
1526         int cpu;
1527
1528         for_each_possible_cpu(cpu) {
1529                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1530
1531                 if (list_empty(&ul->head))
1532                         continue;
1533
1534                 spin_lock_bh(&ul->lock);
1535                 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1536                         if (rt->dst.dev != dev)
1537                                 continue;
1538                         rt->dst.dev = blackhole_netdev;
1539                         netdev_ref_replace(dev, blackhole_netdev,
1540                                            &rt->dst.dev_tracker, GFP_ATOMIC);
1541                         list_move(&rt->dst.rt_uncached, &ul->quarantine);
1542                 }
1543                 spin_unlock_bh(&ul->lock);
1544         }
1545 }
1546
1547 static bool rt_cache_valid(const struct rtable *rt)
1548 {
1549         return  rt &&
1550                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1551                 !rt_is_expired(rt);
1552 }
1553
1554 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1555                            const struct fib_result *res,
1556                            struct fib_nh_exception *fnhe,
1557                            struct fib_info *fi, u16 type, u32 itag,
1558                            const bool do_cache)
1559 {
1560         bool cached = false;
1561
1562         if (fi) {
1563                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1564
1565                 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1566                         rt->rt_uses_gateway = 1;
1567                         rt->rt_gw_family = nhc->nhc_gw_family;
1568                         /* only INET and INET6 are supported */
1569                         if (likely(nhc->nhc_gw_family == AF_INET))
1570                                 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1571                         else
1572                                 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1573                 }
1574
1575                 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1576
1577 #ifdef CONFIG_IP_ROUTE_CLASSID
1578                 if (nhc->nhc_family == AF_INET) {
1579                         struct fib_nh *nh;
1580
1581                         nh = container_of(nhc, struct fib_nh, nh_common);
1582                         rt->dst.tclassid = nh->nh_tclassid;
1583                 }
1584 #endif
1585                 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1586                 if (unlikely(fnhe))
1587                         cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1588                 else if (do_cache)
1589                         cached = rt_cache_route(nhc, rt);
1590                 if (unlikely(!cached)) {
1591                         /* Routes we intend to cache in nexthop exception or
1592                          * FIB nexthop have the DST_NOCACHE bit clear.
1593                          * However, if we are unsuccessful at storing this
1594                          * route into the cache we really need to set it.
1595                          */
1596                         if (!rt->rt_gw4) {
1597                                 rt->rt_gw_family = AF_INET;
1598                                 rt->rt_gw4 = daddr;
1599                         }
1600                         rt_add_uncached_list(rt);
1601                 }
1602         } else
1603                 rt_add_uncached_list(rt);
1604
1605 #ifdef CONFIG_IP_ROUTE_CLASSID
1606 #ifdef CONFIG_IP_MULTIPLE_TABLES
1607         set_class_tag(rt, res->tclassid);
1608 #endif
1609         set_class_tag(rt, itag);
1610 #endif
1611 }
1612
1613 struct rtable *rt_dst_alloc(struct net_device *dev,
1614                             unsigned int flags, u16 type,
1615                             bool noxfrm)
1616 {
1617         struct rtable *rt;
1618
1619         rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1620                        (noxfrm ? DST_NOXFRM : 0));
1621
1622         if (rt) {
1623                 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1624                 rt->rt_flags = flags;
1625                 rt->rt_type = type;
1626                 rt->rt_is_input = 0;
1627                 rt->rt_iif = 0;
1628                 rt->rt_pmtu = 0;
1629                 rt->rt_mtu_locked = 0;
1630                 rt->rt_uses_gateway = 0;
1631                 rt->rt_gw_family = 0;
1632                 rt->rt_gw4 = 0;
1633
1634                 rt->dst.output = ip_output;
1635                 if (flags & RTCF_LOCAL)
1636                         rt->dst.input = ip_local_deliver;
1637         }
1638
1639         return rt;
1640 }
1641 EXPORT_SYMBOL(rt_dst_alloc);
1642
1643 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1644 {
1645         struct rtable *new_rt;
1646
1647         new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1648                            rt->dst.flags);
1649
1650         if (new_rt) {
1651                 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1652                 new_rt->rt_flags = rt->rt_flags;
1653                 new_rt->rt_type = rt->rt_type;
1654                 new_rt->rt_is_input = rt->rt_is_input;
1655                 new_rt->rt_iif = rt->rt_iif;
1656                 new_rt->rt_pmtu = rt->rt_pmtu;
1657                 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1658                 new_rt->rt_gw_family = rt->rt_gw_family;
1659                 if (rt->rt_gw_family == AF_INET)
1660                         new_rt->rt_gw4 = rt->rt_gw4;
1661                 else if (rt->rt_gw_family == AF_INET6)
1662                         new_rt->rt_gw6 = rt->rt_gw6;
1663
1664                 new_rt->dst.input = rt->dst.input;
1665                 new_rt->dst.output = rt->dst.output;
1666                 new_rt->dst.error = rt->dst.error;
1667                 new_rt->dst.lastuse = jiffies;
1668                 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1669         }
1670         return new_rt;
1671 }
1672 EXPORT_SYMBOL(rt_dst_clone);
1673
1674 /* called in rcu_read_lock() section */
1675 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1676                           u8 tos, struct net_device *dev,
1677                           struct in_device *in_dev, u32 *itag)
1678 {
1679         int err;
1680
1681         /* Primary sanity checks. */
1682         if (!in_dev)
1683                 return -EINVAL;
1684
1685         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1686             skb->protocol != htons(ETH_P_IP))
1687                 return -EINVAL;
1688
1689         if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1690                 return -EINVAL;
1691
1692         if (ipv4_is_zeronet(saddr)) {
1693                 if (!ipv4_is_local_multicast(daddr) &&
1694                     ip_hdr(skb)->protocol != IPPROTO_IGMP)
1695                         return -EINVAL;
1696         } else {
1697                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1698                                           in_dev, itag);
1699                 if (err < 0)
1700                         return err;
1701         }
1702         return 0;
1703 }
1704
1705 /* called in rcu_read_lock() section */
1706 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1707                              u8 tos, struct net_device *dev, int our)
1708 {
1709         struct in_device *in_dev = __in_dev_get_rcu(dev);
1710         unsigned int flags = RTCF_MULTICAST;
1711         struct rtable *rth;
1712         u32 itag = 0;
1713         int err;
1714
1715         err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1716         if (err)
1717                 return err;
1718
1719         if (our)
1720                 flags |= RTCF_LOCAL;
1721
1722         if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1723                 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1724
1725         rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1726                            false);
1727         if (!rth)
1728                 return -ENOBUFS;
1729
1730 #ifdef CONFIG_IP_ROUTE_CLASSID
1731         rth->dst.tclassid = itag;
1732 #endif
1733         rth->dst.output = ip_rt_bug;
1734         rth->rt_is_input= 1;
1735
1736 #ifdef CONFIG_IP_MROUTE
1737         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1738                 rth->dst.input = ip_mr_input;
1739 #endif
1740         RT_CACHE_STAT_INC(in_slow_mc);
1741
1742         skb_dst_drop(skb);
1743         skb_dst_set(skb, &rth->dst);
1744         return 0;
1745 }
1746
1747
1748 static void ip_handle_martian_source(struct net_device *dev,
1749                                      struct in_device *in_dev,
1750                                      struct sk_buff *skb,
1751                                      __be32 daddr,
1752                                      __be32 saddr)
1753 {
1754         RT_CACHE_STAT_INC(in_martian_src);
1755 #ifdef CONFIG_IP_ROUTE_VERBOSE
1756         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1757                 /*
1758                  *      RFC1812 recommendation, if source is martian,
1759                  *      the only hint is MAC header.
1760                  */
1761                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1762                         &daddr, &saddr, dev->name);
1763                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1764                         print_hex_dump(KERN_WARNING, "ll header: ",
1765                                        DUMP_PREFIX_OFFSET, 16, 1,
1766                                        skb_mac_header(skb),
1767                                        dev->hard_header_len, false);
1768                 }
1769         }
1770 #endif
1771 }
1772
1773 /* called in rcu_read_lock() section */
1774 static int __mkroute_input(struct sk_buff *skb,
1775                            const struct fib_result *res,
1776                            struct in_device *in_dev,
1777                            __be32 daddr, __be32 saddr, u32 tos)
1778 {
1779         struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1780         struct net_device *dev = nhc->nhc_dev;
1781         struct fib_nh_exception *fnhe;
1782         struct rtable *rth;
1783         int err;
1784         struct in_device *out_dev;
1785         bool do_cache;
1786         u32 itag = 0;
1787
1788         /* get a working reference to the output device */
1789         out_dev = __in_dev_get_rcu(dev);
1790         if (!out_dev) {
1791                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1792                 return -EINVAL;
1793         }
1794
1795         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1796                                   in_dev->dev, in_dev, &itag);
1797         if (err < 0) {
1798                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1799                                          saddr);
1800
1801                 goto cleanup;
1802         }
1803
1804         do_cache = res->fi && !itag;
1805         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1806             skb->protocol == htons(ETH_P_IP)) {
1807                 __be32 gw;
1808
1809                 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1810                 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1811                     inet_addr_onlink(out_dev, saddr, gw))
1812                         IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1813         }
1814
1815         if (skb->protocol != htons(ETH_P_IP)) {
1816                 /* Not IP (i.e. ARP). Do not create route, if it is
1817                  * invalid for proxy arp. DNAT routes are always valid.
1818                  *
1819                  * Proxy arp feature have been extended to allow, ARP
1820                  * replies back to the same interface, to support
1821                  * Private VLAN switch technologies. See arp.c.
1822                  */
1823                 if (out_dev == in_dev &&
1824                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1825                         err = -EINVAL;
1826                         goto cleanup;
1827                 }
1828         }
1829
1830         if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1831                 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1832
1833         fnhe = find_exception(nhc, daddr);
1834         if (do_cache) {
1835                 if (fnhe)
1836                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1837                 else
1838                         rth = rcu_dereference(nhc->nhc_rth_input);
1839                 if (rt_cache_valid(rth)) {
1840                         skb_dst_set_noref(skb, &rth->dst);
1841                         goto out;
1842                 }
1843         }
1844
1845         rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1846                            IN_DEV_ORCONF(out_dev, NOXFRM));
1847         if (!rth) {
1848                 err = -ENOBUFS;
1849                 goto cleanup;
1850         }
1851
1852         rth->rt_is_input = 1;
1853         RT_CACHE_STAT_INC(in_slow_tot);
1854
1855         rth->dst.input = ip_forward;
1856
1857         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1858                        do_cache);
1859         lwtunnel_set_redirect(&rth->dst);
1860         skb_dst_set(skb, &rth->dst);
1861 out:
1862         err = 0;
1863  cleanup:
1864         return err;
1865 }
1866
1867 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1868 /* To make ICMP packets follow the right flow, the multipath hash is
1869  * calculated from the inner IP addresses.
1870  */
1871 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1872                                  struct flow_keys *hash_keys)
1873 {
1874         const struct iphdr *outer_iph = ip_hdr(skb);
1875         const struct iphdr *key_iph = outer_iph;
1876         const struct iphdr *inner_iph;
1877         const struct icmphdr *icmph;
1878         struct iphdr _inner_iph;
1879         struct icmphdr _icmph;
1880
1881         if (likely(outer_iph->protocol != IPPROTO_ICMP))
1882                 goto out;
1883
1884         if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1885                 goto out;
1886
1887         icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1888                                    &_icmph);
1889         if (!icmph)
1890                 goto out;
1891
1892         if (!icmp_is_err(icmph->type))
1893                 goto out;
1894
1895         inner_iph = skb_header_pointer(skb,
1896                                        outer_iph->ihl * 4 + sizeof(_icmph),
1897                                        sizeof(_inner_iph), &_inner_iph);
1898         if (!inner_iph)
1899                 goto out;
1900
1901         key_iph = inner_iph;
1902 out:
1903         hash_keys->addrs.v4addrs.src = key_iph->saddr;
1904         hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1905 }
1906
1907 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1908                                            const struct sk_buff *skb,
1909                                            bool *p_has_inner)
1910 {
1911         u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1912         struct flow_keys keys, hash_keys;
1913
1914         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1915                 return 0;
1916
1917         memset(&hash_keys, 0, sizeof(hash_keys));
1918         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1919
1920         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1921         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1922                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1923         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1924                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1925         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1926                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1927         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1928                 hash_keys.ports.src = keys.ports.src;
1929         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1930                 hash_keys.ports.dst = keys.ports.dst;
1931
1932         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1933         return flow_hash_from_keys(&hash_keys);
1934 }
1935
1936 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1937                                            const struct sk_buff *skb,
1938                                            bool has_inner)
1939 {
1940         u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1941         struct flow_keys keys, hash_keys;
1942
1943         /* We assume the packet carries an encapsulation, but if none was
1944          * encountered during dissection of the outer flow, then there is no
1945          * point in calling the flow dissector again.
1946          */
1947         if (!has_inner)
1948                 return 0;
1949
1950         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1951                 return 0;
1952
1953         memset(&hash_keys, 0, sizeof(hash_keys));
1954         skb_flow_dissect_flow_keys(skb, &keys, 0);
1955
1956         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1957                 return 0;
1958
1959         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1960                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1961                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1962                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1963                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1964                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1965         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1966                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1967                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1968                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1969                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1970                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1971                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1972                         hash_keys.tags.flow_label = keys.tags.flow_label;
1973         }
1974
1975         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1976                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1977         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1978                 hash_keys.ports.src = keys.ports.src;
1979         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1980                 hash_keys.ports.dst = keys.ports.dst;
1981
1982         return flow_hash_from_keys(&hash_keys);
1983 }
1984
1985 static u32 fib_multipath_custom_hash_skb(const struct net *net,
1986                                          const struct sk_buff *skb)
1987 {
1988         u32 mhash, mhash_inner;
1989         bool has_inner = true;
1990
1991         mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
1992         mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
1993
1994         return jhash_2words(mhash, mhash_inner, 0);
1995 }
1996
1997 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
1998                                          const struct flowi4 *fl4)
1999 {
2000         u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2001         struct flow_keys hash_keys;
2002
2003         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2004                 return 0;
2005
2006         memset(&hash_keys, 0, sizeof(hash_keys));
2007         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2008         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2009                 hash_keys.addrs.v4addrs.src = fl4->saddr;
2010         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2011                 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2012         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2013                 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2014         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2015                 hash_keys.ports.src = fl4->fl4_sport;
2016         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2017                 hash_keys.ports.dst = fl4->fl4_dport;
2018
2019         return flow_hash_from_keys(&hash_keys);
2020 }
2021
2022 /* if skb is set it will be used and fl4 can be NULL */
2023 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2024                        const struct sk_buff *skb, struct flow_keys *flkeys)
2025 {
2026         u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2027         struct flow_keys hash_keys;
2028         u32 mhash = 0;
2029
2030         switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2031         case 0:
2032                 memset(&hash_keys, 0, sizeof(hash_keys));
2033                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2034                 if (skb) {
2035                         ip_multipath_l3_keys(skb, &hash_keys);
2036                 } else {
2037                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2038                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2039                 }
2040                 mhash = flow_hash_from_keys(&hash_keys);
2041                 break;
2042         case 1:
2043                 /* skb is currently provided only when forwarding */
2044                 if (skb) {
2045                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2046                         struct flow_keys keys;
2047
2048                         /* short-circuit if we already have L4 hash present */
2049                         if (skb->l4_hash)
2050                                 return skb_get_hash_raw(skb) >> 1;
2051
2052                         memset(&hash_keys, 0, sizeof(hash_keys));
2053
2054                         if (!flkeys) {
2055                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2056                                 flkeys = &keys;
2057                         }
2058
2059                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2060                         hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2061                         hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2062                         hash_keys.ports.src = flkeys->ports.src;
2063                         hash_keys.ports.dst = flkeys->ports.dst;
2064                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2065                 } else {
2066                         memset(&hash_keys, 0, sizeof(hash_keys));
2067                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2068                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2069                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2070                         hash_keys.ports.src = fl4->fl4_sport;
2071                         hash_keys.ports.dst = fl4->fl4_dport;
2072                         hash_keys.basic.ip_proto = fl4->flowi4_proto;
2073                 }
2074                 mhash = flow_hash_from_keys(&hash_keys);
2075                 break;
2076         case 2:
2077                 memset(&hash_keys, 0, sizeof(hash_keys));
2078                 /* skb is currently provided only when forwarding */
2079                 if (skb) {
2080                         struct flow_keys keys;
2081
2082                         skb_flow_dissect_flow_keys(skb, &keys, 0);
2083                         /* Inner can be v4 or v6 */
2084                         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2085                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2086                                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2087                                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2088                         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2089                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2090                                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2091                                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2092                                 hash_keys.tags.flow_label = keys.tags.flow_label;
2093                                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2094                         } else {
2095                                 /* Same as case 0 */
2096                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2097                                 ip_multipath_l3_keys(skb, &hash_keys);
2098                         }
2099                 } else {
2100                         /* Same as case 0 */
2101                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2102                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2103                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2104                 }
2105                 mhash = flow_hash_from_keys(&hash_keys);
2106                 break;
2107         case 3:
2108                 if (skb)
2109                         mhash = fib_multipath_custom_hash_skb(net, skb);
2110                 else
2111                         mhash = fib_multipath_custom_hash_fl4(net, fl4);
2112                 break;
2113         }
2114
2115         if (multipath_hash)
2116                 mhash = jhash_2words(mhash, multipath_hash, 0);
2117
2118         return mhash >> 1;
2119 }
2120 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2121
2122 static int ip_mkroute_input(struct sk_buff *skb,
2123                             struct fib_result *res,
2124                             struct in_device *in_dev,
2125                             __be32 daddr, __be32 saddr, u32 tos,
2126                             struct flow_keys *hkeys)
2127 {
2128 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2129         if (res->fi && fib_info_num_path(res->fi) > 1) {
2130                 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2131
2132                 fib_select_multipath(res, h);
2133                 IPCB(skb)->flags |= IPSKB_MULTIPATH;
2134         }
2135 #endif
2136
2137         /* create a routing cache entry */
2138         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2139 }
2140
2141 /* Implements all the saddr-related checks as ip_route_input_slow(),
2142  * assuming daddr is valid and the destination is not a local broadcast one.
2143  * Uses the provided hint instead of performing a route lookup.
2144  */
2145 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2146                       u8 tos, struct net_device *dev,
2147                       const struct sk_buff *hint)
2148 {
2149         struct in_device *in_dev = __in_dev_get_rcu(dev);
2150         struct rtable *rt = skb_rtable(hint);
2151         struct net *net = dev_net(dev);
2152         int err = -EINVAL;
2153         u32 tag = 0;
2154
2155         if (!in_dev)
2156                 return -EINVAL;
2157
2158         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2159                 goto martian_source;
2160
2161         if (ipv4_is_zeronet(saddr))
2162                 goto martian_source;
2163
2164         if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2165                 goto martian_source;
2166
2167         if (rt->rt_type != RTN_LOCAL)
2168                 goto skip_validate_source;
2169
2170         tos &= IPTOS_RT_MASK;
2171         err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2172         if (err < 0)
2173                 goto martian_source;
2174
2175 skip_validate_source:
2176         skb_dst_copy(skb, hint);
2177         return 0;
2178
2179 martian_source:
2180         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2181         return err;
2182 }
2183
2184 /* get device for dst_alloc with local routes */
2185 static struct net_device *ip_rt_get_dev(struct net *net,
2186                                         const struct fib_result *res)
2187 {
2188         struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2189         struct net_device *dev = NULL;
2190
2191         if (nhc)
2192                 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2193
2194         return dev ? : net->loopback_dev;
2195 }
2196
2197 /*
2198  *      NOTE. We drop all the packets that has local source
2199  *      addresses, because every properly looped back packet
2200  *      must have correct destination already attached by output routine.
2201  *      Changes in the enforced policies must be applied also to
2202  *      ip_route_use_hint().
2203  *
2204  *      Such approach solves two big problems:
2205  *      1. Not simplex devices are handled properly.
2206  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2207  *      called with rcu_read_lock()
2208  */
2209
2210 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2211                                u8 tos, struct net_device *dev,
2212                                struct fib_result *res)
2213 {
2214         struct in_device *in_dev = __in_dev_get_rcu(dev);
2215         struct flow_keys *flkeys = NULL, _flkeys;
2216         struct net    *net = dev_net(dev);
2217         struct ip_tunnel_info *tun_info;
2218         int             err = -EINVAL;
2219         unsigned int    flags = 0;
2220         u32             itag = 0;
2221         struct rtable   *rth;
2222         struct flowi4   fl4;
2223         bool do_cache = true;
2224
2225         /* IP on this device is disabled. */
2226
2227         if (!in_dev)
2228                 goto out;
2229
2230         /* Check for the most weird martians, which can be not detected
2231          * by fib_lookup.
2232          */
2233
2234         tun_info = skb_tunnel_info(skb);
2235         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2236                 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2237         else
2238                 fl4.flowi4_tun_key.tun_id = 0;
2239         skb_dst_drop(skb);
2240
2241         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2242                 goto martian_source;
2243
2244         res->fi = NULL;
2245         res->table = NULL;
2246         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2247                 goto brd_input;
2248
2249         /* Accept zero addresses only to limited broadcast;
2250          * I even do not know to fix it or not. Waiting for complains :-)
2251          */
2252         if (ipv4_is_zeronet(saddr))
2253                 goto martian_source;
2254
2255         if (ipv4_is_zeronet(daddr))
2256                 goto martian_destination;
2257
2258         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2259          * and call it once if daddr or/and saddr are loopback addresses
2260          */
2261         if (ipv4_is_loopback(daddr)) {
2262                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2263                         goto martian_destination;
2264         } else if (ipv4_is_loopback(saddr)) {
2265                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2266                         goto martian_source;
2267         }
2268
2269         /*
2270          *      Now we are ready to route packet.
2271          */
2272         fl4.flowi4_l3mdev = 0;
2273         fl4.flowi4_oif = 0;
2274         fl4.flowi4_iif = dev->ifindex;
2275         fl4.flowi4_mark = skb->mark;
2276         fl4.flowi4_tos = tos;
2277         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2278         fl4.flowi4_flags = 0;
2279         fl4.daddr = daddr;
2280         fl4.saddr = saddr;
2281         fl4.flowi4_uid = sock_net_uid(net, NULL);
2282         fl4.flowi4_multipath_hash = 0;
2283
2284         if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2285                 flkeys = &_flkeys;
2286         } else {
2287                 fl4.flowi4_proto = 0;
2288                 fl4.fl4_sport = 0;
2289                 fl4.fl4_dport = 0;
2290         }
2291
2292         err = fib_lookup(net, &fl4, res, 0);
2293         if (err != 0) {
2294                 if (!IN_DEV_FORWARD(in_dev))
2295                         err = -EHOSTUNREACH;
2296                 goto no_route;
2297         }
2298
2299         if (res->type == RTN_BROADCAST) {
2300                 if (IN_DEV_BFORWARD(in_dev))
2301                         goto make_route;
2302                 /* not do cache if bc_forwarding is enabled */
2303                 if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
2304                         do_cache = false;
2305                 goto brd_input;
2306         }
2307
2308         if (res->type == RTN_LOCAL) {
2309                 err = fib_validate_source(skb, saddr, daddr, tos,
2310                                           0, dev, in_dev, &itag);
2311                 if (err < 0)
2312                         goto martian_source;
2313                 goto local_input;
2314         }
2315
2316         if (!IN_DEV_FORWARD(in_dev)) {
2317                 err = -EHOSTUNREACH;
2318                 goto no_route;
2319         }
2320         if (res->type != RTN_UNICAST)
2321                 goto martian_destination;
2322
2323 make_route:
2324         err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2325 out:    return err;
2326
2327 brd_input:
2328         if (skb->protocol != htons(ETH_P_IP))
2329                 goto e_inval;
2330
2331         if (!ipv4_is_zeronet(saddr)) {
2332                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2333                                           in_dev, &itag);
2334                 if (err < 0)
2335                         goto martian_source;
2336         }
2337         flags |= RTCF_BROADCAST;
2338         res->type = RTN_BROADCAST;
2339         RT_CACHE_STAT_INC(in_brd);
2340
2341 local_input:
2342         if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2343                 IPCB(skb)->flags |= IPSKB_NOPOLICY;
2344
2345         do_cache &= res->fi && !itag;
2346         if (do_cache) {
2347                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2348
2349                 rth = rcu_dereference(nhc->nhc_rth_input);
2350                 if (rt_cache_valid(rth)) {
2351                         skb_dst_set_noref(skb, &rth->dst);
2352                         err = 0;
2353                         goto out;
2354                 }
2355         }
2356
2357         rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2358                            flags | RTCF_LOCAL, res->type, false);
2359         if (!rth)
2360                 goto e_nobufs;
2361
2362         rth->dst.output= ip_rt_bug;
2363 #ifdef CONFIG_IP_ROUTE_CLASSID
2364         rth->dst.tclassid = itag;
2365 #endif
2366         rth->rt_is_input = 1;
2367
2368         RT_CACHE_STAT_INC(in_slow_tot);
2369         if (res->type == RTN_UNREACHABLE) {
2370                 rth->dst.input= ip_error;
2371                 rth->dst.error= -err;
2372                 rth->rt_flags   &= ~RTCF_LOCAL;
2373         }
2374
2375         if (do_cache) {
2376                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2377
2378                 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2379                 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2380                         WARN_ON(rth->dst.input == lwtunnel_input);
2381                         rth->dst.lwtstate->orig_input = rth->dst.input;
2382                         rth->dst.input = lwtunnel_input;
2383                 }
2384
2385                 if (unlikely(!rt_cache_route(nhc, rth)))
2386                         rt_add_uncached_list(rth);
2387         }
2388         skb_dst_set(skb, &rth->dst);
2389         err = 0;
2390         goto out;
2391
2392 no_route:
2393         RT_CACHE_STAT_INC(in_no_route);
2394         res->type = RTN_UNREACHABLE;
2395         res->fi = NULL;
2396         res->table = NULL;
2397         goto local_input;
2398
2399         /*
2400          *      Do not cache martian addresses: they should be logged (RFC1812)
2401          */
2402 martian_destination:
2403         RT_CACHE_STAT_INC(in_martian_dst);
2404 #ifdef CONFIG_IP_ROUTE_VERBOSE
2405         if (IN_DEV_LOG_MARTIANS(in_dev))
2406                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2407                                      &daddr, &saddr, dev->name);
2408 #endif
2409
2410 e_inval:
2411         err = -EINVAL;
2412         goto out;
2413
2414 e_nobufs:
2415         err = -ENOBUFS;
2416         goto out;
2417
2418 martian_source:
2419         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2420         goto out;
2421 }
2422
2423 /* called with rcu_read_lock held */
2424 static int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2425                               u8 tos, struct net_device *dev, struct fib_result *res)
2426 {
2427         /* Multicast recognition logic is moved from route cache to here.
2428          * The problem was that too many Ethernet cards have broken/missing
2429          * hardware multicast filters :-( As result the host on multicasting
2430          * network acquires a lot of useless route cache entries, sort of
2431          * SDR messages from all the world. Now we try to get rid of them.
2432          * Really, provided software IP multicast filter is organized
2433          * reasonably (at least, hashed), it does not result in a slowdown
2434          * comparing with route cache reject entries.
2435          * Note, that multicast routers are not affected, because
2436          * route cache entry is created eventually.
2437          */
2438         if (ipv4_is_multicast(daddr)) {
2439                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2440                 int our = 0;
2441                 int err = -EINVAL;
2442
2443                 if (!in_dev)
2444                         return err;
2445                 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2446                                       ip_hdr(skb)->protocol);
2447
2448                 /* check l3 master if no match yet */
2449                 if (!our && netif_is_l3_slave(dev)) {
2450                         struct in_device *l3_in_dev;
2451
2452                         l3_in_dev = __in_dev_get_rcu(skb->dev);
2453                         if (l3_in_dev)
2454                                 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2455                                                       ip_hdr(skb)->protocol);
2456                 }
2457
2458                 if (our
2459 #ifdef CONFIG_IP_MROUTE
2460                         ||
2461                     (!ipv4_is_local_multicast(daddr) &&
2462                      IN_DEV_MFORWARD(in_dev))
2463 #endif
2464                    ) {
2465                         err = ip_route_input_mc(skb, daddr, saddr,
2466                                                 tos, dev, our);
2467                 }
2468                 return err;
2469         }
2470
2471         return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2472 }
2473
2474 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2475                          u8 tos, struct net_device *dev)
2476 {
2477         struct fib_result res;
2478         int err;
2479
2480         tos &= IPTOS_RT_MASK;
2481         rcu_read_lock();
2482         err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2483         rcu_read_unlock();
2484
2485         return err;
2486 }
2487 EXPORT_SYMBOL(ip_route_input_noref);
2488
2489 /* called with rcu_read_lock() */
2490 static struct rtable *__mkroute_output(const struct fib_result *res,
2491                                        const struct flowi4 *fl4, int orig_oif,
2492                                        struct net_device *dev_out,
2493                                        unsigned int flags)
2494 {
2495         struct fib_info *fi = res->fi;
2496         struct fib_nh_exception *fnhe;
2497         struct in_device *in_dev;
2498         u16 type = res->type;
2499         struct rtable *rth;
2500         bool do_cache;
2501
2502         in_dev = __in_dev_get_rcu(dev_out);
2503         if (!in_dev)
2504                 return ERR_PTR(-EINVAL);
2505
2506         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2507                 if (ipv4_is_loopback(fl4->saddr) &&
2508                     !(dev_out->flags & IFF_LOOPBACK) &&
2509                     !netif_is_l3_master(dev_out))
2510                         return ERR_PTR(-EINVAL);
2511
2512         if (ipv4_is_lbcast(fl4->daddr))
2513                 type = RTN_BROADCAST;
2514         else if (ipv4_is_multicast(fl4->daddr))
2515                 type = RTN_MULTICAST;
2516         else if (ipv4_is_zeronet(fl4->daddr))
2517                 return ERR_PTR(-EINVAL);
2518
2519         if (dev_out->flags & IFF_LOOPBACK)
2520                 flags |= RTCF_LOCAL;
2521
2522         do_cache = true;
2523         if (type == RTN_BROADCAST) {
2524                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2525                 fi = NULL;
2526         } else if (type == RTN_MULTICAST) {
2527                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2528                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2529                                      fl4->flowi4_proto))
2530                         flags &= ~RTCF_LOCAL;
2531                 else
2532                         do_cache = false;
2533                 /* If multicast route do not exist use
2534                  * default one, but do not gateway in this case.
2535                  * Yes, it is hack.
2536                  */
2537                 if (fi && res->prefixlen < 4)
2538                         fi = NULL;
2539         } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2540                    (orig_oif != dev_out->ifindex)) {
2541                 /* For local routes that require a particular output interface
2542                  * we do not want to cache the result.  Caching the result
2543                  * causes incorrect behaviour when there are multiple source
2544                  * addresses on the interface, the end result being that if the
2545                  * intended recipient is waiting on that interface for the
2546                  * packet he won't receive it because it will be delivered on
2547                  * the loopback interface and the IP_PKTINFO ipi_ifindex will
2548                  * be set to the loopback interface as well.
2549                  */
2550                 do_cache = false;
2551         }
2552
2553         fnhe = NULL;
2554         do_cache &= fi != NULL;
2555         if (fi) {
2556                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2557                 struct rtable __rcu **prth;
2558
2559                 fnhe = find_exception(nhc, fl4->daddr);
2560                 if (!do_cache)
2561                         goto add;
2562                 if (fnhe) {
2563                         prth = &fnhe->fnhe_rth_output;
2564                 } else {
2565                         if (unlikely(fl4->flowi4_flags &
2566                                      FLOWI_FLAG_KNOWN_NH &&
2567                                      !(nhc->nhc_gw_family &&
2568                                        nhc->nhc_scope == RT_SCOPE_LINK))) {
2569                                 do_cache = false;
2570                                 goto add;
2571                         }
2572                         prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2573                 }
2574                 rth = rcu_dereference(*prth);
2575                 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2576                         return rth;
2577         }
2578
2579 add:
2580         rth = rt_dst_alloc(dev_out, flags, type,
2581                            IN_DEV_ORCONF(in_dev, NOXFRM));
2582         if (!rth)
2583                 return ERR_PTR(-ENOBUFS);
2584
2585         rth->rt_iif = orig_oif;
2586
2587         RT_CACHE_STAT_INC(out_slow_tot);
2588
2589         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2590                 if (flags & RTCF_LOCAL &&
2591                     !(dev_out->flags & IFF_LOOPBACK)) {
2592                         rth->dst.output = ip_mc_output;
2593                         RT_CACHE_STAT_INC(out_slow_mc);
2594                 }
2595 #ifdef CONFIG_IP_MROUTE
2596                 if (type == RTN_MULTICAST) {
2597                         if (IN_DEV_MFORWARD(in_dev) &&
2598                             !ipv4_is_local_multicast(fl4->daddr)) {
2599                                 rth->dst.input = ip_mr_input;
2600                                 rth->dst.output = ip_mc_output;
2601                         }
2602                 }
2603 #endif
2604         }
2605
2606         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2607         lwtunnel_set_redirect(&rth->dst);
2608
2609         return rth;
2610 }
2611
2612 /*
2613  * Major route resolver routine.
2614  */
2615
2616 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2617                                         const struct sk_buff *skb)
2618 {
2619         struct fib_result res = {
2620                 .type           = RTN_UNSPEC,
2621                 .fi             = NULL,
2622                 .table          = NULL,
2623                 .tclassid       = 0,
2624         };
2625         struct rtable *rth;
2626
2627         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2628         fl4->flowi4_tos &= IPTOS_RT_MASK;
2629
2630         rcu_read_lock();
2631         rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2632         rcu_read_unlock();
2633
2634         return rth;
2635 }
2636 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2637
2638 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2639                                             struct fib_result *res,
2640                                             const struct sk_buff *skb)
2641 {
2642         struct net_device *dev_out = NULL;
2643         int orig_oif = fl4->flowi4_oif;
2644         unsigned int flags = 0;
2645         struct rtable *rth;
2646         int err;
2647
2648         if (fl4->saddr) {
2649                 if (ipv4_is_multicast(fl4->saddr) ||
2650                     ipv4_is_lbcast(fl4->saddr) ||
2651                     ipv4_is_zeronet(fl4->saddr)) {
2652                         rth = ERR_PTR(-EINVAL);
2653                         goto out;
2654                 }
2655
2656                 rth = ERR_PTR(-ENETUNREACH);
2657
2658                 /* I removed check for oif == dev_out->oif here.
2659                  * It was wrong for two reasons:
2660                  * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2661                  *    is assigned to multiple interfaces.
2662                  * 2. Moreover, we are allowed to send packets with saddr
2663                  *    of another iface. --ANK
2664                  */
2665
2666                 if (fl4->flowi4_oif == 0 &&
2667                     (ipv4_is_multicast(fl4->daddr) ||
2668                      ipv4_is_lbcast(fl4->daddr))) {
2669                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2670                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2671                         if (!dev_out)
2672                                 goto out;
2673
2674                         /* Special hack: user can direct multicasts
2675                          * and limited broadcast via necessary interface
2676                          * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2677                          * This hack is not just for fun, it allows
2678                          * vic,vat and friends to work.
2679                          * They bind socket to loopback, set ttl to zero
2680                          * and expect that it will work.
2681                          * From the viewpoint of routing cache they are broken,
2682                          * because we are not allowed to build multicast path
2683                          * with loopback source addr (look, routing cache
2684                          * cannot know, that ttl is zero, so that packet
2685                          * will not leave this host and route is valid).
2686                          * Luckily, this hack is good workaround.
2687                          */
2688
2689                         fl4->flowi4_oif = dev_out->ifindex;
2690                         goto make_route;
2691                 }
2692
2693                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2694                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2695                         if (!__ip_dev_find(net, fl4->saddr, false))
2696                                 goto out;
2697                 }
2698         }
2699
2700
2701         if (fl4->flowi4_oif) {
2702                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2703                 rth = ERR_PTR(-ENODEV);
2704                 if (!dev_out)
2705                         goto out;
2706
2707                 /* RACE: Check return value of inet_select_addr instead. */
2708                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2709                         rth = ERR_PTR(-ENETUNREACH);
2710                         goto out;
2711                 }
2712                 if (ipv4_is_local_multicast(fl4->daddr) ||
2713                     ipv4_is_lbcast(fl4->daddr) ||
2714                     fl4->flowi4_proto == IPPROTO_IGMP) {
2715                         if (!fl4->saddr)
2716                                 fl4->saddr = inet_select_addr(dev_out, 0,
2717                                                               RT_SCOPE_LINK);
2718                         goto make_route;
2719                 }
2720                 if (!fl4->saddr) {
2721                         if (ipv4_is_multicast(fl4->daddr))
2722                                 fl4->saddr = inet_select_addr(dev_out, 0,
2723                                                               fl4->flowi4_scope);
2724                         else if (!fl4->daddr)
2725                                 fl4->saddr = inet_select_addr(dev_out, 0,
2726                                                               RT_SCOPE_HOST);
2727                 }
2728         }
2729
2730         if (!fl4->daddr) {
2731                 fl4->daddr = fl4->saddr;
2732                 if (!fl4->daddr)
2733                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2734                 dev_out = net->loopback_dev;
2735                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2736                 res->type = RTN_LOCAL;
2737                 flags |= RTCF_LOCAL;
2738                 goto make_route;
2739         }
2740
2741         err = fib_lookup(net, fl4, res, 0);
2742         if (err) {
2743                 res->fi = NULL;
2744                 res->table = NULL;
2745                 if (fl4->flowi4_oif &&
2746                     (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2747                         /* Apparently, routing tables are wrong. Assume,
2748                          * that the destination is on link.
2749                          *
2750                          * WHY? DW.
2751                          * Because we are allowed to send to iface
2752                          * even if it has NO routes and NO assigned
2753                          * addresses. When oif is specified, routing
2754                          * tables are looked up with only one purpose:
2755                          * to catch if destination is gatewayed, rather than
2756                          * direct. Moreover, if MSG_DONTROUTE is set,
2757                          * we send packet, ignoring both routing tables
2758                          * and ifaddr state. --ANK
2759                          *
2760                          *
2761                          * We could make it even if oif is unknown,
2762                          * likely IPv6, but we do not.
2763                          */
2764
2765                         if (fl4->saddr == 0)
2766                                 fl4->saddr = inet_select_addr(dev_out, 0,
2767                                                               RT_SCOPE_LINK);
2768                         res->type = RTN_UNICAST;
2769                         goto make_route;
2770                 }
2771                 rth = ERR_PTR(err);
2772                 goto out;
2773         }
2774
2775         if (res->type == RTN_LOCAL) {
2776                 if (!fl4->saddr) {
2777                         if (res->fi->fib_prefsrc)
2778                                 fl4->saddr = res->fi->fib_prefsrc;
2779                         else
2780                                 fl4->saddr = fl4->daddr;
2781                 }
2782
2783                 /* L3 master device is the loopback for that domain */
2784                 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2785                         net->loopback_dev;
2786
2787                 /* make sure orig_oif points to fib result device even
2788                  * though packet rx/tx happens over loopback or l3mdev
2789                  */
2790                 orig_oif = FIB_RES_OIF(*res);
2791
2792                 fl4->flowi4_oif = dev_out->ifindex;
2793                 flags |= RTCF_LOCAL;
2794                 goto make_route;
2795         }
2796
2797         fib_select_path(net, res, fl4, skb);
2798
2799         dev_out = FIB_RES_DEV(*res);
2800
2801 make_route:
2802         rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2803
2804 out:
2805         return rth;
2806 }
2807
2808 static struct dst_ops ipv4_dst_blackhole_ops = {
2809         .family                 = AF_INET,
2810         .default_advmss         = ipv4_default_advmss,
2811         .neigh_lookup           = ipv4_neigh_lookup,
2812         .check                  = dst_blackhole_check,
2813         .cow_metrics            = dst_blackhole_cow_metrics,
2814         .update_pmtu            = dst_blackhole_update_pmtu,
2815         .redirect               = dst_blackhole_redirect,
2816         .mtu                    = dst_blackhole_mtu,
2817 };
2818
2819 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2820 {
2821         struct rtable *ort = dst_rtable(dst_orig);
2822         struct rtable *rt;
2823
2824         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2825         if (rt) {
2826                 struct dst_entry *new = &rt->dst;
2827
2828                 new->__use = 1;
2829                 new->input = dst_discard;
2830                 new->output = dst_discard_out;
2831
2832                 new->dev = net->loopback_dev;
2833                 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2834
2835                 rt->rt_is_input = ort->rt_is_input;
2836                 rt->rt_iif = ort->rt_iif;
2837                 rt->rt_pmtu = ort->rt_pmtu;
2838                 rt->rt_mtu_locked = ort->rt_mtu_locked;
2839
2840                 rt->rt_genid = rt_genid_ipv4(net);
2841                 rt->rt_flags = ort->rt_flags;
2842                 rt->rt_type = ort->rt_type;
2843                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2844                 rt->rt_gw_family = ort->rt_gw_family;
2845                 if (rt->rt_gw_family == AF_INET)
2846                         rt->rt_gw4 = ort->rt_gw4;
2847                 else if (rt->rt_gw_family == AF_INET6)
2848                         rt->rt_gw6 = ort->rt_gw6;
2849         }
2850
2851         dst_release(dst_orig);
2852
2853         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2854 }
2855
2856 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2857                                     const struct sock *sk)
2858 {
2859         struct rtable *rt = __ip_route_output_key(net, flp4);
2860
2861         if (IS_ERR(rt))
2862                 return rt;
2863
2864         if (flp4->flowi4_proto) {
2865                 flp4->flowi4_oif = rt->dst.dev->ifindex;
2866                 rt = dst_rtable(xfrm_lookup_route(net, &rt->dst,
2867                                                   flowi4_to_flowi(flp4),
2868                                                   sk, 0));
2869         }
2870
2871         return rt;
2872 }
2873 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2874
2875 /* called with rcu_read_lock held */
2876 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2877                         struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2878                         struct sk_buff *skb, u32 portid, u32 seq,
2879                         unsigned int flags)
2880 {
2881         struct rtmsg *r;
2882         struct nlmsghdr *nlh;
2883         unsigned long expires = 0;
2884         u32 error;
2885         u32 metrics[RTAX_MAX];
2886
2887         nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2888         if (!nlh)
2889                 return -EMSGSIZE;
2890
2891         r = nlmsg_data(nlh);
2892         r->rtm_family    = AF_INET;
2893         r->rtm_dst_len  = 32;
2894         r->rtm_src_len  = 0;
2895         r->rtm_tos      = fl4 ? fl4->flowi4_tos : 0;
2896         r->rtm_table    = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2897         if (nla_put_u32(skb, RTA_TABLE, table_id))
2898                 goto nla_put_failure;
2899         r->rtm_type     = rt->rt_type;
2900         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2901         r->rtm_protocol = RTPROT_UNSPEC;
2902         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2903         if (rt->rt_flags & RTCF_NOTIFY)
2904                 r->rtm_flags |= RTM_F_NOTIFY;
2905         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2906                 r->rtm_flags |= RTCF_DOREDIRECT;
2907
2908         if (nla_put_in_addr(skb, RTA_DST, dst))
2909                 goto nla_put_failure;
2910         if (src) {
2911                 r->rtm_src_len = 32;
2912                 if (nla_put_in_addr(skb, RTA_SRC, src))
2913                         goto nla_put_failure;
2914         }
2915         if (rt->dst.dev &&
2916             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2917                 goto nla_put_failure;
2918         if (rt->dst.lwtstate &&
2919             lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2920                 goto nla_put_failure;
2921 #ifdef CONFIG_IP_ROUTE_CLASSID
2922         if (rt->dst.tclassid &&
2923             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2924                 goto nla_put_failure;
2925 #endif
2926         if (fl4 && !rt_is_input_route(rt) &&
2927             fl4->saddr != src) {
2928                 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2929                         goto nla_put_failure;
2930         }
2931         if (rt->rt_uses_gateway) {
2932                 if (rt->rt_gw_family == AF_INET &&
2933                     nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2934                         goto nla_put_failure;
2935                 } else if (rt->rt_gw_family == AF_INET6) {
2936                         int alen = sizeof(struct in6_addr);
2937                         struct nlattr *nla;
2938                         struct rtvia *via;
2939
2940                         nla = nla_reserve(skb, RTA_VIA, alen + 2);
2941                         if (!nla)
2942                                 goto nla_put_failure;
2943
2944                         via = nla_data(nla);
2945                         via->rtvia_family = AF_INET6;
2946                         memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2947                 }
2948         }
2949
2950         expires = rt->dst.expires;
2951         if (expires) {
2952                 unsigned long now = jiffies;
2953
2954                 if (time_before(now, expires))
2955                         expires -= now;
2956                 else
2957                         expires = 0;
2958         }
2959
2960         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2961         if (rt->rt_pmtu && expires)
2962                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2963         if (rt->rt_mtu_locked && expires)
2964                 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2965         if (rtnetlink_put_metrics(skb, metrics) < 0)
2966                 goto nla_put_failure;
2967
2968         if (fl4) {
2969                 if (fl4->flowi4_mark &&
2970                     nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2971                         goto nla_put_failure;
2972
2973                 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2974                     nla_put_u32(skb, RTA_UID,
2975                                 from_kuid_munged(current_user_ns(),
2976                                                  fl4->flowi4_uid)))
2977                         goto nla_put_failure;
2978
2979                 if (rt_is_input_route(rt)) {
2980 #ifdef CONFIG_IP_MROUTE
2981                         if (ipv4_is_multicast(dst) &&
2982                             !ipv4_is_local_multicast(dst) &&
2983                             IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
2984                                 int err = ipmr_get_route(net, skb,
2985                                                          fl4->saddr, fl4->daddr,
2986                                                          r, portid);
2987
2988                                 if (err <= 0) {
2989                                         if (err == 0)
2990                                                 return 0;
2991                                         goto nla_put_failure;
2992                                 }
2993                         } else
2994 #endif
2995                                 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
2996                                         goto nla_put_failure;
2997                 }
2998         }
2999
3000         error = rt->dst.error;
3001
3002         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3003                 goto nla_put_failure;
3004
3005         nlmsg_end(skb, nlh);
3006         return 0;
3007
3008 nla_put_failure:
3009         nlmsg_cancel(skb, nlh);
3010         return -EMSGSIZE;
3011 }
3012
3013 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3014                             struct netlink_callback *cb, u32 table_id,
3015                             struct fnhe_hash_bucket *bucket, int genid,
3016                             int *fa_index, int fa_start, unsigned int flags)
3017 {
3018         int i;
3019
3020         for (i = 0; i < FNHE_HASH_SIZE; i++) {
3021                 struct fib_nh_exception *fnhe;
3022
3023                 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3024                      fnhe = rcu_dereference(fnhe->fnhe_next)) {
3025                         struct rtable *rt;
3026                         int err;
3027
3028                         if (*fa_index < fa_start)
3029                                 goto next;
3030
3031                         if (fnhe->fnhe_genid != genid)
3032                                 goto next;
3033
3034                         if (fnhe->fnhe_expires &&
3035                             time_after(jiffies, fnhe->fnhe_expires))
3036                                 goto next;
3037
3038                         rt = rcu_dereference(fnhe->fnhe_rth_input);
3039                         if (!rt)
3040                                 rt = rcu_dereference(fnhe->fnhe_rth_output);
3041                         if (!rt)
3042                                 goto next;
3043
3044                         err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3045                                            table_id, NULL, skb,
3046                                            NETLINK_CB(cb->skb).portid,
3047                                            cb->nlh->nlmsg_seq, flags);
3048                         if (err)
3049                                 return err;
3050 next:
3051                         (*fa_index)++;
3052                 }
3053         }
3054
3055         return 0;
3056 }
3057
3058 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3059                        u32 table_id, struct fib_info *fi,
3060                        int *fa_index, int fa_start, unsigned int flags)
3061 {
3062         struct net *net = sock_net(cb->skb->sk);
3063         int nhsel, genid = fnhe_genid(net);
3064
3065         for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3066                 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3067                 struct fnhe_hash_bucket *bucket;
3068                 int err;
3069
3070                 if (nhc->nhc_flags & RTNH_F_DEAD)
3071                         continue;
3072
3073                 rcu_read_lock();
3074                 bucket = rcu_dereference(nhc->nhc_exceptions);
3075                 err = 0;
3076                 if (bucket)
3077                         err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3078                                                genid, fa_index, fa_start,
3079                                                flags);
3080                 rcu_read_unlock();
3081                 if (err)
3082                         return err;
3083         }
3084
3085         return 0;
3086 }
3087
3088 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3089                                                    u8 ip_proto, __be16 sport,
3090                                                    __be16 dport)
3091 {
3092         struct sk_buff *skb;
3093         struct iphdr *iph;
3094
3095         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3096         if (!skb)
3097                 return NULL;
3098
3099         /* Reserve room for dummy headers, this skb can pass
3100          * through good chunk of routing engine.
3101          */
3102         skb_reset_mac_header(skb);
3103         skb_reset_network_header(skb);
3104         skb->protocol = htons(ETH_P_IP);
3105         iph = skb_put(skb, sizeof(struct iphdr));
3106         iph->protocol = ip_proto;
3107         iph->saddr = src;
3108         iph->daddr = dst;
3109         iph->version = 0x4;
3110         iph->frag_off = 0;
3111         iph->ihl = 0x5;
3112         skb_set_transport_header(skb, skb->len);
3113
3114         switch (iph->protocol) {
3115         case IPPROTO_UDP: {
3116                 struct udphdr *udph;
3117
3118                 udph = skb_put_zero(skb, sizeof(struct udphdr));
3119                 udph->source = sport;
3120                 udph->dest = dport;
3121                 udph->len = htons(sizeof(struct udphdr));
3122                 udph->check = 0;
3123                 break;
3124         }
3125         case IPPROTO_TCP: {
3126                 struct tcphdr *tcph;
3127
3128                 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3129                 tcph->source    = sport;
3130                 tcph->dest      = dport;
3131                 tcph->doff      = sizeof(struct tcphdr) / 4;
3132                 tcph->rst = 1;
3133                 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3134                                             src, dst, 0);
3135                 break;
3136         }
3137         case IPPROTO_ICMP: {
3138                 struct icmphdr *icmph;
3139
3140                 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3141                 icmph->type = ICMP_ECHO;
3142                 icmph->code = 0;
3143         }
3144         }
3145
3146         return skb;
3147 }
3148
3149 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3150                                        const struct nlmsghdr *nlh,
3151                                        struct nlattr **tb,
3152                                        struct netlink_ext_ack *extack)
3153 {
3154         struct rtmsg *rtm;
3155         int i, err;
3156
3157         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3158                 NL_SET_ERR_MSG(extack,
3159                                "ipv4: Invalid header for route get request");
3160                 return -EINVAL;
3161         }
3162
3163         if (!netlink_strict_get_check(skb))
3164                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3165                                               rtm_ipv4_policy, extack);
3166
3167         rtm = nlmsg_data(nlh);
3168         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3169             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3170             rtm->rtm_table || rtm->rtm_protocol ||
3171             rtm->rtm_scope || rtm->rtm_type) {
3172                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3173                 return -EINVAL;
3174         }
3175
3176         if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3177                                RTM_F_LOOKUP_TABLE |
3178                                RTM_F_FIB_MATCH)) {
3179                 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3180                 return -EINVAL;
3181         }
3182
3183         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3184                                             rtm_ipv4_policy, extack);
3185         if (err)
3186                 return err;
3187
3188         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3189             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3190                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3191                 return -EINVAL;
3192         }
3193
3194         for (i = 0; i <= RTA_MAX; i++) {
3195                 if (!tb[i])
3196                         continue;
3197
3198                 switch (i) {
3199                 case RTA_IIF:
3200                 case RTA_OIF:
3201                 case RTA_SRC:
3202                 case RTA_DST:
3203                 case RTA_IP_PROTO:
3204                 case RTA_SPORT:
3205                 case RTA_DPORT:
3206                 case RTA_MARK:
3207                 case RTA_UID:
3208                         break;
3209                 default:
3210                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3211                         return -EINVAL;
3212                 }
3213         }
3214
3215         return 0;
3216 }
3217
3218 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3219                              struct netlink_ext_ack *extack)
3220 {
3221         struct net *net = sock_net(in_skb->sk);
3222         struct nlattr *tb[RTA_MAX+1];
3223         u32 table_id = RT_TABLE_MAIN;
3224         __be16 sport = 0, dport = 0;
3225         struct fib_result res = {};
3226         u8 ip_proto = IPPROTO_UDP;
3227         struct rtable *rt = NULL;
3228         struct sk_buff *skb;
3229         struct rtmsg *rtm;
3230         struct flowi4 fl4 = {};
3231         __be32 dst = 0;
3232         __be32 src = 0;
3233         kuid_t uid;
3234         u32 iif;
3235         int err;
3236         int mark;
3237
3238         err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3239         if (err < 0)
3240                 return err;
3241
3242         rtm = nlmsg_data(nlh);
3243         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3244         dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3245         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3246         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3247         if (tb[RTA_UID])
3248                 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3249         else
3250                 uid = (iif ? INVALID_UID : current_uid());
3251
3252         if (tb[RTA_IP_PROTO]) {
3253                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3254                                                   &ip_proto, AF_INET, extack);
3255                 if (err)
3256                         return err;
3257         }
3258
3259         if (tb[RTA_SPORT])
3260                 sport = nla_get_be16(tb[RTA_SPORT]);
3261
3262         if (tb[RTA_DPORT])
3263                 dport = nla_get_be16(tb[RTA_DPORT]);
3264
3265         skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3266         if (!skb)
3267                 return -ENOBUFS;
3268
3269         fl4.daddr = dst;
3270         fl4.saddr = src;
3271         fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3272         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3273         fl4.flowi4_mark = mark;
3274         fl4.flowi4_uid = uid;
3275         if (sport)
3276                 fl4.fl4_sport = sport;
3277         if (dport)
3278                 fl4.fl4_dport = dport;
3279         fl4.flowi4_proto = ip_proto;
3280
3281         rcu_read_lock();
3282
3283         if (iif) {
3284                 struct net_device *dev;
3285
3286                 dev = dev_get_by_index_rcu(net, iif);
3287                 if (!dev) {
3288                         err = -ENODEV;
3289                         goto errout_rcu;
3290                 }
3291
3292                 fl4.flowi4_iif = iif; /* for rt_fill_info */
3293                 skb->dev        = dev;
3294                 skb->mark       = mark;
3295                 err = ip_route_input_rcu(skb, dst, src,
3296                                          rtm->rtm_tos & IPTOS_RT_MASK, dev,
3297                                          &res);
3298
3299                 rt = skb_rtable(skb);
3300                 if (err == 0 && rt->dst.error)
3301                         err = -rt->dst.error;
3302         } else {
3303                 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3304                 skb->dev = net->loopback_dev;
3305                 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3306                 err = 0;
3307                 if (IS_ERR(rt))
3308                         err = PTR_ERR(rt);
3309                 else
3310                         skb_dst_set(skb, &rt->dst);
3311         }
3312
3313         if (err)
3314                 goto errout_rcu;
3315
3316         if (rtm->rtm_flags & RTM_F_NOTIFY)
3317                 rt->rt_flags |= RTCF_NOTIFY;
3318
3319         if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3320                 table_id = res.table ? res.table->tb_id : 0;
3321
3322         /* reset skb for netlink reply msg */
3323         skb_trim(skb, 0);
3324         skb_reset_network_header(skb);
3325         skb_reset_transport_header(skb);
3326         skb_reset_mac_header(skb);
3327
3328         if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3329                 struct fib_rt_info fri;
3330
3331                 if (!res.fi) {
3332                         err = fib_props[res.type].error;
3333                         if (!err)
3334                                 err = -EHOSTUNREACH;
3335                         goto errout_rcu;
3336                 }
3337                 fri.fi = res.fi;
3338                 fri.tb_id = table_id;
3339                 fri.dst = res.prefix;
3340                 fri.dst_len = res.prefixlen;
3341                 fri.dscp = inet_dsfield_to_dscp(fl4.flowi4_tos);
3342                 fri.type = rt->rt_type;
3343                 fri.offload = 0;
3344                 fri.trap = 0;
3345                 fri.offload_failed = 0;
3346                 if (res.fa_head) {
3347                         struct fib_alias *fa;
3348
3349                         hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3350                                 u8 slen = 32 - fri.dst_len;
3351
3352                                 if (fa->fa_slen == slen &&
3353                                     fa->tb_id == fri.tb_id &&
3354                                     fa->fa_dscp == fri.dscp &&
3355                                     fa->fa_info == res.fi &&
3356                                     fa->fa_type == fri.type) {
3357                                         fri.offload = READ_ONCE(fa->offload);
3358                                         fri.trap = READ_ONCE(fa->trap);
3359                                         fri.offload_failed =
3360                                                 READ_ONCE(fa->offload_failed);
3361                                         break;
3362                                 }
3363                         }
3364                 }
3365                 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3366                                     nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3367         } else {
3368                 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3369                                    NETLINK_CB(in_skb).portid,
3370                                    nlh->nlmsg_seq, 0);
3371         }
3372         if (err < 0)
3373                 goto errout_rcu;
3374
3375         rcu_read_unlock();
3376
3377         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3378
3379 errout_free:
3380         return err;
3381 errout_rcu:
3382         rcu_read_unlock();
3383         kfree_skb(skb);
3384         goto errout_free;
3385 }
3386
3387 void ip_rt_multicast_event(struct in_device *in_dev)
3388 {
3389         rt_cache_flush(dev_net(in_dev->dev));
3390 }
3391
3392 #ifdef CONFIG_SYSCTL
3393 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3394 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
3395 static int ip_rt_gc_elasticity __read_mostly    = 8;
3396 static int ip_min_valid_pmtu __read_mostly      = IPV4_MIN_MTU;
3397
3398 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3399                 void *buffer, size_t *lenp, loff_t *ppos)
3400 {
3401         struct net *net = (struct net *)__ctl->extra1;
3402
3403         if (write) {
3404                 rt_cache_flush(net);
3405                 fnhe_genid_bump(net);
3406                 return 0;
3407         }
3408
3409         return -EINVAL;
3410 }
3411
3412 static struct ctl_table ipv4_route_table[] = {
3413         {
3414                 .procname       = "gc_thresh",
3415                 .data           = &ipv4_dst_ops.gc_thresh,
3416                 .maxlen         = sizeof(int),
3417                 .mode           = 0644,
3418                 .proc_handler   = proc_dointvec,
3419         },
3420         {
3421                 .procname       = "max_size",
3422                 .data           = &ip_rt_max_size,
3423                 .maxlen         = sizeof(int),
3424                 .mode           = 0644,
3425                 .proc_handler   = proc_dointvec,
3426         },
3427         {
3428                 /*  Deprecated. Use gc_min_interval_ms */
3429
3430                 .procname       = "gc_min_interval",
3431                 .data           = &ip_rt_gc_min_interval,
3432                 .maxlen         = sizeof(int),
3433                 .mode           = 0644,
3434                 .proc_handler   = proc_dointvec_jiffies,
3435         },
3436         {
3437                 .procname       = "gc_min_interval_ms",
3438                 .data           = &ip_rt_gc_min_interval,
3439                 .maxlen         = sizeof(int),
3440                 .mode           = 0644,
3441                 .proc_handler   = proc_dointvec_ms_jiffies,
3442         },
3443         {
3444                 .procname       = "gc_timeout",
3445                 .data           = &ip_rt_gc_timeout,
3446                 .maxlen         = sizeof(int),
3447                 .mode           = 0644,
3448                 .proc_handler   = proc_dointvec_jiffies,
3449         },
3450         {
3451                 .procname       = "gc_interval",
3452                 .data           = &ip_rt_gc_interval,
3453                 .maxlen         = sizeof(int),
3454                 .mode           = 0644,
3455                 .proc_handler   = proc_dointvec_jiffies,
3456         },
3457         {
3458                 .procname       = "redirect_load",
3459                 .data           = &ip_rt_redirect_load,
3460                 .maxlen         = sizeof(int),
3461                 .mode           = 0644,
3462                 .proc_handler   = proc_dointvec,
3463         },
3464         {
3465                 .procname       = "redirect_number",
3466                 .data           = &ip_rt_redirect_number,
3467                 .maxlen         = sizeof(int),
3468                 .mode           = 0644,
3469                 .proc_handler   = proc_dointvec,
3470         },
3471         {
3472                 .procname       = "redirect_silence",
3473                 .data           = &ip_rt_redirect_silence,
3474                 .maxlen         = sizeof(int),
3475                 .mode           = 0644,
3476                 .proc_handler   = proc_dointvec,
3477         },
3478         {
3479                 .procname       = "error_cost",
3480                 .data           = &ip_rt_error_cost,
3481                 .maxlen         = sizeof(int),
3482                 .mode           = 0644,
3483                 .proc_handler   = proc_dointvec,
3484         },
3485         {
3486                 .procname       = "error_burst",
3487                 .data           = &ip_rt_error_burst,
3488                 .maxlen         = sizeof(int),
3489                 .mode           = 0644,
3490                 .proc_handler   = proc_dointvec,
3491         },
3492         {
3493                 .procname       = "gc_elasticity",
3494                 .data           = &ip_rt_gc_elasticity,
3495                 .maxlen         = sizeof(int),
3496                 .mode           = 0644,
3497                 .proc_handler   = proc_dointvec,
3498         },
3499 };
3500
3501 static const char ipv4_route_flush_procname[] = "flush";
3502
3503 static struct ctl_table ipv4_route_netns_table[] = {
3504         {
3505                 .procname       = ipv4_route_flush_procname,
3506                 .maxlen         = sizeof(int),
3507                 .mode           = 0200,
3508                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3509         },
3510         {
3511                 .procname       = "min_pmtu",
3512                 .data           = &init_net.ipv4.ip_rt_min_pmtu,
3513                 .maxlen         = sizeof(int),
3514                 .mode           = 0644,
3515                 .proc_handler   = proc_dointvec_minmax,
3516                 .extra1         = &ip_min_valid_pmtu,
3517         },
3518         {
3519                 .procname       = "mtu_expires",
3520                 .data           = &init_net.ipv4.ip_rt_mtu_expires,
3521                 .maxlen         = sizeof(int),
3522                 .mode           = 0644,
3523                 .proc_handler   = proc_dointvec_jiffies,
3524         },
3525         {
3526                 .procname   = "min_adv_mss",
3527                 .data       = &init_net.ipv4.ip_rt_min_advmss,
3528                 .maxlen     = sizeof(int),
3529                 .mode       = 0644,
3530                 .proc_handler   = proc_dointvec,
3531         },
3532 };
3533
3534 static __net_init int sysctl_route_net_init(struct net *net)
3535 {
3536         struct ctl_table *tbl;
3537         size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3538
3539         tbl = ipv4_route_netns_table;
3540         if (!net_eq(net, &init_net)) {
3541                 int i;
3542
3543                 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3544                 if (!tbl)
3545                         goto err_dup;
3546
3547                 /* Don't export non-whitelisted sysctls to unprivileged users */
3548                 if (net->user_ns != &init_user_ns) {
3549                         if (tbl[0].procname != ipv4_route_flush_procname)
3550                                 table_size = 0;
3551                 }
3552
3553                 /* Update the variables to point into the current struct net
3554                  * except for the first element flush
3555                  */
3556                 for (i = 1; i < table_size; i++)
3557                         tbl[i].data += (void *)net - (void *)&init_net;
3558         }
3559         tbl[0].extra1 = net;
3560
3561         net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3562                                                      tbl, table_size);
3563         if (!net->ipv4.route_hdr)
3564                 goto err_reg;
3565         return 0;
3566
3567 err_reg:
3568         if (tbl != ipv4_route_netns_table)
3569                 kfree(tbl);
3570 err_dup:
3571         return -ENOMEM;
3572 }
3573
3574 static __net_exit void sysctl_route_net_exit(struct net *net)
3575 {
3576         const struct ctl_table *tbl;
3577
3578         tbl = net->ipv4.route_hdr->ctl_table_arg;
3579         unregister_net_sysctl_table(net->ipv4.route_hdr);
3580         BUG_ON(tbl == ipv4_route_netns_table);
3581         kfree(tbl);
3582 }
3583
3584 static __net_initdata struct pernet_operations sysctl_route_ops = {
3585         .init = sysctl_route_net_init,
3586         .exit = sysctl_route_net_exit,
3587 };
3588 #endif
3589
3590 static __net_init int netns_ip_rt_init(struct net *net)
3591 {
3592         /* Set default value for namespaceified sysctls */
3593         net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3594         net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3595         net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3596         return 0;
3597 }
3598
3599 static struct pernet_operations __net_initdata ip_rt_ops = {
3600         .init = netns_ip_rt_init,
3601 };
3602
3603 static __net_init int rt_genid_init(struct net *net)
3604 {
3605         atomic_set(&net->ipv4.rt_genid, 0);
3606         atomic_set(&net->fnhe_genid, 0);
3607         atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3608         return 0;
3609 }
3610
3611 static __net_initdata struct pernet_operations rt_genid_ops = {
3612         .init = rt_genid_init,
3613 };
3614
3615 static int __net_init ipv4_inetpeer_init(struct net *net)
3616 {
3617         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3618
3619         if (!bp)
3620                 return -ENOMEM;
3621         inet_peer_base_init(bp);
3622         net->ipv4.peers = bp;
3623         return 0;
3624 }
3625
3626 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3627 {
3628         struct inet_peer_base *bp = net->ipv4.peers;
3629
3630         net->ipv4.peers = NULL;
3631         inetpeer_invalidate_tree(bp);
3632         kfree(bp);
3633 }
3634
3635 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3636         .init   =       ipv4_inetpeer_init,
3637         .exit   =       ipv4_inetpeer_exit,
3638 };
3639
3640 #ifdef CONFIG_IP_ROUTE_CLASSID
3641 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3642 #endif /* CONFIG_IP_ROUTE_CLASSID */
3643
3644 int __init ip_rt_init(void)
3645 {
3646         void *idents_hash;
3647         int cpu;
3648
3649         /* For modern hosts, this will use 2 MB of memory */
3650         idents_hash = alloc_large_system_hash("IP idents",
3651                                               sizeof(*ip_idents) + sizeof(*ip_tstamps),
3652                                               0,
3653                                               16, /* one bucket per 64 KB */
3654                                               HASH_ZERO,
3655                                               NULL,
3656                                               &ip_idents_mask,
3657                                               2048,
3658                                               256*1024);
3659
3660         ip_idents = idents_hash;
3661
3662         get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3663
3664         ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3665
3666         for_each_possible_cpu(cpu) {
3667                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3668
3669                 INIT_LIST_HEAD(&ul->head);
3670                 INIT_LIST_HEAD(&ul->quarantine);
3671                 spin_lock_init(&ul->lock);
3672         }
3673 #ifdef CONFIG_IP_ROUTE_CLASSID
3674         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3675         if (!ip_rt_acct)
3676                 panic("IP: failed to allocate ip_rt_acct\n");
3677 #endif
3678
3679         ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable,
3680                                               SLAB_HWCACHE_ALIGN | SLAB_PANIC);
3681
3682         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3683
3684         if (dst_entries_init(&ipv4_dst_ops) < 0)
3685                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3686
3687         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3688                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3689
3690         ipv4_dst_ops.gc_thresh = ~0;
3691         ip_rt_max_size = INT_MAX;
3692
3693         devinet_init();
3694         ip_fib_init();
3695
3696         if (ip_rt_proc_init())
3697                 pr_err("Unable to create route proc files\n");
3698 #ifdef CONFIG_XFRM
3699         xfrm_init();
3700         xfrm4_init();
3701 #endif
3702         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3703                       RTNL_FLAG_DOIT_UNLOCKED);
3704
3705 #ifdef CONFIG_SYSCTL
3706         register_pernet_subsys(&sysctl_route_ops);
3707 #endif
3708         register_pernet_subsys(&ip_rt_ops);
3709         register_pernet_subsys(&rt_genid_ops);
3710         register_pernet_subsys(&ipv4_inetpeer_ops);
3711         return 0;
3712 }
3713
3714 #ifdef CONFIG_SYSCTL
3715 /*
3716  * We really need to sanitize the damn ipv4 init order, then all
3717  * this nonsense will go away.
3718  */
3719 void __init ip_static_sysctl_init(void)
3720 {
3721         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3722 }
3723 #endif