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_FL_TOS(oldflp4) \
 110         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
 111
 112 #define RT_GC_TIMEOUT (300*HZ)
 113
 114 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
 115 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
 116 #define DEFAULT_MIN_ADVMSS 256
 117 static int ip_rt_max_size;
 118 static int ip_rt_redirect_number __read_mostly  = 9;
 119 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 120 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 121 static int ip_rt_error_cost __read_mostly       = HZ;
 122 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 123
 124 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
 125
 126 /*
 127  *      Interface to generic destination cache.
 128  */
 129
 130 INDIRECT_CALLABLE_SCOPE
 131 struct dst_entry        *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 132 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 133 INDIRECT_CALLABLE_SCOPE
 134 unsigned int            ipv4_mtu(const struct dst_entry *dst);
 135 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 136 static void              ipv4_link_failure(struct sk_buff *skb);
 137 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 138                                            struct sk_buff *skb, u32 mtu,
 139                                            bool confirm_neigh);
 140 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 141                                         struct sk_buff *skb);
 142 static void             ipv4_dst_destroy(struct dst_entry *dst);
 143
 144 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 145 {
 146         WARN_ON(1);
 147         return NULL;
 148 }
 149
 150 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 151                                            struct sk_buff *skb,
 152                                            const void *daddr);
 153 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
 154
 155 static struct dst_ops ipv4_dst_ops = {
 156         .family =               AF_INET,
 157         .check =                ipv4_dst_check,
 158         .default_advmss =       ipv4_default_advmss,
 159         .mtu =                  ipv4_mtu,
 160         .cow_metrics =          ipv4_cow_metrics,
 161         .destroy =              ipv4_dst_destroy,
 162         .negative_advice =      ipv4_negative_advice,
 163         .link_failure =         ipv4_link_failure,
 164         .update_pmtu =          ip_rt_update_pmtu,
 165         .redirect =             ip_do_redirect,
 166         .local_out =            __ip_local_out,
 167         .neigh_lookup =         ipv4_neigh_lookup,
 168         .confirm_neigh =        ipv4_confirm_neigh,
 169 };
 170
 171 #define ECN_OR_COST(class)      TC_PRIO_##class
 172
 173 const __u8 ip_tos2prio[16] = {
 174         TC_PRIO_BESTEFFORT,
 175         ECN_OR_COST(BESTEFFORT),
 176         TC_PRIO_BESTEFFORT,
 177         ECN_OR_COST(BESTEFFORT),
 178         TC_PRIO_BULK,
 179         ECN_OR_COST(BULK),
 180         TC_PRIO_BULK,
 181         ECN_OR_COST(BULK),
 182         TC_PRIO_INTERACTIVE,
 183         ECN_OR_COST(INTERACTIVE),
 184         TC_PRIO_INTERACTIVE,
 185         ECN_OR_COST(INTERACTIVE),
 186         TC_PRIO_INTERACTIVE_BULK,
 187         ECN_OR_COST(INTERACTIVE_BULK),
 188         TC_PRIO_INTERACTIVE_BULK,
 189         ECN_OR_COST(INTERACTIVE_BULK)
 190 };
 191 EXPORT_SYMBOL(ip_tos2prio);
 192
 193 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 194 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
 195
 196 #ifdef CONFIG_PROC_FS
 197 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 198 {
 199         if (*pos)
 200                 return NULL;
 201         return SEQ_START_TOKEN;
 202 }
 203
 204 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 205 {
 206         ++*pos;
 207         return NULL;
 208 }
 209
 210 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 211 {
 212 }
 213
 214 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 215 {
 216         if (v == SEQ_START_TOKEN)
 217                 seq_printf(seq, "%-127s\n",
 218                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 219                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 220                            "HHUptod\tSpecDst");
 221         return 0;
 222 }
 223
 224 static const struct seq_operations rt_cache_seq_ops = {
 225         .start  = rt_cache_seq_start,
 226         .next   = rt_cache_seq_next,
 227         .stop   = rt_cache_seq_stop,
 228         .show   = rt_cache_seq_show,
 229 };
 230
 231 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 232 {
 233         int cpu;
 234
 235         if (*pos == 0)
 236                 return SEQ_START_TOKEN;
 237
 238         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 239                 if (!cpu_possible(cpu))
 240                         continue;
 241                 *pos = cpu+1;
 242                 return &per_cpu(rt_cache_stat, cpu);
 243         }
 244         return NULL;
 245 }
 246
 247 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 248 {
 249         int cpu;
 250
 251         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 252                 if (!cpu_possible(cpu))
 253                         continue;
 254                 *pos = cpu+1;
 255                 return &per_cpu(rt_cache_stat, cpu);
 256         }
 257         (*pos)++;
 258         return NULL;
 259
 260 }
 261
 262 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 263 {
 264
 265 }
 266
 267 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 268 {
 269         struct rt_cache_stat *st = v;
 270
 271         if (v == SEQ_START_TOKEN) {
 272                 seq_puts(seq, "entries  in_hit   in_slow_tot in_slow_mc in_no_route in_brd   in_martian_dst in_martian_src out_hit  out_slow_tot out_slow_mc gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
 273                 return 0;
 274         }
 275
 276         seq_printf(seq, "%08x %08x %08x    %08x   %08x    %08x %08x       "
 277                         "%08x       %08x %08x     %08x    %08x %08x   "
 278                         "%08x     %08x        %08x        %08x\n",
 279                    dst_entries_get_slow(&ipv4_dst_ops),
 280                    0, /* st->in_hit */
 281                    st->in_slow_tot,
 282                    st->in_slow_mc,
 283                    st->in_no_route,
 284                    st->in_brd,
 285                    st->in_martian_dst,
 286                    st->in_martian_src,
 287
 288                    0, /* st->out_hit */
 289                    st->out_slow_tot,
 290                    st->out_slow_mc,
 291
 292                    0, /* st->gc_total */
 293                    0, /* st->gc_ignored */
 294                    0, /* st->gc_goal_miss */
 295                    0, /* st->gc_dst_overflow */
 296                    0, /* st->in_hlist_search */
 297                    0  /* st->out_hlist_search */
 298                 );
 299         return 0;
 300 }
 301
 302 static const struct seq_operations rt_cpu_seq_ops = {
 303         .start  = rt_cpu_seq_start,
 304         .next   = rt_cpu_seq_next,
 305         .stop   = rt_cpu_seq_stop,
 306         .show   = rt_cpu_seq_show,
 307 };
 308
 309 #ifdef CONFIG_IP_ROUTE_CLASSID
 310 static int rt_acct_proc_show(struct seq_file *m, void *v)
 311 {
 312         struct ip_rt_acct *dst, *src;
 313         unsigned int i, j;
 314
 315         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 316         if (!dst)
 317                 return -ENOMEM;
 318
 319         for_each_possible_cpu(i) {
 320                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 321                 for (j = 0; j < 256; j++) {
 322                         dst[j].o_bytes   += src[j].o_bytes;
 323                         dst[j].o_packets += src[j].o_packets;
 324                         dst[j].i_bytes   += src[j].i_bytes;
 325                         dst[j].i_packets += src[j].i_packets;
 326                 }
 327         }
 328
 329         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 330         kfree(dst);
 331         return 0;
 332 }
 333 #endif
 334
 335 static int __net_init ip_rt_do_proc_init(struct net *net)
 336 {
 337         struct proc_dir_entry *pde;
 338
 339         pde = proc_create_seq("rt_cache", 0444, net->proc_net,
 340                               &rt_cache_seq_ops);
 341         if (!pde)
 342                 goto err1;
 343
 344         pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
 345                               &rt_cpu_seq_ops);
 346         if (!pde)
 347                 goto err2;
 348
 349 #ifdef CONFIG_IP_ROUTE_CLASSID
 350         pde = proc_create_single("rt_acct", 0, net->proc_net,
 351                         rt_acct_proc_show);
 352         if (!pde)
 353                 goto err3;
 354 #endif
 355         return 0;
 356
 357 #ifdef CONFIG_IP_ROUTE_CLASSID
 358 err3:
 359         remove_proc_entry("rt_cache", net->proc_net_stat);
 360 #endif
 361 err2:
 362         remove_proc_entry("rt_cache", net->proc_net);
 363 err1:
 364         return -ENOMEM;
 365 }
 366
 367 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 368 {
 369         remove_proc_entry("rt_cache", net->proc_net_stat);
 370         remove_proc_entry("rt_cache", net->proc_net);
 371 #ifdef CONFIG_IP_ROUTE_CLASSID
 372         remove_proc_entry("rt_acct", net->proc_net);
 373 #endif
 374 }
 375
 376 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 377         .init = ip_rt_do_proc_init,
 378         .exit = ip_rt_do_proc_exit,
 379 };
 380
 381 static int __init ip_rt_proc_init(void)
 382 {
 383         return register_pernet_subsys(&ip_rt_proc_ops);
 384 }
 385
 386 #else
 387 static inline int ip_rt_proc_init(void)
 388 {
 389         return 0;
 390 }
 391 #endif /* CONFIG_PROC_FS */
 392
 393 static inline bool rt_is_expired(const struct rtable *rth)
 394 {
 395         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 396 }
 397
 398 void rt_cache_flush(struct net *net)
 399 {
 400         rt_genid_bump_ipv4(net);
 401 }
 402
 403 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 404                                            struct sk_buff *skb,
 405                                            const void *daddr)
 406 {
 407         const struct rtable *rt = container_of(dst, struct rtable, dst);
 408         struct net_device *dev = dst->dev;
 409         struct neighbour *n;
 410
 411         rcu_read_lock();
 412
 413         if (likely(rt->rt_gw_family == AF_INET)) {
 414                 n = ip_neigh_gw4(dev, rt->rt_gw4);
 415         } else if (rt->rt_gw_family == AF_INET6) {
 416                 n = ip_neigh_gw6(dev, &rt->rt_gw6);
 417         } else {
 418                 __be32 pkey;
 419
 420                 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
 421                 n = ip_neigh_gw4(dev, pkey);
 422         }
 423
 424         if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
 425                 n = NULL;
 426
 427         rcu_read_unlock();
 428
 429         return n;
 430 }
 431
 432 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 433 {
 434         const struct rtable *rt = container_of(dst, struct rtable, dst);
 435         struct net_device *dev = dst->dev;
 436         const __be32 *pkey = daddr;
 437
 438         if (rt->rt_gw_family == AF_INET) {
 439                 pkey = (const __be32 *)&rt->rt_gw4;
 440         } else if (rt->rt_gw_family == AF_INET6) {
 441                 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
 442         } else if (!daddr ||
 443                  (rt->rt_flags &
 444                   (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
 445                 return;
 446         }
 447         __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
 448 }
 449
 450 /* Hash tables of size 2048..262144 depending on RAM size.
 451  * Each bucket uses 8 bytes.
 452  */
 453 static u32 ip_idents_mask __read_mostly;
 454 static atomic_t *ip_idents __read_mostly;
 455 static u32 *ip_tstamps __read_mostly;
 456
 457 /* In order to protect privacy, we add a perturbation to identifiers
 458  * if one generator is seldom used. This makes hard for an attacker
 459  * to infer how many packets were sent between two points in time.
 460  */
 461 static u32 ip_idents_reserve(u32 hash, int segs)
 462 {
 463         u32 bucket, old, now = (u32)jiffies;
 464         atomic_t *p_id;
 465         u32 *p_tstamp;
 466         u32 delta = 0;
 467
 468         bucket = hash & ip_idents_mask;
 469         p_tstamp = ip_tstamps + bucket;
 470         p_id = ip_idents + bucket;
 471         old = READ_ONCE(*p_tstamp);
 472
 473         if (old != now && cmpxchg(p_tstamp, old, now) == old)
 474                 delta = get_random_u32_below(now - old);
 475
 476         /* If UBSAN reports an error there, please make sure your compiler
 477          * supports -fno-strict-overflow before reporting it that was a bug
 478          * in UBSAN, and it has been fixed in GCC-8.
 479          */
 480         return atomic_add_return(segs + delta, p_id) - segs;
 481 }
 482
 483 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
 484 {
 485         u32 hash, id;
 486
 487         /* Note the following code is not safe, but this is okay. */
 488         if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
 489                 get_random_bytes(&net->ipv4.ip_id_key,
 490                                  sizeof(net->ipv4.ip_id_key));
 491
 492         hash = siphash_3u32((__force u32)iph->daddr,
 493                             (__force u32)iph->saddr,
 494                             iph->protocol,
 495                             &net->ipv4.ip_id_key);
 496         id = ip_idents_reserve(hash, segs);
 497         iph->id = htons(id);
 498 }
 499 EXPORT_SYMBOL(__ip_select_ident);
 500
 501 static void ip_rt_fix_tos(struct flowi4 *fl4)
 502 {
 503         __u8 tos = RT_FL_TOS(fl4);
 504
 505         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
 506         if (tos & RTO_ONLINK)
 507                 fl4->flowi4_scope = RT_SCOPE_LINK;
 508 }
 509
 510 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
 511                              const struct sock *sk, const struct iphdr *iph,
 512                              int oif, __u8 tos, u8 prot, u32 mark,
 513                              int flow_flags)
 514 {
 515         __u8 scope = RT_SCOPE_UNIVERSE;
 516
 517         if (sk) {
 518                 oif = sk->sk_bound_dev_if;
 519                 mark = READ_ONCE(sk->sk_mark);
 520                 tos = ip_sock_rt_tos(sk);
 521                 scope = ip_sock_rt_scope(sk);
 522                 prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW :
 523                                                     sk->sk_protocol;
 524         }
 525
 526         flowi4_init_output(fl4, oif, mark, tos & IPTOS_RT_MASK, scope,
 527                            prot, flow_flags, iph->daddr, iph->saddr, 0, 0,
 528                            sock_net_uid(net, sk));
 529 }
 530
 531 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 532                                const struct sock *sk)
 533 {
 534         const struct net *net = dev_net(skb->dev);
 535         const struct iphdr *iph = ip_hdr(skb);
 536         int oif = skb->dev->ifindex;
 537         u8 prot = iph->protocol;
 538         u32 mark = skb->mark;
 539         __u8 tos = iph->tos;
 540
 541         __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
 542 }
 543
 544 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 545 {
 546         const struct inet_sock *inet = inet_sk(sk);
 547         const struct ip_options_rcu *inet_opt;
 548         __be32 daddr = inet->inet_daddr;
 549
 550         rcu_read_lock();
 551         inet_opt = rcu_dereference(inet->inet_opt);
 552         if (inet_opt && inet_opt->opt.srr)
 553                 daddr = inet_opt->opt.faddr;
 554         flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark),
 555                            ip_sock_rt_tos(sk) & IPTOS_RT_MASK,
 556                            ip_sock_rt_scope(sk),
 557                            inet_test_bit(HDRINCL, sk) ?
 558                                 IPPROTO_RAW : sk->sk_protocol,
 559                            inet_sk_flowi_flags(sk),
 560                            daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
 561         rcu_read_unlock();
 562 }
 563
 564 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 565                                  const struct sk_buff *skb)
 566 {
 567         if (skb)
 568                 build_skb_flow_key(fl4, skb, sk);
 569         else
 570                 build_sk_flow_key(fl4, sk);
 571 }
 572
 573 static DEFINE_SPINLOCK(fnhe_lock);
 574
 575 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 576 {
 577         struct rtable *rt;
 578
 579         rt = rcu_dereference(fnhe->fnhe_rth_input);
 580         if (rt) {
 581                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 582                 dst_dev_put(&rt->dst);
 583                 dst_release(&rt->dst);
 584         }
 585         rt = rcu_dereference(fnhe->fnhe_rth_output);
 586         if (rt) {
 587                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 588                 dst_dev_put(&rt->dst);
 589                 dst_release(&rt->dst);
 590         }
 591 }
 592
 593 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
 594 {
 595         struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
 596         struct fib_nh_exception *fnhe, *oldest = NULL;
 597
 598         for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
 599                 fnhe = rcu_dereference_protected(*fnhe_p,
 600                                                  lockdep_is_held(&fnhe_lock));
 601                 if (!fnhe)
 602                         break;
 603                 if (!oldest ||
 604                     time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
 605                         oldest = fnhe;
 606                         oldest_p = fnhe_p;
 607                 }
 608         }
 609         fnhe_flush_routes(oldest);
 610         *oldest_p = oldest->fnhe_next;
 611         kfree_rcu(oldest, rcu);
 612 }
 613
 614 static u32 fnhe_hashfun(__be32 daddr)
 615 {
 616         static siphash_aligned_key_t fnhe_hash_key;
 617         u64 hval;
 618
 619         net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
 620         hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
 621         return hash_64(hval, FNHE_HASH_SHIFT);
 622 }
 623
 624 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 625 {
 626         rt->rt_pmtu = fnhe->fnhe_pmtu;
 627         rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
 628         rt->dst.expires = fnhe->fnhe_expires;
 629
 630         if (fnhe->fnhe_gw) {
 631                 rt->rt_flags |= RTCF_REDIRECTED;
 632                 rt->rt_uses_gateway = 1;
 633                 rt->rt_gw_family = AF_INET;
 634                 rt->rt_gw4 = fnhe->fnhe_gw;
 635         }
 636 }
 637
 638 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
 639                                   __be32 gw, u32 pmtu, bool lock,
 640                                   unsigned long expires)
 641 {
 642         struct fnhe_hash_bucket *hash;
 643         struct fib_nh_exception *fnhe;
 644         struct rtable *rt;
 645         u32 genid, hval;
 646         unsigned int i;
 647         int depth;
 648
 649         genid = fnhe_genid(dev_net(nhc->nhc_dev));
 650         hval = fnhe_hashfun(daddr);
 651
 652         spin_lock_bh(&fnhe_lock);
 653
 654         hash = rcu_dereference(nhc->nhc_exceptions);
 655         if (!hash) {
 656                 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
 657                 if (!hash)
 658                         goto out_unlock;
 659                 rcu_assign_pointer(nhc->nhc_exceptions, hash);
 660         }
 661
 662         hash += hval;
 663
 664         depth = 0;
 665         for (fnhe = rcu_dereference(hash->chain); fnhe;
 666              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 667                 if (fnhe->fnhe_daddr == daddr)
 668                         break;
 669                 depth++;
 670         }
 671
 672         if (fnhe) {
 673                 if (fnhe->fnhe_genid != genid)
 674                         fnhe->fnhe_genid = genid;
 675                 if (gw)
 676                         fnhe->fnhe_gw = gw;
 677                 if (pmtu) {
 678                         fnhe->fnhe_pmtu = pmtu;
 679                         fnhe->fnhe_mtu_locked = lock;
 680                 }
 681                 fnhe->fnhe_expires = max(1UL, expires);
 682                 /* Update all cached dsts too */
 683                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 684                 if (rt)
 685                         fill_route_from_fnhe(rt, fnhe);
 686                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 687                 if (rt)
 688                         fill_route_from_fnhe(rt, fnhe);
 689         } else {
 690                 /* Randomize max depth to avoid some side channels attacks. */
 691                 int max_depth = FNHE_RECLAIM_DEPTH +
 692                                 get_random_u32_below(FNHE_RECLAIM_DEPTH);
 693
 694                 while (depth > max_depth) {
 695                         fnhe_remove_oldest(hash);
 696                         depth--;
 697                 }
 698
 699                 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 700                 if (!fnhe)
 701                         goto out_unlock;
 702
 703                 fnhe->fnhe_next = hash->chain;
 704
 705                 fnhe->fnhe_genid = genid;
 706                 fnhe->fnhe_daddr = daddr;
 707                 fnhe->fnhe_gw = gw;
 708                 fnhe->fnhe_pmtu = pmtu;
 709                 fnhe->fnhe_mtu_locked = lock;
 710                 fnhe->fnhe_expires = max(1UL, expires);
 711
 712                 rcu_assign_pointer(hash->chain, fnhe);
 713
 714                 /* Exception created; mark the cached routes for the nexthop
 715                  * stale, so anyone caching it rechecks if this exception
 716                  * applies to them.
 717                  */
 718                 rt = rcu_dereference(nhc->nhc_rth_input);
 719                 if (rt)
 720                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 721
 722                 for_each_possible_cpu(i) {
 723                         struct rtable __rcu **prt;
 724
 725                         prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
 726                         rt = rcu_dereference(*prt);
 727                         if (rt)
 728                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 729                 }
 730         }
 731
 732         fnhe->fnhe_stamp = jiffies;
 733
 734 out_unlock:
 735         spin_unlock_bh(&fnhe_lock);
 736 }
 737
 738 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 739                              bool kill_route)
 740 {
 741         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 742         __be32 old_gw = ip_hdr(skb)->saddr;
 743         struct net_device *dev = skb->dev;
 744         struct in_device *in_dev;
 745         struct fib_result res;
 746         struct neighbour *n;
 747         struct net *net;
 748
 749         switch (icmp_hdr(skb)->code & 7) {
 750         case ICMP_REDIR_NET:
 751         case ICMP_REDIR_NETTOS:
 752         case ICMP_REDIR_HOST:
 753         case ICMP_REDIR_HOSTTOS:
 754                 break;
 755
 756         default:
 757                 return;
 758         }
 759
 760         if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
 761                 return;
 762
 763         in_dev = __in_dev_get_rcu(dev);
 764         if (!in_dev)
 765                 return;
 766
 767         net = dev_net(dev);
 768         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 769             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 770             ipv4_is_zeronet(new_gw))
 771                 goto reject_redirect;
 772
 773         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 774                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 775                         goto reject_redirect;
 776                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 777                         goto reject_redirect;
 778         } else {
 779                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 780                         goto reject_redirect;
 781         }
 782
 783         n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw);
 784         if (!n)
 785                 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
 786         if (!IS_ERR(n)) {
 787                 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) {
 788                         neigh_event_send(n, NULL);
 789                 } else {
 790                         if (fib_lookup(net, fl4, &res, 0) == 0) {
 791                                 struct fib_nh_common *nhc;
 792
 793                                 fib_select_path(net, &res, fl4, skb);
 794                                 nhc = FIB_RES_NHC(res);
 795                                 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
 796                                                 0, false,
 797                                                 jiffies + ip_rt_gc_timeout);
 798                         }
 799                         if (kill_route)
 800                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 801                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 802                 }
 803                 neigh_release(n);
 804         }
 805         return;
 806
 807 reject_redirect:
 808 #ifdef CONFIG_IP_ROUTE_VERBOSE
 809         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 810                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 811                 __be32 daddr = iph->daddr;
 812                 __be32 saddr = iph->saddr;
 813
 814                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 815                                      "  Advised path = %pI4 -> %pI4\n",
 816                                      &old_gw, dev->name, &new_gw,
 817                                      &saddr, &daddr);
 818         }
 819 #endif
 820         ;
 821 }
 822
 823 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 824 {
 825         struct rtable *rt;
 826         struct flowi4 fl4;
 827         const struct iphdr *iph = (const struct iphdr *) skb->data;
 828         struct net *net = dev_net(skb->dev);
 829         int oif = skb->dev->ifindex;
 830         u8 prot = iph->protocol;
 831         u32 mark = skb->mark;
 832         __u8 tos = iph->tos;
 833
 834         rt = (struct rtable *) dst;
 835
 836         __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
 837         __ip_do_redirect(rt, skb, &fl4, true);
 838 }
 839
 840 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 841 {
 842         struct rtable *rt = (struct rtable *)dst;
 843         struct dst_entry *ret = dst;
 844
 845         if (rt) {
 846                 if (dst->obsolete > 0) {
 847                         ip_rt_put(rt);
 848                         ret = NULL;
 849                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 850                            rt->dst.expires) {
 851                         ip_rt_put(rt);
 852                         ret = NULL;
 853                 }
 854         }
 855         return ret;
 856 }
 857
 858 /*
 859  * Algorithm:
 860  *      1. The first ip_rt_redirect_number redirects are sent
 861  *         with exponential backoff, then we stop sending them at all,
 862  *         assuming that the host ignores our redirects.
 863  *      2. If we did not see packets requiring redirects
 864  *         during ip_rt_redirect_silence, we assume that the host
 865  *         forgot redirected route and start to send redirects again.
 866  *
 867  * This algorithm is much cheaper and more intelligent than dumb load limiting
 868  * in icmp.c.
 869  *
 870  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 871  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 872  */
 873
 874 void ip_rt_send_redirect(struct sk_buff *skb)
 875 {
 876         struct rtable *rt = skb_rtable(skb);
 877         struct in_device *in_dev;
 878         struct inet_peer *peer;
 879         struct net *net;
 880         int log_martians;
 881         int vif;
 882
 883         rcu_read_lock();
 884         in_dev = __in_dev_get_rcu(rt->dst.dev);
 885         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 886                 rcu_read_unlock();
 887                 return;
 888         }
 889         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 890         vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
 891         rcu_read_unlock();
 892
 893         net = dev_net(rt->dst.dev);
 894         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
 895         if (!peer) {
 896                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 897                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 898                 return;
 899         }
 900
 901         /* No redirected packets during ip_rt_redirect_silence;
 902          * reset the algorithm.
 903          */
 904         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
 905                 peer->rate_tokens = 0;
 906                 peer->n_redirects = 0;
 907         }
 908
 909         /* Too many ignored redirects; do not send anything
 910          * set dst.rate_last to the last seen redirected packet.
 911          */
 912         if (peer->n_redirects >= ip_rt_redirect_number) {
 913                 peer->rate_last = jiffies;
 914                 goto out_put_peer;
 915         }
 916
 917         /* Check for load limit; set rate_last to the latest sent
 918          * redirect.
 919          */
 920         if (peer->n_redirects == 0 ||
 921             time_after(jiffies,
 922                        (peer->rate_last +
 923                         (ip_rt_redirect_load << peer->n_redirects)))) {
 924                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 925
 926                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 927                 peer->rate_last = jiffies;
 928                 ++peer->n_redirects;
 929 #ifdef CONFIG_IP_ROUTE_VERBOSE
 930                 if (log_martians &&
 931                     peer->n_redirects == ip_rt_redirect_number)
 932                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 933                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 934                                              &ip_hdr(skb)->daddr, &gw);
 935 #endif
 936         }
 937 out_put_peer:
 938         inet_putpeer(peer);
 939 }
 940
 941 static int ip_error(struct sk_buff *skb)
 942 {
 943         struct rtable *rt = skb_rtable(skb);
 944         struct net_device *dev = skb->dev;
 945         struct in_device *in_dev;
 946         struct inet_peer *peer;
 947         unsigned long now;
 948         struct net *net;
 949         SKB_DR(reason);
 950         bool send;
 951         int code;
 952
 953         if (netif_is_l3_master(skb->dev)) {
 954                 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
 955                 if (!dev)
 956                         goto out;
 957         }
 958
 959         in_dev = __in_dev_get_rcu(dev);
 960
 961         /* IP on this device is disabled. */
 962         if (!in_dev)
 963                 goto out;
 964
 965         net = dev_net(rt->dst.dev);
 966         if (!IN_DEV_FORWARD(in_dev)) {
 967                 switch (rt->dst.error) {
 968                 case EHOSTUNREACH:
 969                         SKB_DR_SET(reason, IP_INADDRERRORS);
 970                         __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
 971                         break;
 972
 973                 case ENETUNREACH:
 974                         SKB_DR_SET(reason, IP_INNOROUTES);
 975                         __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 976                         break;
 977                 }
 978                 goto out;
 979         }
 980
 981         switch (rt->dst.error) {
 982         case EINVAL:
 983         default:
 984                 goto out;
 985         case EHOSTUNREACH:
 986                 code = ICMP_HOST_UNREACH;
 987                 break;
 988         case ENETUNREACH:
 989                 code = ICMP_NET_UNREACH;
 990                 SKB_DR_SET(reason, IP_INNOROUTES);
 991                 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 992                 break;
 993         case EACCES:
 994                 code = ICMP_PKT_FILTERED;
 995                 break;
 996         }
 997
 998         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
 999                                l3mdev_master_ifindex(skb->dev), 1);
1000
1001         send = true;
1002         if (peer) {
1003                 now = jiffies;
1004                 peer->rate_tokens += now - peer->rate_last;
1005                 if (peer->rate_tokens > ip_rt_error_burst)
1006                         peer->rate_tokens = ip_rt_error_burst;
1007                 peer->rate_last = now;
1008                 if (peer->rate_tokens >= ip_rt_error_cost)
1009                         peer->rate_tokens -= ip_rt_error_cost;
1010                 else
1011                         send = false;
1012                 inet_putpeer(peer);
1013         }
1014         if (send)
1015                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1016
1017 out:    kfree_skb_reason(skb, reason);
1018         return 0;
1019 }
1020
1021 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1022 {
1023         struct dst_entry *dst = &rt->dst;
1024         struct net *net = dev_net(dst->dev);
1025         struct fib_result res;
1026         bool lock = false;
1027         u32 old_mtu;
1028
1029         if (ip_mtu_locked(dst))
1030                 return;
1031
1032         old_mtu = ipv4_mtu(dst);
1033         if (old_mtu < mtu)
1034                 return;
1035
1036         if (mtu < net->ipv4.ip_rt_min_pmtu) {
1037                 lock = true;
1038                 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1039         }
1040
1041         if (rt->rt_pmtu == mtu && !lock &&
1042             time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1043                 return;
1044
1045         rcu_read_lock();
1046         if (fib_lookup(net, fl4, &res, 0) == 0) {
1047                 struct fib_nh_common *nhc;
1048
1049                 fib_select_path(net, &res, fl4, NULL);
1050                 nhc = FIB_RES_NHC(res);
1051                 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1052                                       jiffies + net->ipv4.ip_rt_mtu_expires);
1053         }
1054         rcu_read_unlock();
1055 }
1056
1057 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1058                               struct sk_buff *skb, u32 mtu,
1059                               bool confirm_neigh)
1060 {
1061         struct rtable *rt = (struct rtable *) dst;
1062         struct flowi4 fl4;
1063
1064         ip_rt_build_flow_key(&fl4, sk, skb);
1065
1066         /* Don't make lookup fail for bridged encapsulations */
1067         if (skb && netif_is_any_bridge_port(skb->dev))
1068                 fl4.flowi4_oif = 0;
1069
1070         __ip_rt_update_pmtu(rt, &fl4, mtu);
1071 }
1072
1073 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1074                       int oif, u8 protocol)
1075 {
1076         const struct iphdr *iph = (const struct iphdr *)skb->data;
1077         struct flowi4 fl4;
1078         struct rtable *rt;
1079         u32 mark = IP4_REPLY_MARK(net, skb->mark);
1080
1081         __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark,
1082                          0);
1083         rt = __ip_route_output_key(net, &fl4);
1084         if (!IS_ERR(rt)) {
1085                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1086                 ip_rt_put(rt);
1087         }
1088 }
1089 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1090
1091 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1092 {
1093         const struct iphdr *iph = (const struct iphdr *)skb->data;
1094         struct flowi4 fl4;
1095         struct rtable *rt;
1096
1097         __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1098
1099         if (!fl4.flowi4_mark)
1100                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1101
1102         rt = __ip_route_output_key(sock_net(sk), &fl4);
1103         if (!IS_ERR(rt)) {
1104                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1105                 ip_rt_put(rt);
1106         }
1107 }
1108
1109 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1110 {
1111         const struct iphdr *iph = (const struct iphdr *)skb->data;
1112         struct flowi4 fl4;
1113         struct rtable *rt;
1114         struct dst_entry *odst = NULL;
1115         bool new = false;
1116         struct net *net = sock_net(sk);
1117
1118         bh_lock_sock(sk);
1119
1120         if (!ip_sk_accept_pmtu(sk))
1121                 goto out;
1122
1123         odst = sk_dst_get(sk);
1124
1125         if (sock_owned_by_user(sk) || !odst) {
1126                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1127                 goto out;
1128         }
1129
1130         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1131
1132         rt = (struct rtable *)odst;
1133         if (odst->obsolete && !odst->ops->check(odst, 0)) {
1134                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1135                 if (IS_ERR(rt))
1136                         goto out;
1137
1138                 new = true;
1139         }
1140
1141         __ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu);
1142
1143         if (!dst_check(&rt->dst, 0)) {
1144                 if (new)
1145                         dst_release(&rt->dst);
1146
1147                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1148                 if (IS_ERR(rt))
1149                         goto out;
1150
1151                 new = true;
1152         }
1153
1154         if (new)
1155                 sk_dst_set(sk, &rt->dst);
1156
1157 out:
1158         bh_unlock_sock(sk);
1159         dst_release(odst);
1160 }
1161 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1162
1163 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1164                    int oif, u8 protocol)
1165 {
1166         const struct iphdr *iph = (const struct iphdr *)skb->data;
1167         struct flowi4 fl4;
1168         struct rtable *rt;
1169
1170         __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0);
1171         rt = __ip_route_output_key(net, &fl4);
1172         if (!IS_ERR(rt)) {
1173                 __ip_do_redirect(rt, skb, &fl4, false);
1174                 ip_rt_put(rt);
1175         }
1176 }
1177 EXPORT_SYMBOL_GPL(ipv4_redirect);
1178
1179 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1180 {
1181         const struct iphdr *iph = (const struct iphdr *)skb->data;
1182         struct flowi4 fl4;
1183         struct rtable *rt;
1184         struct net *net = sock_net(sk);
1185
1186         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1187         rt = __ip_route_output_key(net, &fl4);
1188         if (!IS_ERR(rt)) {
1189                 __ip_do_redirect(rt, skb, &fl4, false);
1190                 ip_rt_put(rt);
1191         }
1192 }
1193 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1194
1195 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1196                                                          u32 cookie)
1197 {
1198         struct rtable *rt = (struct rtable *) dst;
1199
1200         /* All IPV4 dsts are created with ->obsolete set to the value
1201          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1202          * into this function always.
1203          *
1204          * When a PMTU/redirect information update invalidates a route,
1205          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1206          * DST_OBSOLETE_DEAD.
1207          */
1208         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1209                 return NULL;
1210         return dst;
1211 }
1212 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1213
1214 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1215 {
1216         struct net_device *dev;
1217         struct ip_options opt;
1218         int res;
1219
1220         /* Recompile ip options since IPCB may not be valid anymore.
1221          * Also check we have a reasonable ipv4 header.
1222          */
1223         if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1224             ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1225                 return;
1226
1227         memset(&opt, 0, sizeof(opt));
1228         if (ip_hdr(skb)->ihl > 5) {
1229                 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1230                         return;
1231                 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1232
1233                 rcu_read_lock();
1234                 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1235                 res = __ip_options_compile(dev_net(dev), &opt, skb, NULL);
1236                 rcu_read_unlock();
1237
1238                 if (res)
1239                         return;
1240         }
1241         __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1242 }
1243
1244 static void ipv4_link_failure(struct sk_buff *skb)
1245 {
1246         struct rtable *rt;
1247
1248         ipv4_send_dest_unreach(skb);
1249
1250         rt = skb_rtable(skb);
1251         if (rt)
1252                 dst_set_expires(&rt->dst, 0);
1253 }
1254
1255 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1256 {
1257         pr_debug("%s: %pI4 -> %pI4, %s\n",
1258                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1259                  skb->dev ? skb->dev->name : "?");
1260         kfree_skb(skb);
1261         WARN_ON(1);
1262         return 0;
1263 }
1264
1265 /*
1266  * We do not cache source address of outgoing interface,
1267  * because it is used only by IP RR, TS and SRR options,
1268  * so that it out of fast path.
1269  *
1270  * BTW remember: "addr" is allowed to be not aligned
1271  * in IP options!
1272  */
1273
1274 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1275 {
1276         __be32 src;
1277
1278         if (rt_is_output_route(rt))
1279                 src = ip_hdr(skb)->saddr;
1280         else {
1281                 struct fib_result res;
1282                 struct iphdr *iph = ip_hdr(skb);
1283                 struct flowi4 fl4 = {
1284                         .daddr = iph->daddr,
1285                         .saddr = iph->saddr,
1286                         .flowi4_tos = RT_TOS(iph->tos),
1287                         .flowi4_oif = rt->dst.dev->ifindex,
1288                         .flowi4_iif = skb->dev->ifindex,
1289                         .flowi4_mark = skb->mark,
1290                 };
1291
1292                 rcu_read_lock();
1293                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1294                         src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1295                 else
1296                         src = inet_select_addr(rt->dst.dev,
1297                                                rt_nexthop(rt, iph->daddr),
1298                                                RT_SCOPE_UNIVERSE);
1299                 rcu_read_unlock();
1300         }
1301         memcpy(addr, &src, 4);
1302 }
1303
1304 #ifdef CONFIG_IP_ROUTE_CLASSID
1305 static void set_class_tag(struct rtable *rt, u32 tag)
1306 {
1307         if (!(rt->dst.tclassid & 0xFFFF))
1308                 rt->dst.tclassid |= tag & 0xFFFF;
1309         if (!(rt->dst.tclassid & 0xFFFF0000))
1310                 rt->dst.tclassid |= tag & 0xFFFF0000;
1311 }
1312 #endif
1313
1314 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1315 {
1316         struct net *net = dev_net(dst->dev);
1317         unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1318         unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1319                                     net->ipv4.ip_rt_min_advmss);
1320
1321         return min(advmss, IPV4_MAX_PMTU - header_size);
1322 }
1323
1324 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1325 {
1326         return ip_dst_mtu_maybe_forward(dst, false);
1327 }
1328 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1329
1330 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1331 {
1332         struct fnhe_hash_bucket *hash;
1333         struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1334         u32 hval = fnhe_hashfun(daddr);
1335
1336         spin_lock_bh(&fnhe_lock);
1337
1338         hash = rcu_dereference_protected(nhc->nhc_exceptions,
1339                                          lockdep_is_held(&fnhe_lock));
1340         hash += hval;
1341
1342         fnhe_p = &hash->chain;
1343         fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1344         while (fnhe) {
1345                 if (fnhe->fnhe_daddr == daddr) {
1346                         rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1347                                 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1348                         /* set fnhe_daddr to 0 to ensure it won't bind with
1349                          * new dsts in rt_bind_exception().
1350                          */
1351                         fnhe->fnhe_daddr = 0;
1352                         fnhe_flush_routes(fnhe);
1353                         kfree_rcu(fnhe, rcu);
1354                         break;
1355                 }
1356                 fnhe_p = &fnhe->fnhe_next;
1357                 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1358                                                  lockdep_is_held(&fnhe_lock));
1359         }
1360
1361         spin_unlock_bh(&fnhe_lock);
1362 }
1363
1364 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1365                                                __be32 daddr)
1366 {
1367         struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1368         struct fib_nh_exception *fnhe;
1369         u32 hval;
1370
1371         if (!hash)
1372                 return NULL;
1373
1374         hval = fnhe_hashfun(daddr);
1375
1376         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1377              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1378                 if (fnhe->fnhe_daddr == daddr) {
1379                         if (fnhe->fnhe_expires &&
1380                             time_after(jiffies, fnhe->fnhe_expires)) {
1381                                 ip_del_fnhe(nhc, daddr);
1382                                 break;
1383                         }
1384                         return fnhe;
1385                 }
1386         }
1387         return NULL;
1388 }
1389
1390 /* MTU selection:
1391  * 1. mtu on route is locked - use it
1392  * 2. mtu from nexthop exception
1393  * 3. mtu from egress device
1394  */
1395
1396 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1397 {
1398         struct fib_nh_common *nhc = res->nhc;
1399         struct net_device *dev = nhc->nhc_dev;
1400         struct fib_info *fi = res->fi;
1401         u32 mtu = 0;
1402
1403         if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1404             fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1405                 mtu = fi->fib_mtu;
1406
1407         if (likely(!mtu)) {
1408                 struct fib_nh_exception *fnhe;
1409
1410                 fnhe = find_exception(nhc, daddr);
1411                 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1412                         mtu = fnhe->fnhe_pmtu;
1413         }
1414
1415         if (likely(!mtu))
1416                 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1417
1418         return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1419 }
1420
1421 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1422                               __be32 daddr, const bool do_cache)
1423 {
1424         bool ret = false;
1425
1426         spin_lock_bh(&fnhe_lock);
1427
1428         if (daddr == fnhe->fnhe_daddr) {
1429                 struct rtable __rcu **porig;
1430                 struct rtable *orig;
1431                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1432
1433                 if (rt_is_input_route(rt))
1434                         porig = &fnhe->fnhe_rth_input;
1435                 else
1436                         porig = &fnhe->fnhe_rth_output;
1437                 orig = rcu_dereference(*porig);
1438
1439                 if (fnhe->fnhe_genid != genid) {
1440                         fnhe->fnhe_genid = genid;
1441                         fnhe->fnhe_gw = 0;
1442                         fnhe->fnhe_pmtu = 0;
1443                         fnhe->fnhe_expires = 0;
1444                         fnhe->fnhe_mtu_locked = false;
1445                         fnhe_flush_routes(fnhe);
1446                         orig = NULL;
1447                 }
1448                 fill_route_from_fnhe(rt, fnhe);
1449                 if (!rt->rt_gw4) {
1450                         rt->rt_gw4 = daddr;
1451                         rt->rt_gw_family = AF_INET;
1452                 }
1453
1454                 if (do_cache) {
1455                         dst_hold(&rt->dst);
1456                         rcu_assign_pointer(*porig, rt);
1457                         if (orig) {
1458                                 dst_dev_put(&orig->dst);
1459                                 dst_release(&orig->dst);
1460                         }
1461                         ret = true;
1462                 }
1463
1464                 fnhe->fnhe_stamp = jiffies;
1465         }
1466         spin_unlock_bh(&fnhe_lock);
1467
1468         return ret;
1469 }
1470
1471 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1472 {
1473         struct rtable *orig, *prev, **p;
1474         bool ret = true;
1475
1476         if (rt_is_input_route(rt)) {
1477                 p = (struct rtable **)&nhc->nhc_rth_input;
1478         } else {
1479                 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1480         }
1481         orig = *p;
1482
1483         /* hold dst before doing cmpxchg() to avoid race condition
1484          * on this dst
1485          */
1486         dst_hold(&rt->dst);
1487         prev = cmpxchg(p, orig, rt);
1488         if (prev == orig) {
1489                 if (orig) {
1490                         rt_add_uncached_list(orig);
1491                         dst_release(&orig->dst);
1492                 }
1493         } else {
1494                 dst_release(&rt->dst);
1495                 ret = false;
1496         }
1497
1498         return ret;
1499 }
1500
1501 struct uncached_list {
1502         spinlock_t              lock;
1503         struct list_head        head;
1504         struct list_head        quarantine;
1505 };
1506
1507 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1508
1509 void rt_add_uncached_list(struct rtable *rt)
1510 {
1511         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1512
1513         rt->dst.rt_uncached_list = ul;
1514
1515         spin_lock_bh(&ul->lock);
1516         list_add_tail(&rt->dst.rt_uncached, &ul->head);
1517         spin_unlock_bh(&ul->lock);
1518 }
1519
1520 void rt_del_uncached_list(struct rtable *rt)
1521 {
1522         if (!list_empty(&rt->dst.rt_uncached)) {
1523                 struct uncached_list *ul = rt->dst.rt_uncached_list;
1524
1525                 spin_lock_bh(&ul->lock);
1526                 list_del_init(&rt->dst.rt_uncached);
1527                 spin_unlock_bh(&ul->lock);
1528         }
1529 }
1530
1531 static void ipv4_dst_destroy(struct dst_entry *dst)
1532 {
1533         struct rtable *rt = (struct rtable *)dst;
1534
1535         ip_dst_metrics_put(dst);
1536         rt_del_uncached_list(rt);
1537 }
1538
1539 void rt_flush_dev(struct net_device *dev)
1540 {
1541         struct rtable *rt, *safe;
1542         int cpu;
1543
1544         for_each_possible_cpu(cpu) {
1545                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1546
1547                 if (list_empty(&ul->head))
1548                         continue;
1549
1550                 spin_lock_bh(&ul->lock);
1551                 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
1552                         if (rt->dst.dev != dev)
1553                                 continue;
1554                         rt->dst.dev = blackhole_netdev;
1555                         netdev_ref_replace(dev, blackhole_netdev,
1556                                            &rt->dst.dev_tracker, GFP_ATOMIC);
1557                         list_move(&rt->dst.rt_uncached, &ul->quarantine);
1558                 }
1559                 spin_unlock_bh(&ul->lock);
1560         }
1561 }
1562
1563 static bool rt_cache_valid(const struct rtable *rt)
1564 {
1565         return  rt &&
1566                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1567                 !rt_is_expired(rt);
1568 }
1569
1570 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1571                            const struct fib_result *res,
1572                            struct fib_nh_exception *fnhe,
1573                            struct fib_info *fi, u16 type, u32 itag,
1574                            const bool do_cache)
1575 {
1576         bool cached = false;
1577
1578         if (fi) {
1579                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1580
1581                 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1582                         rt->rt_uses_gateway = 1;
1583                         rt->rt_gw_family = nhc->nhc_gw_family;
1584                         /* only INET and INET6 are supported */
1585                         if (likely(nhc->nhc_gw_family == AF_INET))
1586                                 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1587                         else
1588                                 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1589                 }
1590
1591                 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1592
1593 #ifdef CONFIG_IP_ROUTE_CLASSID
1594                 if (nhc->nhc_family == AF_INET) {
1595                         struct fib_nh *nh;
1596
1597                         nh = container_of(nhc, struct fib_nh, nh_common);
1598                         rt->dst.tclassid = nh->nh_tclassid;
1599                 }
1600 #endif
1601                 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1602                 if (unlikely(fnhe))
1603                         cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1604                 else if (do_cache)
1605                         cached = rt_cache_route(nhc, rt);
1606                 if (unlikely(!cached)) {
1607                         /* Routes we intend to cache in nexthop exception or
1608                          * FIB nexthop have the DST_NOCACHE bit clear.
1609                          * However, if we are unsuccessful at storing this
1610                          * route into the cache we really need to set it.
1611                          */
1612                         if (!rt->rt_gw4) {
1613                                 rt->rt_gw_family = AF_INET;
1614                                 rt->rt_gw4 = daddr;
1615                         }
1616                         rt_add_uncached_list(rt);
1617                 }
1618         } else
1619                 rt_add_uncached_list(rt);
1620
1621 #ifdef CONFIG_IP_ROUTE_CLASSID
1622 #ifdef CONFIG_IP_MULTIPLE_TABLES
1623         set_class_tag(rt, res->tclassid);
1624 #endif
1625         set_class_tag(rt, itag);
1626 #endif
1627 }
1628
1629 struct rtable *rt_dst_alloc(struct net_device *dev,
1630                             unsigned int flags, u16 type,
1631                             bool noxfrm)
1632 {
1633         struct rtable *rt;
1634
1635         rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1636                        (noxfrm ? DST_NOXFRM : 0));
1637
1638         if (rt) {
1639                 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1640                 rt->rt_flags = flags;
1641                 rt->rt_type = type;
1642                 rt->rt_is_input = 0;
1643                 rt->rt_iif = 0;
1644                 rt->rt_pmtu = 0;
1645                 rt->rt_mtu_locked = 0;
1646                 rt->rt_uses_gateway = 0;
1647                 rt->rt_gw_family = 0;
1648                 rt->rt_gw4 = 0;
1649
1650                 rt->dst.output = ip_output;
1651                 if (flags & RTCF_LOCAL)
1652                         rt->dst.input = ip_local_deliver;
1653         }
1654
1655         return rt;
1656 }
1657 EXPORT_SYMBOL(rt_dst_alloc);
1658
1659 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1660 {
1661         struct rtable *new_rt;
1662
1663         new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK,
1664                            rt->dst.flags);
1665
1666         if (new_rt) {
1667                 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1668                 new_rt->rt_flags = rt->rt_flags;
1669                 new_rt->rt_type = rt->rt_type;
1670                 new_rt->rt_is_input = rt->rt_is_input;
1671                 new_rt->rt_iif = rt->rt_iif;
1672                 new_rt->rt_pmtu = rt->rt_pmtu;
1673                 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1674                 new_rt->rt_gw_family = rt->rt_gw_family;
1675                 if (rt->rt_gw_family == AF_INET)
1676                         new_rt->rt_gw4 = rt->rt_gw4;
1677                 else if (rt->rt_gw_family == AF_INET6)
1678                         new_rt->rt_gw6 = rt->rt_gw6;
1679
1680                 new_rt->dst.input = rt->dst.input;
1681                 new_rt->dst.output = rt->dst.output;
1682                 new_rt->dst.error = rt->dst.error;
1683                 new_rt->dst.lastuse = jiffies;
1684                 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1685         }
1686         return new_rt;
1687 }
1688 EXPORT_SYMBOL(rt_dst_clone);
1689
1690 /* called in rcu_read_lock() section */
1691 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1692                           u8 tos, struct net_device *dev,
1693                           struct in_device *in_dev, u32 *itag)
1694 {
1695         int err;
1696
1697         /* Primary sanity checks. */
1698         if (!in_dev)
1699                 return -EINVAL;
1700
1701         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1702             skb->protocol != htons(ETH_P_IP))
1703                 return -EINVAL;
1704
1705         if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1706                 return -EINVAL;
1707
1708         if (ipv4_is_zeronet(saddr)) {
1709                 if (!ipv4_is_local_multicast(daddr) &&
1710                     ip_hdr(skb)->protocol != IPPROTO_IGMP)
1711                         return -EINVAL;
1712         } else {
1713                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1714                                           in_dev, itag);
1715                 if (err < 0)
1716                         return err;
1717         }
1718         return 0;
1719 }
1720
1721 /* called in rcu_read_lock() section */
1722 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1723                              u8 tos, struct net_device *dev, int our)
1724 {
1725         struct in_device *in_dev = __in_dev_get_rcu(dev);
1726         unsigned int flags = RTCF_MULTICAST;
1727         struct rtable *rth;
1728         u32 itag = 0;
1729         int err;
1730
1731         err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1732         if (err)
1733                 return err;
1734
1735         if (our)
1736                 flags |= RTCF_LOCAL;
1737
1738         if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1739                 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1740
1741         rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1742                            false);
1743         if (!rth)
1744                 return -ENOBUFS;
1745
1746 #ifdef CONFIG_IP_ROUTE_CLASSID
1747         rth->dst.tclassid = itag;
1748 #endif
1749         rth->dst.output = ip_rt_bug;
1750         rth->rt_is_input= 1;
1751
1752 #ifdef CONFIG_IP_MROUTE
1753         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1754                 rth->dst.input = ip_mr_input;
1755 #endif
1756         RT_CACHE_STAT_INC(in_slow_mc);
1757
1758         skb_dst_drop(skb);
1759         skb_dst_set(skb, &rth->dst);
1760         return 0;
1761 }
1762
1763
1764 static void ip_handle_martian_source(struct net_device *dev,
1765                                      struct in_device *in_dev,
1766                                      struct sk_buff *skb,
1767                                      __be32 daddr,
1768                                      __be32 saddr)
1769 {
1770         RT_CACHE_STAT_INC(in_martian_src);
1771 #ifdef CONFIG_IP_ROUTE_VERBOSE
1772         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1773                 /*
1774                  *      RFC1812 recommendation, if source is martian,
1775                  *      the only hint is MAC header.
1776                  */
1777                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1778                         &daddr, &saddr, dev->name);
1779                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1780                         print_hex_dump(KERN_WARNING, "ll header: ",
1781                                        DUMP_PREFIX_OFFSET, 16, 1,
1782                                        skb_mac_header(skb),
1783                                        dev->hard_header_len, false);
1784                 }
1785         }
1786 #endif
1787 }
1788
1789 /* called in rcu_read_lock() section */
1790 static int __mkroute_input(struct sk_buff *skb,
1791                            const struct fib_result *res,
1792                            struct in_device *in_dev,
1793                            __be32 daddr, __be32 saddr, u32 tos)
1794 {
1795         struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1796         struct net_device *dev = nhc->nhc_dev;
1797         struct fib_nh_exception *fnhe;
1798         struct rtable *rth;
1799         int err;
1800         struct in_device *out_dev;
1801         bool do_cache;
1802         u32 itag = 0;
1803
1804         /* get a working reference to the output device */
1805         out_dev = __in_dev_get_rcu(dev);
1806         if (!out_dev) {
1807                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1808                 return -EINVAL;
1809         }
1810
1811         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1812                                   in_dev->dev, in_dev, &itag);
1813         if (err < 0) {
1814                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1815                                          saddr);
1816
1817                 goto cleanup;
1818         }
1819
1820         do_cache = res->fi && !itag;
1821         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1822             skb->protocol == htons(ETH_P_IP)) {
1823                 __be32 gw;
1824
1825                 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1826                 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1827                     inet_addr_onlink(out_dev, saddr, gw))
1828                         IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1829         }
1830
1831         if (skb->protocol != htons(ETH_P_IP)) {
1832                 /* Not IP (i.e. ARP). Do not create route, if it is
1833                  * invalid for proxy arp. DNAT routes are always valid.
1834                  *
1835                  * Proxy arp feature have been extended to allow, ARP
1836                  * replies back to the same interface, to support
1837                  * Private VLAN switch technologies. See arp.c.
1838                  */
1839                 if (out_dev == in_dev &&
1840                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1841                         err = -EINVAL;
1842                         goto cleanup;
1843                 }
1844         }
1845
1846         if (IN_DEV_ORCONF(in_dev, NOPOLICY))
1847                 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1848
1849         fnhe = find_exception(nhc, daddr);
1850         if (do_cache) {
1851                 if (fnhe)
1852                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1853                 else
1854                         rth = rcu_dereference(nhc->nhc_rth_input);
1855                 if (rt_cache_valid(rth)) {
1856                         skb_dst_set_noref(skb, &rth->dst);
1857                         goto out;
1858                 }
1859         }
1860
1861         rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1862                            IN_DEV_ORCONF(out_dev, NOXFRM));
1863         if (!rth) {
1864                 err = -ENOBUFS;
1865                 goto cleanup;
1866         }
1867
1868         rth->rt_is_input = 1;
1869         RT_CACHE_STAT_INC(in_slow_tot);
1870
1871         rth->dst.input = ip_forward;
1872
1873         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1874                        do_cache);
1875         lwtunnel_set_redirect(&rth->dst);
1876         skb_dst_set(skb, &rth->dst);
1877 out:
1878         err = 0;
1879  cleanup:
1880         return err;
1881 }
1882
1883 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1884 /* To make ICMP packets follow the right flow, the multipath hash is
1885  * calculated from the inner IP addresses.
1886  */
1887 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1888                                  struct flow_keys *hash_keys)
1889 {
1890         const struct iphdr *outer_iph = ip_hdr(skb);
1891         const struct iphdr *key_iph = outer_iph;
1892         const struct iphdr *inner_iph;
1893         const struct icmphdr *icmph;
1894         struct iphdr _inner_iph;
1895         struct icmphdr _icmph;
1896
1897         if (likely(outer_iph->protocol != IPPROTO_ICMP))
1898                 goto out;
1899
1900         if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1901                 goto out;
1902
1903         icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1904                                    &_icmph);
1905         if (!icmph)
1906                 goto out;
1907
1908         if (!icmp_is_err(icmph->type))
1909                 goto out;
1910
1911         inner_iph = skb_header_pointer(skb,
1912                                        outer_iph->ihl * 4 + sizeof(_icmph),
1913                                        sizeof(_inner_iph), &_inner_iph);
1914         if (!inner_iph)
1915                 goto out;
1916
1917         key_iph = inner_iph;
1918 out:
1919         hash_keys->addrs.v4addrs.src = key_iph->saddr;
1920         hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1921 }
1922
1923 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1924                                            const struct sk_buff *skb,
1925                                            bool *p_has_inner)
1926 {
1927         u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1928         struct flow_keys keys, hash_keys;
1929
1930         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1931                 return 0;
1932
1933         memset(&hash_keys, 0, sizeof(hash_keys));
1934         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1935
1936         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1937         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1938                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1939         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1940                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1941         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1942                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1943         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1944                 hash_keys.ports.src = keys.ports.src;
1945         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1946                 hash_keys.ports.dst = keys.ports.dst;
1947
1948         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1949         return flow_hash_from_keys(&hash_keys);
1950 }
1951
1952 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1953                                            const struct sk_buff *skb,
1954                                            bool has_inner)
1955 {
1956         u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1957         struct flow_keys keys, hash_keys;
1958
1959         /* We assume the packet carries an encapsulation, but if none was
1960          * encountered during dissection of the outer flow, then there is no
1961          * point in calling the flow dissector again.
1962          */
1963         if (!has_inner)
1964                 return 0;
1965
1966         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1967                 return 0;
1968
1969         memset(&hash_keys, 0, sizeof(hash_keys));
1970         skb_flow_dissect_flow_keys(skb, &keys, 0);
1971
1972         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1973                 return 0;
1974
1975         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1976                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1977                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1978                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1979                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1980                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1981         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1982                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1983                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1984                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1985                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1986                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1987                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1988                         hash_keys.tags.flow_label = keys.tags.flow_label;
1989         }
1990
1991         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1992                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1993         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1994                 hash_keys.ports.src = keys.ports.src;
1995         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1996                 hash_keys.ports.dst = keys.ports.dst;
1997
1998         return flow_hash_from_keys(&hash_keys);
1999 }
2000
2001 static u32 fib_multipath_custom_hash_skb(const struct net *net,
2002                                          const struct sk_buff *skb)
2003 {
2004         u32 mhash, mhash_inner;
2005         bool has_inner = true;
2006
2007         mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2008         mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2009
2010         return jhash_2words(mhash, mhash_inner, 0);
2011 }
2012
2013 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2014                                          const struct flowi4 *fl4)
2015 {
2016         u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2017         struct flow_keys hash_keys;
2018
2019         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2020                 return 0;
2021
2022         memset(&hash_keys, 0, sizeof(hash_keys));
2023         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2024         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2025                 hash_keys.addrs.v4addrs.src = fl4->saddr;
2026         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2027                 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2028         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2029                 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2030         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2031                 hash_keys.ports.src = fl4->fl4_sport;
2032         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2033                 hash_keys.ports.dst = fl4->fl4_dport;
2034
2035         return flow_hash_from_keys(&hash_keys);
2036 }
2037
2038 /* if skb is set it will be used and fl4 can be NULL */
2039 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2040                        const struct sk_buff *skb, struct flow_keys *flkeys)
2041 {
2042         u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2043         struct flow_keys hash_keys;
2044         u32 mhash = 0;
2045
2046         switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2047         case 0:
2048                 memset(&hash_keys, 0, sizeof(hash_keys));
2049                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2050                 if (skb) {
2051                         ip_multipath_l3_keys(skb, &hash_keys);
2052                 } else {
2053                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2054                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2055                 }
2056                 mhash = flow_hash_from_keys(&hash_keys);
2057                 break;
2058         case 1:
2059                 /* skb is currently provided only when forwarding */
2060                 if (skb) {
2061                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2062                         struct flow_keys keys;
2063
2064                         /* short-circuit if we already have L4 hash present */
2065                         if (skb->l4_hash)
2066                                 return skb_get_hash_raw(skb) >> 1;
2067
2068                         memset(&hash_keys, 0, sizeof(hash_keys));
2069
2070                         if (!flkeys) {
2071                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2072                                 flkeys = &keys;
2073                         }
2074
2075                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2076                         hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2077                         hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2078                         hash_keys.ports.src = flkeys->ports.src;
2079                         hash_keys.ports.dst = flkeys->ports.dst;
2080                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2081                 } else {
2082                         memset(&hash_keys, 0, sizeof(hash_keys));
2083                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2084                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2085                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2086                         hash_keys.ports.src = fl4->fl4_sport;
2087                         hash_keys.ports.dst = fl4->fl4_dport;
2088                         hash_keys.basic.ip_proto = fl4->flowi4_proto;
2089                 }
2090                 mhash = flow_hash_from_keys(&hash_keys);
2091                 break;
2092         case 2:
2093                 memset(&hash_keys, 0, sizeof(hash_keys));
2094                 /* skb is currently provided only when forwarding */
2095                 if (skb) {
2096                         struct flow_keys keys;
2097
2098                         skb_flow_dissect_flow_keys(skb, &keys, 0);
2099                         /* Inner can be v4 or v6 */
2100                         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2101                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2102                                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2103                                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2104                         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2105                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2106                                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2107                                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2108                                 hash_keys.tags.flow_label = keys.tags.flow_label;
2109                                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2110                         } else {
2111                                 /* Same as case 0 */
2112                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2113                                 ip_multipath_l3_keys(skb, &hash_keys);
2114                         }
2115                 } else {
2116                         /* Same as case 0 */
2117                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2118                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2119                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2120                 }
2121                 mhash = flow_hash_from_keys(&hash_keys);
2122                 break;
2123         case 3:
2124                 if (skb)
2125                         mhash = fib_multipath_custom_hash_skb(net, skb);
2126                 else
2127                         mhash = fib_multipath_custom_hash_fl4(net, fl4);
2128                 break;
2129         }
2130
2131         if (multipath_hash)
2132                 mhash = jhash_2words(mhash, multipath_hash, 0);
2133
2134         return mhash >> 1;
2135 }
2136 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2137
2138 static int ip_mkroute_input(struct sk_buff *skb,
2139                             struct fib_result *res,
2140                             struct in_device *in_dev,
2141                             __be32 daddr, __be32 saddr, u32 tos,
2142                             struct flow_keys *hkeys)
2143 {
2144 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2145         if (res->fi && fib_info_num_path(res->fi) > 1) {
2146                 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2147
2148                 fib_select_multipath(res, h);
2149                 IPCB(skb)->flags |= IPSKB_MULTIPATH;
2150         }
2151 #endif
2152
2153         /* create a routing cache entry */
2154         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2155 }
2156
2157 /* Implements all the saddr-related checks as ip_route_input_slow(),
2158  * assuming daddr is valid and the destination is not a local broadcast one.
2159  * Uses the provided hint instead of performing a route lookup.
2160  */
2161 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2162                       u8 tos, struct net_device *dev,
2163                       const struct sk_buff *hint)
2164 {
2165         struct in_device *in_dev = __in_dev_get_rcu(dev);
2166         struct rtable *rt = skb_rtable(hint);
2167         struct net *net = dev_net(dev);
2168         int err = -EINVAL;
2169         u32 tag = 0;
2170
2171         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2172                 goto martian_source;
2173
2174         if (ipv4_is_zeronet(saddr))
2175                 goto martian_source;
2176
2177         if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2178                 goto martian_source;
2179
2180         if (rt->rt_type != RTN_LOCAL)
2181                 goto skip_validate_source;
2182
2183         tos &= IPTOS_RT_MASK;
2184         err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2185         if (err < 0)
2186                 goto martian_source;
2187
2188 skip_validate_source:
2189         skb_dst_copy(skb, hint);
2190         return 0;
2191
2192 martian_source:
2193         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2194         return err;
2195 }
2196
2197 /* get device for dst_alloc with local routes */
2198 static struct net_device *ip_rt_get_dev(struct net *net,
2199                                         const struct fib_result *res)
2200 {
2201         struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2202         struct net_device *dev = NULL;
2203
2204         if (nhc)
2205                 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2206
2207         return dev ? : net->loopback_dev;
2208 }
2209
2210 /*
2211  *      NOTE. We drop all the packets that has local source
2212  *      addresses, because every properly looped back packet
2213  *      must have correct destination already attached by output routine.
2214  *      Changes in the enforced policies must be applied also to
2215  *      ip_route_use_hint().
2216  *
2217  *      Such approach solves two big problems:
2218  *      1. Not simplex devices are handled properly.
2219  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2220  *      called with rcu_read_lock()
2221  */
2222
2223 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2224                                u8 tos, struct net_device *dev,
2225                                struct fib_result *res)
2226 {
2227         struct in_device *in_dev = __in_dev_get_rcu(dev);
2228         struct flow_keys *flkeys = NULL, _flkeys;
2229         struct net    *net = dev_net(dev);
2230         struct ip_tunnel_info *tun_info;
2231         int             err = -EINVAL;
2232         unsigned int    flags = 0;
2233         u32             itag = 0;
2234         struct rtable   *rth;
2235         struct flowi4   fl4;
2236         bool do_cache = true;
2237
2238         /* IP on this device is disabled. */
2239
2240         if (!in_dev)
2241                 goto out;
2242
2243         /* Check for the most weird martians, which can be not detected
2244          * by fib_lookup.
2245          */
2246
2247         tun_info = skb_tunnel_info(skb);
2248         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2249                 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2250         else
2251                 fl4.flowi4_tun_key.tun_id = 0;
2252         skb_dst_drop(skb);
2253
2254         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2255                 goto martian_source;
2256
2257         res->fi = NULL;
2258         res->table = NULL;
2259         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2260                 goto brd_input;
2261
2262         /* Accept zero addresses only to limited broadcast;
2263          * I even do not know to fix it or not. Waiting for complains :-)
2264          */
2265         if (ipv4_is_zeronet(saddr))
2266                 goto martian_source;
2267
2268         if (ipv4_is_zeronet(daddr))
2269                 goto martian_destination;
2270
2271         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2272          * and call it once if daddr or/and saddr are loopback addresses
2273          */
2274         if (ipv4_is_loopback(daddr)) {
2275                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2276                         goto martian_destination;
2277         } else if (ipv4_is_loopback(saddr)) {
2278                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2279                         goto martian_source;
2280         }
2281
2282         /*
2283          *      Now we are ready to route packet.
2284          */
2285         fl4.flowi4_l3mdev = 0;
2286         fl4.flowi4_oif = 0;
2287         fl4.flowi4_iif = dev->ifindex;
2288         fl4.flowi4_mark = skb->mark;
2289         fl4.flowi4_tos = tos;
2290         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2291         fl4.flowi4_flags = 0;
2292         fl4.daddr = daddr;
2293         fl4.saddr = saddr;
2294         fl4.flowi4_uid = sock_net_uid(net, NULL);
2295         fl4.flowi4_multipath_hash = 0;
2296
2297         if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2298                 flkeys = &_flkeys;
2299         } else {
2300                 fl4.flowi4_proto = 0;
2301                 fl4.fl4_sport = 0;
2302                 fl4.fl4_dport = 0;
2303         }
2304
2305         err = fib_lookup(net, &fl4, res, 0);
2306         if (err != 0) {
2307                 if (!IN_DEV_FORWARD(in_dev))
2308                         err = -EHOSTUNREACH;
2309                 goto no_route;
2310         }
2311
2312         if (res->type == RTN_BROADCAST) {
2313                 if (IN_DEV_BFORWARD(in_dev))
2314                         goto make_route;
2315                 /* not do cache if bc_forwarding is enabled */
2316                 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2317                         do_cache = false;
2318                 goto brd_input;
2319         }
2320
2321         if (res->type == RTN_LOCAL) {
2322                 err = fib_validate_source(skb, saddr, daddr, tos,
2323                                           0, dev, in_dev, &itag);
2324                 if (err < 0)
2325                         goto martian_source;
2326                 goto local_input;
2327         }
2328
2329         if (!IN_DEV_FORWARD(in_dev)) {
2330                 err = -EHOSTUNREACH;
2331                 goto no_route;
2332         }
2333         if (res->type != RTN_UNICAST)
2334                 goto martian_destination;
2335
2336 make_route:
2337         err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2338 out:    return err;
2339
2340 brd_input:
2341         if (skb->protocol != htons(ETH_P_IP))
2342                 goto e_inval;
2343
2344         if (!ipv4_is_zeronet(saddr)) {
2345                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2346                                           in_dev, &itag);
2347                 if (err < 0)
2348                         goto martian_source;
2349         }
2350         flags |= RTCF_BROADCAST;
2351         res->type = RTN_BROADCAST;
2352         RT_CACHE_STAT_INC(in_brd);
2353
2354 local_input:
2355         if (IN_DEV_ORCONF(in_dev, NOPOLICY))
2356                 IPCB(skb)->flags |= IPSKB_NOPOLICY;
2357
2358         do_cache &= res->fi && !itag;
2359         if (do_cache) {
2360                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2361
2362                 rth = rcu_dereference(nhc->nhc_rth_input);
2363                 if (rt_cache_valid(rth)) {
2364                         skb_dst_set_noref(skb, &rth->dst);
2365                         err = 0;
2366                         goto out;
2367                 }
2368         }
2369
2370         rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2371                            flags | RTCF_LOCAL, res->type, false);
2372         if (!rth)
2373                 goto e_nobufs;
2374
2375         rth->dst.output= ip_rt_bug;
2376 #ifdef CONFIG_IP_ROUTE_CLASSID
2377         rth->dst.tclassid = itag;
2378 #endif
2379         rth->rt_is_input = 1;
2380
2381         RT_CACHE_STAT_INC(in_slow_tot);
2382         if (res->type == RTN_UNREACHABLE) {
2383                 rth->dst.input= ip_error;
2384                 rth->dst.error= -err;
2385                 rth->rt_flags   &= ~RTCF_LOCAL;
2386         }
2387
2388         if (do_cache) {
2389                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2390
2391                 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2392                 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2393                         WARN_ON(rth->dst.input == lwtunnel_input);
2394                         rth->dst.lwtstate->orig_input = rth->dst.input;
2395                         rth->dst.input = lwtunnel_input;
2396                 }
2397
2398                 if (unlikely(!rt_cache_route(nhc, rth)))
2399                         rt_add_uncached_list(rth);
2400         }
2401         skb_dst_set(skb, &rth->dst);
2402         err = 0;
2403         goto out;
2404
2405 no_route:
2406         RT_CACHE_STAT_INC(in_no_route);
2407         res->type = RTN_UNREACHABLE;
2408         res->fi = NULL;
2409         res->table = NULL;
2410         goto local_input;
2411
2412         /*
2413          *      Do not cache martian addresses: they should be logged (RFC1812)
2414          */
2415 martian_destination:
2416         RT_CACHE_STAT_INC(in_martian_dst);
2417 #ifdef CONFIG_IP_ROUTE_VERBOSE
2418         if (IN_DEV_LOG_MARTIANS(in_dev))
2419                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2420                                      &daddr, &saddr, dev->name);
2421 #endif
2422
2423 e_inval:
2424         err = -EINVAL;
2425         goto out;
2426
2427 e_nobufs:
2428         err = -ENOBUFS;
2429         goto out;
2430
2431 martian_source:
2432         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2433         goto out;
2434 }
2435
2436 /* called with rcu_read_lock held */
2437 static int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2438                               u8 tos, struct net_device *dev, struct fib_result *res)
2439 {
2440         /* Multicast recognition logic is moved from route cache to here.
2441          * The problem was that too many Ethernet cards have broken/missing
2442          * hardware multicast filters :-( As result the host on multicasting
2443          * network acquires a lot of useless route cache entries, sort of
2444          * SDR messages from all the world. Now we try to get rid of them.
2445          * Really, provided software IP multicast filter is organized
2446          * reasonably (at least, hashed), it does not result in a slowdown
2447          * comparing with route cache reject entries.
2448          * Note, that multicast routers are not affected, because
2449          * route cache entry is created eventually.
2450          */
2451         if (ipv4_is_multicast(daddr)) {
2452                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2453                 int our = 0;
2454                 int err = -EINVAL;
2455
2456                 if (!in_dev)
2457                         return err;
2458                 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2459                                       ip_hdr(skb)->protocol);
2460
2461                 /* check l3 master if no match yet */
2462                 if (!our && netif_is_l3_slave(dev)) {
2463                         struct in_device *l3_in_dev;
2464
2465                         l3_in_dev = __in_dev_get_rcu(skb->dev);
2466                         if (l3_in_dev)
2467                                 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2468                                                       ip_hdr(skb)->protocol);
2469                 }
2470
2471                 if (our
2472 #ifdef CONFIG_IP_MROUTE
2473                         ||
2474                     (!ipv4_is_local_multicast(daddr) &&
2475                      IN_DEV_MFORWARD(in_dev))
2476 #endif
2477                    ) {
2478                         err = ip_route_input_mc(skb, daddr, saddr,
2479                                                 tos, dev, our);
2480                 }
2481                 return err;
2482         }
2483
2484         return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2485 }
2486
2487 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2488                          u8 tos, struct net_device *dev)
2489 {
2490         struct fib_result res;
2491         int err;
2492
2493         tos &= IPTOS_RT_MASK;
2494         rcu_read_lock();
2495         err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2496         rcu_read_unlock();
2497
2498         return err;
2499 }
2500 EXPORT_SYMBOL(ip_route_input_noref);
2501
2502 /* called with rcu_read_lock() */
2503 static struct rtable *__mkroute_output(const struct fib_result *res,
2504                                        const struct flowi4 *fl4, int orig_oif,
2505                                        struct net_device *dev_out,
2506                                        unsigned int flags)
2507 {
2508         struct fib_info *fi = res->fi;
2509         struct fib_nh_exception *fnhe;
2510         struct in_device *in_dev;
2511         u16 type = res->type;
2512         struct rtable *rth;
2513         bool do_cache;
2514
2515         in_dev = __in_dev_get_rcu(dev_out);
2516         if (!in_dev)
2517                 return ERR_PTR(-EINVAL);
2518
2519         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2520                 if (ipv4_is_loopback(fl4->saddr) &&
2521                     !(dev_out->flags & IFF_LOOPBACK) &&
2522                     !netif_is_l3_master(dev_out))
2523                         return ERR_PTR(-EINVAL);
2524
2525         if (ipv4_is_lbcast(fl4->daddr))
2526                 type = RTN_BROADCAST;
2527         else if (ipv4_is_multicast(fl4->daddr))
2528                 type = RTN_MULTICAST;
2529         else if (ipv4_is_zeronet(fl4->daddr))
2530                 return ERR_PTR(-EINVAL);
2531
2532         if (dev_out->flags & IFF_LOOPBACK)
2533                 flags |= RTCF_LOCAL;
2534
2535         do_cache = true;
2536         if (type == RTN_BROADCAST) {
2537                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2538                 fi = NULL;
2539         } else if (type == RTN_MULTICAST) {
2540                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2541                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2542                                      fl4->flowi4_proto))
2543                         flags &= ~RTCF_LOCAL;
2544                 else
2545                         do_cache = false;
2546                 /* If multicast route do not exist use
2547                  * default one, but do not gateway in this case.
2548                  * Yes, it is hack.
2549                  */
2550                 if (fi && res->prefixlen < 4)
2551                         fi = NULL;
2552         } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2553                    (orig_oif != dev_out->ifindex)) {
2554                 /* For local routes that require a particular output interface
2555                  * we do not want to cache the result.  Caching the result
2556                  * causes incorrect behaviour when there are multiple source
2557                  * addresses on the interface, the end result being that if the
2558                  * intended recipient is waiting on that interface for the
2559                  * packet he won't receive it because it will be delivered on
2560                  * the loopback interface and the IP_PKTINFO ipi_ifindex will
2561                  * be set to the loopback interface as well.
2562                  */
2563                 do_cache = false;
2564         }
2565
2566         fnhe = NULL;
2567         do_cache &= fi != NULL;
2568         if (fi) {
2569                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2570                 struct rtable __rcu **prth;
2571
2572                 fnhe = find_exception(nhc, fl4->daddr);
2573                 if (!do_cache)
2574                         goto add;
2575                 if (fnhe) {
2576                         prth = &fnhe->fnhe_rth_output;
2577                 } else {
2578                         if (unlikely(fl4->flowi4_flags &
2579                                      FLOWI_FLAG_KNOWN_NH &&
2580                                      !(nhc->nhc_gw_family &&
2581                                        nhc->nhc_scope == RT_SCOPE_LINK))) {
2582                                 do_cache = false;
2583                                 goto add;
2584                         }
2585                         prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2586                 }
2587                 rth = rcu_dereference(*prth);
2588                 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2589                         return rth;
2590         }
2591
2592 add:
2593         rth = rt_dst_alloc(dev_out, flags, type,
2594                            IN_DEV_ORCONF(in_dev, NOXFRM));
2595         if (!rth)
2596                 return ERR_PTR(-ENOBUFS);
2597
2598         rth->rt_iif = orig_oif;
2599
2600         RT_CACHE_STAT_INC(out_slow_tot);
2601
2602         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2603                 if (flags & RTCF_LOCAL &&
2604                     !(dev_out->flags & IFF_LOOPBACK)) {
2605                         rth->dst.output = ip_mc_output;
2606                         RT_CACHE_STAT_INC(out_slow_mc);
2607                 }
2608 #ifdef CONFIG_IP_MROUTE
2609                 if (type == RTN_MULTICAST) {
2610                         if (IN_DEV_MFORWARD(in_dev) &&
2611                             !ipv4_is_local_multicast(fl4->daddr)) {
2612                                 rth->dst.input = ip_mr_input;
2613                                 rth->dst.output = ip_mc_output;
2614                         }
2615                 }
2616 #endif
2617         }
2618
2619         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2620         lwtunnel_set_redirect(&rth->dst);
2621
2622         return rth;
2623 }
2624
2625 /*
2626  * Major route resolver routine.
2627  */
2628
2629 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2630                                         const struct sk_buff *skb)
2631 {
2632         struct fib_result res = {
2633                 .type           = RTN_UNSPEC,
2634                 .fi             = NULL,
2635                 .table          = NULL,
2636                 .tclassid       = 0,
2637         };
2638         struct rtable *rth;
2639
2640         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2641         ip_rt_fix_tos(fl4);
2642
2643         rcu_read_lock();
2644         rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2645         rcu_read_unlock();
2646
2647         return rth;
2648 }
2649 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2650
2651 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2652                                             struct fib_result *res,
2653                                             const struct sk_buff *skb)
2654 {
2655         struct net_device *dev_out = NULL;
2656         int orig_oif = fl4->flowi4_oif;
2657         unsigned int flags = 0;
2658         struct rtable *rth;
2659         int err;
2660
2661         if (fl4->saddr) {
2662                 if (ipv4_is_multicast(fl4->saddr) ||
2663                     ipv4_is_lbcast(fl4->saddr) ||
2664                     ipv4_is_zeronet(fl4->saddr)) {
2665                         rth = ERR_PTR(-EINVAL);
2666                         goto out;
2667                 }
2668
2669                 rth = ERR_PTR(-ENETUNREACH);
2670
2671                 /* I removed check for oif == dev_out->oif here.
2672                  * It was wrong for two reasons:
2673                  * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2674                  *    is assigned to multiple interfaces.
2675                  * 2. Moreover, we are allowed to send packets with saddr
2676                  *    of another iface. --ANK
2677                  */
2678
2679                 if (fl4->flowi4_oif == 0 &&
2680                     (ipv4_is_multicast(fl4->daddr) ||
2681                      ipv4_is_lbcast(fl4->daddr))) {
2682                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2683                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2684                         if (!dev_out)
2685                                 goto out;
2686
2687                         /* Special hack: user can direct multicasts
2688                          * and limited broadcast via necessary interface
2689                          * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2690                          * This hack is not just for fun, it allows
2691                          * vic,vat and friends to work.
2692                          * They bind socket to loopback, set ttl to zero
2693                          * and expect that it will work.
2694                          * From the viewpoint of routing cache they are broken,
2695                          * because we are not allowed to build multicast path
2696                          * with loopback source addr (look, routing cache
2697                          * cannot know, that ttl is zero, so that packet
2698                          * will not leave this host and route is valid).
2699                          * Luckily, this hack is good workaround.
2700                          */
2701
2702                         fl4->flowi4_oif = dev_out->ifindex;
2703                         goto make_route;
2704                 }
2705
2706                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2707                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2708                         if (!__ip_dev_find(net, fl4->saddr, false))
2709                                 goto out;
2710                 }
2711         }
2712
2713
2714         if (fl4->flowi4_oif) {
2715                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2716                 rth = ERR_PTR(-ENODEV);
2717                 if (!dev_out)
2718                         goto out;
2719
2720                 /* RACE: Check return value of inet_select_addr instead. */
2721                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2722                         rth = ERR_PTR(-ENETUNREACH);
2723                         goto out;
2724                 }
2725                 if (ipv4_is_local_multicast(fl4->daddr) ||
2726                     ipv4_is_lbcast(fl4->daddr) ||
2727                     fl4->flowi4_proto == IPPROTO_IGMP) {
2728                         if (!fl4->saddr)
2729                                 fl4->saddr = inet_select_addr(dev_out, 0,
2730                                                               RT_SCOPE_LINK);
2731                         goto make_route;
2732                 }
2733                 if (!fl4->saddr) {
2734                         if (ipv4_is_multicast(fl4->daddr))
2735                                 fl4->saddr = inet_select_addr(dev_out, 0,
2736                                                               fl4->flowi4_scope);
2737                         else if (!fl4->daddr)
2738                                 fl4->saddr = inet_select_addr(dev_out, 0,
2739                                                               RT_SCOPE_HOST);
2740                 }
2741         }
2742
2743         if (!fl4->daddr) {
2744                 fl4->daddr = fl4->saddr;
2745                 if (!fl4->daddr)
2746                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2747                 dev_out = net->loopback_dev;
2748                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2749                 res->type = RTN_LOCAL;
2750                 flags |= RTCF_LOCAL;
2751                 goto make_route;
2752         }
2753
2754         err = fib_lookup(net, fl4, res, 0);
2755         if (err) {
2756                 res->fi = NULL;
2757                 res->table = NULL;
2758                 if (fl4->flowi4_oif &&
2759                     (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2760                         /* Apparently, routing tables are wrong. Assume,
2761                          * that the destination is on link.
2762                          *
2763                          * WHY? DW.
2764                          * Because we are allowed to send to iface
2765                          * even if it has NO routes and NO assigned
2766                          * addresses. When oif is specified, routing
2767                          * tables are looked up with only one purpose:
2768                          * to catch if destination is gatewayed, rather than
2769                          * direct. Moreover, if MSG_DONTROUTE is set,
2770                          * we send packet, ignoring both routing tables
2771                          * and ifaddr state. --ANK
2772                          *
2773                          *
2774                          * We could make it even if oif is unknown,
2775                          * likely IPv6, but we do not.
2776                          */
2777
2778                         if (fl4->saddr == 0)
2779                                 fl4->saddr = inet_select_addr(dev_out, 0,
2780                                                               RT_SCOPE_LINK);
2781                         res->type = RTN_UNICAST;
2782                         goto make_route;
2783                 }
2784                 rth = ERR_PTR(err);
2785                 goto out;
2786         }
2787
2788         if (res->type == RTN_LOCAL) {
2789                 if (!fl4->saddr) {
2790                         if (res->fi->fib_prefsrc)
2791                                 fl4->saddr = res->fi->fib_prefsrc;
2792                         else
2793                                 fl4->saddr = fl4->daddr;
2794                 }
2795
2796                 /* L3 master device is the loopback for that domain */
2797                 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2798                         net->loopback_dev;
2799
2800                 /* make sure orig_oif points to fib result device even
2801                  * though packet rx/tx happens over loopback or l3mdev
2802                  */
2803                 orig_oif = FIB_RES_OIF(*res);
2804
2805                 fl4->flowi4_oif = dev_out->ifindex;
2806                 flags |= RTCF_LOCAL;
2807                 goto make_route;
2808         }
2809
2810         fib_select_path(net, res, fl4, skb);
2811
2812         dev_out = FIB_RES_DEV(*res);
2813
2814 make_route:
2815         rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2816
2817 out:
2818         return rth;
2819 }
2820
2821 static struct dst_ops ipv4_dst_blackhole_ops = {
2822         .family                 = AF_INET,
2823         .default_advmss         = ipv4_default_advmss,
2824         .neigh_lookup           = ipv4_neigh_lookup,
2825         .check                  = dst_blackhole_check,
2826         .cow_metrics            = dst_blackhole_cow_metrics,
2827         .update_pmtu            = dst_blackhole_update_pmtu,
2828         .redirect               = dst_blackhole_redirect,
2829         .mtu                    = dst_blackhole_mtu,
2830 };
2831
2832 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2833 {
2834         struct rtable *ort = (struct rtable *) dst_orig;
2835         struct rtable *rt;
2836
2837         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0);
2838         if (rt) {
2839                 struct dst_entry *new = &rt->dst;
2840
2841                 new->__use = 1;
2842                 new->input = dst_discard;
2843                 new->output = dst_discard_out;
2844
2845                 new->dev = net->loopback_dev;
2846                 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC);
2847
2848                 rt->rt_is_input = ort->rt_is_input;
2849                 rt->rt_iif = ort->rt_iif;
2850                 rt->rt_pmtu = ort->rt_pmtu;
2851                 rt->rt_mtu_locked = ort->rt_mtu_locked;
2852
2853                 rt->rt_genid = rt_genid_ipv4(net);
2854                 rt->rt_flags = ort->rt_flags;
2855                 rt->rt_type = ort->rt_type;
2856                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2857                 rt->rt_gw_family = ort->rt_gw_family;
2858                 if (rt->rt_gw_family == AF_INET)
2859                         rt->rt_gw4 = ort->rt_gw4;
2860                 else if (rt->rt_gw_family == AF_INET6)
2861                         rt->rt_gw6 = ort->rt_gw6;
2862         }
2863
2864         dst_release(dst_orig);
2865
2866         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2867 }
2868
2869 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2870                                     const struct sock *sk)
2871 {
2872         struct rtable *rt = __ip_route_output_key(net, flp4);
2873
2874         if (IS_ERR(rt))
2875                 return rt;
2876
2877         if (flp4->flowi4_proto) {
2878                 flp4->flowi4_oif = rt->dst.dev->ifindex;
2879                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2880                                                         flowi4_to_flowi(flp4),
2881                                                         sk, 0);
2882         }
2883
2884         return rt;
2885 }
2886 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2887
2888 /* called with rcu_read_lock held */
2889 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2890                         struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2891                         struct sk_buff *skb, u32 portid, u32 seq,
2892                         unsigned int flags)
2893 {
2894         struct rtmsg *r;
2895         struct nlmsghdr *nlh;
2896         unsigned long expires = 0;
2897         u32 error;
2898         u32 metrics[RTAX_MAX];
2899
2900         nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2901         if (!nlh)
2902                 return -EMSGSIZE;
2903
2904         r = nlmsg_data(nlh);
2905         r->rtm_family    = AF_INET;
2906         r->rtm_dst_len  = 32;
2907         r->rtm_src_len  = 0;
2908         r->rtm_tos      = fl4 ? fl4->flowi4_tos : 0;
2909         r->rtm_table    = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2910         if (nla_put_u32(skb, RTA_TABLE, table_id))
2911                 goto nla_put_failure;
2912         r->rtm_type     = rt->rt_type;
2913         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2914         r->rtm_protocol = RTPROT_UNSPEC;
2915         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2916         if (rt->rt_flags & RTCF_NOTIFY)
2917                 r->rtm_flags |= RTM_F_NOTIFY;
2918         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2919                 r->rtm_flags |= RTCF_DOREDIRECT;
2920
2921         if (nla_put_in_addr(skb, RTA_DST, dst))
2922                 goto nla_put_failure;
2923         if (src) {
2924                 r->rtm_src_len = 32;
2925                 if (nla_put_in_addr(skb, RTA_SRC, src))
2926                         goto nla_put_failure;
2927         }
2928         if (rt->dst.dev &&
2929             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2930                 goto nla_put_failure;
2931         if (rt->dst.lwtstate &&
2932             lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2933                 goto nla_put_failure;
2934 #ifdef CONFIG_IP_ROUTE_CLASSID
2935         if (rt->dst.tclassid &&
2936             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2937                 goto nla_put_failure;
2938 #endif
2939         if (fl4 && !rt_is_input_route(rt) &&
2940             fl4->saddr != src) {
2941                 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2942                         goto nla_put_failure;
2943         }
2944         if (rt->rt_uses_gateway) {
2945                 if (rt->rt_gw_family == AF_INET &&
2946                     nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2947                         goto nla_put_failure;
2948                 } else if (rt->rt_gw_family == AF_INET6) {
2949                         int alen = sizeof(struct in6_addr);
2950                         struct nlattr *nla;
2951                         struct rtvia *via;
2952
2953                         nla = nla_reserve(skb, RTA_VIA, alen + 2);
2954                         if (!nla)
2955                                 goto nla_put_failure;
2956
2957                         via = nla_data(nla);
2958                         via->rtvia_family = AF_INET6;
2959                         memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2960                 }
2961         }
2962
2963         expires = rt->dst.expires;
2964         if (expires) {
2965                 unsigned long now = jiffies;
2966
2967                 if (time_before(now, expires))
2968                         expires -= now;
2969                 else
2970                         expires = 0;
2971         }
2972
2973         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2974         if (rt->rt_pmtu && expires)
2975                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2976         if (rt->rt_mtu_locked && expires)
2977                 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2978         if (rtnetlink_put_metrics(skb, metrics) < 0)
2979                 goto nla_put_failure;
2980
2981         if (fl4) {
2982                 if (fl4->flowi4_mark &&
2983                     nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2984                         goto nla_put_failure;
2985
2986                 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2987                     nla_put_u32(skb, RTA_UID,
2988                                 from_kuid_munged(current_user_ns(),
2989                                                  fl4->flowi4_uid)))
2990                         goto nla_put_failure;
2991
2992                 if (rt_is_input_route(rt)) {
2993 #ifdef CONFIG_IP_MROUTE
2994                         if (ipv4_is_multicast(dst) &&
2995                             !ipv4_is_local_multicast(dst) &&
2996                             IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2997                                 int err = ipmr_get_route(net, skb,
2998                                                          fl4->saddr, fl4->daddr,
2999                                                          r, portid);
3000
3001                                 if (err <= 0) {
3002                                         if (err == 0)
3003                                                 return 0;
3004                                         goto nla_put_failure;
3005                                 }
3006                         } else
3007 #endif
3008                                 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3009                                         goto nla_put_failure;
3010                 }
3011         }
3012
3013         error = rt->dst.error;
3014
3015         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3016                 goto nla_put_failure;
3017
3018         nlmsg_end(skb, nlh);
3019         return 0;
3020
3021 nla_put_failure:
3022         nlmsg_cancel(skb, nlh);
3023         return -EMSGSIZE;
3024 }
3025
3026 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3027                             struct netlink_callback *cb, u32 table_id,
3028                             struct fnhe_hash_bucket *bucket, int genid,
3029                             int *fa_index, int fa_start, unsigned int flags)
3030 {
3031         int i;
3032
3033         for (i = 0; i < FNHE_HASH_SIZE; i++) {
3034                 struct fib_nh_exception *fnhe;
3035
3036                 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3037                      fnhe = rcu_dereference(fnhe->fnhe_next)) {
3038                         struct rtable *rt;
3039                         int err;
3040
3041                         if (*fa_index < fa_start)
3042                                 goto next;
3043
3044                         if (fnhe->fnhe_genid != genid)
3045                                 goto next;
3046
3047                         if (fnhe->fnhe_expires &&
3048                             time_after(jiffies, fnhe->fnhe_expires))
3049                                 goto next;
3050
3051                         rt = rcu_dereference(fnhe->fnhe_rth_input);
3052                         if (!rt)
3053                                 rt = rcu_dereference(fnhe->fnhe_rth_output);
3054                         if (!rt)
3055                                 goto next;
3056
3057                         err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3058                                            table_id, NULL, skb,
3059                                            NETLINK_CB(cb->skb).portid,
3060                                            cb->nlh->nlmsg_seq, flags);
3061                         if (err)
3062                                 return err;
3063 next:
3064                         (*fa_index)++;
3065                 }
3066         }
3067
3068         return 0;
3069 }
3070
3071 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3072                        u32 table_id, struct fib_info *fi,
3073                        int *fa_index, int fa_start, unsigned int flags)
3074 {
3075         struct net *net = sock_net(cb->skb->sk);
3076         int nhsel, genid = fnhe_genid(net);
3077
3078         for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3079                 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3080                 struct fnhe_hash_bucket *bucket;
3081                 int err;
3082
3083                 if (nhc->nhc_flags & RTNH_F_DEAD)
3084                         continue;
3085
3086                 rcu_read_lock();
3087                 bucket = rcu_dereference(nhc->nhc_exceptions);
3088                 err = 0;
3089                 if (bucket)
3090                         err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3091                                                genid, fa_index, fa_start,
3092                                                flags);
3093                 rcu_read_unlock();
3094                 if (err)
3095                         return err;
3096         }
3097
3098         return 0;
3099 }
3100
3101 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3102                                                    u8 ip_proto, __be16 sport,
3103                                                    __be16 dport)
3104 {
3105         struct sk_buff *skb;
3106         struct iphdr *iph;
3107
3108         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3109         if (!skb)
3110                 return NULL;
3111
3112         /* Reserve room for dummy headers, this skb can pass
3113          * through good chunk of routing engine.
3114          */
3115         skb_reset_mac_header(skb);
3116         skb_reset_network_header(skb);
3117         skb->protocol = htons(ETH_P_IP);
3118         iph = skb_put(skb, sizeof(struct iphdr));
3119         iph->protocol = ip_proto;
3120         iph->saddr = src;
3121         iph->daddr = dst;
3122         iph->version = 0x4;
3123         iph->frag_off = 0;
3124         iph->ihl = 0x5;
3125         skb_set_transport_header(skb, skb->len);
3126
3127         switch (iph->protocol) {
3128         case IPPROTO_UDP: {
3129                 struct udphdr *udph;
3130
3131                 udph = skb_put_zero(skb, sizeof(struct udphdr));
3132                 udph->source = sport;
3133                 udph->dest = dport;
3134                 udph->len = htons(sizeof(struct udphdr));
3135                 udph->check = 0;
3136                 break;
3137         }
3138         case IPPROTO_TCP: {
3139                 struct tcphdr *tcph;
3140
3141                 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3142                 tcph->source    = sport;
3143                 tcph->dest      = dport;
3144                 tcph->doff      = sizeof(struct tcphdr) / 4;
3145                 tcph->rst = 1;
3146                 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3147                                             src, dst, 0);
3148                 break;
3149         }
3150         case IPPROTO_ICMP: {
3151                 struct icmphdr *icmph;
3152
3153                 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3154                 icmph->type = ICMP_ECHO;
3155                 icmph->code = 0;
3156         }
3157         }
3158
3159         return skb;
3160 }
3161
3162 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3163                                        const struct nlmsghdr *nlh,
3164                                        struct nlattr **tb,
3165                                        struct netlink_ext_ack *extack)
3166 {
3167         struct rtmsg *rtm;
3168         int i, err;
3169
3170         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3171                 NL_SET_ERR_MSG(extack,
3172                                "ipv4: Invalid header for route get request");
3173                 return -EINVAL;
3174         }
3175
3176         if (!netlink_strict_get_check(skb))
3177                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3178                                               rtm_ipv4_policy, extack);
3179
3180         rtm = nlmsg_data(nlh);
3181         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3182             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3183             rtm->rtm_table || rtm->rtm_protocol ||
3184             rtm->rtm_scope || rtm->rtm_type) {
3185                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3186                 return -EINVAL;
3187         }
3188
3189         if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3190                                RTM_F_LOOKUP_TABLE |
3191                                RTM_F_FIB_MATCH)) {
3192                 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3193                 return -EINVAL;
3194         }
3195
3196         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3197                                             rtm_ipv4_policy, extack);
3198         if (err)
3199                 return err;
3200
3201         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3202             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3203                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3204                 return -EINVAL;
3205         }
3206
3207         for (i = 0; i <= RTA_MAX; i++) {
3208                 if (!tb[i])
3209                         continue;
3210
3211                 switch (i) {
3212                 case RTA_IIF:
3213                 case RTA_OIF:
3214                 case RTA_SRC:
3215                 case RTA_DST:
3216                 case RTA_IP_PROTO:
3217                 case RTA_SPORT:
3218                 case RTA_DPORT:
3219                 case RTA_MARK:
3220                 case RTA_UID:
3221                         break;
3222                 default:
3223                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3224                         return -EINVAL;
3225                 }
3226         }
3227
3228         return 0;
3229 }
3230
3231 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3232                              struct netlink_ext_ack *extack)
3233 {
3234         struct net *net = sock_net(in_skb->sk);
3235         struct nlattr *tb[RTA_MAX+1];
3236         u32 table_id = RT_TABLE_MAIN;
3237         __be16 sport = 0, dport = 0;
3238         struct fib_result res = {};
3239         u8 ip_proto = IPPROTO_UDP;
3240         struct rtable *rt = NULL;
3241         struct sk_buff *skb;
3242         struct rtmsg *rtm;
3243         struct flowi4 fl4 = {};
3244         __be32 dst = 0;
3245         __be32 src = 0;
3246         kuid_t uid;
3247         u32 iif;
3248         int err;
3249         int mark;
3250
3251         err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3252         if (err < 0)
3253                 return err;
3254
3255         rtm = nlmsg_data(nlh);
3256         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3257         dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3258         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3259         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3260         if (tb[RTA_UID])
3261                 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3262         else
3263                 uid = (iif ? INVALID_UID : current_uid());
3264
3265         if (tb[RTA_IP_PROTO]) {
3266                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3267                                                   &ip_proto, AF_INET, extack);
3268                 if (err)
3269                         return err;
3270         }
3271
3272         if (tb[RTA_SPORT])
3273                 sport = nla_get_be16(tb[RTA_SPORT]);
3274
3275         if (tb[RTA_DPORT])
3276                 dport = nla_get_be16(tb[RTA_DPORT]);
3277
3278         skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3279         if (!skb)
3280                 return -ENOBUFS;
3281
3282         fl4.daddr = dst;
3283         fl4.saddr = src;
3284         fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3285         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3286         fl4.flowi4_mark = mark;
3287         fl4.flowi4_uid = uid;
3288         if (sport)
3289                 fl4.fl4_sport = sport;
3290         if (dport)
3291                 fl4.fl4_dport = dport;
3292         fl4.flowi4_proto = ip_proto;
3293
3294         rcu_read_lock();
3295
3296         if (iif) {
3297                 struct net_device *dev;
3298
3299                 dev = dev_get_by_index_rcu(net, iif);
3300                 if (!dev) {
3301                         err = -ENODEV;
3302                         goto errout_rcu;
3303                 }
3304
3305                 fl4.flowi4_iif = iif; /* for rt_fill_info */
3306                 skb->dev        = dev;
3307                 skb->mark       = mark;
3308                 err = ip_route_input_rcu(skb, dst, src,
3309                                          rtm->rtm_tos & IPTOS_RT_MASK, dev,
3310                                          &res);
3311
3312                 rt = skb_rtable(skb);
3313                 if (err == 0 && rt->dst.error)
3314                         err = -rt->dst.error;
3315         } else {
3316                 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3317                 skb->dev = net->loopback_dev;
3318                 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3319                 err = 0;
3320                 if (IS_ERR(rt))
3321                         err = PTR_ERR(rt);
3322                 else
3323                         skb_dst_set(skb, &rt->dst);
3324         }
3325
3326         if (err)
3327                 goto errout_rcu;
3328
3329         if (rtm->rtm_flags & RTM_F_NOTIFY)
3330                 rt->rt_flags |= RTCF_NOTIFY;
3331
3332         if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3333                 table_id = res.table ? res.table->tb_id : 0;
3334
3335         /* reset skb for netlink reply msg */
3336         skb_trim(skb, 0);
3337         skb_reset_network_header(skb);
3338         skb_reset_transport_header(skb);
3339         skb_reset_mac_header(skb);
3340
3341         if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3342                 struct fib_rt_info fri;
3343
3344                 if (!res.fi) {
3345                         err = fib_props[res.type].error;
3346                         if (!err)
3347                                 err = -EHOSTUNREACH;
3348                         goto errout_rcu;
3349                 }
3350                 fri.fi = res.fi;
3351                 fri.tb_id = table_id;
3352                 fri.dst = res.prefix;
3353                 fri.dst_len = res.prefixlen;
3354                 fri.dscp = inet_dsfield_to_dscp(fl4.flowi4_tos);
3355                 fri.type = rt->rt_type;
3356                 fri.offload = 0;
3357                 fri.trap = 0;
3358                 fri.offload_failed = 0;
3359                 if (res.fa_head) {
3360                         struct fib_alias *fa;
3361
3362                         hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3363                                 u8 slen = 32 - fri.dst_len;
3364
3365                                 if (fa->fa_slen == slen &&
3366                                     fa->tb_id == fri.tb_id &&
3367                                     fa->fa_dscp == fri.dscp &&
3368                                     fa->fa_info == res.fi &&
3369                                     fa->fa_type == fri.type) {
3370                                         fri.offload = READ_ONCE(fa->offload);
3371                                         fri.trap = READ_ONCE(fa->trap);
3372                                         fri.offload_failed =
3373                                                 READ_ONCE(fa->offload_failed);
3374                                         break;
3375                                 }
3376                         }
3377                 }
3378                 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3379                                     nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3380         } else {
3381                 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3382                                    NETLINK_CB(in_skb).portid,
3383                                    nlh->nlmsg_seq, 0);
3384         }
3385         if (err < 0)
3386                 goto errout_rcu;
3387
3388         rcu_read_unlock();
3389
3390         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3391
3392 errout_free:
3393         return err;
3394 errout_rcu:
3395         rcu_read_unlock();
3396         kfree_skb(skb);
3397         goto errout_free;
3398 }
3399
3400 void ip_rt_multicast_event(struct in_device *in_dev)
3401 {
3402         rt_cache_flush(dev_net(in_dev->dev));
3403 }
3404
3405 #ifdef CONFIG_SYSCTL
3406 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3407 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
3408 static int ip_rt_gc_elasticity __read_mostly    = 8;
3409 static int ip_min_valid_pmtu __read_mostly      = IPV4_MIN_MTU;
3410
3411 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3412                 void *buffer, size_t *lenp, loff_t *ppos)
3413 {
3414         struct net *net = (struct net *)__ctl->extra1;
3415
3416         if (write) {
3417                 rt_cache_flush(net);
3418                 fnhe_genid_bump(net);
3419                 return 0;
3420         }
3421
3422         return -EINVAL;
3423 }
3424
3425 static struct ctl_table ipv4_route_table[] = {
3426         {
3427                 .procname       = "gc_thresh",
3428                 .data           = &ipv4_dst_ops.gc_thresh,
3429                 .maxlen         = sizeof(int),
3430                 .mode           = 0644,
3431                 .proc_handler   = proc_dointvec,
3432         },
3433         {
3434                 .procname       = "max_size",
3435                 .data           = &ip_rt_max_size,
3436                 .maxlen         = sizeof(int),
3437                 .mode           = 0644,
3438                 .proc_handler   = proc_dointvec,
3439         },
3440         {
3441                 /*  Deprecated. Use gc_min_interval_ms */
3442
3443                 .procname       = "gc_min_interval",
3444                 .data           = &ip_rt_gc_min_interval,
3445                 .maxlen         = sizeof(int),
3446                 .mode           = 0644,
3447                 .proc_handler   = proc_dointvec_jiffies,
3448         },
3449         {
3450                 .procname       = "gc_min_interval_ms",
3451                 .data           = &ip_rt_gc_min_interval,
3452                 .maxlen         = sizeof(int),
3453                 .mode           = 0644,
3454                 .proc_handler   = proc_dointvec_ms_jiffies,
3455         },
3456         {
3457                 .procname       = "gc_timeout",
3458                 .data           = &ip_rt_gc_timeout,
3459                 .maxlen         = sizeof(int),
3460                 .mode           = 0644,
3461                 .proc_handler   = proc_dointvec_jiffies,
3462         },
3463         {
3464                 .procname       = "gc_interval",
3465                 .data           = &ip_rt_gc_interval,
3466                 .maxlen         = sizeof(int),
3467                 .mode           = 0644,
3468                 .proc_handler   = proc_dointvec_jiffies,
3469         },
3470         {
3471                 .procname       = "redirect_load",
3472                 .data           = &ip_rt_redirect_load,
3473                 .maxlen         = sizeof(int),
3474                 .mode           = 0644,
3475                 .proc_handler   = proc_dointvec,
3476         },
3477         {
3478                 .procname       = "redirect_number",
3479                 .data           = &ip_rt_redirect_number,
3480                 .maxlen         = sizeof(int),
3481                 .mode           = 0644,
3482                 .proc_handler   = proc_dointvec,
3483         },
3484         {
3485                 .procname       = "redirect_silence",
3486                 .data           = &ip_rt_redirect_silence,
3487                 .maxlen         = sizeof(int),
3488                 .mode           = 0644,
3489                 .proc_handler   = proc_dointvec,
3490         },
3491         {
3492                 .procname       = "error_cost",
3493                 .data           = &ip_rt_error_cost,
3494                 .maxlen         = sizeof(int),
3495                 .mode           = 0644,
3496                 .proc_handler   = proc_dointvec,
3497         },
3498         {
3499                 .procname       = "error_burst",
3500                 .data           = &ip_rt_error_burst,
3501                 .maxlen         = sizeof(int),
3502                 .mode           = 0644,
3503                 .proc_handler   = proc_dointvec,
3504         },
3505         {
3506                 .procname       = "gc_elasticity",
3507                 .data           = &ip_rt_gc_elasticity,
3508                 .maxlen         = sizeof(int),
3509                 .mode           = 0644,
3510                 .proc_handler   = proc_dointvec,
3511         },
3512         { }
3513 };
3514
3515 static const char ipv4_route_flush_procname[] = "flush";
3516
3517 static struct ctl_table ipv4_route_netns_table[] = {
3518         {
3519                 .procname       = ipv4_route_flush_procname,
3520                 .maxlen         = sizeof(int),
3521                 .mode           = 0200,
3522                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3523         },
3524         {
3525                 .procname       = "min_pmtu",
3526                 .data           = &init_net.ipv4.ip_rt_min_pmtu,
3527                 .maxlen         = sizeof(int),
3528                 .mode           = 0644,
3529                 .proc_handler   = proc_dointvec_minmax,
3530                 .extra1         = &ip_min_valid_pmtu,
3531         },
3532         {
3533                 .procname       = "mtu_expires",
3534                 .data           = &init_net.ipv4.ip_rt_mtu_expires,
3535                 .maxlen         = sizeof(int),
3536                 .mode           = 0644,
3537                 .proc_handler   = proc_dointvec_jiffies,
3538         },
3539         {
3540                 .procname   = "min_adv_mss",
3541                 .data       = &init_net.ipv4.ip_rt_min_advmss,
3542                 .maxlen     = sizeof(int),
3543                 .mode       = 0644,
3544                 .proc_handler   = proc_dointvec,
3545         },
3546         { },
3547 };
3548
3549 static __net_init int sysctl_route_net_init(struct net *net)
3550 {
3551         struct ctl_table *tbl;
3552         size_t table_size = ARRAY_SIZE(ipv4_route_netns_table);
3553
3554         tbl = ipv4_route_netns_table;
3555         if (!net_eq(net, &init_net)) {
3556                 int i;
3557
3558                 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3559                 if (!tbl)
3560                         goto err_dup;
3561
3562                 /* Don't export non-whitelisted sysctls to unprivileged users */
3563                 if (net->user_ns != &init_user_ns) {
3564                         if (tbl[0].procname != ipv4_route_flush_procname) {
3565                                 tbl[0].procname = NULL;
3566                                 table_size = 0;
3567                         }
3568                 }
3569
3570                 /* Update the variables to point into the current struct net
3571                  * except for the first element flush
3572                  */
3573                 for (i = 1; i < ARRAY_SIZE(ipv4_route_netns_table) - 1; i++)
3574                         tbl[i].data += (void *)net - (void *)&init_net;
3575         }
3576         tbl[0].extra1 = net;
3577
3578         net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route",
3579                                                      tbl, table_size);
3580         if (!net->ipv4.route_hdr)
3581                 goto err_reg;
3582         return 0;
3583
3584 err_reg:
3585         if (tbl != ipv4_route_netns_table)
3586                 kfree(tbl);
3587 err_dup:
3588         return -ENOMEM;
3589 }
3590
3591 static __net_exit void sysctl_route_net_exit(struct net *net)
3592 {
3593         struct ctl_table *tbl;
3594
3595         tbl = net->ipv4.route_hdr->ctl_table_arg;
3596         unregister_net_sysctl_table(net->ipv4.route_hdr);
3597         BUG_ON(tbl == ipv4_route_netns_table);
3598         kfree(tbl);
3599 }
3600
3601 static __net_initdata struct pernet_operations sysctl_route_ops = {
3602         .init = sysctl_route_net_init,
3603         .exit = sysctl_route_net_exit,
3604 };
3605 #endif
3606
3607 static __net_init int netns_ip_rt_init(struct net *net)
3608 {
3609         /* Set default value for namespaceified sysctls */
3610         net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3611         net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3612         net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3613         return 0;
3614 }
3615
3616 static struct pernet_operations __net_initdata ip_rt_ops = {
3617         .init = netns_ip_rt_init,
3618 };
3619
3620 static __net_init int rt_genid_init(struct net *net)
3621 {
3622         atomic_set(&net->ipv4.rt_genid, 0);
3623         atomic_set(&net->fnhe_genid, 0);
3624         atomic_set(&net->ipv4.dev_addr_genid, get_random_u32());
3625         return 0;
3626 }
3627
3628 static __net_initdata struct pernet_operations rt_genid_ops = {
3629         .init = rt_genid_init,
3630 };
3631
3632 static int __net_init ipv4_inetpeer_init(struct net *net)
3633 {
3634         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3635
3636         if (!bp)
3637                 return -ENOMEM;
3638         inet_peer_base_init(bp);
3639         net->ipv4.peers = bp;
3640         return 0;
3641 }
3642
3643 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3644 {
3645         struct inet_peer_base *bp = net->ipv4.peers;
3646
3647         net->ipv4.peers = NULL;
3648         inetpeer_invalidate_tree(bp);
3649         kfree(bp);
3650 }
3651
3652 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3653         .init   =       ipv4_inetpeer_init,
3654         .exit   =       ipv4_inetpeer_exit,
3655 };
3656
3657 #ifdef CONFIG_IP_ROUTE_CLASSID
3658 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3659 #endif /* CONFIG_IP_ROUTE_CLASSID */
3660
3661 int __init ip_rt_init(void)
3662 {
3663         void *idents_hash;
3664         int cpu;
3665
3666         /* For modern hosts, this will use 2 MB of memory */
3667         idents_hash = alloc_large_system_hash("IP idents",
3668                                               sizeof(*ip_idents) + sizeof(*ip_tstamps),
3669                                               0,
3670                                               16, /* one bucket per 64 KB */
3671                                               HASH_ZERO,
3672                                               NULL,
3673                                               &ip_idents_mask,
3674                                               2048,
3675                                               256*1024);
3676
3677         ip_idents = idents_hash;
3678
3679         get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3680
3681         ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3682
3683         for_each_possible_cpu(cpu) {
3684                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3685
3686                 INIT_LIST_HEAD(&ul->head);
3687                 INIT_LIST_HEAD(&ul->quarantine);
3688                 spin_lock_init(&ul->lock);
3689         }
3690 #ifdef CONFIG_IP_ROUTE_CLASSID
3691         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3692         if (!ip_rt_acct)
3693                 panic("IP: failed to allocate ip_rt_acct\n");
3694 #endif
3695
3696         ipv4_dst_ops.kmem_cachep =
3697                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3698                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3699
3700         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3701
3702         if (dst_entries_init(&ipv4_dst_ops) < 0)
3703                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3704
3705         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3706                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3707
3708         ipv4_dst_ops.gc_thresh = ~0;
3709         ip_rt_max_size = INT_MAX;
3710
3711         devinet_init();
3712         ip_fib_init();
3713
3714         if (ip_rt_proc_init())
3715                 pr_err("Unable to create route proc files\n");
3716 #ifdef CONFIG_XFRM
3717         xfrm_init();
3718         xfrm4_init();
3719 #endif
3720         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3721                       RTNL_FLAG_DOIT_UNLOCKED);
3722
3723 #ifdef CONFIG_SYSCTL
3724         register_pernet_subsys(&sysctl_route_ops);
3725 #endif
3726         register_pernet_subsys(&ip_rt_ops);
3727         register_pernet_subsys(&rt_genid_ops);
3728         register_pernet_subsys(&ipv4_inetpeer_ops);
3729         return 0;
3730 }
3731
3732 #ifdef CONFIG_SYSCTL
3733 /*
3734  * We really need to sanitize the damn ipv4 init order, then all
3735  * this nonsense will go away.
3736  */
3737 void __init ip_static_sysctl_init(void)
3738 {
3739         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3740 }
3741 #endif