net/ipv4/route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              ROUTE - implementation of the IP router.
   8  *
   9  * Authors:     Ross Biro
  10  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  12  *              Linus Torvalds, <Linus.Torvalds@helsinki.fi>
  13  *              Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  14  *
  15  * Fixes:
  16  *              Alan Cox        :       Verify area fixes.
  17  *              Alan Cox        :       cli() protects routing changes
  18  *              Rui Oliveira    :       ICMP routing table updates
  19  *              (rco@di.uminho.pt)      Routing table insertion and update
  20  *              Linus Torvalds  :       Rewrote bits to be sensible
  21  *              Alan Cox        :       Added BSD route gw semantics
  22  *              Alan Cox        :       Super /proc >4K
  23  *              Alan Cox        :       MTU in route table
  24  *              Alan Cox        :       MSS actually. Also added the window
  25  *                                      clamper.
  26  *              Sam Lantinga    :       Fixed route matching in rt_del()
  27  *              Alan Cox        :       Routing cache support.
  28  *              Alan Cox        :       Removed compatibility cruft.
  29  *              Alan Cox        :       RTF_REJECT support.
  30  *              Alan Cox        :       TCP irtt support.
  31  *              Jonathan Naylor :       Added Metric support.
  32  *      Miquel van Smoorenburg  :       BSD API fixes.
  33  *      Miquel van Smoorenburg  :       Metrics.
  34  *              Alan Cox        :       Use __u32 properly
  35  *              Alan Cox        :       Aligned routing errors more closely with BSD
  36  *                                      our system is still very different.
  37  *              Alan Cox        :       Faster /proc handling
  38  *      Alexey Kuznetsov        :       Massive rework to support tree based routing,
  39  *                                      routing caches and better behaviour.
  40  *
  41  *              Olaf Erb        :       irtt wasn't being copied right.
  42  *              Bjorn Ekwall    :       Kerneld route support.
  43  *              Alan Cox        :       Multicast fixed (I hope)
  44  *              Pavel Krauz     :       Limited broadcast fixed
  45  *              Mike McLagan    :       Routing by source
  46  *      Alexey Kuznetsov        :       End of old history. Split to fib.c and
  47  *                                      route.c and rewritten from scratch.
  48  *              Andi Kleen      :       Load-limit warning messages.
  49  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  50  *      Vitaly E. Lavrov        :       Race condition in ip_route_input_slow.
  51  *      Tobias Ringstrom        :       Uninitialized res.type in ip_route_output_slow.
  52  *      Vladimir V. Ivanov      :       IP rule info (flowid) is really useful.
  53  *              Marc Boucher    :       routing by fwmark
  54  *      Robert Olsson           :       Added rt_cache statistics
  55  *      Arnaldo C. Melo         :       Convert proc stuff to seq_file
  56  *      Eric Dumazet            :       hashed spinlocks and rt_check_expire() fixes.
  57  *      Ilia Sotnikov           :       Ignore TOS on PMTUD and Redirect
  58  *      Ilia Sotnikov           :       Removed TOS from hash calculations
  59  */
  60
  61 #define pr_fmt(fmt) "IPv4: " fmt
  62
  63 #include <linux/module.h>
  64 #include <linux/bitops.h>
  65 #include <linux/kernel.h>
  66 #include <linux/mm.h>
  67 #include <linux/memblock.h>
  68 #include <linux/socket.h>
  69 #include <linux/errno.h>
  70 #include <linux/in.h>
  71 #include <linux/inet.h>
  72 #include <linux/netdevice.h>
  73 #include <linux/proc_fs.h>
  74 #include <linux/init.h>
  75 #include <linux/skbuff.h>
  76 #include <linux/inetdevice.h>
  77 #include <linux/igmp.h>
  78 #include <linux/pkt_sched.h>
  79 #include <linux/mroute.h>
  80 #include <linux/netfilter_ipv4.h>
  81 #include <linux/random.h>
  82 #include <linux/rcupdate.h>
  83 #include <linux/slab.h>
  84 #include <linux/jhash.h>
  85 #include <net/dst.h>
  86 #include <net/dst_metadata.h>
  87 #include <net/inet_dscp.h>
  88 #include <net/net_namespace.h>
  89 #include <net/ip.h>
  90 #include <net/route.h>
  91 #include <net/inetpeer.h>
  92 #include <net/sock.h>
  93 #include <net/ip_fib.h>
  94 #include <net/nexthop.h>
  95 #include <net/tcp.h>
  96 #include <net/icmp.h>
  97 #include <net/xfrm.h>
  98 #include <net/lwtunnel.h>
  99 #include <net/netevent.h>
 100 #include <net/rtnetlink.h>
 101 #ifdef CONFIG_SYSCTL
 102 #include <linux/sysctl.h>
 103 #endif
 104 #include <net/secure_seq.h>
 105 #include <net/ip_tunnels.h>
 106
 107 #include "fib_lookup.h"
 108
 109 #define RT_FL_TOS(oldflp4) \
 110         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
 111
 112 #define RT_GC_TIMEOUT (300*HZ)
 113
 114 #define DEFAULT_MIN_PMTU (512 + 20 + 20)
 115 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ)
 116 #define DEFAULT_MIN_ADVMSS 256
 117 static int ip_rt_max_size;
 118 static int ip_rt_redirect_number __read_mostly  = 9;
 119 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 120 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 121 static int ip_rt_error_cost __read_mostly       = HZ;
 122 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 123
 124 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
 125
 126 /*
 127  *      Interface to generic destination cache.
 128  */
 129
 130 INDIRECT_CALLABLE_SCOPE
 131 struct dst_entry        *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 132 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 133 INDIRECT_CALLABLE_SCOPE
 134 unsigned int            ipv4_mtu(const struct dst_entry *dst);
 135 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 136 static void              ipv4_link_failure(struct sk_buff *skb);
 137 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 138                                            struct sk_buff *skb, u32 mtu,
 139                                            bool confirm_neigh);
 140 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 141                                         struct sk_buff *skb);
 142 static void             ipv4_dst_destroy(struct dst_entry *dst);
 143
 144 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 145 {
 146         WARN_ON(1);
 147         return NULL;
 148 }
 149
 150 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 151                                            struct sk_buff *skb,
 152                                            const void *daddr);
 153 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
 154
 155 static struct dst_ops ipv4_dst_ops = {
 156         .family =               AF_INET,
 157         .check =                ipv4_dst_check,
 158         .default_advmss =       ipv4_default_advmss,
 159         .mtu =                  ipv4_mtu,
 160         .cow_metrics =          ipv4_cow_metrics,
 161         .destroy =              ipv4_dst_destroy,
 162         .negative_advice =      ipv4_negative_advice,
 163         .link_failure =         ipv4_link_failure,
 164         .update_pmtu =          ip_rt_update_pmtu,
 165         .redirect =             ip_do_redirect,
 166         .local_out =            __ip_local_out,
 167         .neigh_lookup =         ipv4_neigh_lookup,
 168         .confirm_neigh =        ipv4_confirm_neigh,
 169 };
 170
 171 #define ECN_OR_COST(class)      TC_PRIO_##class
 172
 173 const __u8 ip_tos2prio[16] = {
 174         TC_PRIO_BESTEFFORT,
 175         ECN_OR_COST(BESTEFFORT),
 176         TC_PRIO_BESTEFFORT,
 177         ECN_OR_COST(BESTEFFORT),
 178         TC_PRIO_BULK,
 179         ECN_OR_COST(BULK),
 180         TC_PRIO_BULK,
 181         ECN_OR_COST(BULK),
 182         TC_PRIO_INTERACTIVE,
 183         ECN_OR_COST(INTERACTIVE),
 184         TC_PRIO_INTERACTIVE,
 185         ECN_OR_COST(INTERACTIVE),
 186         TC_PRIO_INTERACTIVE_BULK,
 187         ECN_OR_COST(INTERACTIVE_BULK),
 188         TC_PRIO_INTERACTIVE_BULK,
 189         ECN_OR_COST(INTERACTIVE_BULK)
 190 };
 191 EXPORT_SYMBOL(ip_tos2prio);
 192
 193 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 194 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
 195
 196 #ifdef CONFIG_PROC_FS
 197 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 198 {
 199         if (*pos)
 200                 return NULL;
 201         return SEQ_START_TOKEN;
 202 }
 203
 204 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 205 {
 206         ++*pos;
 207         return NULL;
 208 }
 209
 210 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 211 {
 212 }
 213
 214 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 215 {
 216         if (v == SEQ_START_TOKEN)
 217                 seq_printf(seq, "%-127s\n",
 218                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 219                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 220                            "HHUptod\tSpecDst");
 221         return 0;
 222 }
 223
 224 static const struct seq_operations rt_cache_seq_ops = {
 225         .start  = rt_cache_seq_start,
 226         .next   = rt_cache_seq_next,
 227         .stop   = rt_cache_seq_stop,
 228         .show   = rt_cache_seq_show,
 229 };
 230
 231 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 232 {
 233         int cpu;
 234
 235         if (*pos == 0)
 236                 return SEQ_START_TOKEN;
 237
 238         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 239                 if (!cpu_possible(cpu))
 240                         continue;
 241                 *pos = cpu+1;
 242                 return &per_cpu(rt_cache_stat, cpu);
 243         }
 244         return NULL;
 245 }
 246
 247 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 248 {
 249         int cpu;
 250
 251         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 252                 if (!cpu_possible(cpu))
 253                         continue;
 254                 *pos = cpu+1;
 255                 return &per_cpu(rt_cache_stat, cpu);
 256         }
 257         (*pos)++;
 258         return NULL;
 259
 260 }
 261
 262 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 263 {
 264
 265 }
 266
 267 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 268 {
 269         struct rt_cache_stat *st = v;
 270
 271         if (v == SEQ_START_TOKEN) {
 272                 seq_puts(seq, "entries  in_hit   in_slow_tot in_slow_mc in_no_route in_brd   in_martian_dst in_martian_src out_hit  out_slow_tot out_slow_mc gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
 273                 return 0;
 274         }
 275
 276         seq_printf(seq, "%08x %08x %08x    %08x   %08x    %08x %08x       "
 277                         "%08x       %08x %08x     %08x    %08x %08x   "
 278                         "%08x     %08x        %08x        %08x\n",
 279                    dst_entries_get_slow(&ipv4_dst_ops),
 280                    0, /* st->in_hit */
 281                    st->in_slow_tot,
 282                    st->in_slow_mc,
 283                    st->in_no_route,
 284                    st->in_brd,
 285                    st->in_martian_dst,
 286                    st->in_martian_src,
 287
 288                    0, /* st->out_hit */
 289                    st->out_slow_tot,
 290                    st->out_slow_mc,
 291
 292                    0, /* st->gc_total */
 293                    0, /* st->gc_ignored */
 294                    0, /* st->gc_goal_miss */
 295                    0, /* st->gc_dst_overflow */
 296                    0, /* st->in_hlist_search */
 297                    0  /* st->out_hlist_search */
 298                 );
 299         return 0;
 300 }
 301
 302 static const struct seq_operations rt_cpu_seq_ops = {
 303         .start  = rt_cpu_seq_start,
 304         .next   = rt_cpu_seq_next,
 305         .stop   = rt_cpu_seq_stop,
 306         .show   = rt_cpu_seq_show,
 307 };
 308
 309 #ifdef CONFIG_IP_ROUTE_CLASSID
 310 static int rt_acct_proc_show(struct seq_file *m, void *v)
 311 {
 312         struct ip_rt_acct *dst, *src;
 313         unsigned int i, j;
 314
 315         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 316         if (!dst)
 317                 return -ENOMEM;
 318
 319         for_each_possible_cpu(i) {
 320                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 321                 for (j = 0; j < 256; j++) {
 322                         dst[j].o_bytes   += src[j].o_bytes;
 323                         dst[j].o_packets += src[j].o_packets;
 324                         dst[j].i_bytes   += src[j].i_bytes;
 325                         dst[j].i_packets += src[j].i_packets;
 326                 }
 327         }
 328
 329         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 330         kfree(dst);
 331         return 0;
 332 }
 333 #endif
 334
 335 static int __net_init ip_rt_do_proc_init(struct net *net)
 336 {
 337         struct proc_dir_entry *pde;
 338
 339         pde = proc_create_seq("rt_cache", 0444, net->proc_net,
 340                               &rt_cache_seq_ops);
 341         if (!pde)
 342                 goto err1;
 343
 344         pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
 345                               &rt_cpu_seq_ops);
 346         if (!pde)
 347                 goto err2;
 348
 349 #ifdef CONFIG_IP_ROUTE_CLASSID
 350         pde = proc_create_single("rt_acct", 0, net->proc_net,
 351                         rt_acct_proc_show);
 352         if (!pde)
 353                 goto err3;
 354 #endif
 355         return 0;
 356
 357 #ifdef CONFIG_IP_ROUTE_CLASSID
 358 err3:
 359         remove_proc_entry("rt_cache", net->proc_net_stat);
 360 #endif
 361 err2:
 362         remove_proc_entry("rt_cache", net->proc_net);
 363 err1:
 364         return -ENOMEM;
 365 }
 366
 367 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 368 {
 369         remove_proc_entry("rt_cache", net->proc_net_stat);
 370         remove_proc_entry("rt_cache", net->proc_net);
 371 #ifdef CONFIG_IP_ROUTE_CLASSID
 372         remove_proc_entry("rt_acct", net->proc_net);
 373 #endif
 374 }
 375
 376 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 377         .init = ip_rt_do_proc_init,
 378         .exit = ip_rt_do_proc_exit,
 379 };
 380
 381 static int __init ip_rt_proc_init(void)
 382 {
 383         return register_pernet_subsys(&ip_rt_proc_ops);
 384 }
 385
 386 #else
 387 static inline int ip_rt_proc_init(void)
 388 {
 389         return 0;
 390 }
 391 #endif /* CONFIG_PROC_FS */
 392
 393 static inline bool rt_is_expired(const struct rtable *rth)
 394 {
 395         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 396 }
 397
 398 void rt_cache_flush(struct net *net)
 399 {
 400         rt_genid_bump_ipv4(net);
 401 }
 402
 403 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 404                                            struct sk_buff *skb,
 405                                            const void *daddr)
 406 {
 407         const struct rtable *rt = container_of(dst, struct rtable, dst);
 408         struct net_device *dev = dst->dev;
 409         struct neighbour *n;
 410
 411         rcu_read_lock_bh();
 412
 413         if (likely(rt->rt_gw_family == AF_INET)) {
 414                 n = ip_neigh_gw4(dev, rt->rt_gw4);
 415         } else if (rt->rt_gw_family == AF_INET6) {
 416                 n = ip_neigh_gw6(dev, &rt->rt_gw6);
 417         } else {
 418                 __be32 pkey;
 419
 420                 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
 421                 n = ip_neigh_gw4(dev, pkey);
 422         }
 423
 424         if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
 425                 n = NULL;
 426
 427         rcu_read_unlock_bh();
 428
 429         return n;
 430 }
 431
 432 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 433 {
 434         const struct rtable *rt = container_of(dst, struct rtable, dst);
 435         struct net_device *dev = dst->dev;
 436         const __be32 *pkey = daddr;
 437
 438         if (rt->rt_gw_family == AF_INET) {
 439                 pkey = (const __be32 *)&rt->rt_gw4;
 440         } else if (rt->rt_gw_family == AF_INET6) {
 441                 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
 442         } else if (!daddr ||
 443                  (rt->rt_flags &
 444                   (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
 445                 return;
 446         }
 447         __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
 448 }
 449
 450 /* Hash tables of size 2048..262144 depending on RAM size.
 451  * Each bucket uses 8 bytes.
 452  */
 453 static u32 ip_idents_mask __read_mostly;
 454 static atomic_t *ip_idents __read_mostly;
 455 static u32 *ip_tstamps __read_mostly;
 456
 457 /* In order to protect privacy, we add a perturbation to identifiers
 458  * if one generator is seldom used. This makes hard for an attacker
 459  * to infer how many packets were sent between two points in time.
 460  */
 461 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 = prandom_u32_max(now - old);
 475
 476         /* If UBSAN reports an error there, please make sure your compiler
 477          * supports -fno-strict-overflow before reporting it that was a bug
 478          * in UBSAN, and it has been fixed in GCC-8.
 479          */
 480         return atomic_add_return(segs + delta, p_id) - segs;
 481 }
 482
 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         fl4->flowi4_scope = tos & RTO_ONLINK ?
 507                             RT_SCOPE_LINK : RT_SCOPE_UNIVERSE;
 508 }
 509
 510 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
 511                              const struct sock *sk,
 512                              const struct iphdr *iph,
 513                              int oif, u8 tos,
 514                              u8 prot, u32 mark, int flow_flags)
 515 {
 516         if (sk) {
 517                 const struct inet_sock *inet = inet_sk(sk);
 518
 519                 oif = sk->sk_bound_dev_if;
 520                 mark = sk->sk_mark;
 521                 tos = RT_CONN_FLAGS(sk);
 522                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 523         }
 524         flowi4_init_output(fl4, oif, mark, tos,
 525                            RT_SCOPE_UNIVERSE, prot,
 526                            flow_flags,
 527                            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 tos = RT_TOS(iph->tos);
 538         u8 prot = iph->protocol;
 539         u32 mark = skb->mark;
 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, sk->sk_mark,
 555                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 556                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 557                            inet_sk_flowi_flags(sk),
 558                            daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
 559         rcu_read_unlock();
 560 }
 561
 562 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 563                                  const struct sk_buff *skb)
 564 {
 565         if (skb)
 566                 build_skb_flow_key(fl4, skb, sk);
 567         else
 568                 build_sk_flow_key(fl4, sk);
 569 }
 570
 571 static DEFINE_SPINLOCK(fnhe_lock);
 572
 573 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 574 {
 575         struct rtable *rt;
 576
 577         rt = rcu_dereference(fnhe->fnhe_rth_input);
 578         if (rt) {
 579                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 580                 dst_dev_put(&rt->dst);
 581                 dst_release(&rt->dst);
 582         }
 583         rt = rcu_dereference(fnhe->fnhe_rth_output);
 584         if (rt) {
 585                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 586                 dst_dev_put(&rt->dst);
 587                 dst_release(&rt->dst);
 588         }
 589 }
 590
 591 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
 592 {
 593         struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
 594         struct fib_nh_exception *fnhe, *oldest = NULL;
 595
 596         for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
 597                 fnhe = rcu_dereference_protected(*fnhe_p,
 598                                                  lockdep_is_held(&fnhe_lock));
 599                 if (!fnhe)
 600                         break;
 601                 if (!oldest ||
 602                     time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
 603                         oldest = fnhe;
 604                         oldest_p = fnhe_p;
 605                 }
 606         }
 607         fnhe_flush_routes(oldest);
 608         *oldest_p = oldest->fnhe_next;
 609         kfree_rcu(oldest, rcu);
 610 }
 611
 612 static u32 fnhe_hashfun(__be32 daddr)
 613 {
 614         static siphash_aligned_key_t fnhe_hash_key;
 615         u64 hval;
 616
 617         net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
 618         hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
 619         return hash_64(hval, FNHE_HASH_SHIFT);
 620 }
 621
 622 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 623 {
 624         rt->rt_pmtu = fnhe->fnhe_pmtu;
 625         rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
 626         rt->dst.expires = fnhe->fnhe_expires;
 627
 628         if (fnhe->fnhe_gw) {
 629                 rt->rt_flags |= RTCF_REDIRECTED;
 630                 rt->rt_uses_gateway = 1;
 631                 rt->rt_gw_family = AF_INET;
 632                 rt->rt_gw4 = fnhe->fnhe_gw;
 633         }
 634 }
 635
 636 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
 637                                   __be32 gw, u32 pmtu, bool lock,
 638                                   unsigned long expires)
 639 {
 640         struct fnhe_hash_bucket *hash;
 641         struct fib_nh_exception *fnhe;
 642         struct rtable *rt;
 643         u32 genid, hval;
 644         unsigned int i;
 645         int depth;
 646
 647         genid = fnhe_genid(dev_net(nhc->nhc_dev));
 648         hval = fnhe_hashfun(daddr);
 649
 650         spin_lock_bh(&fnhe_lock);
 651
 652         hash = rcu_dereference(nhc->nhc_exceptions);
 653         if (!hash) {
 654                 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
 655                 if (!hash)
 656                         goto out_unlock;
 657                 rcu_assign_pointer(nhc->nhc_exceptions, hash);
 658         }
 659
 660         hash += hval;
 661
 662         depth = 0;
 663         for (fnhe = rcu_dereference(hash->chain); fnhe;
 664              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 665                 if (fnhe->fnhe_daddr == daddr)
 666                         break;
 667                 depth++;
 668         }
 669
 670         if (fnhe) {
 671                 if (fnhe->fnhe_genid != genid)
 672                         fnhe->fnhe_genid = genid;
 673                 if (gw)
 674                         fnhe->fnhe_gw = gw;
 675                 if (pmtu) {
 676                         fnhe->fnhe_pmtu = pmtu;
 677                         fnhe->fnhe_mtu_locked = lock;
 678                 }
 679                 fnhe->fnhe_expires = max(1UL, expires);
 680                 /* Update all cached dsts too */
 681                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 682                 if (rt)
 683                         fill_route_from_fnhe(rt, fnhe);
 684                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 685                 if (rt)
 686                         fill_route_from_fnhe(rt, fnhe);
 687         } else {
 688                 /* Randomize max depth to avoid some side channels attacks. */
 689                 int max_depth = FNHE_RECLAIM_DEPTH +
 690                                 prandom_u32_max(FNHE_RECLAIM_DEPTH);
 691
 692                 while (depth > max_depth) {
 693                         fnhe_remove_oldest(hash);
 694                         depth--;
 695                 }
 696
 697                 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 698                 if (!fnhe)
 699                         goto out_unlock;
 700
 701                 fnhe->fnhe_next = hash->chain;
 702
 703                 fnhe->fnhe_genid = genid;
 704                 fnhe->fnhe_daddr = daddr;
 705                 fnhe->fnhe_gw = gw;
 706                 fnhe->fnhe_pmtu = pmtu;
 707                 fnhe->fnhe_mtu_locked = lock;
 708                 fnhe->fnhe_expires = max(1UL, expires);
 709
 710                 rcu_assign_pointer(hash->chain, fnhe);
 711
 712                 /* Exception created; mark the cached routes for the nexthop
 713                  * stale, so anyone caching it rechecks if this exception
 714                  * applies to them.
 715                  */
 716                 rt = rcu_dereference(nhc->nhc_rth_input);
 717                 if (rt)
 718                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 719
 720                 for_each_possible_cpu(i) {
 721                         struct rtable __rcu **prt;
 722
 723                         prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
 724                         rt = rcu_dereference(*prt);
 725                         if (rt)
 726                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 727                 }
 728         }
 729
 730         fnhe->fnhe_stamp = jiffies;
 731
 732 out_unlock:
 733         spin_unlock_bh(&fnhe_lock);
 734 }
 735
 736 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 737                              bool kill_route)
 738 {
 739         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 740         __be32 old_gw = ip_hdr(skb)->saddr;
 741         struct net_device *dev = skb->dev;
 742         struct in_device *in_dev;
 743         struct fib_result res;
 744         struct neighbour *n;
 745         struct net *net;
 746
 747         switch (icmp_hdr(skb)->code & 7) {
 748         case ICMP_REDIR_NET:
 749         case ICMP_REDIR_NETTOS:
 750         case ICMP_REDIR_HOST:
 751         case ICMP_REDIR_HOSTTOS:
 752                 break;
 753
 754         default:
 755                 return;
 756         }
 757
 758         if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
 759                 return;
 760
 761         in_dev = __in_dev_get_rcu(dev);
 762         if (!in_dev)
 763                 return;
 764
 765         net = dev_net(dev);
 766         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 767             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 768             ipv4_is_zeronet(new_gw))
 769                 goto reject_redirect;
 770
 771         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 772                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 773                         goto reject_redirect;
 774                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 775                         goto reject_redirect;
 776         } else {
 777                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 778                         goto reject_redirect;
 779         }
 780
 781         n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
 782         if (!n)
 783                 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
 784         if (!IS_ERR(n)) {
 785                 if (!(n->nud_state & NUD_VALID)) {
 786                         neigh_event_send(n, NULL);
 787                 } else {
 788                         if (fib_lookup(net, fl4, &res, 0) == 0) {
 789                                 struct fib_nh_common *nhc;
 790
 791                                 fib_select_path(net, &res, fl4, skb);
 792                                 nhc = FIB_RES_NHC(res);
 793                                 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
 794                                                 0, false,
 795                                                 jiffies + ip_rt_gc_timeout);
 796                         }
 797                         if (kill_route)
 798                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 799                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 800                 }
 801                 neigh_release(n);
 802         }
 803         return;
 804
 805 reject_redirect:
 806 #ifdef CONFIG_IP_ROUTE_VERBOSE
 807         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 808                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 809                 __be32 daddr = iph->daddr;
 810                 __be32 saddr = iph->saddr;
 811
 812                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 813                                      "  Advised path = %pI4 -> %pI4\n",
 814                                      &old_gw, dev->name, &new_gw,
 815                                      &saddr, &daddr);
 816         }
 817 #endif
 818         ;
 819 }
 820
 821 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 822 {
 823         struct rtable *rt;
 824         struct flowi4 fl4;
 825         const struct iphdr *iph = (const struct iphdr *) skb->data;
 826         struct net *net = dev_net(skb->dev);
 827         int oif = skb->dev->ifindex;
 828         u8 tos = RT_TOS(iph->tos);
 829         u8 prot = iph->protocol;
 830         u32 mark = skb->mark;
 831
 832         rt = (struct rtable *) dst;
 833
 834         __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
 835         ip_rt_fix_tos(&fl4);
 836         __ip_do_redirect(rt, skb, &fl4, true);
 837 }
 838
 839 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 840 {
 841         struct rtable *rt = (struct rtable *)dst;
 842         struct dst_entry *ret = dst;
 843
 844         if (rt) {
 845                 if (dst->obsolete > 0) {
 846                         ip_rt_put(rt);
 847                         ret = NULL;
 848                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 849                            rt->dst.expires) {
 850                         ip_rt_put(rt);
 851                         ret = NULL;
 852                 }
 853         }
 854         return ret;
 855 }
 856
 857 /*
 858  * Algorithm:
 859  *      1. The first ip_rt_redirect_number redirects are sent
 860  *         with exponential backoff, then we stop sending them at all,
 861  *         assuming that the host ignores our redirects.
 862  *      2. If we did not see packets requiring redirects
 863  *         during ip_rt_redirect_silence, we assume that the host
 864  *         forgot redirected route and start to send redirects again.
 865  *
 866  * This algorithm is much cheaper and more intelligent than dumb load limiting
 867  * in icmp.c.
 868  *
 869  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 870  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 871  */
 872
 873 void ip_rt_send_redirect(struct sk_buff *skb)
 874 {
 875         struct rtable *rt = skb_rtable(skb);
 876         struct in_device *in_dev;
 877         struct inet_peer *peer;
 878         struct net *net;
 879         int log_martians;
 880         int vif;
 881
 882         rcu_read_lock();
 883         in_dev = __in_dev_get_rcu(rt->dst.dev);
 884         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 885                 rcu_read_unlock();
 886                 return;
 887         }
 888         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 889         vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
 890         rcu_read_unlock();
 891
 892         net = dev_net(rt->dst.dev);
 893         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
 894         if (!peer) {
 895                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 896                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 897                 return;
 898         }
 899
 900         /* No redirected packets during ip_rt_redirect_silence;
 901          * reset the algorithm.
 902          */
 903         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
 904                 peer->rate_tokens = 0;
 905                 peer->n_redirects = 0;
 906         }
 907
 908         /* Too many ignored redirects; do not send anything
 909          * set dst.rate_last to the last seen redirected packet.
 910          */
 911         if (peer->n_redirects >= ip_rt_redirect_number) {
 912                 peer->rate_last = jiffies;
 913                 goto out_put_peer;
 914         }
 915
 916         /* Check for load limit; set rate_last to the latest sent
 917          * redirect.
 918          */
 919         if (peer->n_redirects == 0 ||
 920             time_after(jiffies,
 921                        (peer->rate_last +
 922                         (ip_rt_redirect_load << peer->n_redirects)))) {
 923                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 924
 925                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 926                 peer->rate_last = jiffies;
 927                 ++peer->n_redirects;
 928 #ifdef CONFIG_IP_ROUTE_VERBOSE
 929                 if (log_martians &&
 930                     peer->n_redirects == ip_rt_redirect_number)
 931                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 932                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 933                                              &ip_hdr(skb)->daddr, &gw);
 934 #endif
 935         }
 936 out_put_peer:
 937         inet_putpeer(peer);
 938 }
 939
 940 static int ip_error(struct sk_buff *skb)
 941 {
 942         struct rtable *rt = skb_rtable(skb);
 943         struct net_device *dev = skb->dev;
 944         struct in_device *in_dev;
 945         struct inet_peer *peer;
 946         unsigned long now;
 947         struct net *net;
 948         bool send;
 949         int code;
 950
 951         if (netif_is_l3_master(skb->dev)) {
 952                 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
 953                 if (!dev)
 954                         goto out;
 955         }
 956
 957         in_dev = __in_dev_get_rcu(dev);
 958
 959         /* IP on this device is disabled. */
 960         if (!in_dev)
 961                 goto out;
 962
 963         net = dev_net(rt->dst.dev);
 964         if (!IN_DEV_FORWARD(in_dev)) {
 965                 switch (rt->dst.error) {
 966                 case EHOSTUNREACH:
 967                         __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
 968                         break;
 969
 970                 case ENETUNREACH:
 971                         __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 972                         break;
 973                 }
 974                 goto out;
 975         }
 976
 977         switch (rt->dst.error) {
 978         case EINVAL:
 979         default:
 980                 goto out;
 981         case EHOSTUNREACH:
 982                 code = ICMP_HOST_UNREACH;
 983                 break;
 984         case ENETUNREACH:
 985                 code = ICMP_NET_UNREACH;
 986                 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 987                 break;
 988         case EACCES:
 989                 code = ICMP_PKT_FILTERED;
 990                 break;
 991         }
 992
 993         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
 994                                l3mdev_master_ifindex(skb->dev), 1);
 995
 996         send = true;
 997         if (peer) {
 998                 now = jiffies;
 999                 peer->rate_tokens += now - peer->rate_last;
1000                 if (peer->rate_tokens > ip_rt_error_burst)
1001                         peer->rate_tokens = ip_rt_error_burst;
1002                 peer->rate_last = now;
1003                 if (peer->rate_tokens >= ip_rt_error_cost)
1004                         peer->rate_tokens -= ip_rt_error_cost;
1005                 else
1006                         send = false;
1007                 inet_putpeer(peer);
1008         }
1009         if (send)
1010                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1011
1012 out:    kfree_skb(skb);
1013         return 0;
1014 }
1015
1016 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1017 {
1018         struct dst_entry *dst = &rt->dst;
1019         struct net *net = dev_net(dst->dev);
1020         struct fib_result res;
1021         bool lock = false;
1022         u32 old_mtu;
1023
1024         if (ip_mtu_locked(dst))
1025                 return;
1026
1027         old_mtu = ipv4_mtu(dst);
1028         if (old_mtu < mtu)
1029                 return;
1030
1031         if (mtu < net->ipv4.ip_rt_min_pmtu) {
1032                 lock = true;
1033                 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu);
1034         }
1035
1036         if (rt->rt_pmtu == mtu && !lock &&
1037             time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2))
1038                 return;
1039
1040         rcu_read_lock();
1041         if (fib_lookup(net, fl4, &res, 0) == 0) {
1042                 struct fib_nh_common *nhc;
1043
1044                 fib_select_path(net, &res, fl4, NULL);
1045                 nhc = FIB_RES_NHC(res);
1046                 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1047                                       jiffies + net->ipv4.ip_rt_mtu_expires);
1048         }
1049         rcu_read_unlock();
1050 }
1051
1052 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1053                               struct sk_buff *skb, u32 mtu,
1054                               bool confirm_neigh)
1055 {
1056         struct rtable *rt = (struct rtable *) dst;
1057         struct flowi4 fl4;
1058
1059         ip_rt_build_flow_key(&fl4, sk, skb);
1060         ip_rt_fix_tos(&fl4);
1061
1062         /* Don't make lookup fail for bridged encapsulations */
1063         if (skb && netif_is_any_bridge_port(skb->dev))
1064                 fl4.flowi4_oif = 0;
1065
1066         __ip_rt_update_pmtu(rt, &fl4, mtu);
1067 }
1068
1069 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1070                       int oif, u8 protocol)
1071 {
1072         const struct iphdr *iph = (const struct iphdr *)skb->data;
1073         struct flowi4 fl4;
1074         struct rtable *rt;
1075         u32 mark = IP4_REPLY_MARK(net, skb->mark);
1076
1077         __build_flow_key(net, &fl4, NULL, iph, oif,
1078                          RT_TOS(iph->tos), protocol, mark, 0);
1079         rt = __ip_route_output_key(net, &fl4);
1080         if (!IS_ERR(rt)) {
1081                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1082                 ip_rt_put(rt);
1083         }
1084 }
1085 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1086
1087 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1088 {
1089         const struct iphdr *iph = (const struct iphdr *)skb->data;
1090         struct flowi4 fl4;
1091         struct rtable *rt;
1092
1093         __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1094
1095         if (!fl4.flowi4_mark)
1096                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1097
1098         rt = __ip_route_output_key(sock_net(sk), &fl4);
1099         if (!IS_ERR(rt)) {
1100                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1101                 ip_rt_put(rt);
1102         }
1103 }
1104
1105 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1106 {
1107         const struct iphdr *iph = (const struct iphdr *)skb->data;
1108         struct flowi4 fl4;
1109         struct rtable *rt;
1110         struct dst_entry *odst = NULL;
1111         bool new = false;
1112         struct net *net = sock_net(sk);
1113
1114         bh_lock_sock(sk);
1115
1116         if (!ip_sk_accept_pmtu(sk))
1117                 goto out;
1118
1119         odst = sk_dst_get(sk);
1120
1121         if (sock_owned_by_user(sk) || !odst) {
1122                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1123                 goto out;
1124         }
1125
1126         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1127
1128         rt = (struct rtable *)odst;
1129         if (odst->obsolete && !odst->ops->check(odst, 0)) {
1130                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1131                 if (IS_ERR(rt))
1132                         goto out;
1133
1134                 new = true;
1135         } else {
1136                 ip_rt_fix_tos(&fl4);
1137         }
1138
1139         __ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu);
1140
1141         if (!dst_check(&rt->dst, 0)) {
1142                 if (new)
1143                         dst_release(&rt->dst);
1144
1145                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1146                 if (IS_ERR(rt))
1147                         goto out;
1148
1149                 new = true;
1150         }
1151
1152         if (new)
1153                 sk_dst_set(sk, &rt->dst);
1154
1155 out:
1156         bh_unlock_sock(sk);
1157         dst_release(odst);
1158 }
1159 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1160
1161 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1162                    int oif, u8 protocol)
1163 {
1164         const struct iphdr *iph = (const struct iphdr *)skb->data;
1165         struct flowi4 fl4;
1166         struct rtable *rt;
1167
1168         __build_flow_key(net, &fl4, NULL, iph, oif,
1169                          RT_TOS(iph->tos), protocol, 0, 0);
1170         rt = __ip_route_output_key(net, &fl4);
1171         if (!IS_ERR(rt)) {
1172                 __ip_do_redirect(rt, skb, &fl4, false);
1173                 ip_rt_put(rt);
1174         }
1175 }
1176 EXPORT_SYMBOL_GPL(ipv4_redirect);
1177
1178 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1179 {
1180         const struct iphdr *iph = (const struct iphdr *)skb->data;
1181         struct flowi4 fl4;
1182         struct rtable *rt;
1183         struct net *net = sock_net(sk);
1184
1185         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1186         rt = __ip_route_output_key(net, &fl4);
1187         if (!IS_ERR(rt)) {
1188                 __ip_do_redirect(rt, skb, &fl4, false);
1189                 ip_rt_put(rt);
1190         }
1191 }
1192 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1193
1194 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1195                                                          u32 cookie)
1196 {
1197         struct rtable *rt = (struct rtable *) dst;
1198
1199         /* All IPV4 dsts are created with ->obsolete set to the value
1200          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1201          * into this function always.
1202          *
1203          * When a PMTU/redirect information update invalidates a route,
1204          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1205          * DST_OBSOLETE_DEAD.
1206          */
1207         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1208                 return NULL;
1209         return dst;
1210 }
1211 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1212
1213 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1214 {
1215         struct ip_options opt;
1216         int res;
1217
1218         /* Recompile ip options since IPCB may not be valid anymore.
1219          * Also check we have a reasonable ipv4 header.
1220          */
1221         if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1222             ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1223                 return;
1224
1225         memset(&opt, 0, sizeof(opt));
1226         if (ip_hdr(skb)->ihl > 5) {
1227                 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1228                         return;
1229                 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1230
1231                 rcu_read_lock();
1232                 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1233                 rcu_read_unlock();
1234
1235                 if (res)
1236                         return;
1237         }
1238         __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1239 }
1240
1241 static void ipv4_link_failure(struct sk_buff *skb)
1242 {
1243         struct rtable *rt;
1244
1245         ipv4_send_dest_unreach(skb);
1246
1247         rt = skb_rtable(skb);
1248         if (rt)
1249                 dst_set_expires(&rt->dst, 0);
1250 }
1251
1252 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1253 {
1254         pr_debug("%s: %pI4 -> %pI4, %s\n",
1255                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1256                  skb->dev ? skb->dev->name : "?");
1257         kfree_skb(skb);
1258         WARN_ON(1);
1259         return 0;
1260 }
1261
1262 /*
1263  * We do not cache source address of outgoing interface,
1264  * because it is used only by IP RR, TS and SRR options,
1265  * so that it out of fast path.
1266  *
1267  * BTW remember: "addr" is allowed to be not aligned
1268  * in IP options!
1269  */
1270
1271 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1272 {
1273         __be32 src;
1274
1275         if (rt_is_output_route(rt))
1276                 src = ip_hdr(skb)->saddr;
1277         else {
1278                 struct fib_result res;
1279                 struct iphdr *iph = ip_hdr(skb);
1280                 struct flowi4 fl4 = {
1281                         .daddr = iph->daddr,
1282                         .saddr = iph->saddr,
1283                         .flowi4_tos = RT_TOS(iph->tos),
1284                         .flowi4_oif = rt->dst.dev->ifindex,
1285                         .flowi4_iif = skb->dev->ifindex,
1286                         .flowi4_mark = skb->mark,
1287                 };
1288
1289                 rcu_read_lock();
1290                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1291                         src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1292                 else
1293                         src = inet_select_addr(rt->dst.dev,
1294                                                rt_nexthop(rt, iph->daddr),
1295                                                RT_SCOPE_UNIVERSE);
1296                 rcu_read_unlock();
1297         }
1298         memcpy(addr, &src, 4);
1299 }
1300
1301 #ifdef CONFIG_IP_ROUTE_CLASSID
1302 static void set_class_tag(struct rtable *rt, u32 tag)
1303 {
1304         if (!(rt->dst.tclassid & 0xFFFF))
1305                 rt->dst.tclassid |= tag & 0xFFFF;
1306         if (!(rt->dst.tclassid & 0xFFFF0000))
1307                 rt->dst.tclassid |= tag & 0xFFFF0000;
1308 }
1309 #endif
1310
1311 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1312 {
1313         struct net *net = dev_net(dst->dev);
1314         unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1315         unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1316                                     net->ipv4.ip_rt_min_advmss);
1317
1318         return min(advmss, IPV4_MAX_PMTU - header_size);
1319 }
1320
1321 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1322 {
1323         return ip_dst_mtu_maybe_forward(dst, false);
1324 }
1325 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1326
1327 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1328 {
1329         struct fnhe_hash_bucket *hash;
1330         struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1331         u32 hval = fnhe_hashfun(daddr);
1332
1333         spin_lock_bh(&fnhe_lock);
1334
1335         hash = rcu_dereference_protected(nhc->nhc_exceptions,
1336                                          lockdep_is_held(&fnhe_lock));
1337         hash += hval;
1338
1339         fnhe_p = &hash->chain;
1340         fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1341         while (fnhe) {
1342                 if (fnhe->fnhe_daddr == daddr) {
1343                         rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1344                                 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1345                         /* set fnhe_daddr to 0 to ensure it won't bind with
1346                          * new dsts in rt_bind_exception().
1347                          */
1348                         fnhe->fnhe_daddr = 0;
1349                         fnhe_flush_routes(fnhe);
1350                         kfree_rcu(fnhe, rcu);
1351                         break;
1352                 }
1353                 fnhe_p = &fnhe->fnhe_next;
1354                 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1355                                                  lockdep_is_held(&fnhe_lock));
1356         }
1357
1358         spin_unlock_bh(&fnhe_lock);
1359 }
1360
1361 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1362                                                __be32 daddr)
1363 {
1364         struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1365         struct fib_nh_exception *fnhe;
1366         u32 hval;
1367
1368         if (!hash)
1369                 return NULL;
1370
1371         hval = fnhe_hashfun(daddr);
1372
1373         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1374              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1375                 if (fnhe->fnhe_daddr == daddr) {
1376                         if (fnhe->fnhe_expires &&
1377                             time_after(jiffies, fnhe->fnhe_expires)) {
1378                                 ip_del_fnhe(nhc, daddr);
1379                                 break;
1380                         }
1381                         return fnhe;
1382                 }
1383         }
1384         return NULL;
1385 }
1386
1387 /* MTU selection:
1388  * 1. mtu on route is locked - use it
1389  * 2. mtu from nexthop exception
1390  * 3. mtu from egress device
1391  */
1392
1393 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1394 {
1395         struct fib_nh_common *nhc = res->nhc;
1396         struct net_device *dev = nhc->nhc_dev;
1397         struct fib_info *fi = res->fi;
1398         u32 mtu = 0;
1399
1400         if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1401             fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1402                 mtu = fi->fib_mtu;
1403
1404         if (likely(!mtu)) {
1405                 struct fib_nh_exception *fnhe;
1406
1407                 fnhe = find_exception(nhc, daddr);
1408                 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1409                         mtu = fnhe->fnhe_pmtu;
1410         }
1411
1412         if (likely(!mtu))
1413                 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1414
1415         return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1416 }
1417
1418 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1419                               __be32 daddr, const bool do_cache)
1420 {
1421         bool ret = false;
1422
1423         spin_lock_bh(&fnhe_lock);
1424
1425         if (daddr == fnhe->fnhe_daddr) {
1426                 struct rtable __rcu **porig;
1427                 struct rtable *orig;
1428                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1429
1430                 if (rt_is_input_route(rt))
1431                         porig = &fnhe->fnhe_rth_input;
1432                 else
1433                         porig = &fnhe->fnhe_rth_output;
1434                 orig = rcu_dereference(*porig);
1435
1436                 if (fnhe->fnhe_genid != genid) {
1437                         fnhe->fnhe_genid = genid;
1438                         fnhe->fnhe_gw = 0;
1439                         fnhe->fnhe_pmtu = 0;
1440                         fnhe->fnhe_expires = 0;
1441                         fnhe->fnhe_mtu_locked = false;
1442                         fnhe_flush_routes(fnhe);
1443                         orig = NULL;
1444                 }
1445                 fill_route_from_fnhe(rt, fnhe);
1446                 if (!rt->rt_gw4) {
1447                         rt->rt_gw4 = daddr;
1448                         rt->rt_gw_family = AF_INET;
1449                 }
1450
1451                 if (do_cache) {
1452                         dst_hold(&rt->dst);
1453                         rcu_assign_pointer(*porig, rt);
1454                         if (orig) {
1455                                 dst_dev_put(&orig->dst);
1456                                 dst_release(&orig->dst);
1457                         }
1458                         ret = true;
1459                 }
1460
1461                 fnhe->fnhe_stamp = jiffies;
1462         }
1463         spin_unlock_bh(&fnhe_lock);
1464
1465         return ret;
1466 }
1467
1468 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1469 {
1470         struct rtable *orig, *prev, **p;
1471         bool ret = true;
1472
1473         if (rt_is_input_route(rt)) {
1474                 p = (struct rtable **)&nhc->nhc_rth_input;
1475         } else {
1476                 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1477         }
1478         orig = *p;
1479
1480         /* hold dst before doing cmpxchg() to avoid race condition
1481          * on this dst
1482          */
1483         dst_hold(&rt->dst);
1484         prev = cmpxchg(p, orig, rt);
1485         if (prev == orig) {
1486                 if (orig) {
1487                         rt_add_uncached_list(orig);
1488                         dst_release(&orig->dst);
1489                 }
1490         } else {
1491                 dst_release(&rt->dst);
1492                 ret = false;
1493         }
1494
1495         return ret;
1496 }
1497
1498 struct uncached_list {
1499         spinlock_t              lock;
1500         struct list_head        head;
1501         struct list_head        quarantine;
1502 };
1503
1504 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1505
1506 void rt_add_uncached_list(struct rtable *rt)
1507 {
1508         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1509
1510         rt->rt_uncached_list = ul;
1511
1512         spin_lock_bh(&ul->lock);
1513         list_add_tail(&rt->rt_uncached, &ul->head);
1514         spin_unlock_bh(&ul->lock);
1515 }
1516
1517 void rt_del_uncached_list(struct rtable *rt)
1518 {
1519         if (!list_empty(&rt->rt_uncached)) {
1520                 struct uncached_list *ul = rt->rt_uncached_list;
1521
1522                 spin_lock_bh(&ul->lock);
1523                 list_del_init(&rt->rt_uncached);
1524                 spin_unlock_bh(&ul->lock);
1525         }
1526 }
1527
1528 static void ipv4_dst_destroy(struct dst_entry *dst)
1529 {
1530         struct rtable *rt = (struct rtable *)dst;
1531
1532         ip_dst_metrics_put(dst);
1533         rt_del_uncached_list(rt);
1534 }
1535
1536 void rt_flush_dev(struct net_device *dev)
1537 {
1538         struct rtable *rt, *safe;
1539         int cpu;
1540
1541         for_each_possible_cpu(cpu) {
1542                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1543
1544                 if (list_empty(&ul->head))
1545                         continue;
1546
1547                 spin_lock_bh(&ul->lock);
1548                 list_for_each_entry_safe(rt, safe, &ul->head, rt_uncached) {
1549                         if (rt->dst.dev != dev)
1550                                 continue;
1551                         rt->dst.dev = blackhole_netdev;
1552                         dev_replace_track(dev, blackhole_netdev,
1553                                           &rt->dst.dev_tracker,
1554                                           GFP_ATOMIC);
1555                         list_move(&rt->rt_uncached, &ul->quarantine);
1556                 }
1557                 spin_unlock_bh(&ul->lock);
1558         }
1559 }
1560
1561 static bool rt_cache_valid(const struct rtable *rt)
1562 {
1563         return  rt &&
1564                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1565                 !rt_is_expired(rt);
1566 }
1567
1568 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1569                            const struct fib_result *res,
1570                            struct fib_nh_exception *fnhe,
1571                            struct fib_info *fi, u16 type, u32 itag,
1572                            const bool do_cache)
1573 {
1574         bool cached = false;
1575
1576         if (fi) {
1577                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1578
1579                 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1580                         rt->rt_uses_gateway = 1;
1581                         rt->rt_gw_family = nhc->nhc_gw_family;
1582                         /* only INET and INET6 are supported */
1583                         if (likely(nhc->nhc_gw_family == AF_INET))
1584                                 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1585                         else
1586                                 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1587                 }
1588
1589                 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1590
1591 #ifdef CONFIG_IP_ROUTE_CLASSID
1592                 if (nhc->nhc_family == AF_INET) {
1593                         struct fib_nh *nh;
1594
1595                         nh = container_of(nhc, struct fib_nh, nh_common);
1596                         rt->dst.tclassid = nh->nh_tclassid;
1597                 }
1598 #endif
1599                 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1600                 if (unlikely(fnhe))
1601                         cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1602                 else if (do_cache)
1603                         cached = rt_cache_route(nhc, rt);
1604                 if (unlikely(!cached)) {
1605                         /* Routes we intend to cache in nexthop exception or
1606                          * FIB nexthop have the DST_NOCACHE bit clear.
1607                          * However, if we are unsuccessful at storing this
1608                          * route into the cache we really need to set it.
1609                          */
1610                         if (!rt->rt_gw4) {
1611                                 rt->rt_gw_family = AF_INET;
1612                                 rt->rt_gw4 = daddr;
1613                         }
1614                         rt_add_uncached_list(rt);
1615                 }
1616         } else
1617                 rt_add_uncached_list(rt);
1618
1619 #ifdef CONFIG_IP_ROUTE_CLASSID
1620 #ifdef CONFIG_IP_MULTIPLE_TABLES
1621         set_class_tag(rt, res->tclassid);
1622 #endif
1623         set_class_tag(rt, itag);
1624 #endif
1625 }
1626
1627 struct rtable *rt_dst_alloc(struct net_device *dev,
1628                             unsigned int flags, u16 type,
1629                             bool nopolicy, bool noxfrm)
1630 {
1631         struct rtable *rt;
1632
1633         rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1634                        (nopolicy ? DST_NOPOLICY : 0) |
1635                        (noxfrm ? DST_NOXFRM : 0));
1636
1637         if (rt) {
1638                 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1639                 rt->rt_flags = flags;
1640                 rt->rt_type = type;
1641                 rt->rt_is_input = 0;
1642                 rt->rt_iif = 0;
1643                 rt->rt_pmtu = 0;
1644                 rt->rt_mtu_locked = 0;
1645                 rt->rt_uses_gateway = 0;
1646                 rt->rt_gw_family = 0;
1647                 rt->rt_gw4 = 0;
1648                 INIT_LIST_HEAD(&rt->rt_uncached);
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, 1, 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                 INIT_LIST_HEAD(&new_rt->rt_uncached);
1680
1681                 new_rt->dst.input = rt->dst.input;
1682                 new_rt->dst.output = rt->dst.output;
1683                 new_rt->dst.error = rt->dst.error;
1684                 new_rt->dst.lastuse = jiffies;
1685                 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1686         }
1687         return new_rt;
1688 }
1689 EXPORT_SYMBOL(rt_dst_clone);
1690
1691 /* called in rcu_read_lock() section */
1692 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1693                           u8 tos, struct net_device *dev,
1694                           struct in_device *in_dev, u32 *itag)
1695 {
1696         int err;
1697
1698         /* Primary sanity checks. */
1699         if (!in_dev)
1700                 return -EINVAL;
1701
1702         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1703             skb->protocol != htons(ETH_P_IP))
1704                 return -EINVAL;
1705
1706         if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1707                 return -EINVAL;
1708
1709         if (ipv4_is_zeronet(saddr)) {
1710                 if (!ipv4_is_local_multicast(daddr) &&
1711                     ip_hdr(skb)->protocol != IPPROTO_IGMP)
1712                         return -EINVAL;
1713         } else {
1714                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1715                                           in_dev, itag);
1716                 if (err < 0)
1717                         return err;
1718         }
1719         return 0;
1720 }
1721
1722 /* called in rcu_read_lock() section */
1723 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1724                              u8 tos, struct net_device *dev, int our)
1725 {
1726         struct in_device *in_dev = __in_dev_get_rcu(dev);
1727         unsigned int flags = RTCF_MULTICAST;
1728         struct rtable *rth;
1729         u32 itag = 0;
1730         int err;
1731
1732         err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1733         if (err)
1734                 return err;
1735
1736         if (our)
1737                 flags |= RTCF_LOCAL;
1738
1739         rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1740                            IN_DEV_ORCONF(in_dev, NOPOLICY), false);
1741         if (!rth)
1742                 return -ENOBUFS;
1743
1744 #ifdef CONFIG_IP_ROUTE_CLASSID
1745         rth->dst.tclassid = itag;
1746 #endif
1747         rth->dst.output = ip_rt_bug;
1748         rth->rt_is_input= 1;
1749
1750 #ifdef CONFIG_IP_MROUTE
1751         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1752                 rth->dst.input = ip_mr_input;
1753 #endif
1754         RT_CACHE_STAT_INC(in_slow_mc);
1755
1756         skb_dst_set(skb, &rth->dst);
1757         return 0;
1758 }
1759
1760
1761 static void ip_handle_martian_source(struct net_device *dev,
1762                                      struct in_device *in_dev,
1763                                      struct sk_buff *skb,
1764                                      __be32 daddr,
1765                                      __be32 saddr)
1766 {
1767         RT_CACHE_STAT_INC(in_martian_src);
1768 #ifdef CONFIG_IP_ROUTE_VERBOSE
1769         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1770                 /*
1771                  *      RFC1812 recommendation, if source is martian,
1772                  *      the only hint is MAC header.
1773                  */
1774                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1775                         &daddr, &saddr, dev->name);
1776                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1777                         print_hex_dump(KERN_WARNING, "ll header: ",
1778                                        DUMP_PREFIX_OFFSET, 16, 1,
1779                                        skb_mac_header(skb),
1780                                        dev->hard_header_len, false);
1781                 }
1782         }
1783 #endif
1784 }
1785
1786 /* called in rcu_read_lock() section */
1787 static int __mkroute_input(struct sk_buff *skb,
1788                            const struct fib_result *res,
1789                            struct in_device *in_dev,
1790                            __be32 daddr, __be32 saddr, u32 tos)
1791 {
1792         struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1793         struct net_device *dev = nhc->nhc_dev;
1794         struct fib_nh_exception *fnhe;
1795         struct rtable *rth;
1796         int err;
1797         struct in_device *out_dev;
1798         bool do_cache;
1799         u32 itag = 0;
1800
1801         /* get a working reference to the output device */
1802         out_dev = __in_dev_get_rcu(dev);
1803         if (!out_dev) {
1804                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1805                 return -EINVAL;
1806         }
1807
1808         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1809                                   in_dev->dev, in_dev, &itag);
1810         if (err < 0) {
1811                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1812                                          saddr);
1813
1814                 goto cleanup;
1815         }
1816
1817         do_cache = res->fi && !itag;
1818         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1819             skb->protocol == htons(ETH_P_IP)) {
1820                 __be32 gw;
1821
1822                 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1823                 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1824                     inet_addr_onlink(out_dev, saddr, gw))
1825                         IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1826         }
1827
1828         if (skb->protocol != htons(ETH_P_IP)) {
1829                 /* Not IP (i.e. ARP). Do not create route, if it is
1830                  * invalid for proxy arp. DNAT routes are always valid.
1831                  *
1832                  * Proxy arp feature have been extended to allow, ARP
1833                  * replies back to the same interface, to support
1834                  * Private VLAN switch technologies. See arp.c.
1835                  */
1836                 if (out_dev == in_dev &&
1837                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1838                         err = -EINVAL;
1839                         goto cleanup;
1840                 }
1841         }
1842
1843         fnhe = find_exception(nhc, daddr);
1844         if (do_cache) {
1845                 if (fnhe)
1846                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1847                 else
1848                         rth = rcu_dereference(nhc->nhc_rth_input);
1849                 if (rt_cache_valid(rth)) {
1850                         skb_dst_set_noref(skb, &rth->dst);
1851                         goto out;
1852                 }
1853         }
1854
1855         rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1856                            IN_DEV_ORCONF(in_dev, NOPOLICY),
1857                            IN_DEV_ORCONF(out_dev, NOXFRM));
1858         if (!rth) {
1859                 err = -ENOBUFS;
1860                 goto cleanup;
1861         }
1862
1863         rth->rt_is_input = 1;
1864         RT_CACHE_STAT_INC(in_slow_tot);
1865
1866         rth->dst.input = ip_forward;
1867
1868         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1869                        do_cache);
1870         lwtunnel_set_redirect(&rth->dst);
1871         skb_dst_set(skb, &rth->dst);
1872 out:
1873         err = 0;
1874  cleanup:
1875         return err;
1876 }
1877
1878 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1879 /* To make ICMP packets follow the right flow, the multipath hash is
1880  * calculated from the inner IP addresses.
1881  */
1882 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1883                                  struct flow_keys *hash_keys)
1884 {
1885         const struct iphdr *outer_iph = ip_hdr(skb);
1886         const struct iphdr *key_iph = outer_iph;
1887         const struct iphdr *inner_iph;
1888         const struct icmphdr *icmph;
1889         struct iphdr _inner_iph;
1890         struct icmphdr _icmph;
1891
1892         if (likely(outer_iph->protocol != IPPROTO_ICMP))
1893                 goto out;
1894
1895         if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1896                 goto out;
1897
1898         icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1899                                    &_icmph);
1900         if (!icmph)
1901                 goto out;
1902
1903         if (!icmp_is_err(icmph->type))
1904                 goto out;
1905
1906         inner_iph = skb_header_pointer(skb,
1907                                        outer_iph->ihl * 4 + sizeof(_icmph),
1908                                        sizeof(_inner_iph), &_inner_iph);
1909         if (!inner_iph)
1910                 goto out;
1911
1912         key_iph = inner_iph;
1913 out:
1914         hash_keys->addrs.v4addrs.src = key_iph->saddr;
1915         hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1916 }
1917
1918 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1919                                            const struct sk_buff *skb,
1920                                            bool *p_has_inner)
1921 {
1922         u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
1923         struct flow_keys keys, hash_keys;
1924
1925         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1926                 return 0;
1927
1928         memset(&hash_keys, 0, sizeof(hash_keys));
1929         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1930
1931         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1932         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1933                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1934         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1935                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1936         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1937                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1938         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1939                 hash_keys.ports.src = keys.ports.src;
1940         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1941                 hash_keys.ports.dst = keys.ports.dst;
1942
1943         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1944         return flow_hash_from_keys(&hash_keys);
1945 }
1946
1947 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1948                                            const struct sk_buff *skb,
1949                                            bool has_inner)
1950 {
1951         u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
1952         struct flow_keys keys, hash_keys;
1953
1954         /* We assume the packet carries an encapsulation, but if none was
1955          * encountered during dissection of the outer flow, then there is no
1956          * point in calling the flow dissector again.
1957          */
1958         if (!has_inner)
1959                 return 0;
1960
1961         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1962                 return 0;
1963
1964         memset(&hash_keys, 0, sizeof(hash_keys));
1965         skb_flow_dissect_flow_keys(skb, &keys, 0);
1966
1967         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1968                 return 0;
1969
1970         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1971                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1972                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1973                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1974                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1975                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1976         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1977                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1978                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1979                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1980                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1981                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1982                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1983                         hash_keys.tags.flow_label = keys.tags.flow_label;
1984         }
1985
1986         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1987                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1988         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1989                 hash_keys.ports.src = keys.ports.src;
1990         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1991                 hash_keys.ports.dst = keys.ports.dst;
1992
1993         return flow_hash_from_keys(&hash_keys);
1994 }
1995
1996 static u32 fib_multipath_custom_hash_skb(const struct net *net,
1997                                          const struct sk_buff *skb)
1998 {
1999         u32 mhash, mhash_inner;
2000         bool has_inner = true;
2001
2002         mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2003         mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2004
2005         return jhash_2words(mhash, mhash_inner, 0);
2006 }
2007
2008 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2009                                          const struct flowi4 *fl4)
2010 {
2011         u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
2012         struct flow_keys hash_keys;
2013
2014         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2015                 return 0;
2016
2017         memset(&hash_keys, 0, sizeof(hash_keys));
2018         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2019         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2020                 hash_keys.addrs.v4addrs.src = fl4->saddr;
2021         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2022                 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2023         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2024                 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2025         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2026                 hash_keys.ports.src = fl4->fl4_sport;
2027         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2028                 hash_keys.ports.dst = fl4->fl4_dport;
2029
2030         return flow_hash_from_keys(&hash_keys);
2031 }
2032
2033 /* if skb is set it will be used and fl4 can be NULL */
2034 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2035                        const struct sk_buff *skb, struct flow_keys *flkeys)
2036 {
2037         u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2038         struct flow_keys hash_keys;
2039         u32 mhash = 0;
2040
2041         switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
2042         case 0:
2043                 memset(&hash_keys, 0, sizeof(hash_keys));
2044                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2045                 if (skb) {
2046                         ip_multipath_l3_keys(skb, &hash_keys);
2047                 } else {
2048                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2049                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2050                 }
2051                 mhash = flow_hash_from_keys(&hash_keys);
2052                 break;
2053         case 1:
2054                 /* skb is currently provided only when forwarding */
2055                 if (skb) {
2056                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2057                         struct flow_keys keys;
2058
2059                         /* short-circuit if we already have L4 hash present */
2060                         if (skb->l4_hash)
2061                                 return skb_get_hash_raw(skb) >> 1;
2062
2063                         memset(&hash_keys, 0, sizeof(hash_keys));
2064
2065                         if (!flkeys) {
2066                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2067                                 flkeys = &keys;
2068                         }
2069
2070                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2071                         hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2072                         hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2073                         hash_keys.ports.src = flkeys->ports.src;
2074                         hash_keys.ports.dst = flkeys->ports.dst;
2075                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2076                 } else {
2077                         memset(&hash_keys, 0, sizeof(hash_keys));
2078                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2079                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2080                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2081                         hash_keys.ports.src = fl4->fl4_sport;
2082                         hash_keys.ports.dst = fl4->fl4_dport;
2083                         hash_keys.basic.ip_proto = fl4->flowi4_proto;
2084                 }
2085                 mhash = flow_hash_from_keys(&hash_keys);
2086                 break;
2087         case 2:
2088                 memset(&hash_keys, 0, sizeof(hash_keys));
2089                 /* skb is currently provided only when forwarding */
2090                 if (skb) {
2091                         struct flow_keys keys;
2092
2093                         skb_flow_dissect_flow_keys(skb, &keys, 0);
2094                         /* Inner can be v4 or v6 */
2095                         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2096                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2097                                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2098                                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2099                         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2100                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2101                                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2102                                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2103                                 hash_keys.tags.flow_label = keys.tags.flow_label;
2104                                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2105                         } else {
2106                                 /* Same as case 0 */
2107                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2108                                 ip_multipath_l3_keys(skb, &hash_keys);
2109                         }
2110                 } else {
2111                         /* Same as case 0 */
2112                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2113                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2114                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2115                 }
2116                 mhash = flow_hash_from_keys(&hash_keys);
2117                 break;
2118         case 3:
2119                 if (skb)
2120                         mhash = fib_multipath_custom_hash_skb(net, skb);
2121                 else
2122                         mhash = fib_multipath_custom_hash_fl4(net, fl4);
2123                 break;
2124         }
2125
2126         if (multipath_hash)
2127                 mhash = jhash_2words(mhash, multipath_hash, 0);
2128
2129         return mhash >> 1;
2130 }
2131 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2132
2133 static int ip_mkroute_input(struct sk_buff *skb,
2134                             struct fib_result *res,
2135                             struct in_device *in_dev,
2136                             __be32 daddr, __be32 saddr, u32 tos,
2137                             struct flow_keys *hkeys)
2138 {
2139 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2140         if (res->fi && fib_info_num_path(res->fi) > 1) {
2141                 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2142
2143                 fib_select_multipath(res, h);
2144         }
2145 #endif
2146
2147         /* create a routing cache entry */
2148         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2149 }
2150
2151 /* Implements all the saddr-related checks as ip_route_input_slow(),
2152  * assuming daddr is valid and the destination is not a local broadcast one.
2153  * Uses the provided hint instead of performing a route lookup.
2154  */
2155 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2156                       u8 tos, struct net_device *dev,
2157                       const struct sk_buff *hint)
2158 {
2159         struct in_device *in_dev = __in_dev_get_rcu(dev);
2160         struct rtable *rt = skb_rtable(hint);
2161         struct net *net = dev_net(dev);
2162         int err = -EINVAL;
2163         u32 tag = 0;
2164
2165         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2166                 goto martian_source;
2167
2168         if (ipv4_is_zeronet(saddr))
2169                 goto martian_source;
2170
2171         if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2172                 goto martian_source;
2173
2174         if (rt->rt_type != RTN_LOCAL)
2175                 goto skip_validate_source;
2176
2177         tos &= IPTOS_RT_MASK;
2178         err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2179         if (err < 0)
2180                 goto martian_source;
2181
2182 skip_validate_source:
2183         skb_dst_copy(skb, hint);
2184         return 0;
2185
2186 martian_source:
2187         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2188         return err;
2189 }
2190
2191 /* get device for dst_alloc with local routes */
2192 static struct net_device *ip_rt_get_dev(struct net *net,
2193                                         const struct fib_result *res)
2194 {
2195         struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2196         struct net_device *dev = NULL;
2197
2198         if (nhc)
2199                 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2200
2201         return dev ? : net->loopback_dev;
2202 }
2203
2204 /*
2205  *      NOTE. We drop all the packets that has local source
2206  *      addresses, because every properly looped back packet
2207  *      must have correct destination already attached by output routine.
2208  *      Changes in the enforced policies must be applied also to
2209  *      ip_route_use_hint().
2210  *
2211  *      Such approach solves two big problems:
2212  *      1. Not simplex devices are handled properly.
2213  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2214  *      called with rcu_read_lock()
2215  */
2216
2217 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2218                                u8 tos, struct net_device *dev,
2219                                struct fib_result *res)
2220 {
2221         struct in_device *in_dev = __in_dev_get_rcu(dev);
2222         struct flow_keys *flkeys = NULL, _flkeys;
2223         struct net    *net = dev_net(dev);
2224         struct ip_tunnel_info *tun_info;
2225         int             err = -EINVAL;
2226         unsigned int    flags = 0;
2227         u32             itag = 0;
2228         struct rtable   *rth;
2229         struct flowi4   fl4;
2230         bool do_cache = true;
2231
2232         /* IP on this device is disabled. */
2233
2234         if (!in_dev)
2235                 goto out;
2236
2237         /* Check for the most weird martians, which can be not detected
2238          * by fib_lookup.
2239          */
2240
2241         tun_info = skb_tunnel_info(skb);
2242         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2243                 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2244         else
2245                 fl4.flowi4_tun_key.tun_id = 0;
2246         skb_dst_drop(skb);
2247
2248         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2249                 goto martian_source;
2250
2251         res->fi = NULL;
2252         res->table = NULL;
2253         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2254                 goto brd_input;
2255
2256         /* Accept zero addresses only to limited broadcast;
2257          * I even do not know to fix it or not. Waiting for complains :-)
2258          */
2259         if (ipv4_is_zeronet(saddr))
2260                 goto martian_source;
2261
2262         if (ipv4_is_zeronet(daddr))
2263                 goto martian_destination;
2264
2265         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2266          * and call it once if daddr or/and saddr are loopback addresses
2267          */
2268         if (ipv4_is_loopback(daddr)) {
2269                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2270                         goto martian_destination;
2271         } else if (ipv4_is_loopback(saddr)) {
2272                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2273                         goto martian_source;
2274         }
2275
2276         /*
2277          *      Now we are ready to route packet.
2278          */
2279         fl4.flowi4_l3mdev = 0;
2280         fl4.flowi4_oif = 0;
2281         fl4.flowi4_iif = dev->ifindex;
2282         fl4.flowi4_mark = skb->mark;
2283         fl4.flowi4_tos = tos;
2284         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2285         fl4.flowi4_flags = 0;
2286         fl4.daddr = daddr;
2287         fl4.saddr = saddr;
2288         fl4.flowi4_uid = sock_net_uid(net, NULL);
2289         fl4.flowi4_multipath_hash = 0;
2290
2291         if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2292                 flkeys = &_flkeys;
2293         } else {
2294                 fl4.flowi4_proto = 0;
2295                 fl4.fl4_sport = 0;
2296                 fl4.fl4_dport = 0;
2297         }
2298
2299         err = fib_lookup(net, &fl4, res, 0);
2300         if (err != 0) {
2301                 if (!IN_DEV_FORWARD(in_dev))
2302                         err = -EHOSTUNREACH;
2303                 goto no_route;
2304         }
2305
2306         if (res->type == RTN_BROADCAST) {
2307                 if (IN_DEV_BFORWARD(in_dev))
2308                         goto make_route;
2309                 /* not do cache if bc_forwarding is enabled */
2310                 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2311                         do_cache = false;
2312                 goto brd_input;
2313         }
2314
2315         if (res->type == RTN_LOCAL) {
2316                 err = fib_validate_source(skb, saddr, daddr, tos,
2317                                           0, dev, in_dev, &itag);
2318                 if (err < 0)
2319                         goto martian_source;
2320                 goto local_input;
2321         }
2322
2323         if (!IN_DEV_FORWARD(in_dev)) {
2324                 err = -EHOSTUNREACH;
2325                 goto no_route;
2326         }
2327         if (res->type != RTN_UNICAST)
2328                 goto martian_destination;
2329
2330 make_route:
2331         err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2332 out:    return err;
2333
2334 brd_input:
2335         if (skb->protocol != htons(ETH_P_IP))
2336                 goto e_inval;
2337
2338         if (!ipv4_is_zeronet(saddr)) {
2339                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2340                                           in_dev, &itag);
2341                 if (err < 0)
2342                         goto martian_source;
2343         }
2344         flags |= RTCF_BROADCAST;
2345         res->type = RTN_BROADCAST;
2346         RT_CACHE_STAT_INC(in_brd);
2347
2348 local_input:
2349         do_cache &= res->fi && !itag;
2350         if (do_cache) {
2351                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2352
2353                 rth = rcu_dereference(nhc->nhc_rth_input);
2354                 if (rt_cache_valid(rth)) {
2355                         skb_dst_set_noref(skb, &rth->dst);
2356                         err = 0;
2357                         goto out;
2358                 }
2359         }
2360
2361         rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2362                            flags | RTCF_LOCAL, res->type,
2363                            IN_DEV_ORCONF(in_dev, NOPOLICY), false);
2364         if (!rth)
2365                 goto e_nobufs;
2366
2367         rth->dst.output= ip_rt_bug;
2368 #ifdef CONFIG_IP_ROUTE_CLASSID
2369         rth->dst.tclassid = itag;
2370 #endif
2371         rth->rt_is_input = 1;
2372
2373         RT_CACHE_STAT_INC(in_slow_tot);
2374         if (res->type == RTN_UNREACHABLE) {
2375                 rth->dst.input= ip_error;
2376                 rth->dst.error= -err;
2377                 rth->rt_flags   &= ~RTCF_LOCAL;
2378         }
2379
2380         if (do_cache) {
2381                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2382
2383                 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2384                 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2385                         WARN_ON(rth->dst.input == lwtunnel_input);
2386                         rth->dst.lwtstate->orig_input = rth->dst.input;
2387                         rth->dst.input = lwtunnel_input;
2388                 }
2389
2390                 if (unlikely(!rt_cache_route(nhc, rth)))
2391                         rt_add_uncached_list(rth);
2392         }
2393         skb_dst_set(skb, &rth->dst);
2394         err = 0;
2395         goto out;
2396
2397 no_route:
2398         RT_CACHE_STAT_INC(in_no_route);
2399         res->type = RTN_UNREACHABLE;
2400         res->fi = NULL;
2401         res->table = NULL;
2402         goto local_input;
2403
2404         /*
2405          *      Do not cache martian addresses: they should be logged (RFC1812)
2406          */
2407 martian_destination:
2408         RT_CACHE_STAT_INC(in_martian_dst);
2409 #ifdef CONFIG_IP_ROUTE_VERBOSE
2410         if (IN_DEV_LOG_MARTIANS(in_dev))
2411                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2412                                      &daddr, &saddr, dev->name);
2413 #endif
2414
2415 e_inval:
2416         err = -EINVAL;
2417         goto out;
2418
2419 e_nobufs:
2420         err = -ENOBUFS;
2421         goto out;
2422
2423 martian_source:
2424         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2425         goto out;
2426 }
2427
2428 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2429                          u8 tos, struct net_device *dev)
2430 {
2431         struct fib_result res;
2432         int err;
2433
2434         tos &= IPTOS_RT_MASK;
2435         rcu_read_lock();
2436         err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2437         rcu_read_unlock();
2438
2439         return err;
2440 }
2441 EXPORT_SYMBOL(ip_route_input_noref);
2442
2443 /* called with rcu_read_lock held */
2444 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2445                        u8 tos, struct net_device *dev, struct fib_result *res)
2446 {
2447         /* Multicast recognition logic is moved from route cache to here.
2448          * The problem was that too many Ethernet cards have broken/missing
2449          * hardware multicast filters :-( As result the host on multicasting
2450          * network acquires a lot of useless route cache entries, sort of
2451          * SDR messages from all the world. Now we try to get rid of them.
2452          * Really, provided software IP multicast filter is organized
2453          * reasonably (at least, hashed), it does not result in a slowdown
2454          * comparing with route cache reject entries.
2455          * Note, that multicast routers are not affected, because
2456          * route cache entry is created eventually.
2457          */
2458         if (ipv4_is_multicast(daddr)) {
2459                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2460                 int our = 0;
2461                 int err = -EINVAL;
2462
2463                 if (!in_dev)
2464                         return err;
2465                 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2466                                       ip_hdr(skb)->protocol);
2467
2468                 /* check l3 master if no match yet */
2469                 if (!our && netif_is_l3_slave(dev)) {
2470                         struct in_device *l3_in_dev;
2471
2472                         l3_in_dev = __in_dev_get_rcu(skb->dev);
2473                         if (l3_in_dev)
2474                                 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2475                                                       ip_hdr(skb)->protocol);
2476                 }
2477
2478                 if (our
2479 #ifdef CONFIG_IP_MROUTE
2480                         ||
2481                     (!ipv4_is_local_multicast(daddr) &&
2482                      IN_DEV_MFORWARD(in_dev))
2483 #endif
2484                    ) {
2485                         err = ip_route_input_mc(skb, daddr, saddr,
2486                                                 tos, dev, our);
2487                 }
2488                 return err;
2489         }
2490
2491         return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2492 }
2493
2494 /* called with rcu_read_lock() */
2495 static struct rtable *__mkroute_output(const struct fib_result *res,
2496                                        const struct flowi4 *fl4, int orig_oif,
2497                                        struct net_device *dev_out,
2498                                        unsigned int flags)
2499 {
2500         struct fib_info *fi = res->fi;
2501         struct fib_nh_exception *fnhe;
2502         struct in_device *in_dev;
2503         u16 type = res->type;
2504         struct rtable *rth;
2505         bool do_cache;
2506
2507         in_dev = __in_dev_get_rcu(dev_out);
2508         if (!in_dev)
2509                 return ERR_PTR(-EINVAL);
2510
2511         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2512                 if (ipv4_is_loopback(fl4->saddr) &&
2513                     !(dev_out->flags & IFF_LOOPBACK) &&
2514                     !netif_is_l3_master(dev_out))
2515                         return ERR_PTR(-EINVAL);
2516
2517         if (ipv4_is_lbcast(fl4->daddr))
2518                 type = RTN_BROADCAST;
2519         else if (ipv4_is_multicast(fl4->daddr))
2520                 type = RTN_MULTICAST;
2521         else if (ipv4_is_zeronet(fl4->daddr))
2522                 return ERR_PTR(-EINVAL);
2523
2524         if (dev_out->flags & IFF_LOOPBACK)
2525                 flags |= RTCF_LOCAL;
2526
2527         do_cache = true;
2528         if (type == RTN_BROADCAST) {
2529                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2530                 fi = NULL;
2531         } else if (type == RTN_MULTICAST) {
2532                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2533                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2534                                      fl4->flowi4_proto))
2535                         flags &= ~RTCF_LOCAL;
2536                 else
2537                         do_cache = false;
2538                 /* If multicast route do not exist use
2539                  * default one, but do not gateway in this case.
2540                  * Yes, it is hack.
2541                  */
2542                 if (fi && res->prefixlen < 4)
2543                         fi = NULL;
2544         } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2545                    (orig_oif != dev_out->ifindex)) {
2546                 /* For local routes that require a particular output interface
2547                  * we do not want to cache the result.  Caching the result
2548                  * causes incorrect behaviour when there are multiple source
2549                  * addresses on the interface, the end result being that if the
2550                  * intended recipient is waiting on that interface for the
2551                  * packet he won't receive it because it will be delivered on
2552                  * the loopback interface and the IP_PKTINFO ipi_ifindex will
2553                  * be set to the loopback interface as well.
2554                  */
2555                 do_cache = false;
2556         }
2557
2558         fnhe = NULL;
2559         do_cache &= fi != NULL;
2560         if (fi) {
2561                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2562                 struct rtable __rcu **prth;
2563
2564                 fnhe = find_exception(nhc, fl4->daddr);
2565                 if (!do_cache)
2566                         goto add;
2567                 if (fnhe) {
2568                         prth = &fnhe->fnhe_rth_output;
2569                 } else {
2570                         if (unlikely(fl4->flowi4_flags &
2571                                      FLOWI_FLAG_KNOWN_NH &&
2572                                      !(nhc->nhc_gw_family &&
2573                                        nhc->nhc_scope == RT_SCOPE_LINK))) {
2574                                 do_cache = false;
2575                                 goto add;
2576                         }
2577                         prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2578                 }
2579                 rth = rcu_dereference(*prth);
2580                 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2581                         return rth;
2582         }
2583
2584 add:
2585         rth = rt_dst_alloc(dev_out, flags, type,
2586                            IN_DEV_ORCONF(in_dev, NOPOLICY),
2587                            IN_DEV_ORCONF(in_dev, NOXFRM));
2588         if (!rth)
2589                 return ERR_PTR(-ENOBUFS);
2590
2591         rth->rt_iif = orig_oif;
2592
2593         RT_CACHE_STAT_INC(out_slow_tot);
2594
2595         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2596                 if (flags & RTCF_LOCAL &&
2597                     !(dev_out->flags & IFF_LOOPBACK)) {
2598                         rth->dst.output = ip_mc_output;
2599                         RT_CACHE_STAT_INC(out_slow_mc);
2600                 }
2601 #ifdef CONFIG_IP_MROUTE
2602                 if (type == RTN_MULTICAST) {
2603                         if (IN_DEV_MFORWARD(in_dev) &&
2604                             !ipv4_is_local_multicast(fl4->daddr)) {
2605                                 rth->dst.input = ip_mr_input;
2606                                 rth->dst.output = ip_mc_output;
2607                         }
2608                 }
2609 #endif
2610         }
2611
2612         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2613         lwtunnel_set_redirect(&rth->dst);
2614
2615         return rth;
2616 }
2617
2618 /*
2619  * Major route resolver routine.
2620  */
2621
2622 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2623                                         const struct sk_buff *skb)
2624 {
2625         struct fib_result res = {
2626                 .type           = RTN_UNSPEC,
2627                 .fi             = NULL,
2628                 .table          = NULL,
2629                 .tclassid       = 0,
2630         };
2631         struct rtable *rth;
2632
2633         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2634         ip_rt_fix_tos(fl4);
2635
2636         rcu_read_lock();
2637         rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2638         rcu_read_unlock();
2639
2640         return rth;
2641 }
2642 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2643
2644 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2645                                             struct fib_result *res,
2646                                             const struct sk_buff *skb)
2647 {
2648         struct net_device *dev_out = NULL;
2649         int orig_oif = fl4->flowi4_oif;
2650         unsigned int flags = 0;
2651         struct rtable *rth;
2652         int err;
2653
2654         if (fl4->saddr) {
2655                 if (ipv4_is_multicast(fl4->saddr) ||
2656                     ipv4_is_lbcast(fl4->saddr) ||
2657                     ipv4_is_zeronet(fl4->saddr)) {
2658                         rth = ERR_PTR(-EINVAL);
2659                         goto out;
2660                 }
2661
2662                 rth = ERR_PTR(-ENETUNREACH);
2663
2664                 /* I removed check for oif == dev_out->oif here.
2665                  * It was wrong for two reasons:
2666                  * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2667                  *    is assigned to multiple interfaces.
2668                  * 2. Moreover, we are allowed to send packets with saddr
2669                  *    of another iface. --ANK
2670                  */
2671
2672                 if (fl4->flowi4_oif == 0 &&
2673                     (ipv4_is_multicast(fl4->daddr) ||
2674                      ipv4_is_lbcast(fl4->daddr))) {
2675                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2676                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2677                         if (!dev_out)
2678                                 goto out;
2679
2680                         /* Special hack: user can direct multicasts
2681                          * and limited broadcast via necessary interface
2682                          * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2683                          * This hack is not just for fun, it allows
2684                          * vic,vat and friends to work.
2685                          * They bind socket to loopback, set ttl to zero
2686                          * and expect that it will work.
2687                          * From the viewpoint of routing cache they are broken,
2688                          * because we are not allowed to build multicast path
2689                          * with loopback source addr (look, routing cache
2690                          * cannot know, that ttl is zero, so that packet
2691                          * will not leave this host and route is valid).
2692                          * Luckily, this hack is good workaround.
2693                          */
2694
2695                         fl4->flowi4_oif = dev_out->ifindex;
2696                         goto make_route;
2697                 }
2698
2699                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2700                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2701                         if (!__ip_dev_find(net, fl4->saddr, false))
2702                                 goto out;
2703                 }
2704         }
2705
2706
2707         if (fl4->flowi4_oif) {
2708                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2709                 rth = ERR_PTR(-ENODEV);
2710                 if (!dev_out)
2711                         goto out;
2712
2713                 /* RACE: Check return value of inet_select_addr instead. */
2714                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2715                         rth = ERR_PTR(-ENETUNREACH);
2716                         goto out;
2717                 }
2718                 if (ipv4_is_local_multicast(fl4->daddr) ||
2719                     ipv4_is_lbcast(fl4->daddr) ||
2720                     fl4->flowi4_proto == IPPROTO_IGMP) {
2721                         if (!fl4->saddr)
2722                                 fl4->saddr = inet_select_addr(dev_out, 0,
2723                                                               RT_SCOPE_LINK);
2724                         goto make_route;
2725                 }
2726                 if (!fl4->saddr) {
2727                         if (ipv4_is_multicast(fl4->daddr))
2728                                 fl4->saddr = inet_select_addr(dev_out, 0,
2729                                                               fl4->flowi4_scope);
2730                         else if (!fl4->daddr)
2731                                 fl4->saddr = inet_select_addr(dev_out, 0,
2732                                                               RT_SCOPE_HOST);
2733                 }
2734         }
2735
2736         if (!fl4->daddr) {
2737                 fl4->daddr = fl4->saddr;
2738                 if (!fl4->daddr)
2739                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2740                 dev_out = net->loopback_dev;
2741                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2742                 res->type = RTN_LOCAL;
2743                 flags |= RTCF_LOCAL;
2744                 goto make_route;
2745         }
2746
2747         err = fib_lookup(net, fl4, res, 0);
2748         if (err) {
2749                 res->fi = NULL;
2750                 res->table = NULL;
2751                 if (fl4->flowi4_oif &&
2752                     (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) {
2753                         /* Apparently, routing tables are wrong. Assume,
2754                          * that the destination is on link.
2755                          *
2756                          * WHY? DW.
2757                          * Because we are allowed to send to iface
2758                          * even if it has NO routes and NO assigned
2759                          * addresses. When oif is specified, routing
2760                          * tables are looked up with only one purpose:
2761                          * to catch if destination is gatewayed, rather than
2762                          * direct. Moreover, if MSG_DONTROUTE is set,
2763                          * we send packet, ignoring both routing tables
2764                          * and ifaddr state. --ANK
2765                          *
2766                          *
2767                          * We could make it even if oif is unknown,
2768                          * likely IPv6, but we do not.
2769                          */
2770
2771                         if (fl4->saddr == 0)
2772                                 fl4->saddr = inet_select_addr(dev_out, 0,
2773                                                               RT_SCOPE_LINK);
2774                         res->type = RTN_UNICAST;
2775                         goto make_route;
2776                 }
2777                 rth = ERR_PTR(err);
2778                 goto out;
2779         }
2780
2781         if (res->type == RTN_LOCAL) {
2782                 if (!fl4->saddr) {
2783                         if (res->fi->fib_prefsrc)
2784                                 fl4->saddr = res->fi->fib_prefsrc;
2785                         else
2786                                 fl4->saddr = fl4->daddr;
2787                 }
2788
2789                 /* L3 master device is the loopback for that domain */
2790                 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2791                         net->loopback_dev;
2792
2793                 /* make sure orig_oif points to fib result device even
2794                  * though packet rx/tx happens over loopback or l3mdev
2795                  */
2796                 orig_oif = FIB_RES_OIF(*res);
2797
2798                 fl4->flowi4_oif = dev_out->ifindex;
2799                 flags |= RTCF_LOCAL;
2800                 goto make_route;
2801         }
2802
2803         fib_select_path(net, res, fl4, skb);
2804
2805         dev_out = FIB_RES_DEV(*res);
2806
2807 make_route:
2808         rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2809
2810 out:
2811         return rth;
2812 }
2813
2814 static struct dst_ops ipv4_dst_blackhole_ops = {
2815         .family                 = AF_INET,
2816         .default_advmss         = ipv4_default_advmss,
2817         .neigh_lookup           = ipv4_neigh_lookup,
2818         .check                  = dst_blackhole_check,
2819         .cow_metrics            = dst_blackhole_cow_metrics,
2820         .update_pmtu            = dst_blackhole_update_pmtu,
2821         .redirect               = dst_blackhole_redirect,
2822         .mtu                    = dst_blackhole_mtu,
2823 };
2824
2825 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2826 {
2827         struct rtable *ort = (struct rtable *) dst_orig;
2828         struct rtable *rt;
2829
2830         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2831         if (rt) {
2832                 struct dst_entry *new = &rt->dst;
2833
2834                 new->__use = 1;
2835                 new->input = dst_discard;
2836                 new->output = dst_discard_out;
2837
2838                 new->dev = net->loopback_dev;
2839                 dev_hold_track(new->dev, &new->dev_tracker, GFP_ATOMIC);
2840
2841                 rt->rt_is_input = ort->rt_is_input;
2842                 rt->rt_iif = ort->rt_iif;
2843                 rt->rt_pmtu = ort->rt_pmtu;
2844                 rt->rt_mtu_locked = ort->rt_mtu_locked;
2845
2846                 rt->rt_genid = rt_genid_ipv4(net);
2847                 rt->rt_flags = ort->rt_flags;
2848                 rt->rt_type = ort->rt_type;
2849                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2850                 rt->rt_gw_family = ort->rt_gw_family;
2851                 if (rt->rt_gw_family == AF_INET)
2852                         rt->rt_gw4 = ort->rt_gw4;
2853                 else if (rt->rt_gw_family == AF_INET6)
2854                         rt->rt_gw6 = ort->rt_gw6;
2855
2856                 INIT_LIST_HEAD(&rt->rt_uncached);
2857         }
2858
2859         dst_release(dst_orig);
2860
2861         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2862 }
2863
2864 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2865                                     const struct sock *sk)
2866 {
2867         struct rtable *rt = __ip_route_output_key(net, flp4);
2868
2869         if (IS_ERR(rt))
2870                 return rt;
2871
2872         if (flp4->flowi4_proto) {
2873                 flp4->flowi4_oif = rt->dst.dev->ifindex;
2874                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2875                                                         flowi4_to_flowi(flp4),
2876                                                         sk, 0);
2877         }
2878
2879         return rt;
2880 }
2881 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2882
2883 struct rtable *ip_route_output_tunnel(struct sk_buff *skb,
2884                                       struct net_device *dev,
2885                                       struct net *net, __be32 *saddr,
2886                                       const struct ip_tunnel_info *info,
2887                                       u8 protocol, bool use_cache)
2888 {
2889 #ifdef CONFIG_DST_CACHE
2890         struct dst_cache *dst_cache;
2891 #endif
2892         struct rtable *rt = NULL;
2893         struct flowi4 fl4;
2894         __u8 tos;
2895
2896 #ifdef CONFIG_DST_CACHE
2897         dst_cache = (struct dst_cache *)&info->dst_cache;
2898         if (use_cache) {
2899                 rt = dst_cache_get_ip4(dst_cache, saddr);
2900                 if (rt)
2901                         return rt;
2902         }
2903 #endif
2904         memset(&fl4, 0, sizeof(fl4));
2905         fl4.flowi4_mark = skb->mark;
2906         fl4.flowi4_proto = protocol;
2907         fl4.daddr = info->key.u.ipv4.dst;
2908         fl4.saddr = info->key.u.ipv4.src;
2909         tos = info->key.tos;
2910         fl4.flowi4_tos = RT_TOS(tos);
2911
2912         rt = ip_route_output_key(net, &fl4);
2913         if (IS_ERR(rt)) {
2914                 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
2915                 return ERR_PTR(-ENETUNREACH);
2916         }
2917         if (rt->dst.dev == dev) { /* is this necessary? */
2918                 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
2919                 ip_rt_put(rt);
2920                 return ERR_PTR(-ELOOP);
2921         }
2922 #ifdef CONFIG_DST_CACHE
2923         if (use_cache)
2924                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2925 #endif
2926         *saddr = fl4.saddr;
2927         return rt;
2928 }
2929 EXPORT_SYMBOL_GPL(ip_route_output_tunnel);
2930
2931 /* called with rcu_read_lock held */
2932 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2933                         struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2934                         struct sk_buff *skb, u32 portid, u32 seq,
2935                         unsigned int flags)
2936 {
2937         struct rtmsg *r;
2938         struct nlmsghdr *nlh;
2939         unsigned long expires = 0;
2940         u32 error;
2941         u32 metrics[RTAX_MAX];
2942
2943         nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2944         if (!nlh)
2945                 return -EMSGSIZE;
2946
2947         r = nlmsg_data(nlh);
2948         r->rtm_family    = AF_INET;
2949         r->rtm_dst_len  = 32;
2950         r->rtm_src_len  = 0;
2951         r->rtm_tos      = fl4 ? fl4->flowi4_tos : 0;
2952         r->rtm_table    = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2953         if (nla_put_u32(skb, RTA_TABLE, table_id))
2954                 goto nla_put_failure;
2955         r->rtm_type     = rt->rt_type;
2956         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2957         r->rtm_protocol = RTPROT_UNSPEC;
2958         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2959         if (rt->rt_flags & RTCF_NOTIFY)
2960                 r->rtm_flags |= RTM_F_NOTIFY;
2961         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2962                 r->rtm_flags |= RTCF_DOREDIRECT;
2963
2964         if (nla_put_in_addr(skb, RTA_DST, dst))
2965                 goto nla_put_failure;
2966         if (src) {
2967                 r->rtm_src_len = 32;
2968                 if (nla_put_in_addr(skb, RTA_SRC, src))
2969                         goto nla_put_failure;
2970         }
2971         if (rt->dst.dev &&
2972             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2973                 goto nla_put_failure;
2974         if (rt->dst.lwtstate &&
2975             lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2976                 goto nla_put_failure;
2977 #ifdef CONFIG_IP_ROUTE_CLASSID
2978         if (rt->dst.tclassid &&
2979             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2980                 goto nla_put_failure;
2981 #endif
2982         if (fl4 && !rt_is_input_route(rt) &&
2983             fl4->saddr != src) {
2984                 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2985                         goto nla_put_failure;
2986         }
2987         if (rt->rt_uses_gateway) {
2988                 if (rt->rt_gw_family == AF_INET &&
2989                     nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2990                         goto nla_put_failure;
2991                 } else if (rt->rt_gw_family == AF_INET6) {
2992                         int alen = sizeof(struct in6_addr);
2993                         struct nlattr *nla;
2994                         struct rtvia *via;
2995
2996                         nla = nla_reserve(skb, RTA_VIA, alen + 2);
2997                         if (!nla)
2998                                 goto nla_put_failure;
2999
3000                         via = nla_data(nla);
3001                         via->rtvia_family = AF_INET6;
3002                         memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3003                 }
3004         }
3005
3006         expires = rt->dst.expires;
3007         if (expires) {
3008                 unsigned long now = jiffies;
3009
3010                 if (time_before(now, expires))
3011                         expires -= now;
3012                 else
3013                         expires = 0;
3014         }
3015
3016         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3017         if (rt->rt_pmtu && expires)
3018                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3019         if (rt->rt_mtu_locked && expires)
3020                 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3021         if (rtnetlink_put_metrics(skb, metrics) < 0)
3022                 goto nla_put_failure;
3023
3024         if (fl4) {
3025                 if (fl4->flowi4_mark &&
3026                     nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3027                         goto nla_put_failure;
3028
3029                 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3030                     nla_put_u32(skb, RTA_UID,
3031                                 from_kuid_munged(current_user_ns(),
3032                                                  fl4->flowi4_uid)))
3033                         goto nla_put_failure;
3034
3035                 if (rt_is_input_route(rt)) {
3036 #ifdef CONFIG_IP_MROUTE
3037                         if (ipv4_is_multicast(dst) &&
3038                             !ipv4_is_local_multicast(dst) &&
3039                             IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
3040                                 int err = ipmr_get_route(net, skb,
3041                                                          fl4->saddr, fl4->daddr,
3042                                                          r, portid);
3043
3044                                 if (err <= 0) {
3045                                         if (err == 0)
3046                                                 return 0;
3047                                         goto nla_put_failure;
3048                                 }
3049                         } else
3050 #endif
3051                                 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3052                                         goto nla_put_failure;
3053                 }
3054         }
3055
3056         error = rt->dst.error;
3057
3058         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3059                 goto nla_put_failure;
3060
3061         nlmsg_end(skb, nlh);
3062         return 0;
3063
3064 nla_put_failure:
3065         nlmsg_cancel(skb, nlh);
3066         return -EMSGSIZE;
3067 }
3068
3069 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3070                             struct netlink_callback *cb, u32 table_id,
3071                             struct fnhe_hash_bucket *bucket, int genid,
3072                             int *fa_index, int fa_start, unsigned int flags)
3073 {
3074         int i;
3075
3076         for (i = 0; i < FNHE_HASH_SIZE; i++) {
3077                 struct fib_nh_exception *fnhe;
3078
3079                 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3080                      fnhe = rcu_dereference(fnhe->fnhe_next)) {
3081                         struct rtable *rt;
3082                         int err;
3083
3084                         if (*fa_index < fa_start)
3085                                 goto next;
3086
3087                         if (fnhe->fnhe_genid != genid)
3088                                 goto next;
3089
3090                         if (fnhe->fnhe_expires &&
3091                             time_after(jiffies, fnhe->fnhe_expires))
3092                                 goto next;
3093
3094                         rt = rcu_dereference(fnhe->fnhe_rth_input);
3095                         if (!rt)
3096                                 rt = rcu_dereference(fnhe->fnhe_rth_output);
3097                         if (!rt)
3098                                 goto next;
3099
3100                         err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3101                                            table_id, NULL, skb,
3102                                            NETLINK_CB(cb->skb).portid,
3103                                            cb->nlh->nlmsg_seq, flags);
3104                         if (err)
3105                                 return err;
3106 next:
3107                         (*fa_index)++;
3108                 }
3109         }
3110
3111         return 0;
3112 }
3113
3114 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3115                        u32 table_id, struct fib_info *fi,
3116                        int *fa_index, int fa_start, unsigned int flags)
3117 {
3118         struct net *net = sock_net(cb->skb->sk);
3119         int nhsel, genid = fnhe_genid(net);
3120
3121         for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3122                 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3123                 struct fnhe_hash_bucket *bucket;
3124                 int err;
3125
3126                 if (nhc->nhc_flags & RTNH_F_DEAD)
3127                         continue;
3128
3129                 rcu_read_lock();
3130                 bucket = rcu_dereference(nhc->nhc_exceptions);
3131                 err = 0;
3132                 if (bucket)
3133                         err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3134                                                genid, fa_index, fa_start,
3135                                                flags);
3136                 rcu_read_unlock();
3137                 if (err)
3138                         return err;
3139         }
3140
3141         return 0;
3142 }
3143
3144 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3145                                                    u8 ip_proto, __be16 sport,
3146                                                    __be16 dport)
3147 {
3148         struct sk_buff *skb;
3149         struct iphdr *iph;
3150
3151         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3152         if (!skb)
3153                 return NULL;
3154
3155         /* Reserve room for dummy headers, this skb can pass
3156          * through good chunk of routing engine.
3157          */
3158         skb_reset_mac_header(skb);
3159         skb_reset_network_header(skb);
3160         skb->protocol = htons(ETH_P_IP);
3161         iph = skb_put(skb, sizeof(struct iphdr));
3162         iph->protocol = ip_proto;
3163         iph->saddr = src;
3164         iph->daddr = dst;
3165         iph->version = 0x4;
3166         iph->frag_off = 0;
3167         iph->ihl = 0x5;
3168         skb_set_transport_header(skb, skb->len);
3169
3170         switch (iph->protocol) {
3171         case IPPROTO_UDP: {
3172                 struct udphdr *udph;
3173
3174                 udph = skb_put_zero(skb, sizeof(struct udphdr));
3175                 udph->source = sport;
3176                 udph->dest = dport;
3177                 udph->len = htons(sizeof(struct udphdr));
3178                 udph->check = 0;
3179                 break;
3180         }
3181         case IPPROTO_TCP: {
3182                 struct tcphdr *tcph;
3183
3184                 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3185                 tcph->source    = sport;
3186                 tcph->dest      = dport;
3187                 tcph->doff      = sizeof(struct tcphdr) / 4;
3188                 tcph->rst = 1;
3189                 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3190                                             src, dst, 0);
3191                 break;
3192         }
3193         case IPPROTO_ICMP: {
3194                 struct icmphdr *icmph;
3195
3196                 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3197                 icmph->type = ICMP_ECHO;
3198                 icmph->code = 0;
3199         }
3200         }
3201
3202         return skb;
3203 }
3204
3205 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3206                                        const struct nlmsghdr *nlh,
3207                                        struct nlattr **tb,
3208                                        struct netlink_ext_ack *extack)
3209 {
3210         struct rtmsg *rtm;
3211         int i, err;
3212
3213         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3214                 NL_SET_ERR_MSG(extack,
3215                                "ipv4: Invalid header for route get request");
3216                 return -EINVAL;
3217         }
3218
3219         if (!netlink_strict_get_check(skb))
3220                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3221                                               rtm_ipv4_policy, extack);
3222
3223         rtm = nlmsg_data(nlh);
3224         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3225             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3226             rtm->rtm_table || rtm->rtm_protocol ||
3227             rtm->rtm_scope || rtm->rtm_type) {
3228                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3229                 return -EINVAL;
3230         }
3231
3232         if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3233                                RTM_F_LOOKUP_TABLE |
3234                                RTM_F_FIB_MATCH)) {
3235                 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3236                 return -EINVAL;
3237         }
3238
3239         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3240                                             rtm_ipv4_policy, extack);
3241         if (err)
3242                 return err;
3243
3244         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3245             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3246                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3247                 return -EINVAL;
3248         }
3249
3250         for (i = 0; i <= RTA_MAX; i++) {
3251                 if (!tb[i])
3252                         continue;
3253
3254                 switch (i) {
3255                 case RTA_IIF:
3256                 case RTA_OIF:
3257                 case RTA_SRC:
3258                 case RTA_DST:
3259                 case RTA_IP_PROTO:
3260                 case RTA_SPORT:
3261                 case RTA_DPORT:
3262                 case RTA_MARK:
3263                 case RTA_UID:
3264                         break;
3265                 default:
3266                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3267                         return -EINVAL;
3268                 }
3269         }
3270
3271         return 0;
3272 }
3273
3274 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3275                              struct netlink_ext_ack *extack)
3276 {
3277         struct net *net = sock_net(in_skb->sk);
3278         struct nlattr *tb[RTA_MAX+1];
3279         u32 table_id = RT_TABLE_MAIN;
3280         __be16 sport = 0, dport = 0;
3281         struct fib_result res = {};
3282         u8 ip_proto = IPPROTO_UDP;
3283         struct rtable *rt = NULL;
3284         struct sk_buff *skb;
3285         struct rtmsg *rtm;
3286         struct flowi4 fl4 = {};
3287         __be32 dst = 0;
3288         __be32 src = 0;
3289         kuid_t uid;
3290         u32 iif;
3291         int err;
3292         int mark;
3293
3294         err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3295         if (err < 0)
3296                 return err;
3297
3298         rtm = nlmsg_data(nlh);
3299         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3300         dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3301         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3302         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3303         if (tb[RTA_UID])
3304                 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3305         else
3306                 uid = (iif ? INVALID_UID : current_uid());
3307
3308         if (tb[RTA_IP_PROTO]) {
3309                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3310                                                   &ip_proto, AF_INET, extack);
3311                 if (err)
3312                         return err;
3313         }
3314
3315         if (tb[RTA_SPORT])
3316                 sport = nla_get_be16(tb[RTA_SPORT]);
3317
3318         if (tb[RTA_DPORT])
3319                 dport = nla_get_be16(tb[RTA_DPORT]);
3320
3321         skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3322         if (!skb)
3323                 return -ENOBUFS;
3324
3325         fl4.daddr = dst;
3326         fl4.saddr = src;
3327         fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3328         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3329         fl4.flowi4_mark = mark;
3330         fl4.flowi4_uid = uid;
3331         if (sport)
3332                 fl4.fl4_sport = sport;
3333         if (dport)
3334                 fl4.fl4_dport = dport;
3335         fl4.flowi4_proto = ip_proto;
3336
3337         rcu_read_lock();
3338
3339         if (iif) {
3340                 struct net_device *dev;
3341
3342                 dev = dev_get_by_index_rcu(net, iif);
3343                 if (!dev) {
3344                         err = -ENODEV;
3345                         goto errout_rcu;
3346                 }
3347
3348                 fl4.flowi4_iif = iif; /* for rt_fill_info */
3349                 skb->dev        = dev;
3350                 skb->mark       = mark;
3351                 err = ip_route_input_rcu(skb, dst, src,
3352                                          rtm->rtm_tos & IPTOS_RT_MASK, dev,
3353                                          &res);
3354
3355                 rt = skb_rtable(skb);
3356                 if (err == 0 && rt->dst.error)
3357                         err = -rt->dst.error;
3358         } else {
3359                 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3360                 skb->dev = net->loopback_dev;
3361                 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3362                 err = 0;
3363                 if (IS_ERR(rt))
3364                         err = PTR_ERR(rt);
3365                 else
3366                         skb_dst_set(skb, &rt->dst);
3367         }
3368
3369         if (err)
3370                 goto errout_rcu;
3371
3372         if (rtm->rtm_flags & RTM_F_NOTIFY)
3373                 rt->rt_flags |= RTCF_NOTIFY;
3374
3375         if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3376                 table_id = res.table ? res.table->tb_id : 0;
3377
3378         /* reset skb for netlink reply msg */
3379         skb_trim(skb, 0);
3380         skb_reset_network_header(skb);
3381         skb_reset_transport_header(skb);
3382         skb_reset_mac_header(skb);
3383
3384         if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3385                 struct fib_rt_info fri;
3386
3387                 if (!res.fi) {
3388                         err = fib_props[res.type].error;
3389                         if (!err)
3390                                 err = -EHOSTUNREACH;
3391                         goto errout_rcu;
3392                 }
3393                 fri.fi = res.fi;
3394                 fri.tb_id = table_id;
3395                 fri.dst = res.prefix;
3396                 fri.dst_len = res.prefixlen;
3397                 fri.tos = fl4.flowi4_tos;
3398                 fri.type = rt->rt_type;
3399                 fri.offload = 0;
3400                 fri.trap = 0;
3401                 fri.offload_failed = 0;
3402                 if (res.fa_head) {
3403                         struct fib_alias *fa;
3404
3405                         hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3406                                 u8 slen = 32 - fri.dst_len;
3407
3408                                 if (fa->fa_slen == slen &&
3409                                     fa->tb_id == fri.tb_id &&
3410                                     fa->fa_dscp == inet_dsfield_to_dscp(fri.tos) &&
3411                                     fa->fa_info == res.fi &&
3412                                     fa->fa_type == fri.type) {
3413                                         fri.offload = READ_ONCE(fa->offload);
3414                                         fri.trap = READ_ONCE(fa->trap);
3415                                         break;
3416                                 }
3417                         }
3418                 }
3419                 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3420                                     nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3421         } else {
3422                 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3423                                    NETLINK_CB(in_skb).portid,
3424                                    nlh->nlmsg_seq, 0);
3425         }
3426         if (err < 0)
3427                 goto errout_rcu;
3428
3429         rcu_read_unlock();
3430
3431         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3432
3433 errout_free:
3434         return err;
3435 errout_rcu:
3436         rcu_read_unlock();
3437         kfree_skb(skb);
3438         goto errout_free;
3439 }
3440
3441 void ip_rt_multicast_event(struct in_device *in_dev)
3442 {
3443         rt_cache_flush(dev_net(in_dev->dev));
3444 }
3445
3446 #ifdef CONFIG_SYSCTL
3447 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3448 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
3449 static int ip_rt_gc_elasticity __read_mostly    = 8;
3450 static int ip_min_valid_pmtu __read_mostly      = IPV4_MIN_MTU;
3451
3452 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3453                 void *buffer, size_t *lenp, loff_t *ppos)
3454 {
3455         struct net *net = (struct net *)__ctl->extra1;
3456
3457         if (write) {
3458                 rt_cache_flush(net);
3459                 fnhe_genid_bump(net);
3460                 return 0;
3461         }
3462
3463         return -EINVAL;
3464 }
3465
3466 static struct ctl_table ipv4_route_table[] = {
3467         {
3468                 .procname       = "gc_thresh",
3469                 .data           = &ipv4_dst_ops.gc_thresh,
3470                 .maxlen         = sizeof(int),
3471                 .mode           = 0644,
3472                 .proc_handler   = proc_dointvec,
3473         },
3474         {
3475                 .procname       = "max_size",
3476                 .data           = &ip_rt_max_size,
3477                 .maxlen         = sizeof(int),
3478                 .mode           = 0644,
3479                 .proc_handler   = proc_dointvec,
3480         },
3481         {
3482                 /*  Deprecated. Use gc_min_interval_ms */
3483
3484                 .procname       = "gc_min_interval",
3485                 .data           = &ip_rt_gc_min_interval,
3486                 .maxlen         = sizeof(int),
3487                 .mode           = 0644,
3488                 .proc_handler   = proc_dointvec_jiffies,
3489         },
3490         {
3491                 .procname       = "gc_min_interval_ms",
3492                 .data           = &ip_rt_gc_min_interval,
3493                 .maxlen         = sizeof(int),
3494                 .mode           = 0644,
3495                 .proc_handler   = proc_dointvec_ms_jiffies,
3496         },
3497         {
3498                 .procname       = "gc_timeout",
3499                 .data           = &ip_rt_gc_timeout,
3500                 .maxlen         = sizeof(int),
3501                 .mode           = 0644,
3502                 .proc_handler   = proc_dointvec_jiffies,
3503         },
3504         {
3505                 .procname       = "gc_interval",
3506                 .data           = &ip_rt_gc_interval,
3507                 .maxlen         = sizeof(int),
3508                 .mode           = 0644,
3509                 .proc_handler   = proc_dointvec_jiffies,
3510         },
3511         {
3512                 .procname       = "redirect_load",
3513                 .data           = &ip_rt_redirect_load,
3514                 .maxlen         = sizeof(int),
3515                 .mode           = 0644,
3516                 .proc_handler   = proc_dointvec,
3517         },
3518         {
3519                 .procname       = "redirect_number",
3520                 .data           = &ip_rt_redirect_number,
3521                 .maxlen         = sizeof(int),
3522                 .mode           = 0644,
3523                 .proc_handler   = proc_dointvec,
3524         },
3525         {
3526                 .procname       = "redirect_silence",
3527                 .data           = &ip_rt_redirect_silence,
3528                 .maxlen         = sizeof(int),
3529                 .mode           = 0644,
3530                 .proc_handler   = proc_dointvec,
3531         },
3532         {
3533                 .procname       = "error_cost",
3534                 .data           = &ip_rt_error_cost,
3535                 .maxlen         = sizeof(int),
3536                 .mode           = 0644,
3537                 .proc_handler   = proc_dointvec,
3538         },
3539         {
3540                 .procname       = "error_burst",
3541                 .data           = &ip_rt_error_burst,
3542                 .maxlen         = sizeof(int),
3543                 .mode           = 0644,
3544                 .proc_handler   = proc_dointvec,
3545         },
3546         {
3547                 .procname       = "gc_elasticity",
3548                 .data           = &ip_rt_gc_elasticity,
3549                 .maxlen         = sizeof(int),
3550                 .mode           = 0644,
3551                 .proc_handler   = proc_dointvec,
3552         },
3553         { }
3554 };
3555
3556 static const char ipv4_route_flush_procname[] = "flush";
3557
3558 static struct ctl_table ipv4_route_netns_table[] = {
3559         {
3560                 .procname       = ipv4_route_flush_procname,
3561                 .maxlen         = sizeof(int),
3562                 .mode           = 0200,
3563                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3564         },
3565         {
3566                 .procname       = "min_pmtu",
3567                 .data           = &init_net.ipv4.ip_rt_min_pmtu,
3568                 .maxlen         = sizeof(int),
3569                 .mode           = 0644,
3570                 .proc_handler   = proc_dointvec_minmax,
3571                 .extra1         = &ip_min_valid_pmtu,
3572         },
3573         {
3574                 .procname       = "mtu_expires",
3575                 .data           = &init_net.ipv4.ip_rt_mtu_expires,
3576                 .maxlen         = sizeof(int),
3577                 .mode           = 0644,
3578                 .proc_handler   = proc_dointvec_jiffies,
3579         },
3580         {
3581                 .procname   = "min_adv_mss",
3582                 .data       = &init_net.ipv4.ip_rt_min_advmss,
3583                 .maxlen     = sizeof(int),
3584                 .mode       = 0644,
3585                 .proc_handler   = proc_dointvec,
3586         },
3587         { },
3588 };
3589
3590 static __net_init int sysctl_route_net_init(struct net *net)
3591 {
3592         struct ctl_table *tbl;
3593
3594         tbl = ipv4_route_netns_table;
3595         if (!net_eq(net, &init_net)) {
3596                 int i;
3597
3598                 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL);
3599                 if (!tbl)
3600                         goto err_dup;
3601
3602                 /* Don't export non-whitelisted sysctls to unprivileged users */
3603                 if (net->user_ns != &init_user_ns) {
3604                         if (tbl[0].procname != ipv4_route_flush_procname)
3605                                 tbl[0].procname = NULL;
3606                 }
3607
3608                 /* Update the variables to point into the current struct net
3609                  * except for the first element flush
3610                  */
3611                 for (i = 1; i < ARRAY_SIZE(ipv4_route_netns_table) - 1; i++)
3612                         tbl[i].data += (void *)net - (void *)&init_net;
3613         }
3614         tbl[0].extra1 = net;
3615
3616         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3617         if (!net->ipv4.route_hdr)
3618                 goto err_reg;
3619         return 0;
3620
3621 err_reg:
3622         if (tbl != ipv4_route_netns_table)
3623                 kfree(tbl);
3624 err_dup:
3625         return -ENOMEM;
3626 }
3627
3628 static __net_exit void sysctl_route_net_exit(struct net *net)
3629 {
3630         struct ctl_table *tbl;
3631
3632         tbl = net->ipv4.route_hdr->ctl_table_arg;
3633         unregister_net_sysctl_table(net->ipv4.route_hdr);
3634         BUG_ON(tbl == ipv4_route_netns_table);
3635         kfree(tbl);
3636 }
3637
3638 static __net_initdata struct pernet_operations sysctl_route_ops = {
3639         .init = sysctl_route_net_init,
3640         .exit = sysctl_route_net_exit,
3641 };
3642 #endif
3643
3644 static __net_init int netns_ip_rt_init(struct net *net)
3645 {
3646         /* Set default value for namespaceified sysctls */
3647         net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU;
3648         net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES;
3649         net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS;
3650         return 0;
3651 }
3652
3653 static struct pernet_operations __net_initdata ip_rt_ops = {
3654         .init = netns_ip_rt_init,
3655 };
3656
3657 static __net_init int rt_genid_init(struct net *net)
3658 {
3659         atomic_set(&net->ipv4.rt_genid, 0);
3660         atomic_set(&net->fnhe_genid, 0);
3661         atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3662         return 0;
3663 }
3664
3665 static __net_initdata struct pernet_operations rt_genid_ops = {
3666         .init = rt_genid_init,
3667 };
3668
3669 static int __net_init ipv4_inetpeer_init(struct net *net)
3670 {
3671         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3672
3673         if (!bp)
3674                 return -ENOMEM;
3675         inet_peer_base_init(bp);
3676         net->ipv4.peers = bp;
3677         return 0;
3678 }
3679
3680 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3681 {
3682         struct inet_peer_base *bp = net->ipv4.peers;
3683
3684         net->ipv4.peers = NULL;
3685         inetpeer_invalidate_tree(bp);
3686         kfree(bp);
3687 }
3688
3689 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3690         .init   =       ipv4_inetpeer_init,
3691         .exit   =       ipv4_inetpeer_exit,
3692 };
3693
3694 #ifdef CONFIG_IP_ROUTE_CLASSID
3695 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3696 #endif /* CONFIG_IP_ROUTE_CLASSID */
3697
3698 int __init ip_rt_init(void)
3699 {
3700         void *idents_hash;
3701         int cpu;
3702
3703         /* For modern hosts, this will use 2 MB of memory */
3704         idents_hash = alloc_large_system_hash("IP idents",
3705                                               sizeof(*ip_idents) + sizeof(*ip_tstamps),
3706                                               0,
3707                                               16, /* one bucket per 64 KB */
3708                                               HASH_ZERO,
3709                                               NULL,
3710                                               &ip_idents_mask,
3711                                               2048,
3712                                               256*1024);
3713
3714         ip_idents = idents_hash;
3715
3716         prandom_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3717
3718         ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3719
3720         for_each_possible_cpu(cpu) {
3721                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3722
3723                 INIT_LIST_HEAD(&ul->head);
3724                 INIT_LIST_HEAD(&ul->quarantine);
3725                 spin_lock_init(&ul->lock);
3726         }
3727 #ifdef CONFIG_IP_ROUTE_CLASSID
3728         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3729         if (!ip_rt_acct)
3730                 panic("IP: failed to allocate ip_rt_acct\n");
3731 #endif
3732
3733         ipv4_dst_ops.kmem_cachep =
3734                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3735                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3736
3737         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3738
3739         if (dst_entries_init(&ipv4_dst_ops) < 0)
3740                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3741
3742         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3743                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3744
3745         ipv4_dst_ops.gc_thresh = ~0;
3746         ip_rt_max_size = INT_MAX;
3747
3748         devinet_init();
3749         ip_fib_init();
3750
3751         if (ip_rt_proc_init())
3752                 pr_err("Unable to create route proc files\n");
3753 #ifdef CONFIG_XFRM
3754         xfrm_init();
3755         xfrm4_init();
3756 #endif
3757         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3758                       RTNL_FLAG_DOIT_UNLOCKED);
3759
3760 #ifdef CONFIG_SYSCTL
3761         register_pernet_subsys(&sysctl_route_ops);
3762 #endif
3763         register_pernet_subsys(&ip_rt_ops);
3764         register_pernet_subsys(&rt_genid_ops);
3765         register_pernet_subsys(&ipv4_inetpeer_ops);
3766         return 0;
3767 }
3768
3769 #ifdef CONFIG_SYSCTL
3770 /*
3771  * We really need to sanitize the damn ipv4 init order, then all
3772  * this nonsense will go away.
3773  */
3774 void __init ip_static_sysctl_init(void)
3775 {
3776         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3777 }
3778 #endif