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/uaccess.h>
  65 #include <linux/bitops.h>
  66 #include <linux/types.h>
  67 #include <linux/kernel.h>
  68 #include <linux/mm.h>
  69 #include <linux/memblock.h>
  70 #include <linux/string.h>
  71 #include <linux/socket.h>
  72 #include <linux/sockios.h>
  73 #include <linux/errno.h>
  74 #include <linux/in.h>
  75 #include <linux/inet.h>
  76 #include <linux/netdevice.h>
  77 #include <linux/proc_fs.h>
  78 #include <linux/init.h>
  79 #include <linux/skbuff.h>
  80 #include <linux/inetdevice.h>
  81 #include <linux/igmp.h>
  82 #include <linux/pkt_sched.h>
  83 #include <linux/mroute.h>
  84 #include <linux/netfilter_ipv4.h>
  85 #include <linux/random.h>
  86 #include <linux/rcupdate.h>
  87 #include <linux/times.h>
  88 #include <linux/slab.h>
  89 #include <linux/jhash.h>
  90 #include <net/dst.h>
  91 #include <net/dst_metadata.h>
  92 #include <net/net_namespace.h>
  93 #include <net/protocol.h>
  94 #include <net/ip.h>
  95 #include <net/route.h>
  96 #include <net/inetpeer.h>
  97 #include <net/sock.h>
  98 #include <net/ip_fib.h>
  99 #include <net/nexthop.h>
 100 #include <net/arp.h>
 101 #include <net/tcp.h>
 102 #include <net/icmp.h>
 103 #include <net/xfrm.h>
 104 #include <net/lwtunnel.h>
 105 #include <net/netevent.h>
 106 #include <net/rtnetlink.h>
 107 #ifdef CONFIG_SYSCTL
 108 #include <linux/sysctl.h>
 109 #endif
 110 #include <net/secure_seq.h>
 111 #include <net/ip_tunnels.h>
 112 #include <net/l3mdev.h>
 113
 114 #include "fib_lookup.h"
 115
 116 #define RT_FL_TOS(oldflp4) \
 117         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
 118
 119 #define RT_GC_TIMEOUT (300*HZ)
 120
 121 static int ip_rt_max_size;
 122 static int ip_rt_redirect_number __read_mostly  = 9;
 123 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 124 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 125 static int ip_rt_error_cost __read_mostly       = HZ;
 126 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 127 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
 128 static u32 ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
 129 static int ip_rt_min_advmss __read_mostly       = 256;
 130
 131 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
 132
 133 /*
 134  *      Interface to generic destination cache.
 135  */
 136
 137 INDIRECT_CALLABLE_SCOPE
 138 struct dst_entry        *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 139 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 140 INDIRECT_CALLABLE_SCOPE
 141 unsigned int            ipv4_mtu(const struct dst_entry *dst);
 142 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 143 static void              ipv4_link_failure(struct sk_buff *skb);
 144 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 145                                            struct sk_buff *skb, u32 mtu,
 146                                            bool confirm_neigh);
 147 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 148                                         struct sk_buff *skb);
 149 static void             ipv4_dst_destroy(struct dst_entry *dst);
 150
 151 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 152 {
 153         WARN_ON(1);
 154         return NULL;
 155 }
 156
 157 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 158                                            struct sk_buff *skb,
 159                                            const void *daddr);
 160 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
 161
 162 static struct dst_ops ipv4_dst_ops = {
 163         .family =               AF_INET,
 164         .check =                ipv4_dst_check,
 165         .default_advmss =       ipv4_default_advmss,
 166         .mtu =                  ipv4_mtu,
 167         .cow_metrics =          ipv4_cow_metrics,
 168         .destroy =              ipv4_dst_destroy,
 169         .negative_advice =      ipv4_negative_advice,
 170         .link_failure =         ipv4_link_failure,
 171         .update_pmtu =          ip_rt_update_pmtu,
 172         .redirect =             ip_do_redirect,
 173         .local_out =            __ip_local_out,
 174         .neigh_lookup =         ipv4_neigh_lookup,
 175         .confirm_neigh =        ipv4_confirm_neigh,
 176 };
 177
 178 #define ECN_OR_COST(class)      TC_PRIO_##class
 179
 180 const __u8 ip_tos2prio[16] = {
 181         TC_PRIO_BESTEFFORT,
 182         ECN_OR_COST(BESTEFFORT),
 183         TC_PRIO_BESTEFFORT,
 184         ECN_OR_COST(BESTEFFORT),
 185         TC_PRIO_BULK,
 186         ECN_OR_COST(BULK),
 187         TC_PRIO_BULK,
 188         ECN_OR_COST(BULK),
 189         TC_PRIO_INTERACTIVE,
 190         ECN_OR_COST(INTERACTIVE),
 191         TC_PRIO_INTERACTIVE,
 192         ECN_OR_COST(INTERACTIVE),
 193         TC_PRIO_INTERACTIVE_BULK,
 194         ECN_OR_COST(INTERACTIVE_BULK),
 195         TC_PRIO_INTERACTIVE_BULK,
 196         ECN_OR_COST(INTERACTIVE_BULK)
 197 };
 198 EXPORT_SYMBOL(ip_tos2prio);
 199
 200 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 201 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
 202
 203 #ifdef CONFIG_PROC_FS
 204 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 205 {
 206         if (*pos)
 207                 return NULL;
 208         return SEQ_START_TOKEN;
 209 }
 210
 211 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 212 {
 213         ++*pos;
 214         return NULL;
 215 }
 216
 217 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 218 {
 219 }
 220
 221 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 222 {
 223         if (v == SEQ_START_TOKEN)
 224                 seq_printf(seq, "%-127s\n",
 225                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 226                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 227                            "HHUptod\tSpecDst");
 228         return 0;
 229 }
 230
 231 static const struct seq_operations rt_cache_seq_ops = {
 232         .start  = rt_cache_seq_start,
 233         .next   = rt_cache_seq_next,
 234         .stop   = rt_cache_seq_stop,
 235         .show   = rt_cache_seq_show,
 236 };
 237
 238 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 239 {
 240         int cpu;
 241
 242         if (*pos == 0)
 243                 return SEQ_START_TOKEN;
 244
 245         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 246                 if (!cpu_possible(cpu))
 247                         continue;
 248                 *pos = cpu+1;
 249                 return &per_cpu(rt_cache_stat, cpu);
 250         }
 251         return NULL;
 252 }
 253
 254 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 255 {
 256         int cpu;
 257
 258         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 259                 if (!cpu_possible(cpu))
 260                         continue;
 261                 *pos = cpu+1;
 262                 return &per_cpu(rt_cache_stat, cpu);
 263         }
 264         (*pos)++;
 265         return NULL;
 266
 267 }
 268
 269 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 270 {
 271
 272 }
 273
 274 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 275 {
 276         struct rt_cache_stat *st = v;
 277
 278         if (v == SEQ_START_TOKEN) {
 279                 seq_printf(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");
 280                 return 0;
 281         }
 282
 283         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 284                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 285                    dst_entries_get_slow(&ipv4_dst_ops),
 286                    0, /* st->in_hit */
 287                    st->in_slow_tot,
 288                    st->in_slow_mc,
 289                    st->in_no_route,
 290                    st->in_brd,
 291                    st->in_martian_dst,
 292                    st->in_martian_src,
 293
 294                    0, /* st->out_hit */
 295                    st->out_slow_tot,
 296                    st->out_slow_mc,
 297
 298                    0, /* st->gc_total */
 299                    0, /* st->gc_ignored */
 300                    0, /* st->gc_goal_miss */
 301                    0, /* st->gc_dst_overflow */
 302                    0, /* st->in_hlist_search */
 303                    0  /* st->out_hlist_search */
 304                 );
 305         return 0;
 306 }
 307
 308 static const struct seq_operations rt_cpu_seq_ops = {
 309         .start  = rt_cpu_seq_start,
 310         .next   = rt_cpu_seq_next,
 311         .stop   = rt_cpu_seq_stop,
 312         .show   = rt_cpu_seq_show,
 313 };
 314
 315 #ifdef CONFIG_IP_ROUTE_CLASSID
 316 static int rt_acct_proc_show(struct seq_file *m, void *v)
 317 {
 318         struct ip_rt_acct *dst, *src;
 319         unsigned int i, j;
 320
 321         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 322         if (!dst)
 323                 return -ENOMEM;
 324
 325         for_each_possible_cpu(i) {
 326                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 327                 for (j = 0; j < 256; j++) {
 328                         dst[j].o_bytes   += src[j].o_bytes;
 329                         dst[j].o_packets += src[j].o_packets;
 330                         dst[j].i_bytes   += src[j].i_bytes;
 331                         dst[j].i_packets += src[j].i_packets;
 332                 }
 333         }
 334
 335         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 336         kfree(dst);
 337         return 0;
 338 }
 339 #endif
 340
 341 static int __net_init ip_rt_do_proc_init(struct net *net)
 342 {
 343         struct proc_dir_entry *pde;
 344
 345         pde = proc_create_seq("rt_cache", 0444, net->proc_net,
 346                               &rt_cache_seq_ops);
 347         if (!pde)
 348                 goto err1;
 349
 350         pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
 351                               &rt_cpu_seq_ops);
 352         if (!pde)
 353                 goto err2;
 354
 355 #ifdef CONFIG_IP_ROUTE_CLASSID
 356         pde = proc_create_single("rt_acct", 0, net->proc_net,
 357                         rt_acct_proc_show);
 358         if (!pde)
 359                 goto err3;
 360 #endif
 361         return 0;
 362
 363 #ifdef CONFIG_IP_ROUTE_CLASSID
 364 err3:
 365         remove_proc_entry("rt_cache", net->proc_net_stat);
 366 #endif
 367 err2:
 368         remove_proc_entry("rt_cache", net->proc_net);
 369 err1:
 370         return -ENOMEM;
 371 }
 372
 373 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 374 {
 375         remove_proc_entry("rt_cache", net->proc_net_stat);
 376         remove_proc_entry("rt_cache", net->proc_net);
 377 #ifdef CONFIG_IP_ROUTE_CLASSID
 378         remove_proc_entry("rt_acct", net->proc_net);
 379 #endif
 380 }
 381
 382 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 383         .init = ip_rt_do_proc_init,
 384         .exit = ip_rt_do_proc_exit,
 385 };
 386
 387 static int __init ip_rt_proc_init(void)
 388 {
 389         return register_pernet_subsys(&ip_rt_proc_ops);
 390 }
 391
 392 #else
 393 static inline int ip_rt_proc_init(void)
 394 {
 395         return 0;
 396 }
 397 #endif /* CONFIG_PROC_FS */
 398
 399 static inline bool rt_is_expired(const struct rtable *rth)
 400 {
 401         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 402 }
 403
 404 void rt_cache_flush(struct net *net)
 405 {
 406         rt_genid_bump_ipv4(net);
 407 }
 408
 409 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 410                                            struct sk_buff *skb,
 411                                            const void *daddr)
 412 {
 413         const struct rtable *rt = container_of(dst, struct rtable, dst);
 414         struct net_device *dev = dst->dev;
 415         struct neighbour *n;
 416
 417         rcu_read_lock_bh();
 418
 419         if (likely(rt->rt_gw_family == AF_INET)) {
 420                 n = ip_neigh_gw4(dev, rt->rt_gw4);
 421         } else if (rt->rt_gw_family == AF_INET6) {
 422                 n = ip_neigh_gw6(dev, &rt->rt_gw6);
 423         } else {
 424                 __be32 pkey;
 425
 426                 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
 427                 n = ip_neigh_gw4(dev, pkey);
 428         }
 429
 430         if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
 431                 n = NULL;
 432
 433         rcu_read_unlock_bh();
 434
 435         return n;
 436 }
 437
 438 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 439 {
 440         const struct rtable *rt = container_of(dst, struct rtable, dst);
 441         struct net_device *dev = dst->dev;
 442         const __be32 *pkey = daddr;
 443
 444         if (rt->rt_gw_family == AF_INET) {
 445                 pkey = (const __be32 *)&rt->rt_gw4;
 446         } else if (rt->rt_gw_family == AF_INET6) {
 447                 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
 448         } else if (!daddr ||
 449                  (rt->rt_flags &
 450                   (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
 451                 return;
 452         }
 453         __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
 454 }
 455
 456 /* Hash tables of size 2048..262144 depending on RAM size.
 457  * Each bucket uses 8 bytes.
 458  */
 459 static u32 ip_idents_mask __read_mostly;
 460 static atomic_t *ip_idents __read_mostly;
 461 static u32 *ip_tstamps __read_mostly;
 462
 463 /* In order to protect privacy, we add a perturbation to identifiers
 464  * if one generator is seldom used. This makes hard for an attacker
 465  * to infer how many packets were sent between two points in time.
 466  */
 467 u32 ip_idents_reserve(u32 hash, int segs)
 468 {
 469         u32 bucket, old, now = (u32)jiffies;
 470         atomic_t *p_id;
 471         u32 *p_tstamp;
 472         u32 delta = 0;
 473
 474         bucket = hash & ip_idents_mask;
 475         p_tstamp = ip_tstamps + bucket;
 476         p_id = ip_idents + bucket;
 477         old = READ_ONCE(*p_tstamp);
 478
 479         if (old != now && cmpxchg(p_tstamp, old, now) == old)
 480                 delta = prandom_u32_max(now - old);
 481
 482         /* If UBSAN reports an error there, please make sure your compiler
 483          * supports -fno-strict-overflow before reporting it that was a bug
 484          * in UBSAN, and it has been fixed in GCC-8.
 485          */
 486         return atomic_add_return(segs + delta, p_id) - segs;
 487 }
 488 EXPORT_SYMBOL(ip_idents_reserve);
 489
 490 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
 491 {
 492         u32 hash, id;
 493
 494         /* Note the following code is not safe, but this is okay. */
 495         if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
 496                 get_random_bytes(&net->ipv4.ip_id_key,
 497                                  sizeof(net->ipv4.ip_id_key));
 498
 499         hash = siphash_3u32((__force u32)iph->daddr,
 500                             (__force u32)iph->saddr,
 501                             iph->protocol,
 502                             &net->ipv4.ip_id_key);
 503         id = ip_idents_reserve(hash, segs);
 504         iph->id = htons(id);
 505 }
 506 EXPORT_SYMBOL(__ip_select_ident);
 507
 508 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
 509                              const struct sock *sk,
 510                              const struct iphdr *iph,
 511                              int oif, u8 tos,
 512                              u8 prot, u32 mark, int flow_flags)
 513 {
 514         if (sk) {
 515                 const struct inet_sock *inet = inet_sk(sk);
 516
 517                 oif = sk->sk_bound_dev_if;
 518                 mark = sk->sk_mark;
 519                 tos = RT_CONN_FLAGS(sk);
 520                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 521         }
 522         flowi4_init_output(fl4, oif, mark, tos,
 523                            RT_SCOPE_UNIVERSE, prot,
 524                            flow_flags,
 525                            iph->daddr, iph->saddr, 0, 0,
 526                            sock_net_uid(net, sk));
 527 }
 528
 529 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 530                                const struct sock *sk)
 531 {
 532         const struct net *net = dev_net(skb->dev);
 533         const struct iphdr *iph = ip_hdr(skb);
 534         int oif = skb->dev->ifindex;
 535         u8 tos = RT_TOS(iph->tos);
 536         u8 prot = iph->protocol;
 537         u32 mark = skb->mark;
 538
 539         __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
 540 }
 541
 542 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 543 {
 544         const struct inet_sock *inet = inet_sk(sk);
 545         const struct ip_options_rcu *inet_opt;
 546         __be32 daddr = inet->inet_daddr;
 547
 548         rcu_read_lock();
 549         inet_opt = rcu_dereference(inet->inet_opt);
 550         if (inet_opt && inet_opt->opt.srr)
 551                 daddr = inet_opt->opt.faddr;
 552         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 553                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 554                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 555                            inet_sk_flowi_flags(sk),
 556                            daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
 557         rcu_read_unlock();
 558 }
 559
 560 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 561                                  const struct sk_buff *skb)
 562 {
 563         if (skb)
 564                 build_skb_flow_key(fl4, skb, sk);
 565         else
 566                 build_sk_flow_key(fl4, sk);
 567 }
 568
 569 static DEFINE_SPINLOCK(fnhe_lock);
 570
 571 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 572 {
 573         struct rtable *rt;
 574
 575         rt = rcu_dereference(fnhe->fnhe_rth_input);
 576         if (rt) {
 577                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 578                 dst_dev_put(&rt->dst);
 579                 dst_release(&rt->dst);
 580         }
 581         rt = rcu_dereference(fnhe->fnhe_rth_output);
 582         if (rt) {
 583                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 584                 dst_dev_put(&rt->dst);
 585                 dst_release(&rt->dst);
 586         }
 587 }
 588
 589 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
 590 {
 591         struct fib_nh_exception *fnhe, *oldest;
 592
 593         oldest = rcu_dereference(hash->chain);
 594         for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
 595              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 596                 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
 597                         oldest = fnhe;
 598         }
 599         fnhe_flush_routes(oldest);
 600         return oldest;
 601 }
 602
 603 static inline u32 fnhe_hashfun(__be32 daddr)
 604 {
 605         static u32 fnhe_hashrnd __read_mostly;
 606         u32 hval;
 607
 608         net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
 609         hval = jhash_1word((__force u32)daddr, fnhe_hashrnd);
 610         return hash_32(hval, FNHE_HASH_SHIFT);
 611 }
 612
 613 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 614 {
 615         rt->rt_pmtu = fnhe->fnhe_pmtu;
 616         rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
 617         rt->dst.expires = fnhe->fnhe_expires;
 618
 619         if (fnhe->fnhe_gw) {
 620                 rt->rt_flags |= RTCF_REDIRECTED;
 621                 rt->rt_uses_gateway = 1;
 622                 rt->rt_gw_family = AF_INET;
 623                 rt->rt_gw4 = fnhe->fnhe_gw;
 624         }
 625 }
 626
 627 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
 628                                   __be32 gw, u32 pmtu, bool lock,
 629                                   unsigned long expires)
 630 {
 631         struct fnhe_hash_bucket *hash;
 632         struct fib_nh_exception *fnhe;
 633         struct rtable *rt;
 634         u32 genid, hval;
 635         unsigned int i;
 636         int depth;
 637
 638         genid = fnhe_genid(dev_net(nhc->nhc_dev));
 639         hval = fnhe_hashfun(daddr);
 640
 641         spin_lock_bh(&fnhe_lock);
 642
 643         hash = rcu_dereference(nhc->nhc_exceptions);
 644         if (!hash) {
 645                 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
 646                 if (!hash)
 647                         goto out_unlock;
 648                 rcu_assign_pointer(nhc->nhc_exceptions, hash);
 649         }
 650
 651         hash += hval;
 652
 653         depth = 0;
 654         for (fnhe = rcu_dereference(hash->chain); fnhe;
 655              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 656                 if (fnhe->fnhe_daddr == daddr)
 657                         break;
 658                 depth++;
 659         }
 660
 661         if (fnhe) {
 662                 if (fnhe->fnhe_genid != genid)
 663                         fnhe->fnhe_genid = genid;
 664                 if (gw)
 665                         fnhe->fnhe_gw = gw;
 666                 if (pmtu) {
 667                         fnhe->fnhe_pmtu = pmtu;
 668                         fnhe->fnhe_mtu_locked = lock;
 669                 }
 670                 fnhe->fnhe_expires = max(1UL, expires);
 671                 /* Update all cached dsts too */
 672                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 673                 if (rt)
 674                         fill_route_from_fnhe(rt, fnhe);
 675                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 676                 if (rt)
 677                         fill_route_from_fnhe(rt, fnhe);
 678         } else {
 679                 if (depth > FNHE_RECLAIM_DEPTH)
 680                         fnhe = fnhe_oldest(hash);
 681                 else {
 682                         fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 683                         if (!fnhe)
 684                                 goto out_unlock;
 685
 686                         fnhe->fnhe_next = hash->chain;
 687                         rcu_assign_pointer(hash->chain, fnhe);
 688                 }
 689                 fnhe->fnhe_genid = genid;
 690                 fnhe->fnhe_daddr = daddr;
 691                 fnhe->fnhe_gw = gw;
 692                 fnhe->fnhe_pmtu = pmtu;
 693                 fnhe->fnhe_mtu_locked = lock;
 694                 fnhe->fnhe_expires = max(1UL, expires);
 695
 696                 /* Exception created; mark the cached routes for the nexthop
 697                  * stale, so anyone caching it rechecks if this exception
 698                  * applies to them.
 699                  */
 700                 rt = rcu_dereference(nhc->nhc_rth_input);
 701                 if (rt)
 702                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 703
 704                 for_each_possible_cpu(i) {
 705                         struct rtable __rcu **prt;
 706
 707                         prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
 708                         rt = rcu_dereference(*prt);
 709                         if (rt)
 710                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 711                 }
 712         }
 713
 714         fnhe->fnhe_stamp = jiffies;
 715
 716 out_unlock:
 717         spin_unlock_bh(&fnhe_lock);
 718 }
 719
 720 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 721                              bool kill_route)
 722 {
 723         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 724         __be32 old_gw = ip_hdr(skb)->saddr;
 725         struct net_device *dev = skb->dev;
 726         struct in_device *in_dev;
 727         struct fib_result res;
 728         struct neighbour *n;
 729         struct net *net;
 730
 731         switch (icmp_hdr(skb)->code & 7) {
 732         case ICMP_REDIR_NET:
 733         case ICMP_REDIR_NETTOS:
 734         case ICMP_REDIR_HOST:
 735         case ICMP_REDIR_HOSTTOS:
 736                 break;
 737
 738         default:
 739                 return;
 740         }
 741
 742         if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
 743                 return;
 744
 745         in_dev = __in_dev_get_rcu(dev);
 746         if (!in_dev)
 747                 return;
 748
 749         net = dev_net(dev);
 750         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 751             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 752             ipv4_is_zeronet(new_gw))
 753                 goto reject_redirect;
 754
 755         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 756                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 757                         goto reject_redirect;
 758                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 759                         goto reject_redirect;
 760         } else {
 761                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 762                         goto reject_redirect;
 763         }
 764
 765         n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
 766         if (!n)
 767                 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
 768         if (!IS_ERR(n)) {
 769                 if (!(n->nud_state & NUD_VALID)) {
 770                         neigh_event_send(n, NULL);
 771                 } else {
 772                         if (fib_lookup(net, fl4, &res, 0) == 0) {
 773                                 struct fib_nh_common *nhc;
 774
 775                                 fib_select_path(net, &res, fl4, skb);
 776                                 nhc = FIB_RES_NHC(res);
 777                                 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
 778                                                 0, false,
 779                                                 jiffies + ip_rt_gc_timeout);
 780                         }
 781                         if (kill_route)
 782                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 783                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 784                 }
 785                 neigh_release(n);
 786         }
 787         return;
 788
 789 reject_redirect:
 790 #ifdef CONFIG_IP_ROUTE_VERBOSE
 791         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 792                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 793                 __be32 daddr = iph->daddr;
 794                 __be32 saddr = iph->saddr;
 795
 796                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 797                                      "  Advised path = %pI4 -> %pI4\n",
 798                                      &old_gw, dev->name, &new_gw,
 799                                      &saddr, &daddr);
 800         }
 801 #endif
 802         ;
 803 }
 804
 805 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 806 {
 807         struct rtable *rt;
 808         struct flowi4 fl4;
 809         const struct iphdr *iph = (const struct iphdr *) skb->data;
 810         struct net *net = dev_net(skb->dev);
 811         int oif = skb->dev->ifindex;
 812         u8 tos = RT_TOS(iph->tos);
 813         u8 prot = iph->protocol;
 814         u32 mark = skb->mark;
 815
 816         rt = (struct rtable *) dst;
 817
 818         __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
 819         __ip_do_redirect(rt, skb, &fl4, true);
 820 }
 821
 822 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 823 {
 824         struct rtable *rt = (struct rtable *)dst;
 825         struct dst_entry *ret = dst;
 826
 827         if (rt) {
 828                 if (dst->obsolete > 0) {
 829                         ip_rt_put(rt);
 830                         ret = NULL;
 831                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 832                            rt->dst.expires) {
 833                         ip_rt_put(rt);
 834                         ret = NULL;
 835                 }
 836         }
 837         return ret;
 838 }
 839
 840 /*
 841  * Algorithm:
 842  *      1. The first ip_rt_redirect_number redirects are sent
 843  *         with exponential backoff, then we stop sending them at all,
 844  *         assuming that the host ignores our redirects.
 845  *      2. If we did not see packets requiring redirects
 846  *         during ip_rt_redirect_silence, we assume that the host
 847  *         forgot redirected route and start to send redirects again.
 848  *
 849  * This algorithm is much cheaper and more intelligent than dumb load limiting
 850  * in icmp.c.
 851  *
 852  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 853  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 854  */
 855
 856 void ip_rt_send_redirect(struct sk_buff *skb)
 857 {
 858         struct rtable *rt = skb_rtable(skb);
 859         struct in_device *in_dev;
 860         struct inet_peer *peer;
 861         struct net *net;
 862         int log_martians;
 863         int vif;
 864
 865         rcu_read_lock();
 866         in_dev = __in_dev_get_rcu(rt->dst.dev);
 867         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 868                 rcu_read_unlock();
 869                 return;
 870         }
 871         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 872         vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
 873         rcu_read_unlock();
 874
 875         net = dev_net(rt->dst.dev);
 876         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
 877         if (!peer) {
 878                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 879                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 880                 return;
 881         }
 882
 883         /* No redirected packets during ip_rt_redirect_silence;
 884          * reset the algorithm.
 885          */
 886         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
 887                 peer->rate_tokens = 0;
 888                 peer->n_redirects = 0;
 889         }
 890
 891         /* Too many ignored redirects; do not send anything
 892          * set dst.rate_last to the last seen redirected packet.
 893          */
 894         if (peer->n_redirects >= ip_rt_redirect_number) {
 895                 peer->rate_last = jiffies;
 896                 goto out_put_peer;
 897         }
 898
 899         /* Check for load limit; set rate_last to the latest sent
 900          * redirect.
 901          */
 902         if (peer->n_redirects == 0 ||
 903             time_after(jiffies,
 904                        (peer->rate_last +
 905                         (ip_rt_redirect_load << peer->n_redirects)))) {
 906                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 907
 908                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 909                 peer->rate_last = jiffies;
 910                 ++peer->n_redirects;
 911 #ifdef CONFIG_IP_ROUTE_VERBOSE
 912                 if (log_martians &&
 913                     peer->n_redirects == ip_rt_redirect_number)
 914                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 915                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 916                                              &ip_hdr(skb)->daddr, &gw);
 917 #endif
 918         }
 919 out_put_peer:
 920         inet_putpeer(peer);
 921 }
 922
 923 static int ip_error(struct sk_buff *skb)
 924 {
 925         struct rtable *rt = skb_rtable(skb);
 926         struct net_device *dev = skb->dev;
 927         struct in_device *in_dev;
 928         struct inet_peer *peer;
 929         unsigned long now;
 930         struct net *net;
 931         bool send;
 932         int code;
 933
 934         if (netif_is_l3_master(skb->dev)) {
 935                 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
 936                 if (!dev)
 937                         goto out;
 938         }
 939
 940         in_dev = __in_dev_get_rcu(dev);
 941
 942         /* IP on this device is disabled. */
 943         if (!in_dev)
 944                 goto out;
 945
 946         net = dev_net(rt->dst.dev);
 947         if (!IN_DEV_FORWARD(in_dev)) {
 948                 switch (rt->dst.error) {
 949                 case EHOSTUNREACH:
 950                         __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
 951                         break;
 952
 953                 case ENETUNREACH:
 954                         __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 955                         break;
 956                 }
 957                 goto out;
 958         }
 959
 960         switch (rt->dst.error) {
 961         case EINVAL:
 962         default:
 963                 goto out;
 964         case EHOSTUNREACH:
 965                 code = ICMP_HOST_UNREACH;
 966                 break;
 967         case ENETUNREACH:
 968                 code = ICMP_NET_UNREACH;
 969                 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 970                 break;
 971         case EACCES:
 972                 code = ICMP_PKT_FILTERED;
 973                 break;
 974         }
 975
 976         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
 977                                l3mdev_master_ifindex(skb->dev), 1);
 978
 979         send = true;
 980         if (peer) {
 981                 now = jiffies;
 982                 peer->rate_tokens += now - peer->rate_last;
 983                 if (peer->rate_tokens > ip_rt_error_burst)
 984                         peer->rate_tokens = ip_rt_error_burst;
 985                 peer->rate_last = now;
 986                 if (peer->rate_tokens >= ip_rt_error_cost)
 987                         peer->rate_tokens -= ip_rt_error_cost;
 988                 else
 989                         send = false;
 990                 inet_putpeer(peer);
 991         }
 992         if (send)
 993                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
 994
 995 out:    kfree_skb(skb);
 996         return 0;
 997 }
 998
 999 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1000 {
1001         struct dst_entry *dst = &rt->dst;
1002         struct net *net = dev_net(dst->dev);
1003         struct fib_result res;
1004         bool lock = false;
1005         u32 old_mtu;
1006
1007         if (ip_mtu_locked(dst))
1008                 return;
1009
1010         old_mtu = ipv4_mtu(dst);
1011         if (old_mtu < mtu)
1012                 return;
1013
1014         if (mtu < ip_rt_min_pmtu) {
1015                 lock = true;
1016                 mtu = min(old_mtu, ip_rt_min_pmtu);
1017         }
1018
1019         if (rt->rt_pmtu == mtu && !lock &&
1020             time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1021                 return;
1022
1023         rcu_read_lock();
1024         if (fib_lookup(net, fl4, &res, 0) == 0) {
1025                 struct fib_nh_common *nhc;
1026
1027                 fib_select_path(net, &res, fl4, NULL);
1028                 nhc = FIB_RES_NHC(res);
1029                 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1030                                       jiffies + ip_rt_mtu_expires);
1031         }
1032         rcu_read_unlock();
1033 }
1034
1035 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1036                               struct sk_buff *skb, u32 mtu,
1037                               bool confirm_neigh)
1038 {
1039         struct rtable *rt = (struct rtable *) dst;
1040         struct flowi4 fl4;
1041
1042         ip_rt_build_flow_key(&fl4, sk, skb);
1043
1044         /* Don't make lookup fail for bridged encapsulations */
1045         if (skb && netif_is_any_bridge_port(skb->dev))
1046                 fl4.flowi4_oif = 0;
1047
1048         __ip_rt_update_pmtu(rt, &fl4, mtu);
1049 }
1050
1051 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1052                       int oif, u8 protocol)
1053 {
1054         const struct iphdr *iph = (const struct iphdr *)skb->data;
1055         struct flowi4 fl4;
1056         struct rtable *rt;
1057         u32 mark = IP4_REPLY_MARK(net, skb->mark);
1058
1059         __build_flow_key(net, &fl4, NULL, iph, oif,
1060                          RT_TOS(iph->tos), protocol, mark, 0);
1061         rt = __ip_route_output_key(net, &fl4);
1062         if (!IS_ERR(rt)) {
1063                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1064                 ip_rt_put(rt);
1065         }
1066 }
1067 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1068
1069 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1070 {
1071         const struct iphdr *iph = (const struct iphdr *)skb->data;
1072         struct flowi4 fl4;
1073         struct rtable *rt;
1074
1075         __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1076
1077         if (!fl4.flowi4_mark)
1078                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1079
1080         rt = __ip_route_output_key(sock_net(sk), &fl4);
1081         if (!IS_ERR(rt)) {
1082                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1083                 ip_rt_put(rt);
1084         }
1085 }
1086
1087 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         struct dst_entry *odst = NULL;
1093         bool new = false;
1094         struct net *net = sock_net(sk);
1095
1096         bh_lock_sock(sk);
1097
1098         if (!ip_sk_accept_pmtu(sk))
1099                 goto out;
1100
1101         odst = sk_dst_get(sk);
1102
1103         if (sock_owned_by_user(sk) || !odst) {
1104                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1105                 goto out;
1106         }
1107
1108         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1109
1110         rt = (struct rtable *)odst;
1111         if (odst->obsolete && !odst->ops->check(odst, 0)) {
1112                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1113                 if (IS_ERR(rt))
1114                         goto out;
1115
1116                 new = true;
1117         }
1118
1119         __ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu);
1120
1121         if (!dst_check(&rt->dst, 0)) {
1122                 if (new)
1123                         dst_release(&rt->dst);
1124
1125                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1126                 if (IS_ERR(rt))
1127                         goto out;
1128
1129                 new = true;
1130         }
1131
1132         if (new)
1133                 sk_dst_set(sk, &rt->dst);
1134
1135 out:
1136         bh_unlock_sock(sk);
1137         dst_release(odst);
1138 }
1139 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1140
1141 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1142                    int oif, u8 protocol)
1143 {
1144         const struct iphdr *iph = (const struct iphdr *)skb->data;
1145         struct flowi4 fl4;
1146         struct rtable *rt;
1147
1148         __build_flow_key(net, &fl4, NULL, iph, oif,
1149                          RT_TOS(iph->tos), protocol, 0, 0);
1150         rt = __ip_route_output_key(net, &fl4);
1151         if (!IS_ERR(rt)) {
1152                 __ip_do_redirect(rt, skb, &fl4, false);
1153                 ip_rt_put(rt);
1154         }
1155 }
1156 EXPORT_SYMBOL_GPL(ipv4_redirect);
1157
1158 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1159 {
1160         const struct iphdr *iph = (const struct iphdr *)skb->data;
1161         struct flowi4 fl4;
1162         struct rtable *rt;
1163         struct net *net = sock_net(sk);
1164
1165         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1166         rt = __ip_route_output_key(net, &fl4);
1167         if (!IS_ERR(rt)) {
1168                 __ip_do_redirect(rt, skb, &fl4, false);
1169                 ip_rt_put(rt);
1170         }
1171 }
1172 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1173
1174 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1175                                                          u32 cookie)
1176 {
1177         struct rtable *rt = (struct rtable *) dst;
1178
1179         /* All IPV4 dsts are created with ->obsolete set to the value
1180          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1181          * into this function always.
1182          *
1183          * When a PMTU/redirect information update invalidates a route,
1184          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1185          * DST_OBSOLETE_DEAD.
1186          */
1187         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1188                 return NULL;
1189         return dst;
1190 }
1191 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1192
1193 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1194 {
1195         struct ip_options opt;
1196         int res;
1197
1198         /* Recompile ip options since IPCB may not be valid anymore.
1199          * Also check we have a reasonable ipv4 header.
1200          */
1201         if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1202             ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1203                 return;
1204
1205         memset(&opt, 0, sizeof(opt));
1206         if (ip_hdr(skb)->ihl > 5) {
1207                 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1208                         return;
1209                 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1210
1211                 rcu_read_lock();
1212                 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1213                 rcu_read_unlock();
1214
1215                 if (res)
1216                         return;
1217         }
1218         __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1219 }
1220
1221 static void ipv4_link_failure(struct sk_buff *skb)
1222 {
1223         struct rtable *rt;
1224
1225         ipv4_send_dest_unreach(skb);
1226
1227         rt = skb_rtable(skb);
1228         if (rt)
1229                 dst_set_expires(&rt->dst, 0);
1230 }
1231
1232 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1233 {
1234         pr_debug("%s: %pI4 -> %pI4, %s\n",
1235                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1236                  skb->dev ? skb->dev->name : "?");
1237         kfree_skb(skb);
1238         WARN_ON(1);
1239         return 0;
1240 }
1241
1242 /*
1243  * We do not cache source address of outgoing interface,
1244  * because it is used only by IP RR, TS and SRR options,
1245  * so that it out of fast path.
1246  *
1247  * BTW remember: "addr" is allowed to be not aligned
1248  * in IP options!
1249  */
1250
1251 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1252 {
1253         __be32 src;
1254
1255         if (rt_is_output_route(rt))
1256                 src = ip_hdr(skb)->saddr;
1257         else {
1258                 struct fib_result res;
1259                 struct iphdr *iph = ip_hdr(skb);
1260                 struct flowi4 fl4 = {
1261                         .daddr = iph->daddr,
1262                         .saddr = iph->saddr,
1263                         .flowi4_tos = RT_TOS(iph->tos),
1264                         .flowi4_oif = rt->dst.dev->ifindex,
1265                         .flowi4_iif = skb->dev->ifindex,
1266                         .flowi4_mark = skb->mark,
1267                 };
1268
1269                 rcu_read_lock();
1270                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1271                         src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1272                 else
1273                         src = inet_select_addr(rt->dst.dev,
1274                                                rt_nexthop(rt, iph->daddr),
1275                                                RT_SCOPE_UNIVERSE);
1276                 rcu_read_unlock();
1277         }
1278         memcpy(addr, &src, 4);
1279 }
1280
1281 #ifdef CONFIG_IP_ROUTE_CLASSID
1282 static void set_class_tag(struct rtable *rt, u32 tag)
1283 {
1284         if (!(rt->dst.tclassid & 0xFFFF))
1285                 rt->dst.tclassid |= tag & 0xFFFF;
1286         if (!(rt->dst.tclassid & 0xFFFF0000))
1287                 rt->dst.tclassid |= tag & 0xFFFF0000;
1288 }
1289 #endif
1290
1291 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1292 {
1293         unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1294         unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1295                                     ip_rt_min_advmss);
1296
1297         return min(advmss, IPV4_MAX_PMTU - header_size);
1298 }
1299
1300 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1301 {
1302         const struct rtable *rt = (const struct rtable *)dst;
1303         unsigned int mtu = rt->rt_pmtu;
1304
1305         if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1306                 mtu = dst_metric_raw(dst, RTAX_MTU);
1307
1308         if (mtu)
1309                 goto out;
1310
1311         mtu = READ_ONCE(dst->dev->mtu);
1312
1313         if (unlikely(ip_mtu_locked(dst))) {
1314                 if (rt->rt_uses_gateway && mtu > 576)
1315                         mtu = 576;
1316         }
1317
1318 out:
1319         mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1320
1321         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1322 }
1323 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1324
1325 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1326 {
1327         struct fnhe_hash_bucket *hash;
1328         struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1329         u32 hval = fnhe_hashfun(daddr);
1330
1331         spin_lock_bh(&fnhe_lock);
1332
1333         hash = rcu_dereference_protected(nhc->nhc_exceptions,
1334                                          lockdep_is_held(&fnhe_lock));
1335         hash += hval;
1336
1337         fnhe_p = &hash->chain;
1338         fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1339         while (fnhe) {
1340                 if (fnhe->fnhe_daddr == daddr) {
1341                         rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1342                                 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1343                         /* set fnhe_daddr to 0 to ensure it won't bind with
1344                          * new dsts in rt_bind_exception().
1345                          */
1346                         fnhe->fnhe_daddr = 0;
1347                         fnhe_flush_routes(fnhe);
1348                         kfree_rcu(fnhe, rcu);
1349                         break;
1350                 }
1351                 fnhe_p = &fnhe->fnhe_next;
1352                 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1353                                                  lockdep_is_held(&fnhe_lock));
1354         }
1355
1356         spin_unlock_bh(&fnhe_lock);
1357 }
1358
1359 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1360                                                __be32 daddr)
1361 {
1362         struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1363         struct fib_nh_exception *fnhe;
1364         u32 hval;
1365
1366         if (!hash)
1367                 return NULL;
1368
1369         hval = fnhe_hashfun(daddr);
1370
1371         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1372              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1373                 if (fnhe->fnhe_daddr == daddr) {
1374                         if (fnhe->fnhe_expires &&
1375                             time_after(jiffies, fnhe->fnhe_expires)) {
1376                                 ip_del_fnhe(nhc, daddr);
1377                                 break;
1378                         }
1379                         return fnhe;
1380                 }
1381         }
1382         return NULL;
1383 }
1384
1385 /* MTU selection:
1386  * 1. mtu on route is locked - use it
1387  * 2. mtu from nexthop exception
1388  * 3. mtu from egress device
1389  */
1390
1391 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1392 {
1393         struct fib_nh_common *nhc = res->nhc;
1394         struct net_device *dev = nhc->nhc_dev;
1395         struct fib_info *fi = res->fi;
1396         u32 mtu = 0;
1397
1398         if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1399             fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1400                 mtu = fi->fib_mtu;
1401
1402         if (likely(!mtu)) {
1403                 struct fib_nh_exception *fnhe;
1404
1405                 fnhe = find_exception(nhc, daddr);
1406                 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1407                         mtu = fnhe->fnhe_pmtu;
1408         }
1409
1410         if (likely(!mtu))
1411                 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1412
1413         return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1414 }
1415
1416 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1417                               __be32 daddr, const bool do_cache)
1418 {
1419         bool ret = false;
1420
1421         spin_lock_bh(&fnhe_lock);
1422
1423         if (daddr == fnhe->fnhe_daddr) {
1424                 struct rtable __rcu **porig;
1425                 struct rtable *orig;
1426                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1427
1428                 if (rt_is_input_route(rt))
1429                         porig = &fnhe->fnhe_rth_input;
1430                 else
1431                         porig = &fnhe->fnhe_rth_output;
1432                 orig = rcu_dereference(*porig);
1433
1434                 if (fnhe->fnhe_genid != genid) {
1435                         fnhe->fnhe_genid = genid;
1436                         fnhe->fnhe_gw = 0;
1437                         fnhe->fnhe_pmtu = 0;
1438                         fnhe->fnhe_expires = 0;
1439                         fnhe->fnhe_mtu_locked = false;
1440                         fnhe_flush_routes(fnhe);
1441                         orig = NULL;
1442                 }
1443                 fill_route_from_fnhe(rt, fnhe);
1444                 if (!rt->rt_gw4) {
1445                         rt->rt_gw4 = daddr;
1446                         rt->rt_gw_family = AF_INET;
1447                 }
1448
1449                 if (do_cache) {
1450                         dst_hold(&rt->dst);
1451                         rcu_assign_pointer(*porig, rt);
1452                         if (orig) {
1453                                 dst_dev_put(&orig->dst);
1454                                 dst_release(&orig->dst);
1455                         }
1456                         ret = true;
1457                 }
1458
1459                 fnhe->fnhe_stamp = jiffies;
1460         }
1461         spin_unlock_bh(&fnhe_lock);
1462
1463         return ret;
1464 }
1465
1466 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1467 {
1468         struct rtable *orig, *prev, **p;
1469         bool ret = true;
1470
1471         if (rt_is_input_route(rt)) {
1472                 p = (struct rtable **)&nhc->nhc_rth_input;
1473         } else {
1474                 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1475         }
1476         orig = *p;
1477
1478         /* hold dst before doing cmpxchg() to avoid race condition
1479          * on this dst
1480          */
1481         dst_hold(&rt->dst);
1482         prev = cmpxchg(p, orig, rt);
1483         if (prev == orig) {
1484                 if (orig) {
1485                         rt_add_uncached_list(orig);
1486                         dst_release(&orig->dst);
1487                 }
1488         } else {
1489                 dst_release(&rt->dst);
1490                 ret = false;
1491         }
1492
1493         return ret;
1494 }
1495
1496 struct uncached_list {
1497         spinlock_t              lock;
1498         struct list_head        head;
1499 };
1500
1501 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1502
1503 void rt_add_uncached_list(struct rtable *rt)
1504 {
1505         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1506
1507         rt->rt_uncached_list = ul;
1508
1509         spin_lock_bh(&ul->lock);
1510         list_add_tail(&rt->rt_uncached, &ul->head);
1511         spin_unlock_bh(&ul->lock);
1512 }
1513
1514 void rt_del_uncached_list(struct rtable *rt)
1515 {
1516         if (!list_empty(&rt->rt_uncached)) {
1517                 struct uncached_list *ul = rt->rt_uncached_list;
1518
1519                 spin_lock_bh(&ul->lock);
1520                 list_del(&rt->rt_uncached);
1521                 spin_unlock_bh(&ul->lock);
1522         }
1523 }
1524
1525 static void ipv4_dst_destroy(struct dst_entry *dst)
1526 {
1527         struct rtable *rt = (struct rtable *)dst;
1528
1529         ip_dst_metrics_put(dst);
1530         rt_del_uncached_list(rt);
1531 }
1532
1533 void rt_flush_dev(struct net_device *dev)
1534 {
1535         struct rtable *rt;
1536         int cpu;
1537
1538         for_each_possible_cpu(cpu) {
1539                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1540
1541                 spin_lock_bh(&ul->lock);
1542                 list_for_each_entry(rt, &ul->head, rt_uncached) {
1543                         if (rt->dst.dev != dev)
1544                                 continue;
1545                         rt->dst.dev = blackhole_netdev;
1546                         dev_hold(rt->dst.dev);
1547                         dev_put(dev);
1548                 }
1549                 spin_unlock_bh(&ul->lock);
1550         }
1551 }
1552
1553 static bool rt_cache_valid(const struct rtable *rt)
1554 {
1555         return  rt &&
1556                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1557                 !rt_is_expired(rt);
1558 }
1559
1560 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1561                            const struct fib_result *res,
1562                            struct fib_nh_exception *fnhe,
1563                            struct fib_info *fi, u16 type, u32 itag,
1564                            const bool do_cache)
1565 {
1566         bool cached = false;
1567
1568         if (fi) {
1569                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1570
1571                 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1572                         rt->rt_uses_gateway = 1;
1573                         rt->rt_gw_family = nhc->nhc_gw_family;
1574                         /* only INET and INET6 are supported */
1575                         if (likely(nhc->nhc_gw_family == AF_INET))
1576                                 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1577                         else
1578                                 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1579                 }
1580
1581                 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1582
1583 #ifdef CONFIG_IP_ROUTE_CLASSID
1584                 if (nhc->nhc_family == AF_INET) {
1585                         struct fib_nh *nh;
1586
1587                         nh = container_of(nhc, struct fib_nh, nh_common);
1588                         rt->dst.tclassid = nh->nh_tclassid;
1589                 }
1590 #endif
1591                 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1592                 if (unlikely(fnhe))
1593                         cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1594                 else if (do_cache)
1595                         cached = rt_cache_route(nhc, rt);
1596                 if (unlikely(!cached)) {
1597                         /* Routes we intend to cache in nexthop exception or
1598                          * FIB nexthop have the DST_NOCACHE bit clear.
1599                          * However, if we are unsuccessful at storing this
1600                          * route into the cache we really need to set it.
1601                          */
1602                         if (!rt->rt_gw4) {
1603                                 rt->rt_gw_family = AF_INET;
1604                                 rt->rt_gw4 = daddr;
1605                         }
1606                         rt_add_uncached_list(rt);
1607                 }
1608         } else
1609                 rt_add_uncached_list(rt);
1610
1611 #ifdef CONFIG_IP_ROUTE_CLASSID
1612 #ifdef CONFIG_IP_MULTIPLE_TABLES
1613         set_class_tag(rt, res->tclassid);
1614 #endif
1615         set_class_tag(rt, itag);
1616 #endif
1617 }
1618
1619 struct rtable *rt_dst_alloc(struct net_device *dev,
1620                             unsigned int flags, u16 type,
1621                             bool nopolicy, bool noxfrm)
1622 {
1623         struct rtable *rt;
1624
1625         rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1626                        (nopolicy ? DST_NOPOLICY : 0) |
1627                        (noxfrm ? DST_NOXFRM : 0));
1628
1629         if (rt) {
1630                 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1631                 rt->rt_flags = flags;
1632                 rt->rt_type = type;
1633                 rt->rt_is_input = 0;
1634                 rt->rt_iif = 0;
1635                 rt->rt_pmtu = 0;
1636                 rt->rt_mtu_locked = 0;
1637                 rt->rt_uses_gateway = 0;
1638                 rt->rt_gw_family = 0;
1639                 rt->rt_gw4 = 0;
1640                 INIT_LIST_HEAD(&rt->rt_uncached);
1641
1642                 rt->dst.output = ip_output;
1643                 if (flags & RTCF_LOCAL)
1644                         rt->dst.input = ip_local_deliver;
1645         }
1646
1647         return rt;
1648 }
1649 EXPORT_SYMBOL(rt_dst_alloc);
1650
1651 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1652 {
1653         struct rtable *new_rt;
1654
1655         new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1656                            rt->dst.flags);
1657
1658         if (new_rt) {
1659                 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1660                 new_rt->rt_flags = rt->rt_flags;
1661                 new_rt->rt_type = rt->rt_type;
1662                 new_rt->rt_is_input = rt->rt_is_input;
1663                 new_rt->rt_iif = rt->rt_iif;
1664                 new_rt->rt_pmtu = rt->rt_pmtu;
1665                 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1666                 new_rt->rt_gw_family = rt->rt_gw_family;
1667                 if (rt->rt_gw_family == AF_INET)
1668                         new_rt->rt_gw4 = rt->rt_gw4;
1669                 else if (rt->rt_gw_family == AF_INET6)
1670                         new_rt->rt_gw6 = rt->rt_gw6;
1671                 INIT_LIST_HEAD(&new_rt->rt_uncached);
1672
1673                 new_rt->dst.input = rt->dst.input;
1674                 new_rt->dst.output = rt->dst.output;
1675                 new_rt->dst.error = rt->dst.error;
1676                 new_rt->dst.lastuse = jiffies;
1677                 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1678         }
1679         return new_rt;
1680 }
1681 EXPORT_SYMBOL(rt_dst_clone);
1682
1683 /* called in rcu_read_lock() section */
1684 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1685                           u8 tos, struct net_device *dev,
1686                           struct in_device *in_dev, u32 *itag)
1687 {
1688         int err;
1689
1690         /* Primary sanity checks. */
1691         if (!in_dev)
1692                 return -EINVAL;
1693
1694         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1695             skb->protocol != htons(ETH_P_IP))
1696                 return -EINVAL;
1697
1698         if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1699                 return -EINVAL;
1700
1701         if (ipv4_is_zeronet(saddr)) {
1702                 if (!ipv4_is_local_multicast(daddr) &&
1703                     ip_hdr(skb)->protocol != IPPROTO_IGMP)
1704                         return -EINVAL;
1705         } else {
1706                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1707                                           in_dev, itag);
1708                 if (err < 0)
1709                         return err;
1710         }
1711         return 0;
1712 }
1713
1714 /* called in rcu_read_lock() section */
1715 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1716                              u8 tos, struct net_device *dev, int our)
1717 {
1718         struct in_device *in_dev = __in_dev_get_rcu(dev);
1719         unsigned int flags = RTCF_MULTICAST;
1720         struct rtable *rth;
1721         u32 itag = 0;
1722         int err;
1723
1724         err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1725         if (err)
1726                 return err;
1727
1728         if (our)
1729                 flags |= RTCF_LOCAL;
1730
1731         rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1732                            IN_DEV_ORCONF(in_dev, NOPOLICY), false);
1733         if (!rth)
1734                 return -ENOBUFS;
1735
1736 #ifdef CONFIG_IP_ROUTE_CLASSID
1737         rth->dst.tclassid = itag;
1738 #endif
1739         rth->dst.output = ip_rt_bug;
1740         rth->rt_is_input= 1;
1741
1742 #ifdef CONFIG_IP_MROUTE
1743         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1744                 rth->dst.input = ip_mr_input;
1745 #endif
1746         RT_CACHE_STAT_INC(in_slow_mc);
1747
1748         skb_dst_set(skb, &rth->dst);
1749         return 0;
1750 }
1751
1752
1753 static void ip_handle_martian_source(struct net_device *dev,
1754                                      struct in_device *in_dev,
1755                                      struct sk_buff *skb,
1756                                      __be32 daddr,
1757                                      __be32 saddr)
1758 {
1759         RT_CACHE_STAT_INC(in_martian_src);
1760 #ifdef CONFIG_IP_ROUTE_VERBOSE
1761         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1762                 /*
1763                  *      RFC1812 recommendation, if source is martian,
1764                  *      the only hint is MAC header.
1765                  */
1766                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1767                         &daddr, &saddr, dev->name);
1768                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1769                         print_hex_dump(KERN_WARNING, "ll header: ",
1770                                        DUMP_PREFIX_OFFSET, 16, 1,
1771                                        skb_mac_header(skb),
1772                                        dev->hard_header_len, false);
1773                 }
1774         }
1775 #endif
1776 }
1777
1778 /* called in rcu_read_lock() section */
1779 static int __mkroute_input(struct sk_buff *skb,
1780                            const struct fib_result *res,
1781                            struct in_device *in_dev,
1782                            __be32 daddr, __be32 saddr, u32 tos)
1783 {
1784         struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1785         struct net_device *dev = nhc->nhc_dev;
1786         struct fib_nh_exception *fnhe;
1787         struct rtable *rth;
1788         int err;
1789         struct in_device *out_dev;
1790         bool do_cache;
1791         u32 itag = 0;
1792
1793         /* get a working reference to the output device */
1794         out_dev = __in_dev_get_rcu(dev);
1795         if (!out_dev) {
1796                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1797                 return -EINVAL;
1798         }
1799
1800         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1801                                   in_dev->dev, in_dev, &itag);
1802         if (err < 0) {
1803                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1804                                          saddr);
1805
1806                 goto cleanup;
1807         }
1808
1809         do_cache = res->fi && !itag;
1810         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1811             skb->protocol == htons(ETH_P_IP)) {
1812                 __be32 gw;
1813
1814                 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1815                 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1816                     inet_addr_onlink(out_dev, saddr, gw))
1817                         IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1818         }
1819
1820         if (skb->protocol != htons(ETH_P_IP)) {
1821                 /* Not IP (i.e. ARP). Do not create route, if it is
1822                  * invalid for proxy arp. DNAT routes are always valid.
1823                  *
1824                  * Proxy arp feature have been extended to allow, ARP
1825                  * replies back to the same interface, to support
1826                  * Private VLAN switch technologies. See arp.c.
1827                  */
1828                 if (out_dev == in_dev &&
1829                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1830                         err = -EINVAL;
1831                         goto cleanup;
1832                 }
1833         }
1834
1835         fnhe = find_exception(nhc, daddr);
1836         if (do_cache) {
1837                 if (fnhe)
1838                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1839                 else
1840                         rth = rcu_dereference(nhc->nhc_rth_input);
1841                 if (rt_cache_valid(rth)) {
1842                         skb_dst_set_noref(skb, &rth->dst);
1843                         goto out;
1844                 }
1845         }
1846
1847         rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1848                            IN_DEV_ORCONF(in_dev, NOPOLICY),
1849                            IN_DEV_ORCONF(out_dev, NOXFRM));
1850         if (!rth) {
1851                 err = -ENOBUFS;
1852                 goto cleanup;
1853         }
1854
1855         rth->rt_is_input = 1;
1856         RT_CACHE_STAT_INC(in_slow_tot);
1857
1858         rth->dst.input = ip_forward;
1859
1860         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1861                        do_cache);
1862         lwtunnel_set_redirect(&rth->dst);
1863         skb_dst_set(skb, &rth->dst);
1864 out:
1865         err = 0;
1866  cleanup:
1867         return err;
1868 }
1869
1870 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1871 /* To make ICMP packets follow the right flow, the multipath hash is
1872  * calculated from the inner IP addresses.
1873  */
1874 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1875                                  struct flow_keys *hash_keys)
1876 {
1877         const struct iphdr *outer_iph = ip_hdr(skb);
1878         const struct iphdr *key_iph = outer_iph;
1879         const struct iphdr *inner_iph;
1880         const struct icmphdr *icmph;
1881         struct iphdr _inner_iph;
1882         struct icmphdr _icmph;
1883
1884         if (likely(outer_iph->protocol != IPPROTO_ICMP))
1885                 goto out;
1886
1887         if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1888                 goto out;
1889
1890         icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1891                                    &_icmph);
1892         if (!icmph)
1893                 goto out;
1894
1895         if (!icmp_is_err(icmph->type))
1896                 goto out;
1897
1898         inner_iph = skb_header_pointer(skb,
1899                                        outer_iph->ihl * 4 + sizeof(_icmph),
1900                                        sizeof(_inner_iph), &_inner_iph);
1901         if (!inner_iph)
1902                 goto out;
1903
1904         key_iph = inner_iph;
1905 out:
1906         hash_keys->addrs.v4addrs.src = key_iph->saddr;
1907         hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1908 }
1909
1910 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1911                                            const struct sk_buff *skb,
1912                                            bool *p_has_inner)
1913 {
1914         u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
1915         struct flow_keys keys, hash_keys;
1916
1917         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1918                 return 0;
1919
1920         memset(&hash_keys, 0, sizeof(hash_keys));
1921         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1922
1923         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1924         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1925                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1926         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1927                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1928         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1929                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1930         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1931                 hash_keys.ports.src = keys.ports.src;
1932         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1933                 hash_keys.ports.dst = keys.ports.dst;
1934
1935         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1936         return flow_hash_from_keys(&hash_keys);
1937 }
1938
1939 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1940                                            const struct sk_buff *skb,
1941                                            bool has_inner)
1942 {
1943         u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
1944         struct flow_keys keys, hash_keys;
1945
1946         /* We assume the packet carries an encapsulation, but if none was
1947          * encountered during dissection of the outer flow, then there is no
1948          * point in calling the flow dissector again.
1949          */
1950         if (!has_inner)
1951                 return 0;
1952
1953         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1954                 return 0;
1955
1956         memset(&hash_keys, 0, sizeof(hash_keys));
1957         skb_flow_dissect_flow_keys(skb, &keys, 0);
1958
1959         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1960                 return 0;
1961
1962         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1963                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1964                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1965                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1966                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1967                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1968         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1969                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1970                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1971                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1972                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1973                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1974                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1975                         hash_keys.tags.flow_label = keys.tags.flow_label;
1976         }
1977
1978         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1979                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1980         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1981                 hash_keys.ports.src = keys.ports.src;
1982         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1983                 hash_keys.ports.dst = keys.ports.dst;
1984
1985         return flow_hash_from_keys(&hash_keys);
1986 }
1987
1988 static u32 fib_multipath_custom_hash_skb(const struct net *net,
1989                                          const struct sk_buff *skb)
1990 {
1991         u32 mhash, mhash_inner;
1992         bool has_inner = true;
1993
1994         mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
1995         mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
1996
1997         return jhash_2words(mhash, mhash_inner, 0);
1998 }
1999
2000 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2001                                          const struct flowi4 *fl4)
2002 {
2003         u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
2004         struct flow_keys hash_keys;
2005
2006         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2007                 return 0;
2008
2009         memset(&hash_keys, 0, sizeof(hash_keys));
2010         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2011         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2012                 hash_keys.addrs.v4addrs.src = fl4->saddr;
2013         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2014                 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2015         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2016                 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2017         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2018                 hash_keys.ports.src = fl4->fl4_sport;
2019         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2020                 hash_keys.ports.dst = fl4->fl4_dport;
2021
2022         return flow_hash_from_keys(&hash_keys);
2023 }
2024
2025 /* if skb is set it will be used and fl4 can be NULL */
2026 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2027                        const struct sk_buff *skb, struct flow_keys *flkeys)
2028 {
2029         u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2030         struct flow_keys hash_keys;
2031         u32 mhash = 0;
2032
2033         switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
2034         case 0:
2035                 memset(&hash_keys, 0, sizeof(hash_keys));
2036                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2037                 if (skb) {
2038                         ip_multipath_l3_keys(skb, &hash_keys);
2039                 } else {
2040                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2041                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2042                 }
2043                 mhash = flow_hash_from_keys(&hash_keys);
2044                 break;
2045         case 1:
2046                 /* skb is currently provided only when forwarding */
2047                 if (skb) {
2048                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2049                         struct flow_keys keys;
2050
2051                         /* short-circuit if we already have L4 hash present */
2052                         if (skb->l4_hash)
2053                                 return skb_get_hash_raw(skb) >> 1;
2054
2055                         memset(&hash_keys, 0, sizeof(hash_keys));
2056
2057                         if (!flkeys) {
2058                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2059                                 flkeys = &keys;
2060                         }
2061
2062                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2063                         hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2064                         hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2065                         hash_keys.ports.src = flkeys->ports.src;
2066                         hash_keys.ports.dst = flkeys->ports.dst;
2067                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2068                 } else {
2069                         memset(&hash_keys, 0, sizeof(hash_keys));
2070                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2071                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2072                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2073                         hash_keys.ports.src = fl4->fl4_sport;
2074                         hash_keys.ports.dst = fl4->fl4_dport;
2075                         hash_keys.basic.ip_proto = fl4->flowi4_proto;
2076                 }
2077                 mhash = flow_hash_from_keys(&hash_keys);
2078                 break;
2079         case 2:
2080                 memset(&hash_keys, 0, sizeof(hash_keys));
2081                 /* skb is currently provided only when forwarding */
2082                 if (skb) {
2083                         struct flow_keys keys;
2084
2085                         skb_flow_dissect_flow_keys(skb, &keys, 0);
2086                         /* Inner can be v4 or v6 */
2087                         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2088                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2089                                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2090                                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2091                         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2092                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2093                                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2094                                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2095                                 hash_keys.tags.flow_label = keys.tags.flow_label;
2096                                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2097                         } else {
2098                                 /* Same as case 0 */
2099                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2100                                 ip_multipath_l3_keys(skb, &hash_keys);
2101                         }
2102                 } else {
2103                         /* Same as case 0 */
2104                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2105                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2106                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2107                 }
2108                 mhash = flow_hash_from_keys(&hash_keys);
2109                 break;
2110         case 3:
2111                 if (skb)
2112                         mhash = fib_multipath_custom_hash_skb(net, skb);
2113                 else
2114                         mhash = fib_multipath_custom_hash_fl4(net, fl4);
2115                 break;
2116         }
2117
2118         if (multipath_hash)
2119                 mhash = jhash_2words(mhash, multipath_hash, 0);
2120
2121         return mhash >> 1;
2122 }
2123 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2124
2125 static int ip_mkroute_input(struct sk_buff *skb,
2126                             struct fib_result *res,
2127                             struct in_device *in_dev,
2128                             __be32 daddr, __be32 saddr, u32 tos,
2129                             struct flow_keys *hkeys)
2130 {
2131 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2132         if (res->fi && fib_info_num_path(res->fi) > 1) {
2133                 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2134
2135                 fib_select_multipath(res, h);
2136         }
2137 #endif
2138
2139         /* create a routing cache entry */
2140         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2141 }
2142
2143 /* Implements all the saddr-related checks as ip_route_input_slow(),
2144  * assuming daddr is valid and the destination is not a local broadcast one.
2145  * Uses the provided hint instead of performing a route lookup.
2146  */
2147 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2148                       u8 tos, struct net_device *dev,
2149                       const struct sk_buff *hint)
2150 {
2151         struct in_device *in_dev = __in_dev_get_rcu(dev);
2152         struct rtable *rt = skb_rtable(hint);
2153         struct net *net = dev_net(dev);
2154         int err = -EINVAL;
2155         u32 tag = 0;
2156
2157         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2158                 goto martian_source;
2159
2160         if (ipv4_is_zeronet(saddr))
2161                 goto martian_source;
2162
2163         if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2164                 goto martian_source;
2165
2166         if (rt->rt_type != RTN_LOCAL)
2167                 goto skip_validate_source;
2168
2169         tos &= IPTOS_RT_MASK;
2170         err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2171         if (err < 0)
2172                 goto martian_source;
2173
2174 skip_validate_source:
2175         skb_dst_copy(skb, hint);
2176         return 0;
2177
2178 martian_source:
2179         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2180         return err;
2181 }
2182
2183 /* get device for dst_alloc with local routes */
2184 static struct net_device *ip_rt_get_dev(struct net *net,
2185                                         const struct fib_result *res)
2186 {
2187         struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2188         struct net_device *dev = NULL;
2189
2190         if (nhc)
2191                 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2192
2193         return dev ? : net->loopback_dev;
2194 }
2195
2196 /*
2197  *      NOTE. We drop all the packets that has local source
2198  *      addresses, because every properly looped back packet
2199  *      must have correct destination already attached by output routine.
2200  *      Changes in the enforced policies must be applied also to
2201  *      ip_route_use_hint().
2202  *
2203  *      Such approach solves two big problems:
2204  *      1. Not simplex devices are handled properly.
2205  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2206  *      called with rcu_read_lock()
2207  */
2208
2209 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2210                                u8 tos, struct net_device *dev,
2211                                struct fib_result *res)
2212 {
2213         struct in_device *in_dev = __in_dev_get_rcu(dev);
2214         struct flow_keys *flkeys = NULL, _flkeys;
2215         struct net    *net = dev_net(dev);
2216         struct ip_tunnel_info *tun_info;
2217         int             err = -EINVAL;
2218         unsigned int    flags = 0;
2219         u32             itag = 0;
2220         struct rtable   *rth;
2221         struct flowi4   fl4;
2222         bool do_cache = true;
2223
2224         /* IP on this device is disabled. */
2225
2226         if (!in_dev)
2227                 goto out;
2228
2229         /* Check for the most weird martians, which can be not detected
2230          * by fib_lookup.
2231          */
2232
2233         tun_info = skb_tunnel_info(skb);
2234         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2235                 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2236         else
2237                 fl4.flowi4_tun_key.tun_id = 0;
2238         skb_dst_drop(skb);
2239
2240         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2241                 goto martian_source;
2242
2243         res->fi = NULL;
2244         res->table = NULL;
2245         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2246                 goto brd_input;
2247
2248         /* Accept zero addresses only to limited broadcast;
2249          * I even do not know to fix it or not. Waiting for complains :-)
2250          */
2251         if (ipv4_is_zeronet(saddr))
2252                 goto martian_source;
2253
2254         if (ipv4_is_zeronet(daddr))
2255                 goto martian_destination;
2256
2257         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2258          * and call it once if daddr or/and saddr are loopback addresses
2259          */
2260         if (ipv4_is_loopback(daddr)) {
2261                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2262                         goto martian_destination;
2263         } else if (ipv4_is_loopback(saddr)) {
2264                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2265                         goto martian_source;
2266         }
2267
2268         /*
2269          *      Now we are ready to route packet.
2270          */
2271         fl4.flowi4_oif = 0;
2272         fl4.flowi4_iif = dev->ifindex;
2273         fl4.flowi4_mark = skb->mark;
2274         fl4.flowi4_tos = tos;
2275         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2276         fl4.flowi4_flags = 0;
2277         fl4.daddr = daddr;
2278         fl4.saddr = saddr;
2279         fl4.flowi4_uid = sock_net_uid(net, NULL);
2280         fl4.flowi4_multipath_hash = 0;
2281
2282         if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2283                 flkeys = &_flkeys;
2284         } else {
2285                 fl4.flowi4_proto = 0;
2286                 fl4.fl4_sport = 0;
2287                 fl4.fl4_dport = 0;
2288         }
2289
2290         err = fib_lookup(net, &fl4, res, 0);
2291         if (err != 0) {
2292                 if (!IN_DEV_FORWARD(in_dev))
2293                         err = -EHOSTUNREACH;
2294                 goto no_route;
2295         }
2296
2297         if (res->type == RTN_BROADCAST) {
2298                 if (IN_DEV_BFORWARD(in_dev))
2299                         goto make_route;
2300                 /* not do cache if bc_forwarding is enabled */
2301                 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2302                         do_cache = false;
2303                 goto brd_input;
2304         }
2305
2306         if (res->type == RTN_LOCAL) {
2307                 err = fib_validate_source(skb, saddr, daddr, tos,
2308                                           0, dev, in_dev, &itag);
2309                 if (err < 0)
2310                         goto martian_source;
2311                 goto local_input;
2312         }
2313
2314         if (!IN_DEV_FORWARD(in_dev)) {
2315                 err = -EHOSTUNREACH;
2316                 goto no_route;
2317         }
2318         if (res->type != RTN_UNICAST)
2319                 goto martian_destination;
2320
2321 make_route:
2322         err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2323 out:    return err;
2324
2325 brd_input:
2326         if (skb->protocol != htons(ETH_P_IP))
2327                 goto e_inval;
2328
2329         if (!ipv4_is_zeronet(saddr)) {
2330                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2331                                           in_dev, &itag);
2332                 if (err < 0)
2333                         goto martian_source;
2334         }
2335         flags |= RTCF_BROADCAST;
2336         res->type = RTN_BROADCAST;
2337         RT_CACHE_STAT_INC(in_brd);
2338
2339 local_input:
2340         do_cache &= res->fi && !itag;
2341         if (do_cache) {
2342                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2343
2344                 rth = rcu_dereference(nhc->nhc_rth_input);
2345                 if (rt_cache_valid(rth)) {
2346                         skb_dst_set_noref(skb, &rth->dst);
2347                         err = 0;
2348                         goto out;
2349                 }
2350         }
2351
2352         rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2353                            flags | RTCF_LOCAL, res->type,
2354                            IN_DEV_ORCONF(in_dev, NOPOLICY), false);
2355         if (!rth)
2356                 goto e_nobufs;
2357
2358         rth->dst.output= ip_rt_bug;
2359 #ifdef CONFIG_IP_ROUTE_CLASSID
2360         rth->dst.tclassid = itag;
2361 #endif
2362         rth->rt_is_input = 1;
2363
2364         RT_CACHE_STAT_INC(in_slow_tot);
2365         if (res->type == RTN_UNREACHABLE) {
2366                 rth->dst.input= ip_error;
2367                 rth->dst.error= -err;
2368                 rth->rt_flags   &= ~RTCF_LOCAL;
2369         }
2370
2371         if (do_cache) {
2372                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2373
2374                 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2375                 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2376                         WARN_ON(rth->dst.input == lwtunnel_input);
2377                         rth->dst.lwtstate->orig_input = rth->dst.input;
2378                         rth->dst.input = lwtunnel_input;
2379                 }
2380
2381                 if (unlikely(!rt_cache_route(nhc, rth)))
2382                         rt_add_uncached_list(rth);
2383         }
2384         skb_dst_set(skb, &rth->dst);
2385         err = 0;
2386         goto out;
2387
2388 no_route:
2389         RT_CACHE_STAT_INC(in_no_route);
2390         res->type = RTN_UNREACHABLE;
2391         res->fi = NULL;
2392         res->table = NULL;
2393         goto local_input;
2394
2395         /*
2396          *      Do not cache martian addresses: they should be logged (RFC1812)
2397          */
2398 martian_destination:
2399         RT_CACHE_STAT_INC(in_martian_dst);
2400 #ifdef CONFIG_IP_ROUTE_VERBOSE
2401         if (IN_DEV_LOG_MARTIANS(in_dev))
2402                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2403                                      &daddr, &saddr, dev->name);
2404 #endif
2405
2406 e_inval:
2407         err = -EINVAL;
2408         goto out;
2409
2410 e_nobufs:
2411         err = -ENOBUFS;
2412         goto out;
2413
2414 martian_source:
2415         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2416         goto out;
2417 }
2418
2419 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2420                          u8 tos, struct net_device *dev)
2421 {
2422         struct fib_result res;
2423         int err;
2424
2425         tos &= IPTOS_RT_MASK;
2426         rcu_read_lock();
2427         err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2428         rcu_read_unlock();
2429
2430         return err;
2431 }
2432 EXPORT_SYMBOL(ip_route_input_noref);
2433
2434 /* called with rcu_read_lock held */
2435 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2436                        u8 tos, struct net_device *dev, struct fib_result *res)
2437 {
2438         /* Multicast recognition logic is moved from route cache to here.
2439          * The problem was that too many Ethernet cards have broken/missing
2440          * hardware multicast filters :-( As result the host on multicasting
2441          * network acquires a lot of useless route cache entries, sort of
2442          * SDR messages from all the world. Now we try to get rid of them.
2443          * Really, provided software IP multicast filter is organized
2444          * reasonably (at least, hashed), it does not result in a slowdown
2445          * comparing with route cache reject entries.
2446          * Note, that multicast routers are not affected, because
2447          * route cache entry is created eventually.
2448          */
2449         if (ipv4_is_multicast(daddr)) {
2450                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2451                 int our = 0;
2452                 int err = -EINVAL;
2453
2454                 if (!in_dev)
2455                         return err;
2456                 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2457                                       ip_hdr(skb)->protocol);
2458
2459                 /* check l3 master if no match yet */
2460                 if (!our && netif_is_l3_slave(dev)) {
2461                         struct in_device *l3_in_dev;
2462
2463                         l3_in_dev = __in_dev_get_rcu(skb->dev);
2464                         if (l3_in_dev)
2465                                 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2466                                                       ip_hdr(skb)->protocol);
2467                 }
2468
2469                 if (our
2470 #ifdef CONFIG_IP_MROUTE
2471                         ||
2472                     (!ipv4_is_local_multicast(daddr) &&
2473                      IN_DEV_MFORWARD(in_dev))
2474 #endif
2475                    ) {
2476                         err = ip_route_input_mc(skb, daddr, saddr,
2477                                                 tos, dev, our);
2478                 }
2479                 return err;
2480         }
2481
2482         return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2483 }
2484
2485 /* called with rcu_read_lock() */
2486 static struct rtable *__mkroute_output(const struct fib_result *res,
2487                                        const struct flowi4 *fl4, int orig_oif,
2488                                        struct net_device *dev_out,
2489                                        unsigned int flags)
2490 {
2491         struct fib_info *fi = res->fi;
2492         struct fib_nh_exception *fnhe;
2493         struct in_device *in_dev;
2494         u16 type = res->type;
2495         struct rtable *rth;
2496         bool do_cache;
2497
2498         in_dev = __in_dev_get_rcu(dev_out);
2499         if (!in_dev)
2500                 return ERR_PTR(-EINVAL);
2501
2502         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2503                 if (ipv4_is_loopback(fl4->saddr) &&
2504                     !(dev_out->flags & IFF_LOOPBACK) &&
2505                     !netif_is_l3_master(dev_out))
2506                         return ERR_PTR(-EINVAL);
2507
2508         if (ipv4_is_lbcast(fl4->daddr))
2509                 type = RTN_BROADCAST;
2510         else if (ipv4_is_multicast(fl4->daddr))
2511                 type = RTN_MULTICAST;
2512         else if (ipv4_is_zeronet(fl4->daddr))
2513                 return ERR_PTR(-EINVAL);
2514
2515         if (dev_out->flags & IFF_LOOPBACK)
2516                 flags |= RTCF_LOCAL;
2517
2518         do_cache = true;
2519         if (type == RTN_BROADCAST) {
2520                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2521                 fi = NULL;
2522         } else if (type == RTN_MULTICAST) {
2523                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2524                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2525                                      fl4->flowi4_proto))
2526                         flags &= ~RTCF_LOCAL;
2527                 else
2528                         do_cache = false;
2529                 /* If multicast route do not exist use
2530                  * default one, but do not gateway in this case.
2531                  * Yes, it is hack.
2532                  */
2533                 if (fi && res->prefixlen < 4)
2534                         fi = NULL;
2535         } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2536                    (orig_oif != dev_out->ifindex)) {
2537                 /* For local routes that require a particular output interface
2538                  * we do not want to cache the result.  Caching the result
2539                  * causes incorrect behaviour when there are multiple source
2540                  * addresses on the interface, the end result being that if the
2541                  * intended recipient is waiting on that interface for the
2542                  * packet he won't receive it because it will be delivered on
2543                  * the loopback interface and the IP_PKTINFO ipi_ifindex will
2544                  * be set to the loopback interface as well.
2545                  */
2546                 do_cache = false;
2547         }
2548
2549         fnhe = NULL;
2550         do_cache &= fi != NULL;
2551         if (fi) {
2552                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2553                 struct rtable __rcu **prth;
2554
2555                 fnhe = find_exception(nhc, fl4->daddr);
2556                 if (!do_cache)
2557                         goto add;
2558                 if (fnhe) {
2559                         prth = &fnhe->fnhe_rth_output;
2560                 } else {
2561                         if (unlikely(fl4->flowi4_flags &
2562                                      FLOWI_FLAG_KNOWN_NH &&
2563                                      !(nhc->nhc_gw_family &&
2564                                        nhc->nhc_scope == RT_SCOPE_LINK))) {
2565                                 do_cache = false;
2566                                 goto add;
2567                         }
2568                         prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2569                 }
2570                 rth = rcu_dereference(*prth);
2571                 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2572                         return rth;
2573         }
2574
2575 add:
2576         rth = rt_dst_alloc(dev_out, flags, type,
2577                            IN_DEV_ORCONF(in_dev, NOPOLICY),
2578                            IN_DEV_ORCONF(in_dev, NOXFRM));
2579         if (!rth)
2580                 return ERR_PTR(-ENOBUFS);
2581
2582         rth->rt_iif = orig_oif;
2583
2584         RT_CACHE_STAT_INC(out_slow_tot);
2585
2586         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2587                 if (flags & RTCF_LOCAL &&
2588                     !(dev_out->flags & IFF_LOOPBACK)) {
2589                         rth->dst.output = ip_mc_output;
2590                         RT_CACHE_STAT_INC(out_slow_mc);
2591                 }
2592 #ifdef CONFIG_IP_MROUTE
2593                 if (type == RTN_MULTICAST) {
2594                         if (IN_DEV_MFORWARD(in_dev) &&
2595                             !ipv4_is_local_multicast(fl4->daddr)) {
2596                                 rth->dst.input = ip_mr_input;
2597                                 rth->dst.output = ip_mc_output;
2598                         }
2599                 }
2600 #endif
2601         }
2602
2603         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2604         lwtunnel_set_redirect(&rth->dst);
2605
2606         return rth;
2607 }
2608
2609 /*
2610  * Major route resolver routine.
2611  */
2612
2613 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2614                                         const struct sk_buff *skb)
2615 {
2616         __u8 tos = RT_FL_TOS(fl4);
2617         struct fib_result res = {
2618                 .type           = RTN_UNSPEC,
2619                 .fi             = NULL,
2620                 .table          = NULL,
2621                 .tclassid       = 0,
2622         };
2623         struct rtable *rth;
2624
2625         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2626         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2627         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2628                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2629
2630         rcu_read_lock();
2631         rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2632         rcu_read_unlock();
2633
2634         return rth;
2635 }
2636 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2637
2638 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2639                                             struct fib_result *res,
2640                                             const struct sk_buff *skb)
2641 {
2642         struct net_device *dev_out = NULL;
2643         int orig_oif = fl4->flowi4_oif;
2644         unsigned int flags = 0;
2645         struct rtable *rth;
2646         int err;
2647
2648         if (fl4->saddr) {
2649                 if (ipv4_is_multicast(fl4->saddr) ||
2650                     ipv4_is_lbcast(fl4->saddr) ||
2651                     ipv4_is_zeronet(fl4->saddr)) {
2652                         rth = ERR_PTR(-EINVAL);
2653                         goto out;
2654                 }
2655
2656                 rth = ERR_PTR(-ENETUNREACH);
2657
2658                 /* I removed check for oif == dev_out->oif here.
2659                  * It was wrong for two reasons:
2660                  * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2661                  *    is assigned to multiple interfaces.
2662                  * 2. Moreover, we are allowed to send packets with saddr
2663                  *    of another iface. --ANK
2664                  */
2665
2666                 if (fl4->flowi4_oif == 0 &&
2667                     (ipv4_is_multicast(fl4->daddr) ||
2668                      ipv4_is_lbcast(fl4->daddr))) {
2669                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2670                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2671                         if (!dev_out)
2672                                 goto out;
2673
2674                         /* Special hack: user can direct multicasts
2675                          * and limited broadcast via necessary interface
2676                          * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2677                          * This hack is not just for fun, it allows
2678                          * vic,vat and friends to work.
2679                          * They bind socket to loopback, set ttl to zero
2680                          * and expect that it will work.
2681                          * From the viewpoint of routing cache they are broken,
2682                          * because we are not allowed to build multicast path
2683                          * with loopback source addr (look, routing cache
2684                          * cannot know, that ttl is zero, so that packet
2685                          * will not leave this host and route is valid).
2686                          * Luckily, this hack is good workaround.
2687                          */
2688
2689                         fl4->flowi4_oif = dev_out->ifindex;
2690                         goto make_route;
2691                 }
2692
2693                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2694                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2695                         if (!__ip_dev_find(net, fl4->saddr, false))
2696                                 goto out;
2697                 }
2698         }
2699
2700
2701         if (fl4->flowi4_oif) {
2702                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2703                 rth = ERR_PTR(-ENODEV);
2704                 if (!dev_out)
2705                         goto out;
2706
2707                 /* RACE: Check return value of inet_select_addr instead. */
2708                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2709                         rth = ERR_PTR(-ENETUNREACH);
2710                         goto out;
2711                 }
2712                 if (ipv4_is_local_multicast(fl4->daddr) ||
2713                     ipv4_is_lbcast(fl4->daddr) ||
2714                     fl4->flowi4_proto == IPPROTO_IGMP) {
2715                         if (!fl4->saddr)
2716                                 fl4->saddr = inet_select_addr(dev_out, 0,
2717                                                               RT_SCOPE_LINK);
2718                         goto make_route;
2719                 }
2720                 if (!fl4->saddr) {
2721                         if (ipv4_is_multicast(fl4->daddr))
2722                                 fl4->saddr = inet_select_addr(dev_out, 0,
2723                                                               fl4->flowi4_scope);
2724                         else if (!fl4->daddr)
2725                                 fl4->saddr = inet_select_addr(dev_out, 0,
2726                                                               RT_SCOPE_HOST);
2727                 }
2728         }
2729
2730         if (!fl4->daddr) {
2731                 fl4->daddr = fl4->saddr;
2732                 if (!fl4->daddr)
2733                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2734                 dev_out = net->loopback_dev;
2735                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2736                 res->type = RTN_LOCAL;
2737                 flags |= RTCF_LOCAL;
2738                 goto make_route;
2739         }
2740
2741         err = fib_lookup(net, fl4, res, 0);
2742         if (err) {
2743                 res->fi = NULL;
2744                 res->table = NULL;
2745                 if (fl4->flowi4_oif &&
2746                     (ipv4_is_multicast(fl4->daddr) ||
2747                     !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2748                         /* Apparently, routing tables are wrong. Assume,
2749                          * that the destination is on link.
2750                          *
2751                          * WHY? DW.
2752                          * Because we are allowed to send to iface
2753                          * even if it has NO routes and NO assigned
2754                          * addresses. When oif is specified, routing
2755                          * tables are looked up with only one purpose:
2756                          * to catch if destination is gatewayed, rather than
2757                          * direct. Moreover, if MSG_DONTROUTE is set,
2758                          * we send packet, ignoring both routing tables
2759                          * and ifaddr state. --ANK
2760                          *
2761                          *
2762                          * We could make it even if oif is unknown,
2763                          * likely IPv6, but we do not.
2764                          */
2765
2766                         if (fl4->saddr == 0)
2767                                 fl4->saddr = inet_select_addr(dev_out, 0,
2768                                                               RT_SCOPE_LINK);
2769                         res->type = RTN_UNICAST;
2770                         goto make_route;
2771                 }
2772                 rth = ERR_PTR(err);
2773                 goto out;
2774         }
2775
2776         if (res->type == RTN_LOCAL) {
2777                 if (!fl4->saddr) {
2778                         if (res->fi->fib_prefsrc)
2779                                 fl4->saddr = res->fi->fib_prefsrc;
2780                         else
2781                                 fl4->saddr = fl4->daddr;
2782                 }
2783
2784                 /* L3 master device is the loopback for that domain */
2785                 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2786                         net->loopback_dev;
2787
2788                 /* make sure orig_oif points to fib result device even
2789                  * though packet rx/tx happens over loopback or l3mdev
2790                  */
2791                 orig_oif = FIB_RES_OIF(*res);
2792
2793                 fl4->flowi4_oif = dev_out->ifindex;
2794                 flags |= RTCF_LOCAL;
2795                 goto make_route;
2796         }
2797
2798         fib_select_path(net, res, fl4, skb);
2799
2800         dev_out = FIB_RES_DEV(*res);
2801
2802 make_route:
2803         rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2804
2805 out:
2806         return rth;
2807 }
2808
2809 static struct dst_ops ipv4_dst_blackhole_ops = {
2810         .family                 = AF_INET,
2811         .default_advmss         = ipv4_default_advmss,
2812         .neigh_lookup           = ipv4_neigh_lookup,
2813         .check                  = dst_blackhole_check,
2814         .cow_metrics            = dst_blackhole_cow_metrics,
2815         .update_pmtu            = dst_blackhole_update_pmtu,
2816         .redirect               = dst_blackhole_redirect,
2817         .mtu                    = dst_blackhole_mtu,
2818 };
2819
2820 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2821 {
2822         struct rtable *ort = (struct rtable *) dst_orig;
2823         struct rtable *rt;
2824
2825         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2826         if (rt) {
2827                 struct dst_entry *new = &rt->dst;
2828
2829                 new->__use = 1;
2830                 new->input = dst_discard;
2831                 new->output = dst_discard_out;
2832
2833                 new->dev = net->loopback_dev;
2834                 if (new->dev)
2835                         dev_hold(new->dev);
2836
2837                 rt->rt_is_input = ort->rt_is_input;
2838                 rt->rt_iif = ort->rt_iif;
2839                 rt->rt_pmtu = ort->rt_pmtu;
2840                 rt->rt_mtu_locked = ort->rt_mtu_locked;
2841
2842                 rt->rt_genid = rt_genid_ipv4(net);
2843                 rt->rt_flags = ort->rt_flags;
2844                 rt->rt_type = ort->rt_type;
2845                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2846                 rt->rt_gw_family = ort->rt_gw_family;
2847                 if (rt->rt_gw_family == AF_INET)
2848                         rt->rt_gw4 = ort->rt_gw4;
2849                 else if (rt->rt_gw_family == AF_INET6)
2850                         rt->rt_gw6 = ort->rt_gw6;
2851
2852                 INIT_LIST_HEAD(&rt->rt_uncached);
2853         }
2854
2855         dst_release(dst_orig);
2856
2857         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2858 }
2859
2860 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2861                                     const struct sock *sk)
2862 {
2863         struct rtable *rt = __ip_route_output_key(net, flp4);
2864
2865         if (IS_ERR(rt))
2866                 return rt;
2867
2868         if (flp4->flowi4_proto) {
2869                 flp4->flowi4_oif = rt->dst.dev->ifindex;
2870                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2871                                                         flowi4_to_flowi(flp4),
2872                                                         sk, 0);
2873         }
2874
2875         return rt;
2876 }
2877 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2878
2879 struct rtable *ip_route_output_tunnel(struct sk_buff *skb,
2880                                       struct net_device *dev,
2881                                       struct net *net, __be32 *saddr,
2882                                       const struct ip_tunnel_info *info,
2883                                       u8 protocol, bool use_cache)
2884 {
2885 #ifdef CONFIG_DST_CACHE
2886         struct dst_cache *dst_cache;
2887 #endif
2888         struct rtable *rt = NULL;
2889         struct flowi4 fl4;
2890         __u8 tos;
2891
2892 #ifdef CONFIG_DST_CACHE
2893         dst_cache = (struct dst_cache *)&info->dst_cache;
2894         if (use_cache) {
2895                 rt = dst_cache_get_ip4(dst_cache, saddr);
2896                 if (rt)
2897                         return rt;
2898         }
2899 #endif
2900         memset(&fl4, 0, sizeof(fl4));
2901         fl4.flowi4_mark = skb->mark;
2902         fl4.flowi4_proto = protocol;
2903         fl4.daddr = info->key.u.ipv4.dst;
2904         fl4.saddr = info->key.u.ipv4.src;
2905         tos = info->key.tos;
2906         fl4.flowi4_tos = RT_TOS(tos);
2907
2908         rt = ip_route_output_key(net, &fl4);
2909         if (IS_ERR(rt)) {
2910                 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
2911                 return ERR_PTR(-ENETUNREACH);
2912         }
2913         if (rt->dst.dev == dev) { /* is this necessary? */
2914                 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
2915                 ip_rt_put(rt);
2916                 return ERR_PTR(-ELOOP);
2917         }
2918 #ifdef CONFIG_DST_CACHE
2919         if (use_cache)
2920                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2921 #endif
2922         *saddr = fl4.saddr;
2923         return rt;
2924 }
2925 EXPORT_SYMBOL_GPL(ip_route_output_tunnel);
2926
2927 /* called with rcu_read_lock held */
2928 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2929                         struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2930                         struct sk_buff *skb, u32 portid, u32 seq,
2931                         unsigned int flags)
2932 {
2933         struct rtmsg *r;
2934         struct nlmsghdr *nlh;
2935         unsigned long expires = 0;
2936         u32 error;
2937         u32 metrics[RTAX_MAX];
2938
2939         nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2940         if (!nlh)
2941                 return -EMSGSIZE;
2942
2943         r = nlmsg_data(nlh);
2944         r->rtm_family    = AF_INET;
2945         r->rtm_dst_len  = 32;
2946         r->rtm_src_len  = 0;
2947         r->rtm_tos      = fl4 ? fl4->flowi4_tos : 0;
2948         r->rtm_table    = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2949         if (nla_put_u32(skb, RTA_TABLE, table_id))
2950                 goto nla_put_failure;
2951         r->rtm_type     = rt->rt_type;
2952         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2953         r->rtm_protocol = RTPROT_UNSPEC;
2954         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2955         if (rt->rt_flags & RTCF_NOTIFY)
2956                 r->rtm_flags |= RTM_F_NOTIFY;
2957         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2958                 r->rtm_flags |= RTCF_DOREDIRECT;
2959
2960         if (nla_put_in_addr(skb, RTA_DST, dst))
2961                 goto nla_put_failure;
2962         if (src) {
2963                 r->rtm_src_len = 32;
2964                 if (nla_put_in_addr(skb, RTA_SRC, src))
2965                         goto nla_put_failure;
2966         }
2967         if (rt->dst.dev &&
2968             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2969                 goto nla_put_failure;
2970         if (rt->dst.lwtstate &&
2971             lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2972                 goto nla_put_failure;
2973 #ifdef CONFIG_IP_ROUTE_CLASSID
2974         if (rt->dst.tclassid &&
2975             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2976                 goto nla_put_failure;
2977 #endif
2978         if (fl4 && !rt_is_input_route(rt) &&
2979             fl4->saddr != src) {
2980                 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2981                         goto nla_put_failure;
2982         }
2983         if (rt->rt_uses_gateway) {
2984                 if (rt->rt_gw_family == AF_INET &&
2985                     nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2986                         goto nla_put_failure;
2987                 } else if (rt->rt_gw_family == AF_INET6) {
2988                         int alen = sizeof(struct in6_addr);
2989                         struct nlattr *nla;
2990                         struct rtvia *via;
2991
2992                         nla = nla_reserve(skb, RTA_VIA, alen + 2);
2993                         if (!nla)
2994                                 goto nla_put_failure;
2995
2996                         via = nla_data(nla);
2997                         via->rtvia_family = AF_INET6;
2998                         memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2999                 }
3000         }
3001
3002         expires = rt->dst.expires;
3003         if (expires) {
3004                 unsigned long now = jiffies;
3005
3006                 if (time_before(now, expires))
3007                         expires -= now;
3008                 else
3009                         expires = 0;
3010         }
3011
3012         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3013         if (rt->rt_pmtu && expires)
3014                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3015         if (rt->rt_mtu_locked && expires)
3016                 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3017         if (rtnetlink_put_metrics(skb, metrics) < 0)
3018                 goto nla_put_failure;
3019
3020         if (fl4) {
3021                 if (fl4->flowi4_mark &&
3022                     nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3023                         goto nla_put_failure;
3024
3025                 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3026                     nla_put_u32(skb, RTA_UID,
3027                                 from_kuid_munged(current_user_ns(),
3028                                                  fl4->flowi4_uid)))
3029                         goto nla_put_failure;
3030
3031                 if (rt_is_input_route(rt)) {
3032 #ifdef CONFIG_IP_MROUTE
3033                         if (ipv4_is_multicast(dst) &&
3034                             !ipv4_is_local_multicast(dst) &&
3035                             IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
3036                                 int err = ipmr_get_route(net, skb,
3037                                                          fl4->saddr, fl4->daddr,
3038                                                          r, portid);
3039
3040                                 if (err <= 0) {
3041                                         if (err == 0)
3042                                                 return 0;
3043                                         goto nla_put_failure;
3044                                 }
3045                         } else
3046 #endif
3047                                 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3048                                         goto nla_put_failure;
3049                 }
3050         }
3051
3052         error = rt->dst.error;
3053
3054         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3055                 goto nla_put_failure;
3056
3057         nlmsg_end(skb, nlh);
3058         return 0;
3059
3060 nla_put_failure:
3061         nlmsg_cancel(skb, nlh);
3062         return -EMSGSIZE;
3063 }
3064
3065 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3066                             struct netlink_callback *cb, u32 table_id,
3067                             struct fnhe_hash_bucket *bucket, int genid,
3068                             int *fa_index, int fa_start, unsigned int flags)
3069 {
3070         int i;
3071
3072         for (i = 0; i < FNHE_HASH_SIZE; i++) {
3073                 struct fib_nh_exception *fnhe;
3074
3075                 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3076                      fnhe = rcu_dereference(fnhe->fnhe_next)) {
3077                         struct rtable *rt;
3078                         int err;
3079
3080                         if (*fa_index < fa_start)
3081                                 goto next;
3082
3083                         if (fnhe->fnhe_genid != genid)
3084                                 goto next;
3085
3086                         if (fnhe->fnhe_expires &&
3087                             time_after(jiffies, fnhe->fnhe_expires))
3088                                 goto next;
3089
3090                         rt = rcu_dereference(fnhe->fnhe_rth_input);
3091                         if (!rt)
3092                                 rt = rcu_dereference(fnhe->fnhe_rth_output);
3093                         if (!rt)
3094                                 goto next;
3095
3096                         err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3097                                            table_id, NULL, skb,
3098                                            NETLINK_CB(cb->skb).portid,
3099                                            cb->nlh->nlmsg_seq, flags);
3100                         if (err)
3101                                 return err;
3102 next:
3103                         (*fa_index)++;
3104                 }
3105         }
3106
3107         return 0;
3108 }
3109
3110 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3111                        u32 table_id, struct fib_info *fi,
3112                        int *fa_index, int fa_start, unsigned int flags)
3113 {
3114         struct net *net = sock_net(cb->skb->sk);
3115         int nhsel, genid = fnhe_genid(net);
3116
3117         for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3118                 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3119                 struct fnhe_hash_bucket *bucket;
3120                 int err;
3121
3122                 if (nhc->nhc_flags & RTNH_F_DEAD)
3123                         continue;
3124
3125                 rcu_read_lock();
3126                 bucket = rcu_dereference(nhc->nhc_exceptions);
3127                 err = 0;
3128                 if (bucket)
3129                         err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3130                                                genid, fa_index, fa_start,
3131                                                flags);
3132                 rcu_read_unlock();
3133                 if (err)
3134                         return err;
3135         }
3136
3137         return 0;
3138 }
3139
3140 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3141                                                    u8 ip_proto, __be16 sport,
3142                                                    __be16 dport)
3143 {
3144         struct sk_buff *skb;
3145         struct iphdr *iph;
3146
3147         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3148         if (!skb)
3149                 return NULL;
3150
3151         /* Reserve room for dummy headers, this skb can pass
3152          * through good chunk of routing engine.
3153          */
3154         skb_reset_mac_header(skb);
3155         skb_reset_network_header(skb);
3156         skb->protocol = htons(ETH_P_IP);
3157         iph = skb_put(skb, sizeof(struct iphdr));
3158         iph->protocol = ip_proto;
3159         iph->saddr = src;
3160         iph->daddr = dst;
3161         iph->version = 0x4;
3162         iph->frag_off = 0;
3163         iph->ihl = 0x5;
3164         skb_set_transport_header(skb, skb->len);
3165
3166         switch (iph->protocol) {
3167         case IPPROTO_UDP: {
3168                 struct udphdr *udph;
3169
3170                 udph = skb_put_zero(skb, sizeof(struct udphdr));
3171                 udph->source = sport;
3172                 udph->dest = dport;
3173                 udph->len = sizeof(struct udphdr);
3174                 udph->check = 0;
3175                 break;
3176         }
3177         case IPPROTO_TCP: {
3178                 struct tcphdr *tcph;
3179
3180                 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3181                 tcph->source    = sport;
3182                 tcph->dest      = dport;
3183                 tcph->doff      = sizeof(struct tcphdr) / 4;
3184                 tcph->rst = 1;
3185                 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3186                                             src, dst, 0);
3187                 break;
3188         }
3189         case IPPROTO_ICMP: {
3190                 struct icmphdr *icmph;
3191
3192                 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3193                 icmph->type = ICMP_ECHO;
3194                 icmph->code = 0;
3195         }
3196         }
3197
3198         return skb;
3199 }
3200
3201 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3202                                        const struct nlmsghdr *nlh,
3203                                        struct nlattr **tb,
3204                                        struct netlink_ext_ack *extack)
3205 {
3206         struct rtmsg *rtm;
3207         int i, err;
3208
3209         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3210                 NL_SET_ERR_MSG(extack,
3211                                "ipv4: Invalid header for route get request");
3212                 return -EINVAL;
3213         }
3214
3215         if (!netlink_strict_get_check(skb))
3216                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3217                                               rtm_ipv4_policy, extack);
3218
3219         rtm = nlmsg_data(nlh);
3220         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3221             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3222             rtm->rtm_table || rtm->rtm_protocol ||
3223             rtm->rtm_scope || rtm->rtm_type) {
3224                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3225                 return -EINVAL;
3226         }
3227
3228         if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3229                                RTM_F_LOOKUP_TABLE |
3230                                RTM_F_FIB_MATCH)) {
3231                 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3232                 return -EINVAL;
3233         }
3234
3235         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3236                                             rtm_ipv4_policy, extack);
3237         if (err)
3238                 return err;
3239
3240         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3241             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3242                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3243                 return -EINVAL;
3244         }
3245
3246         for (i = 0; i <= RTA_MAX; i++) {
3247                 if (!tb[i])
3248                         continue;
3249
3250                 switch (i) {
3251                 case RTA_IIF:
3252                 case RTA_OIF:
3253                 case RTA_SRC:
3254                 case RTA_DST:
3255                 case RTA_IP_PROTO:
3256                 case RTA_SPORT:
3257                 case RTA_DPORT:
3258                 case RTA_MARK:
3259                 case RTA_UID:
3260                         break;
3261                 default:
3262                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3263                         return -EINVAL;
3264                 }
3265         }
3266
3267         return 0;
3268 }
3269
3270 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3271                              struct netlink_ext_ack *extack)
3272 {
3273         struct net *net = sock_net(in_skb->sk);
3274         struct nlattr *tb[RTA_MAX+1];
3275         u32 table_id = RT_TABLE_MAIN;
3276         __be16 sport = 0, dport = 0;
3277         struct fib_result res = {};
3278         u8 ip_proto = IPPROTO_UDP;
3279         struct rtable *rt = NULL;
3280         struct sk_buff *skb;
3281         struct rtmsg *rtm;
3282         struct flowi4 fl4 = {};
3283         __be32 dst = 0;
3284         __be32 src = 0;
3285         kuid_t uid;
3286         u32 iif;
3287         int err;
3288         int mark;
3289
3290         err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3291         if (err < 0)
3292                 return err;
3293
3294         rtm = nlmsg_data(nlh);
3295         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3296         dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3297         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3298         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3299         if (tb[RTA_UID])
3300                 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3301         else
3302                 uid = (iif ? INVALID_UID : current_uid());
3303
3304         if (tb[RTA_IP_PROTO]) {
3305                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3306                                                   &ip_proto, AF_INET, extack);
3307                 if (err)
3308                         return err;
3309         }
3310
3311         if (tb[RTA_SPORT])
3312                 sport = nla_get_be16(tb[RTA_SPORT]);
3313
3314         if (tb[RTA_DPORT])
3315                 dport = nla_get_be16(tb[RTA_DPORT]);
3316
3317         skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3318         if (!skb)
3319                 return -ENOBUFS;
3320
3321         fl4.daddr = dst;
3322         fl4.saddr = src;
3323         fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3324         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3325         fl4.flowi4_mark = mark;
3326         fl4.flowi4_uid = uid;
3327         if (sport)
3328                 fl4.fl4_sport = sport;
3329         if (dport)
3330                 fl4.fl4_dport = dport;
3331         fl4.flowi4_proto = ip_proto;
3332
3333         rcu_read_lock();
3334
3335         if (iif) {
3336                 struct net_device *dev;
3337
3338                 dev = dev_get_by_index_rcu(net, iif);
3339                 if (!dev) {
3340                         err = -ENODEV;
3341                         goto errout_rcu;
3342                 }
3343
3344                 fl4.flowi4_iif = iif; /* for rt_fill_info */
3345                 skb->dev        = dev;
3346                 skb->mark       = mark;
3347                 err = ip_route_input_rcu(skb, dst, src,
3348                                          rtm->rtm_tos & IPTOS_RT_MASK, dev,
3349                                          &res);
3350
3351                 rt = skb_rtable(skb);
3352                 if (err == 0 && rt->dst.error)
3353                         err = -rt->dst.error;
3354         } else {
3355                 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3356                 skb->dev = net->loopback_dev;
3357                 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3358                 err = 0;
3359                 if (IS_ERR(rt))
3360                         err = PTR_ERR(rt);
3361                 else
3362                         skb_dst_set(skb, &rt->dst);
3363         }
3364
3365         if (err)
3366                 goto errout_rcu;
3367
3368         if (rtm->rtm_flags & RTM_F_NOTIFY)
3369                 rt->rt_flags |= RTCF_NOTIFY;
3370
3371         if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3372                 table_id = res.table ? res.table->tb_id : 0;
3373
3374         /* reset skb for netlink reply msg */
3375         skb_trim(skb, 0);
3376         skb_reset_network_header(skb);
3377         skb_reset_transport_header(skb);
3378         skb_reset_mac_header(skb);
3379
3380         if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3381                 struct fib_rt_info fri;
3382
3383                 if (!res.fi) {
3384                         err = fib_props[res.type].error;
3385                         if (!err)
3386                                 err = -EHOSTUNREACH;
3387                         goto errout_rcu;
3388                 }
3389                 fri.fi = res.fi;
3390                 fri.tb_id = table_id;
3391                 fri.dst = res.prefix;
3392                 fri.dst_len = res.prefixlen;
3393                 fri.tos = fl4.flowi4_tos;
3394                 fri.type = rt->rt_type;
3395                 fri.offload = 0;
3396                 fri.trap = 0;
3397                 fri.offload_failed = 0;
3398                 if (res.fa_head) {
3399                         struct fib_alias *fa;
3400
3401                         hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3402                                 u8 slen = 32 - fri.dst_len;
3403
3404                                 if (fa->fa_slen == slen &&
3405                                     fa->tb_id == fri.tb_id &&
3406                                     fa->fa_tos == fri.tos &&
3407                                     fa->fa_info == res.fi &&
3408                                     fa->fa_type == fri.type) {
3409                                         fri.offload = fa->offload;
3410                                         fri.trap = fa->trap;
3411                                         break;
3412                                 }
3413                         }
3414                 }
3415                 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3416                                     nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3417         } else {
3418                 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3419                                    NETLINK_CB(in_skb).portid,
3420                                    nlh->nlmsg_seq, 0);
3421         }
3422         if (err < 0)
3423                 goto errout_rcu;
3424
3425         rcu_read_unlock();
3426
3427         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3428
3429 errout_free:
3430         return err;
3431 errout_rcu:
3432         rcu_read_unlock();
3433         kfree_skb(skb);
3434         goto errout_free;
3435 }
3436
3437 void ip_rt_multicast_event(struct in_device *in_dev)
3438 {
3439         rt_cache_flush(dev_net(in_dev->dev));
3440 }
3441
3442 #ifdef CONFIG_SYSCTL
3443 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3444 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
3445 static int ip_rt_gc_elasticity __read_mostly    = 8;
3446 static int ip_min_valid_pmtu __read_mostly      = IPV4_MIN_MTU;
3447
3448 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3449                 void *buffer, size_t *lenp, loff_t *ppos)
3450 {
3451         struct net *net = (struct net *)__ctl->extra1;
3452
3453         if (write) {
3454                 rt_cache_flush(net);
3455                 fnhe_genid_bump(net);
3456                 return 0;
3457         }
3458
3459         return -EINVAL;
3460 }
3461
3462 static struct ctl_table ipv4_route_table[] = {
3463         {
3464                 .procname       = "gc_thresh",
3465                 .data           = &ipv4_dst_ops.gc_thresh,
3466                 .maxlen         = sizeof(int),
3467                 .mode           = 0644,
3468                 .proc_handler   = proc_dointvec,
3469         },
3470         {
3471                 .procname       = "max_size",
3472                 .data           = &ip_rt_max_size,
3473                 .maxlen         = sizeof(int),
3474                 .mode           = 0644,
3475                 .proc_handler   = proc_dointvec,
3476         },
3477         {
3478                 /*  Deprecated. Use gc_min_interval_ms */
3479
3480                 .procname       = "gc_min_interval",
3481                 .data           = &ip_rt_gc_min_interval,
3482                 .maxlen         = sizeof(int),
3483                 .mode           = 0644,
3484                 .proc_handler   = proc_dointvec_jiffies,
3485         },
3486         {
3487                 .procname       = "gc_min_interval_ms",
3488                 .data           = &ip_rt_gc_min_interval,
3489                 .maxlen         = sizeof(int),
3490                 .mode           = 0644,
3491                 .proc_handler   = proc_dointvec_ms_jiffies,
3492         },
3493         {
3494                 .procname       = "gc_timeout",
3495                 .data           = &ip_rt_gc_timeout,
3496                 .maxlen         = sizeof(int),
3497                 .mode           = 0644,
3498                 .proc_handler   = proc_dointvec_jiffies,
3499         },
3500         {
3501                 .procname       = "gc_interval",
3502                 .data           = &ip_rt_gc_interval,
3503                 .maxlen         = sizeof(int),
3504                 .mode           = 0644,
3505                 .proc_handler   = proc_dointvec_jiffies,
3506         },
3507         {
3508                 .procname       = "redirect_load",
3509                 .data           = &ip_rt_redirect_load,
3510                 .maxlen         = sizeof(int),
3511                 .mode           = 0644,
3512                 .proc_handler   = proc_dointvec,
3513         },
3514         {
3515                 .procname       = "redirect_number",
3516                 .data           = &ip_rt_redirect_number,
3517                 .maxlen         = sizeof(int),
3518                 .mode           = 0644,
3519                 .proc_handler   = proc_dointvec,
3520         },
3521         {
3522                 .procname       = "redirect_silence",
3523                 .data           = &ip_rt_redirect_silence,
3524                 .maxlen         = sizeof(int),
3525                 .mode           = 0644,
3526                 .proc_handler   = proc_dointvec,
3527         },
3528         {
3529                 .procname       = "error_cost",
3530                 .data           = &ip_rt_error_cost,
3531                 .maxlen         = sizeof(int),
3532                 .mode           = 0644,
3533                 .proc_handler   = proc_dointvec,
3534         },
3535         {
3536                 .procname       = "error_burst",
3537                 .data           = &ip_rt_error_burst,
3538                 .maxlen         = sizeof(int),
3539                 .mode           = 0644,
3540                 .proc_handler   = proc_dointvec,
3541         },
3542         {
3543                 .procname       = "gc_elasticity",
3544                 .data           = &ip_rt_gc_elasticity,
3545                 .maxlen         = sizeof(int),
3546                 .mode           = 0644,
3547                 .proc_handler   = proc_dointvec,
3548         },
3549         {
3550                 .procname       = "mtu_expires",
3551                 .data           = &ip_rt_mtu_expires,
3552                 .maxlen         = sizeof(int),
3553                 .mode           = 0644,
3554                 .proc_handler   = proc_dointvec_jiffies,
3555         },
3556         {
3557                 .procname       = "min_pmtu",
3558                 .data           = &ip_rt_min_pmtu,
3559                 .maxlen         = sizeof(int),
3560                 .mode           = 0644,
3561                 .proc_handler   = proc_dointvec_minmax,
3562                 .extra1         = &ip_min_valid_pmtu,
3563         },
3564         {
3565                 .procname       = "min_adv_mss",
3566                 .data           = &ip_rt_min_advmss,
3567                 .maxlen         = sizeof(int),
3568                 .mode           = 0644,
3569                 .proc_handler   = proc_dointvec,
3570         },
3571         { }
3572 };
3573
3574 static const char ipv4_route_flush_procname[] = "flush";
3575
3576 static struct ctl_table ipv4_route_flush_table[] = {
3577         {
3578                 .procname       = ipv4_route_flush_procname,
3579                 .maxlen         = sizeof(int),
3580                 .mode           = 0200,
3581                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3582         },
3583         { },
3584 };
3585
3586 static __net_init int sysctl_route_net_init(struct net *net)
3587 {
3588         struct ctl_table *tbl;
3589
3590         tbl = ipv4_route_flush_table;
3591         if (!net_eq(net, &init_net)) {
3592                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3593                 if (!tbl)
3594                         goto err_dup;
3595
3596                 /* Don't export non-whitelisted sysctls to unprivileged users */
3597                 if (net->user_ns != &init_user_ns) {
3598                         if (tbl[0].procname != ipv4_route_flush_procname)
3599                                 tbl[0].procname = NULL;
3600                 }
3601         }
3602         tbl[0].extra1 = net;
3603
3604         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3605         if (!net->ipv4.route_hdr)
3606                 goto err_reg;
3607         return 0;
3608
3609 err_reg:
3610         if (tbl != ipv4_route_flush_table)
3611                 kfree(tbl);
3612 err_dup:
3613         return -ENOMEM;
3614 }
3615
3616 static __net_exit void sysctl_route_net_exit(struct net *net)
3617 {
3618         struct ctl_table *tbl;
3619
3620         tbl = net->ipv4.route_hdr->ctl_table_arg;
3621         unregister_net_sysctl_table(net->ipv4.route_hdr);
3622         BUG_ON(tbl == ipv4_route_flush_table);
3623         kfree(tbl);
3624 }
3625
3626 static __net_initdata struct pernet_operations sysctl_route_ops = {
3627         .init = sysctl_route_net_init,
3628         .exit = sysctl_route_net_exit,
3629 };
3630 #endif
3631
3632 static __net_init int rt_genid_init(struct net *net)
3633 {
3634         atomic_set(&net->ipv4.rt_genid, 0);
3635         atomic_set(&net->fnhe_genid, 0);
3636         atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3637         return 0;
3638 }
3639
3640 static __net_initdata struct pernet_operations rt_genid_ops = {
3641         .init = rt_genid_init,
3642 };
3643
3644 static int __net_init ipv4_inetpeer_init(struct net *net)
3645 {
3646         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3647
3648         if (!bp)
3649                 return -ENOMEM;
3650         inet_peer_base_init(bp);
3651         net->ipv4.peers = bp;
3652         return 0;
3653 }
3654
3655 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3656 {
3657         struct inet_peer_base *bp = net->ipv4.peers;
3658
3659         net->ipv4.peers = NULL;
3660         inetpeer_invalidate_tree(bp);
3661         kfree(bp);
3662 }
3663
3664 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3665         .init   =       ipv4_inetpeer_init,
3666         .exit   =       ipv4_inetpeer_exit,
3667 };
3668
3669 #ifdef CONFIG_IP_ROUTE_CLASSID
3670 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3671 #endif /* CONFIG_IP_ROUTE_CLASSID */
3672
3673 int __init ip_rt_init(void)
3674 {
3675         void *idents_hash;
3676         int cpu;
3677
3678         /* For modern hosts, this will use 2 MB of memory */
3679         idents_hash = alloc_large_system_hash("IP idents",
3680                                               sizeof(*ip_idents) + sizeof(*ip_tstamps),
3681                                               0,
3682                                               16, /* one bucket per 64 KB */
3683                                               HASH_ZERO,
3684                                               NULL,
3685                                               &ip_idents_mask,
3686                                               2048,
3687                                               256*1024);
3688
3689         ip_idents = idents_hash;
3690
3691         prandom_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3692
3693         ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3694
3695         for_each_possible_cpu(cpu) {
3696                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3697
3698                 INIT_LIST_HEAD(&ul->head);
3699                 spin_lock_init(&ul->lock);
3700         }
3701 #ifdef CONFIG_IP_ROUTE_CLASSID
3702         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3703         if (!ip_rt_acct)
3704                 panic("IP: failed to allocate ip_rt_acct\n");
3705 #endif
3706
3707         ipv4_dst_ops.kmem_cachep =
3708                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3709                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3710
3711         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3712
3713         if (dst_entries_init(&ipv4_dst_ops) < 0)
3714                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3715
3716         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3717                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3718
3719         ipv4_dst_ops.gc_thresh = ~0;
3720         ip_rt_max_size = INT_MAX;
3721
3722         devinet_init();
3723         ip_fib_init();
3724
3725         if (ip_rt_proc_init())
3726                 pr_err("Unable to create route proc files\n");
3727 #ifdef CONFIG_XFRM
3728         xfrm_init();
3729         xfrm4_init();
3730 #endif
3731         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3732                       RTNL_FLAG_DOIT_UNLOCKED);
3733
3734 #ifdef CONFIG_SYSCTL
3735         register_pernet_subsys(&sysctl_route_ops);
3736 #endif
3737         register_pernet_subsys(&rt_genid_ops);
3738         register_pernet_subsys(&ipv4_inetpeer_ops);
3739         return 0;
3740 }
3741
3742 #ifdef CONFIG_SYSCTL
3743 /*
3744  * We really need to sanitize the damn ipv4 init order, then all
3745  * this nonsense will go away.
3746  */
3747 void __init ip_static_sysctl_init(void)
3748 {
3749         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3750 }
3751 #endif