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_puts(seq, "entries  in_hit   in_slow_tot in_slow_mc in_no_route in_brd   in_martian_dst in_martian_src out_hit  out_slow_tot out_slow_mc gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
 280                 return 0;
 281         }
 282
 283         seq_printf(seq, "%08x %08x %08x    %08x   %08x    %08x %08x       "
 284                         "%08x       %08x %08x     %08x    %08x %08x   "
 285                         "%08x     %08x        %08x        %08x\n",
 286                    dst_entries_get_slow(&ipv4_dst_ops),
 287                    0, /* st->in_hit */
 288                    st->in_slow_tot,
 289                    st->in_slow_mc,
 290                    st->in_no_route,
 291                    st->in_brd,
 292                    st->in_martian_dst,
 293                    st->in_martian_src,
 294
 295                    0, /* st->out_hit */
 296                    st->out_slow_tot,
 297                    st->out_slow_mc,
 298
 299                    0, /* st->gc_total */
 300                    0, /* st->gc_ignored */
 301                    0, /* st->gc_goal_miss */
 302                    0, /* st->gc_dst_overflow */
 303                    0, /* st->in_hlist_search */
 304                    0  /* st->out_hlist_search */
 305                 );
 306         return 0;
 307 }
 308
 309 static const struct seq_operations rt_cpu_seq_ops = {
 310         .start  = rt_cpu_seq_start,
 311         .next   = rt_cpu_seq_next,
 312         .stop   = rt_cpu_seq_stop,
 313         .show   = rt_cpu_seq_show,
 314 };
 315
 316 #ifdef CONFIG_IP_ROUTE_CLASSID
 317 static int rt_acct_proc_show(struct seq_file *m, void *v)
 318 {
 319         struct ip_rt_acct *dst, *src;
 320         unsigned int i, j;
 321
 322         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 323         if (!dst)
 324                 return -ENOMEM;
 325
 326         for_each_possible_cpu(i) {
 327                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 328                 for (j = 0; j < 256; j++) {
 329                         dst[j].o_bytes   += src[j].o_bytes;
 330                         dst[j].o_packets += src[j].o_packets;
 331                         dst[j].i_bytes   += src[j].i_bytes;
 332                         dst[j].i_packets += src[j].i_packets;
 333                 }
 334         }
 335
 336         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 337         kfree(dst);
 338         return 0;
 339 }
 340 #endif
 341
 342 static int __net_init ip_rt_do_proc_init(struct net *net)
 343 {
 344         struct proc_dir_entry *pde;
 345
 346         pde = proc_create_seq("rt_cache", 0444, net->proc_net,
 347                               &rt_cache_seq_ops);
 348         if (!pde)
 349                 goto err1;
 350
 351         pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
 352                               &rt_cpu_seq_ops);
 353         if (!pde)
 354                 goto err2;
 355
 356 #ifdef CONFIG_IP_ROUTE_CLASSID
 357         pde = proc_create_single("rt_acct", 0, net->proc_net,
 358                         rt_acct_proc_show);
 359         if (!pde)
 360                 goto err3;
 361 #endif
 362         return 0;
 363
 364 #ifdef CONFIG_IP_ROUTE_CLASSID
 365 err3:
 366         remove_proc_entry("rt_cache", net->proc_net_stat);
 367 #endif
 368 err2:
 369         remove_proc_entry("rt_cache", net->proc_net);
 370 err1:
 371         return -ENOMEM;
 372 }
 373
 374 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 375 {
 376         remove_proc_entry("rt_cache", net->proc_net_stat);
 377         remove_proc_entry("rt_cache", net->proc_net);
 378 #ifdef CONFIG_IP_ROUTE_CLASSID
 379         remove_proc_entry("rt_acct", net->proc_net);
 380 #endif
 381 }
 382
 383 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 384         .init = ip_rt_do_proc_init,
 385         .exit = ip_rt_do_proc_exit,
 386 };
 387
 388 static int __init ip_rt_proc_init(void)
 389 {
 390         return register_pernet_subsys(&ip_rt_proc_ops);
 391 }
 392
 393 #else
 394 static inline int ip_rt_proc_init(void)
 395 {
 396         return 0;
 397 }
 398 #endif /* CONFIG_PROC_FS */
 399
 400 static inline bool rt_is_expired(const struct rtable *rth)
 401 {
 402         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 403 }
 404
 405 void rt_cache_flush(struct net *net)
 406 {
 407         rt_genid_bump_ipv4(net);
 408 }
 409
 410 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 411                                            struct sk_buff *skb,
 412                                            const void *daddr)
 413 {
 414         const struct rtable *rt = container_of(dst, struct rtable, dst);
 415         struct net_device *dev = dst->dev;
 416         struct neighbour *n;
 417
 418         rcu_read_lock_bh();
 419
 420         if (likely(rt->rt_gw_family == AF_INET)) {
 421                 n = ip_neigh_gw4(dev, rt->rt_gw4);
 422         } else if (rt->rt_gw_family == AF_INET6) {
 423                 n = ip_neigh_gw6(dev, &rt->rt_gw6);
 424         } else {
 425                 __be32 pkey;
 426
 427                 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
 428                 n = ip_neigh_gw4(dev, pkey);
 429         }
 430
 431         if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
 432                 n = NULL;
 433
 434         rcu_read_unlock_bh();
 435
 436         return n;
 437 }
 438
 439 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 440 {
 441         const struct rtable *rt = container_of(dst, struct rtable, dst);
 442         struct net_device *dev = dst->dev;
 443         const __be32 *pkey = daddr;
 444
 445         if (rt->rt_gw_family == AF_INET) {
 446                 pkey = (const __be32 *)&rt->rt_gw4;
 447         } else if (rt->rt_gw_family == AF_INET6) {
 448                 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
 449         } else if (!daddr ||
 450                  (rt->rt_flags &
 451                   (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
 452                 return;
 453         }
 454         __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
 455 }
 456
 457 /* Hash tables of size 2048..262144 depending on RAM size.
 458  * Each bucket uses 8 bytes.
 459  */
 460 static u32 ip_idents_mask __read_mostly;
 461 static atomic_t *ip_idents __read_mostly;
 462 static u32 *ip_tstamps __read_mostly;
 463
 464 /* In order to protect privacy, we add a perturbation to identifiers
 465  * if one generator is seldom used. This makes hard for an attacker
 466  * to infer how many packets were sent between two points in time.
 467  */
 468 u32 ip_idents_reserve(u32 hash, int segs)
 469 {
 470         u32 bucket, old, now = (u32)jiffies;
 471         atomic_t *p_id;
 472         u32 *p_tstamp;
 473         u32 delta = 0;
 474
 475         bucket = hash & ip_idents_mask;
 476         p_tstamp = ip_tstamps + bucket;
 477         p_id = ip_idents + bucket;
 478         old = READ_ONCE(*p_tstamp);
 479
 480         if (old != now && cmpxchg(p_tstamp, old, now) == old)
 481                 delta = prandom_u32_max(now - old);
 482
 483         /* If UBSAN reports an error there, please make sure your compiler
 484          * supports -fno-strict-overflow before reporting it that was a bug
 485          * in UBSAN, and it has been fixed in GCC-8.
 486          */
 487         return atomic_add_return(segs + delta, p_id) - segs;
 488 }
 489 EXPORT_SYMBOL(ip_idents_reserve);
 490
 491 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
 492 {
 493         u32 hash, id;
 494
 495         /* Note the following code is not safe, but this is okay. */
 496         if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
 497                 get_random_bytes(&net->ipv4.ip_id_key,
 498                                  sizeof(net->ipv4.ip_id_key));
 499
 500         hash = siphash_3u32((__force u32)iph->daddr,
 501                             (__force u32)iph->saddr,
 502                             iph->protocol,
 503                             &net->ipv4.ip_id_key);
 504         id = ip_idents_reserve(hash, segs);
 505         iph->id = htons(id);
 506 }
 507 EXPORT_SYMBOL(__ip_select_ident);
 508
 509 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
 510                              const struct sock *sk,
 511                              const struct iphdr *iph,
 512                              int oif, u8 tos,
 513                              u8 prot, u32 mark, int flow_flags)
 514 {
 515         if (sk) {
 516                 const struct inet_sock *inet = inet_sk(sk);
 517
 518                 oif = sk->sk_bound_dev_if;
 519                 mark = sk->sk_mark;
 520                 tos = RT_CONN_FLAGS(sk);
 521                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 522         }
 523         flowi4_init_output(fl4, oif, mark, tos,
 524                            RT_SCOPE_UNIVERSE, prot,
 525                            flow_flags,
 526                            iph->daddr, iph->saddr, 0, 0,
 527                            sock_net_uid(net, sk));
 528 }
 529
 530 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 531                                const struct sock *sk)
 532 {
 533         const struct net *net = dev_net(skb->dev);
 534         const struct iphdr *iph = ip_hdr(skb);
 535         int oif = skb->dev->ifindex;
 536         u8 tos = RT_TOS(iph->tos);
 537         u8 prot = iph->protocol;
 538         u32 mark = skb->mark;
 539
 540         __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
 541 }
 542
 543 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 544 {
 545         const struct inet_sock *inet = inet_sk(sk);
 546         const struct ip_options_rcu *inet_opt;
 547         __be32 daddr = inet->inet_daddr;
 548
 549         rcu_read_lock();
 550         inet_opt = rcu_dereference(inet->inet_opt);
 551         if (inet_opt && inet_opt->opt.srr)
 552                 daddr = inet_opt->opt.faddr;
 553         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 554                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 555                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 556                            inet_sk_flowi_flags(sk),
 557                            daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
 558         rcu_read_unlock();
 559 }
 560
 561 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 562                                  const struct sk_buff *skb)
 563 {
 564         if (skb)
 565                 build_skb_flow_key(fl4, skb, sk);
 566         else
 567                 build_sk_flow_key(fl4, sk);
 568 }
 569
 570 static DEFINE_SPINLOCK(fnhe_lock);
 571
 572 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 573 {
 574         struct rtable *rt;
 575
 576         rt = rcu_dereference(fnhe->fnhe_rth_input);
 577         if (rt) {
 578                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 579                 dst_dev_put(&rt->dst);
 580                 dst_release(&rt->dst);
 581         }
 582         rt = rcu_dereference(fnhe->fnhe_rth_output);
 583         if (rt) {
 584                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 585                 dst_dev_put(&rt->dst);
 586                 dst_release(&rt->dst);
 587         }
 588 }
 589
 590 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
 591 {
 592         struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
 593         struct fib_nh_exception *fnhe, *oldest = NULL;
 594
 595         for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
 596                 fnhe = rcu_dereference_protected(*fnhe_p,
 597                                                  lockdep_is_held(&fnhe_lock));
 598                 if (!fnhe)
 599                         break;
 600                 if (!oldest ||
 601                     time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
 602                         oldest = fnhe;
 603                         oldest_p = fnhe_p;
 604                 }
 605         }
 606         fnhe_flush_routes(oldest);
 607         *oldest_p = oldest->fnhe_next;
 608         kfree_rcu(oldest, rcu);
 609 }
 610
 611 static u32 fnhe_hashfun(__be32 daddr)
 612 {
 613         static siphash_key_t fnhe_hash_key __read_mostly;
 614         u64 hval;
 615
 616         net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
 617         hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
 618         return hash_64(hval, FNHE_HASH_SHIFT);
 619 }
 620
 621 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 622 {
 623         rt->rt_pmtu = fnhe->fnhe_pmtu;
 624         rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
 625         rt->dst.expires = fnhe->fnhe_expires;
 626
 627         if (fnhe->fnhe_gw) {
 628                 rt->rt_flags |= RTCF_REDIRECTED;
 629                 rt->rt_uses_gateway = 1;
 630                 rt->rt_gw_family = AF_INET;
 631                 rt->rt_gw4 = fnhe->fnhe_gw;
 632         }
 633 }
 634
 635 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
 636                                   __be32 gw, u32 pmtu, bool lock,
 637                                   unsigned long expires)
 638 {
 639         struct fnhe_hash_bucket *hash;
 640         struct fib_nh_exception *fnhe;
 641         struct rtable *rt;
 642         u32 genid, hval;
 643         unsigned int i;
 644         int depth;
 645
 646         genid = fnhe_genid(dev_net(nhc->nhc_dev));
 647         hval = fnhe_hashfun(daddr);
 648
 649         spin_lock_bh(&fnhe_lock);
 650
 651         hash = rcu_dereference(nhc->nhc_exceptions);
 652         if (!hash) {
 653                 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
 654                 if (!hash)
 655                         goto out_unlock;
 656                 rcu_assign_pointer(nhc->nhc_exceptions, hash);
 657         }
 658
 659         hash += hval;
 660
 661         depth = 0;
 662         for (fnhe = rcu_dereference(hash->chain); fnhe;
 663              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 664                 if (fnhe->fnhe_daddr == daddr)
 665                         break;
 666                 depth++;
 667         }
 668
 669         if (fnhe) {
 670                 if (fnhe->fnhe_genid != genid)
 671                         fnhe->fnhe_genid = genid;
 672                 if (gw)
 673                         fnhe->fnhe_gw = gw;
 674                 if (pmtu) {
 675                         fnhe->fnhe_pmtu = pmtu;
 676                         fnhe->fnhe_mtu_locked = lock;
 677                 }
 678                 fnhe->fnhe_expires = max(1UL, expires);
 679                 /* Update all cached dsts too */
 680                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 681                 if (rt)
 682                         fill_route_from_fnhe(rt, fnhe);
 683                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 684                 if (rt)
 685                         fill_route_from_fnhe(rt, fnhe);
 686         } else {
 687                 /* Randomize max depth to avoid some side channels attacks. */
 688                 int max_depth = FNHE_RECLAIM_DEPTH +
 689                                 prandom_u32_max(FNHE_RECLAIM_DEPTH);
 690
 691                 while (depth > max_depth) {
 692                         fnhe_remove_oldest(hash);
 693                         depth--;
 694                 }
 695
 696                 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 697                 if (!fnhe)
 698                         goto out_unlock;
 699
 700                 fnhe->fnhe_next = hash->chain;
 701
 702                 fnhe->fnhe_genid = genid;
 703                 fnhe->fnhe_daddr = daddr;
 704                 fnhe->fnhe_gw = gw;
 705                 fnhe->fnhe_pmtu = pmtu;
 706                 fnhe->fnhe_mtu_locked = lock;
 707                 fnhe->fnhe_expires = max(1UL, expires);
 708
 709                 rcu_assign_pointer(hash->chain, fnhe);
 710
 711                 /* Exception created; mark the cached routes for the nexthop
 712                  * stale, so anyone caching it rechecks if this exception
 713                  * applies to them.
 714                  */
 715                 rt = rcu_dereference(nhc->nhc_rth_input);
 716                 if (rt)
 717                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 718
 719                 for_each_possible_cpu(i) {
 720                         struct rtable __rcu **prt;
 721
 722                         prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
 723                         rt = rcu_dereference(*prt);
 724                         if (rt)
 725                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 726                 }
 727         }
 728
 729         fnhe->fnhe_stamp = jiffies;
 730
 731 out_unlock:
 732         spin_unlock_bh(&fnhe_lock);
 733 }
 734
 735 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 736                              bool kill_route)
 737 {
 738         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 739         __be32 old_gw = ip_hdr(skb)->saddr;
 740         struct net_device *dev = skb->dev;
 741         struct in_device *in_dev;
 742         struct fib_result res;
 743         struct neighbour *n;
 744         struct net *net;
 745
 746         switch (icmp_hdr(skb)->code & 7) {
 747         case ICMP_REDIR_NET:
 748         case ICMP_REDIR_NETTOS:
 749         case ICMP_REDIR_HOST:
 750         case ICMP_REDIR_HOSTTOS:
 751                 break;
 752
 753         default:
 754                 return;
 755         }
 756
 757         if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
 758                 return;
 759
 760         in_dev = __in_dev_get_rcu(dev);
 761         if (!in_dev)
 762                 return;
 763
 764         net = dev_net(dev);
 765         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 766             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 767             ipv4_is_zeronet(new_gw))
 768                 goto reject_redirect;
 769
 770         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 771                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 772                         goto reject_redirect;
 773                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 774                         goto reject_redirect;
 775         } else {
 776                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 777                         goto reject_redirect;
 778         }
 779
 780         n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
 781         if (!n)
 782                 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
 783         if (!IS_ERR(n)) {
 784                 if (!(n->nud_state & NUD_VALID)) {
 785                         neigh_event_send(n, NULL);
 786                 } else {
 787                         if (fib_lookup(net, fl4, &res, 0) == 0) {
 788                                 struct fib_nh_common *nhc;
 789
 790                                 fib_select_path(net, &res, fl4, skb);
 791                                 nhc = FIB_RES_NHC(res);
 792                                 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
 793                                                 0, false,
 794                                                 jiffies + ip_rt_gc_timeout);
 795                         }
 796                         if (kill_route)
 797                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 798                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 799                 }
 800                 neigh_release(n);
 801         }
 802         return;
 803
 804 reject_redirect:
 805 #ifdef CONFIG_IP_ROUTE_VERBOSE
 806         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 807                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 808                 __be32 daddr = iph->daddr;
 809                 __be32 saddr = iph->saddr;
 810
 811                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 812                                      "  Advised path = %pI4 -> %pI4\n",
 813                                      &old_gw, dev->name, &new_gw,
 814                                      &saddr, &daddr);
 815         }
 816 #endif
 817         ;
 818 }
 819
 820 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 821 {
 822         struct rtable *rt;
 823         struct flowi4 fl4;
 824         const struct iphdr *iph = (const struct iphdr *) skb->data;
 825         struct net *net = dev_net(skb->dev);
 826         int oif = skb->dev->ifindex;
 827         u8 tos = RT_TOS(iph->tos);
 828         u8 prot = iph->protocol;
 829         u32 mark = skb->mark;
 830
 831         rt = (struct rtable *) dst;
 832
 833         __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
 834         __ip_do_redirect(rt, skb, &fl4, true);
 835 }
 836
 837 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 838 {
 839         struct rtable *rt = (struct rtable *)dst;
 840         struct dst_entry *ret = dst;
 841
 842         if (rt) {
 843                 if (dst->obsolete > 0) {
 844                         ip_rt_put(rt);
 845                         ret = NULL;
 846                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 847                            rt->dst.expires) {
 848                         ip_rt_put(rt);
 849                         ret = NULL;
 850                 }
 851         }
 852         return ret;
 853 }
 854
 855 /*
 856  * Algorithm:
 857  *      1. The first ip_rt_redirect_number redirects are sent
 858  *         with exponential backoff, then we stop sending them at all,
 859  *         assuming that the host ignores our redirects.
 860  *      2. If we did not see packets requiring redirects
 861  *         during ip_rt_redirect_silence, we assume that the host
 862  *         forgot redirected route and start to send redirects again.
 863  *
 864  * This algorithm is much cheaper and more intelligent than dumb load limiting
 865  * in icmp.c.
 866  *
 867  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 868  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 869  */
 870
 871 void ip_rt_send_redirect(struct sk_buff *skb)
 872 {
 873         struct rtable *rt = skb_rtable(skb);
 874         struct in_device *in_dev;
 875         struct inet_peer *peer;
 876         struct net *net;
 877         int log_martians;
 878         int vif;
 879
 880         rcu_read_lock();
 881         in_dev = __in_dev_get_rcu(rt->dst.dev);
 882         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 883                 rcu_read_unlock();
 884                 return;
 885         }
 886         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 887         vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
 888         rcu_read_unlock();
 889
 890         net = dev_net(rt->dst.dev);
 891         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
 892         if (!peer) {
 893                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 894                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 895                 return;
 896         }
 897
 898         /* No redirected packets during ip_rt_redirect_silence;
 899          * reset the algorithm.
 900          */
 901         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
 902                 peer->rate_tokens = 0;
 903                 peer->n_redirects = 0;
 904         }
 905
 906         /* Too many ignored redirects; do not send anything
 907          * set dst.rate_last to the last seen redirected packet.
 908          */
 909         if (peer->n_redirects >= ip_rt_redirect_number) {
 910                 peer->rate_last = jiffies;
 911                 goto out_put_peer;
 912         }
 913
 914         /* Check for load limit; set rate_last to the latest sent
 915          * redirect.
 916          */
 917         if (peer->n_redirects == 0 ||
 918             time_after(jiffies,
 919                        (peer->rate_last +
 920                         (ip_rt_redirect_load << peer->n_redirects)))) {
 921                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 922
 923                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 924                 peer->rate_last = jiffies;
 925                 ++peer->n_redirects;
 926 #ifdef CONFIG_IP_ROUTE_VERBOSE
 927                 if (log_martians &&
 928                     peer->n_redirects == ip_rt_redirect_number)
 929                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 930                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 931                                              &ip_hdr(skb)->daddr, &gw);
 932 #endif
 933         }
 934 out_put_peer:
 935         inet_putpeer(peer);
 936 }
 937
 938 static int ip_error(struct sk_buff *skb)
 939 {
 940         struct rtable *rt = skb_rtable(skb);
 941         struct net_device *dev = skb->dev;
 942         struct in_device *in_dev;
 943         struct inet_peer *peer;
 944         unsigned long now;
 945         struct net *net;
 946         bool send;
 947         int code;
 948
 949         if (netif_is_l3_master(skb->dev)) {
 950                 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
 951                 if (!dev)
 952                         goto out;
 953         }
 954
 955         in_dev = __in_dev_get_rcu(dev);
 956
 957         /* IP on this device is disabled. */
 958         if (!in_dev)
 959                 goto out;
 960
 961         net = dev_net(rt->dst.dev);
 962         if (!IN_DEV_FORWARD(in_dev)) {
 963                 switch (rt->dst.error) {
 964                 case EHOSTUNREACH:
 965                         __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
 966                         break;
 967
 968                 case ENETUNREACH:
 969                         __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 970                         break;
 971                 }
 972                 goto out;
 973         }
 974
 975         switch (rt->dst.error) {
 976         case EINVAL:
 977         default:
 978                 goto out;
 979         case EHOSTUNREACH:
 980                 code = ICMP_HOST_UNREACH;
 981                 break;
 982         case ENETUNREACH:
 983                 code = ICMP_NET_UNREACH;
 984                 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 985                 break;
 986         case EACCES:
 987                 code = ICMP_PKT_FILTERED;
 988                 break;
 989         }
 990
 991         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
 992                                l3mdev_master_ifindex(skb->dev), 1);
 993
 994         send = true;
 995         if (peer) {
 996                 now = jiffies;
 997                 peer->rate_tokens += now - peer->rate_last;
 998                 if (peer->rate_tokens > ip_rt_error_burst)
 999                         peer->rate_tokens = ip_rt_error_burst;
1000                 peer->rate_last = now;
1001                 if (peer->rate_tokens >= ip_rt_error_cost)
1002                         peer->rate_tokens -= ip_rt_error_cost;
1003                 else
1004                         send = false;
1005                 inet_putpeer(peer);
1006         }
1007         if (send)
1008                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1009
1010 out:    kfree_skb(skb);
1011         return 0;
1012 }
1013
1014 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1015 {
1016         struct dst_entry *dst = &rt->dst;
1017         struct net *net = dev_net(dst->dev);
1018         struct fib_result res;
1019         bool lock = false;
1020         u32 old_mtu;
1021
1022         if (ip_mtu_locked(dst))
1023                 return;
1024
1025         old_mtu = ipv4_mtu(dst);
1026         if (old_mtu < mtu)
1027                 return;
1028
1029         if (mtu < ip_rt_min_pmtu) {
1030                 lock = true;
1031                 mtu = min(old_mtu, ip_rt_min_pmtu);
1032         }
1033
1034         if (rt->rt_pmtu == mtu && !lock &&
1035             time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1036                 return;
1037
1038         rcu_read_lock();
1039         if (fib_lookup(net, fl4, &res, 0) == 0) {
1040                 struct fib_nh_common *nhc;
1041
1042                 fib_select_path(net, &res, fl4, NULL);
1043                 nhc = FIB_RES_NHC(res);
1044                 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1045                                       jiffies + ip_rt_mtu_expires);
1046         }
1047         rcu_read_unlock();
1048 }
1049
1050 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1051                               struct sk_buff *skb, u32 mtu,
1052                               bool confirm_neigh)
1053 {
1054         struct rtable *rt = (struct rtable *) dst;
1055         struct flowi4 fl4;
1056
1057         ip_rt_build_flow_key(&fl4, sk, skb);
1058
1059         /* Don't make lookup fail for bridged encapsulations */
1060         if (skb && netif_is_any_bridge_port(skb->dev))
1061                 fl4.flowi4_oif = 0;
1062
1063         __ip_rt_update_pmtu(rt, &fl4, mtu);
1064 }
1065
1066 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1067                       int oif, u8 protocol)
1068 {
1069         const struct iphdr *iph = (const struct iphdr *)skb->data;
1070         struct flowi4 fl4;
1071         struct rtable *rt;
1072         u32 mark = IP4_REPLY_MARK(net, skb->mark);
1073
1074         __build_flow_key(net, &fl4, NULL, iph, oif,
1075                          RT_TOS(iph->tos), protocol, mark, 0);
1076         rt = __ip_route_output_key(net, &fl4);
1077         if (!IS_ERR(rt)) {
1078                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1079                 ip_rt_put(rt);
1080         }
1081 }
1082 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1083
1084 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1085 {
1086         const struct iphdr *iph = (const struct iphdr *)skb->data;
1087         struct flowi4 fl4;
1088         struct rtable *rt;
1089
1090         __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1091
1092         if (!fl4.flowi4_mark)
1093                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1094
1095         rt = __ip_route_output_key(sock_net(sk), &fl4);
1096         if (!IS_ERR(rt)) {
1097                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1098                 ip_rt_put(rt);
1099         }
1100 }
1101
1102 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1103 {
1104         const struct iphdr *iph = (const struct iphdr *)skb->data;
1105         struct flowi4 fl4;
1106         struct rtable *rt;
1107         struct dst_entry *odst = NULL;
1108         bool new = false;
1109         struct net *net = sock_net(sk);
1110
1111         bh_lock_sock(sk);
1112
1113         if (!ip_sk_accept_pmtu(sk))
1114                 goto out;
1115
1116         odst = sk_dst_get(sk);
1117
1118         if (sock_owned_by_user(sk) || !odst) {
1119                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1120                 goto out;
1121         }
1122
1123         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1124
1125         rt = (struct rtable *)odst;
1126         if (odst->obsolete && !odst->ops->check(odst, 0)) {
1127                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1128                 if (IS_ERR(rt))
1129                         goto out;
1130
1131                 new = true;
1132         }
1133
1134         __ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu);
1135
1136         if (!dst_check(&rt->dst, 0)) {
1137                 if (new)
1138                         dst_release(&rt->dst);
1139
1140                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1141                 if (IS_ERR(rt))
1142                         goto out;
1143
1144                 new = true;
1145         }
1146
1147         if (new)
1148                 sk_dst_set(sk, &rt->dst);
1149
1150 out:
1151         bh_unlock_sock(sk);
1152         dst_release(odst);
1153 }
1154 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1155
1156 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1157                    int oif, u8 protocol)
1158 {
1159         const struct iphdr *iph = (const struct iphdr *)skb->data;
1160         struct flowi4 fl4;
1161         struct rtable *rt;
1162
1163         __build_flow_key(net, &fl4, NULL, iph, oif,
1164                          RT_TOS(iph->tos), protocol, 0, 0);
1165         rt = __ip_route_output_key(net, &fl4);
1166         if (!IS_ERR(rt)) {
1167                 __ip_do_redirect(rt, skb, &fl4, false);
1168                 ip_rt_put(rt);
1169         }
1170 }
1171 EXPORT_SYMBOL_GPL(ipv4_redirect);
1172
1173 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1174 {
1175         const struct iphdr *iph = (const struct iphdr *)skb->data;
1176         struct flowi4 fl4;
1177         struct rtable *rt;
1178         struct net *net = sock_net(sk);
1179
1180         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1181         rt = __ip_route_output_key(net, &fl4);
1182         if (!IS_ERR(rt)) {
1183                 __ip_do_redirect(rt, skb, &fl4, false);
1184                 ip_rt_put(rt);
1185         }
1186 }
1187 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1188
1189 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1190                                                          u32 cookie)
1191 {
1192         struct rtable *rt = (struct rtable *) dst;
1193
1194         /* All IPV4 dsts are created with ->obsolete set to the value
1195          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1196          * into this function always.
1197          *
1198          * When a PMTU/redirect information update invalidates a route,
1199          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1200          * DST_OBSOLETE_DEAD.
1201          */
1202         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1203                 return NULL;
1204         return dst;
1205 }
1206 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1207
1208 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1209 {
1210         struct ip_options opt;
1211         int res;
1212
1213         /* Recompile ip options since IPCB may not be valid anymore.
1214          * Also check we have a reasonable ipv4 header.
1215          */
1216         if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1217             ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1218                 return;
1219
1220         memset(&opt, 0, sizeof(opt));
1221         if (ip_hdr(skb)->ihl > 5) {
1222                 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1223                         return;
1224                 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1225
1226                 rcu_read_lock();
1227                 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1228                 rcu_read_unlock();
1229
1230                 if (res)
1231                         return;
1232         }
1233         __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1234 }
1235
1236 static void ipv4_link_failure(struct sk_buff *skb)
1237 {
1238         struct rtable *rt;
1239
1240         ipv4_send_dest_unreach(skb);
1241
1242         rt = skb_rtable(skb);
1243         if (rt)
1244                 dst_set_expires(&rt->dst, 0);
1245 }
1246
1247 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1248 {
1249         pr_debug("%s: %pI4 -> %pI4, %s\n",
1250                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1251                  skb->dev ? skb->dev->name : "?");
1252         kfree_skb(skb);
1253         WARN_ON(1);
1254         return 0;
1255 }
1256
1257 /*
1258  * We do not cache source address of outgoing interface,
1259  * because it is used only by IP RR, TS and SRR options,
1260  * so that it out of fast path.
1261  *
1262  * BTW remember: "addr" is allowed to be not aligned
1263  * in IP options!
1264  */
1265
1266 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1267 {
1268         __be32 src;
1269
1270         if (rt_is_output_route(rt))
1271                 src = ip_hdr(skb)->saddr;
1272         else {
1273                 struct fib_result res;
1274                 struct iphdr *iph = ip_hdr(skb);
1275                 struct flowi4 fl4 = {
1276                         .daddr = iph->daddr,
1277                         .saddr = iph->saddr,
1278                         .flowi4_tos = RT_TOS(iph->tos),
1279                         .flowi4_oif = rt->dst.dev->ifindex,
1280                         .flowi4_iif = skb->dev->ifindex,
1281                         .flowi4_mark = skb->mark,
1282                 };
1283
1284                 rcu_read_lock();
1285                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1286                         src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1287                 else
1288                         src = inet_select_addr(rt->dst.dev,
1289                                                rt_nexthop(rt, iph->daddr),
1290                                                RT_SCOPE_UNIVERSE);
1291                 rcu_read_unlock();
1292         }
1293         memcpy(addr, &src, 4);
1294 }
1295
1296 #ifdef CONFIG_IP_ROUTE_CLASSID
1297 static void set_class_tag(struct rtable *rt, u32 tag)
1298 {
1299         if (!(rt->dst.tclassid & 0xFFFF))
1300                 rt->dst.tclassid |= tag & 0xFFFF;
1301         if (!(rt->dst.tclassid & 0xFFFF0000))
1302                 rt->dst.tclassid |= tag & 0xFFFF0000;
1303 }
1304 #endif
1305
1306 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1307 {
1308         unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1309         unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1310                                     ip_rt_min_advmss);
1311
1312         return min(advmss, IPV4_MAX_PMTU - header_size);
1313 }
1314
1315 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1316 {
1317         return ip_dst_mtu_maybe_forward(dst, false);
1318 }
1319 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1320
1321 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1322 {
1323         struct fnhe_hash_bucket *hash;
1324         struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1325         u32 hval = fnhe_hashfun(daddr);
1326
1327         spin_lock_bh(&fnhe_lock);
1328
1329         hash = rcu_dereference_protected(nhc->nhc_exceptions,
1330                                          lockdep_is_held(&fnhe_lock));
1331         hash += hval;
1332
1333         fnhe_p = &hash->chain;
1334         fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1335         while (fnhe) {
1336                 if (fnhe->fnhe_daddr == daddr) {
1337                         rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1338                                 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1339                         /* set fnhe_daddr to 0 to ensure it won't bind with
1340                          * new dsts in rt_bind_exception().
1341                          */
1342                         fnhe->fnhe_daddr = 0;
1343                         fnhe_flush_routes(fnhe);
1344                         kfree_rcu(fnhe, rcu);
1345                         break;
1346                 }
1347                 fnhe_p = &fnhe->fnhe_next;
1348                 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1349                                                  lockdep_is_held(&fnhe_lock));
1350         }
1351
1352         spin_unlock_bh(&fnhe_lock);
1353 }
1354
1355 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1356                                                __be32 daddr)
1357 {
1358         struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1359         struct fib_nh_exception *fnhe;
1360         u32 hval;
1361
1362         if (!hash)
1363                 return NULL;
1364
1365         hval = fnhe_hashfun(daddr);
1366
1367         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1368              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1369                 if (fnhe->fnhe_daddr == daddr) {
1370                         if (fnhe->fnhe_expires &&
1371                             time_after(jiffies, fnhe->fnhe_expires)) {
1372                                 ip_del_fnhe(nhc, daddr);
1373                                 break;
1374                         }
1375                         return fnhe;
1376                 }
1377         }
1378         return NULL;
1379 }
1380
1381 /* MTU selection:
1382  * 1. mtu on route is locked - use it
1383  * 2. mtu from nexthop exception
1384  * 3. mtu from egress device
1385  */
1386
1387 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1388 {
1389         struct fib_nh_common *nhc = res->nhc;
1390         struct net_device *dev = nhc->nhc_dev;
1391         struct fib_info *fi = res->fi;
1392         u32 mtu = 0;
1393
1394         if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1395             fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1396                 mtu = fi->fib_mtu;
1397
1398         if (likely(!mtu)) {
1399                 struct fib_nh_exception *fnhe;
1400
1401                 fnhe = find_exception(nhc, daddr);
1402                 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1403                         mtu = fnhe->fnhe_pmtu;
1404         }
1405
1406         if (likely(!mtu))
1407                 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1408
1409         return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1410 }
1411
1412 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1413                               __be32 daddr, const bool do_cache)
1414 {
1415         bool ret = false;
1416
1417         spin_lock_bh(&fnhe_lock);
1418
1419         if (daddr == fnhe->fnhe_daddr) {
1420                 struct rtable __rcu **porig;
1421                 struct rtable *orig;
1422                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1423
1424                 if (rt_is_input_route(rt))
1425                         porig = &fnhe->fnhe_rth_input;
1426                 else
1427                         porig = &fnhe->fnhe_rth_output;
1428                 orig = rcu_dereference(*porig);
1429
1430                 if (fnhe->fnhe_genid != genid) {
1431                         fnhe->fnhe_genid = genid;
1432                         fnhe->fnhe_gw = 0;
1433                         fnhe->fnhe_pmtu = 0;
1434                         fnhe->fnhe_expires = 0;
1435                         fnhe->fnhe_mtu_locked = false;
1436                         fnhe_flush_routes(fnhe);
1437                         orig = NULL;
1438                 }
1439                 fill_route_from_fnhe(rt, fnhe);
1440                 if (!rt->rt_gw4) {
1441                         rt->rt_gw4 = daddr;
1442                         rt->rt_gw_family = AF_INET;
1443                 }
1444
1445                 if (do_cache) {
1446                         dst_hold(&rt->dst);
1447                         rcu_assign_pointer(*porig, rt);
1448                         if (orig) {
1449                                 dst_dev_put(&orig->dst);
1450                                 dst_release(&orig->dst);
1451                         }
1452                         ret = true;
1453                 }
1454
1455                 fnhe->fnhe_stamp = jiffies;
1456         }
1457         spin_unlock_bh(&fnhe_lock);
1458
1459         return ret;
1460 }
1461
1462 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1463 {
1464         struct rtable *orig, *prev, **p;
1465         bool ret = true;
1466
1467         if (rt_is_input_route(rt)) {
1468                 p = (struct rtable **)&nhc->nhc_rth_input;
1469         } else {
1470                 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1471         }
1472         orig = *p;
1473
1474         /* hold dst before doing cmpxchg() to avoid race condition
1475          * on this dst
1476          */
1477         dst_hold(&rt->dst);
1478         prev = cmpxchg(p, orig, rt);
1479         if (prev == orig) {
1480                 if (orig) {
1481                         rt_add_uncached_list(orig);
1482                         dst_release(&orig->dst);
1483                 }
1484         } else {
1485                 dst_release(&rt->dst);
1486                 ret = false;
1487         }
1488
1489         return ret;
1490 }
1491
1492 struct uncached_list {
1493         spinlock_t              lock;
1494         struct list_head        head;
1495 };
1496
1497 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1498
1499 void rt_add_uncached_list(struct rtable *rt)
1500 {
1501         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1502
1503         rt->rt_uncached_list = ul;
1504
1505         spin_lock_bh(&ul->lock);
1506         list_add_tail(&rt->rt_uncached, &ul->head);
1507         spin_unlock_bh(&ul->lock);
1508 }
1509
1510 void rt_del_uncached_list(struct rtable *rt)
1511 {
1512         if (!list_empty(&rt->rt_uncached)) {
1513                 struct uncached_list *ul = rt->rt_uncached_list;
1514
1515                 spin_lock_bh(&ul->lock);
1516                 list_del(&rt->rt_uncached);
1517                 spin_unlock_bh(&ul->lock);
1518         }
1519 }
1520
1521 static void ipv4_dst_destroy(struct dst_entry *dst)
1522 {
1523         struct rtable *rt = (struct rtable *)dst;
1524
1525         ip_dst_metrics_put(dst);
1526         rt_del_uncached_list(rt);
1527 }
1528
1529 void rt_flush_dev(struct net_device *dev)
1530 {
1531         struct rtable *rt;
1532         int cpu;
1533
1534         for_each_possible_cpu(cpu) {
1535                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1536
1537                 spin_lock_bh(&ul->lock);
1538                 list_for_each_entry(rt, &ul->head, rt_uncached) {
1539                         if (rt->dst.dev != dev)
1540                                 continue;
1541                         rt->dst.dev = blackhole_netdev;
1542                         dev_hold(rt->dst.dev);
1543                         dev_put(dev);
1544                 }
1545                 spin_unlock_bh(&ul->lock);
1546         }
1547 }
1548
1549 static bool rt_cache_valid(const struct rtable *rt)
1550 {
1551         return  rt &&
1552                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1553                 !rt_is_expired(rt);
1554 }
1555
1556 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1557                            const struct fib_result *res,
1558                            struct fib_nh_exception *fnhe,
1559                            struct fib_info *fi, u16 type, u32 itag,
1560                            const bool do_cache)
1561 {
1562         bool cached = false;
1563
1564         if (fi) {
1565                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1566
1567                 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1568                         rt->rt_uses_gateway = 1;
1569                         rt->rt_gw_family = nhc->nhc_gw_family;
1570                         /* only INET and INET6 are supported */
1571                         if (likely(nhc->nhc_gw_family == AF_INET))
1572                                 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1573                         else
1574                                 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1575                 }
1576
1577                 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1578
1579 #ifdef CONFIG_IP_ROUTE_CLASSID
1580                 if (nhc->nhc_family == AF_INET) {
1581                         struct fib_nh *nh;
1582
1583                         nh = container_of(nhc, struct fib_nh, nh_common);
1584                         rt->dst.tclassid = nh->nh_tclassid;
1585                 }
1586 #endif
1587                 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1588                 if (unlikely(fnhe))
1589                         cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1590                 else if (do_cache)
1591                         cached = rt_cache_route(nhc, rt);
1592                 if (unlikely(!cached)) {
1593                         /* Routes we intend to cache in nexthop exception or
1594                          * FIB nexthop have the DST_NOCACHE bit clear.
1595                          * However, if we are unsuccessful at storing this
1596                          * route into the cache we really need to set it.
1597                          */
1598                         if (!rt->rt_gw4) {
1599                                 rt->rt_gw_family = AF_INET;
1600                                 rt->rt_gw4 = daddr;
1601                         }
1602                         rt_add_uncached_list(rt);
1603                 }
1604         } else
1605                 rt_add_uncached_list(rt);
1606
1607 #ifdef CONFIG_IP_ROUTE_CLASSID
1608 #ifdef CONFIG_IP_MULTIPLE_TABLES
1609         set_class_tag(rt, res->tclassid);
1610 #endif
1611         set_class_tag(rt, itag);
1612 #endif
1613 }
1614
1615 struct rtable *rt_dst_alloc(struct net_device *dev,
1616                             unsigned int flags, u16 type,
1617                             bool nopolicy, bool noxfrm)
1618 {
1619         struct rtable *rt;
1620
1621         rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1622                        (nopolicy ? DST_NOPOLICY : 0) |
1623                        (noxfrm ? DST_NOXFRM : 0));
1624
1625         if (rt) {
1626                 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1627                 rt->rt_flags = flags;
1628                 rt->rt_type = type;
1629                 rt->rt_is_input = 0;
1630                 rt->rt_iif = 0;
1631                 rt->rt_pmtu = 0;
1632                 rt->rt_mtu_locked = 0;
1633                 rt->rt_uses_gateway = 0;
1634                 rt->rt_gw_family = 0;
1635                 rt->rt_gw4 = 0;
1636                 INIT_LIST_HEAD(&rt->rt_uncached);
1637
1638                 rt->dst.output = ip_output;
1639                 if (flags & RTCF_LOCAL)
1640                         rt->dst.input = ip_local_deliver;
1641         }
1642
1643         return rt;
1644 }
1645 EXPORT_SYMBOL(rt_dst_alloc);
1646
1647 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1648 {
1649         struct rtable *new_rt;
1650
1651         new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1652                            rt->dst.flags);
1653
1654         if (new_rt) {
1655                 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1656                 new_rt->rt_flags = rt->rt_flags;
1657                 new_rt->rt_type = rt->rt_type;
1658                 new_rt->rt_is_input = rt->rt_is_input;
1659                 new_rt->rt_iif = rt->rt_iif;
1660                 new_rt->rt_pmtu = rt->rt_pmtu;
1661                 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1662                 new_rt->rt_gw_family = rt->rt_gw_family;
1663                 if (rt->rt_gw_family == AF_INET)
1664                         new_rt->rt_gw4 = rt->rt_gw4;
1665                 else if (rt->rt_gw_family == AF_INET6)
1666                         new_rt->rt_gw6 = rt->rt_gw6;
1667                 INIT_LIST_HEAD(&new_rt->rt_uncached);
1668
1669                 new_rt->dst.input = rt->dst.input;
1670                 new_rt->dst.output = rt->dst.output;
1671                 new_rt->dst.error = rt->dst.error;
1672                 new_rt->dst.lastuse = jiffies;
1673                 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1674         }
1675         return new_rt;
1676 }
1677 EXPORT_SYMBOL(rt_dst_clone);
1678
1679 /* called in rcu_read_lock() section */
1680 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1681                           u8 tos, struct net_device *dev,
1682                           struct in_device *in_dev, u32 *itag)
1683 {
1684         int err;
1685
1686         /* Primary sanity checks. */
1687         if (!in_dev)
1688                 return -EINVAL;
1689
1690         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1691             skb->protocol != htons(ETH_P_IP))
1692                 return -EINVAL;
1693
1694         if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1695                 return -EINVAL;
1696
1697         if (ipv4_is_zeronet(saddr)) {
1698                 if (!ipv4_is_local_multicast(daddr) &&
1699                     ip_hdr(skb)->protocol != IPPROTO_IGMP)
1700                         return -EINVAL;
1701         } else {
1702                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1703                                           in_dev, itag);
1704                 if (err < 0)
1705                         return err;
1706         }
1707         return 0;
1708 }
1709
1710 /* called in rcu_read_lock() section */
1711 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1712                              u8 tos, struct net_device *dev, int our)
1713 {
1714         struct in_device *in_dev = __in_dev_get_rcu(dev);
1715         unsigned int flags = RTCF_MULTICAST;
1716         struct rtable *rth;
1717         u32 itag = 0;
1718         int err;
1719
1720         err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1721         if (err)
1722                 return err;
1723
1724         if (our)
1725                 flags |= RTCF_LOCAL;
1726
1727         rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1728                            IN_DEV_ORCONF(in_dev, NOPOLICY), false);
1729         if (!rth)
1730                 return -ENOBUFS;
1731
1732 #ifdef CONFIG_IP_ROUTE_CLASSID
1733         rth->dst.tclassid = itag;
1734 #endif
1735         rth->dst.output = ip_rt_bug;
1736         rth->rt_is_input= 1;
1737
1738 #ifdef CONFIG_IP_MROUTE
1739         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1740                 rth->dst.input = ip_mr_input;
1741 #endif
1742         RT_CACHE_STAT_INC(in_slow_mc);
1743
1744         skb_dst_set(skb, &rth->dst);
1745         return 0;
1746 }
1747
1748
1749 static void ip_handle_martian_source(struct net_device *dev,
1750                                      struct in_device *in_dev,
1751                                      struct sk_buff *skb,
1752                                      __be32 daddr,
1753                                      __be32 saddr)
1754 {
1755         RT_CACHE_STAT_INC(in_martian_src);
1756 #ifdef CONFIG_IP_ROUTE_VERBOSE
1757         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1758                 /*
1759                  *      RFC1812 recommendation, if source is martian,
1760                  *      the only hint is MAC header.
1761                  */
1762                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1763                         &daddr, &saddr, dev->name);
1764                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1765                         print_hex_dump(KERN_WARNING, "ll header: ",
1766                                        DUMP_PREFIX_OFFSET, 16, 1,
1767                                        skb_mac_header(skb),
1768                                        dev->hard_header_len, false);
1769                 }
1770         }
1771 #endif
1772 }
1773
1774 /* called in rcu_read_lock() section */
1775 static int __mkroute_input(struct sk_buff *skb,
1776                            const struct fib_result *res,
1777                            struct in_device *in_dev,
1778                            __be32 daddr, __be32 saddr, u32 tos)
1779 {
1780         struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1781         struct net_device *dev = nhc->nhc_dev;
1782         struct fib_nh_exception *fnhe;
1783         struct rtable *rth;
1784         int err;
1785         struct in_device *out_dev;
1786         bool do_cache;
1787         u32 itag = 0;
1788
1789         /* get a working reference to the output device */
1790         out_dev = __in_dev_get_rcu(dev);
1791         if (!out_dev) {
1792                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1793                 return -EINVAL;
1794         }
1795
1796         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1797                                   in_dev->dev, in_dev, &itag);
1798         if (err < 0) {
1799                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1800                                          saddr);
1801
1802                 goto cleanup;
1803         }
1804
1805         do_cache = res->fi && !itag;
1806         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1807             skb->protocol == htons(ETH_P_IP)) {
1808                 __be32 gw;
1809
1810                 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1811                 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1812                     inet_addr_onlink(out_dev, saddr, gw))
1813                         IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1814         }
1815
1816         if (skb->protocol != htons(ETH_P_IP)) {
1817                 /* Not IP (i.e. ARP). Do not create route, if it is
1818                  * invalid for proxy arp. DNAT routes are always valid.
1819                  *
1820                  * Proxy arp feature have been extended to allow, ARP
1821                  * replies back to the same interface, to support
1822                  * Private VLAN switch technologies. See arp.c.
1823                  */
1824                 if (out_dev == in_dev &&
1825                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1826                         err = -EINVAL;
1827                         goto cleanup;
1828                 }
1829         }
1830
1831         fnhe = find_exception(nhc, daddr);
1832         if (do_cache) {
1833                 if (fnhe)
1834                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1835                 else
1836                         rth = rcu_dereference(nhc->nhc_rth_input);
1837                 if (rt_cache_valid(rth)) {
1838                         skb_dst_set_noref(skb, &rth->dst);
1839                         goto out;
1840                 }
1841         }
1842
1843         rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1844                            IN_DEV_ORCONF(in_dev, NOPOLICY),
1845                            IN_DEV_ORCONF(out_dev, NOXFRM));
1846         if (!rth) {
1847                 err = -ENOBUFS;
1848                 goto cleanup;
1849         }
1850
1851         rth->rt_is_input = 1;
1852         RT_CACHE_STAT_INC(in_slow_tot);
1853
1854         rth->dst.input = ip_forward;
1855
1856         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1857                        do_cache);
1858         lwtunnel_set_redirect(&rth->dst);
1859         skb_dst_set(skb, &rth->dst);
1860 out:
1861         err = 0;
1862  cleanup:
1863         return err;
1864 }
1865
1866 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1867 /* To make ICMP packets follow the right flow, the multipath hash is
1868  * calculated from the inner IP addresses.
1869  */
1870 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1871                                  struct flow_keys *hash_keys)
1872 {
1873         const struct iphdr *outer_iph = ip_hdr(skb);
1874         const struct iphdr *key_iph = outer_iph;
1875         const struct iphdr *inner_iph;
1876         const struct icmphdr *icmph;
1877         struct iphdr _inner_iph;
1878         struct icmphdr _icmph;
1879
1880         if (likely(outer_iph->protocol != IPPROTO_ICMP))
1881                 goto out;
1882
1883         if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1884                 goto out;
1885
1886         icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1887                                    &_icmph);
1888         if (!icmph)
1889                 goto out;
1890
1891         if (!icmp_is_err(icmph->type))
1892                 goto out;
1893
1894         inner_iph = skb_header_pointer(skb,
1895                                        outer_iph->ihl * 4 + sizeof(_icmph),
1896                                        sizeof(_inner_iph), &_inner_iph);
1897         if (!inner_iph)
1898                 goto out;
1899
1900         key_iph = inner_iph;
1901 out:
1902         hash_keys->addrs.v4addrs.src = key_iph->saddr;
1903         hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1904 }
1905
1906 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1907                                            const struct sk_buff *skb,
1908                                            bool *p_has_inner)
1909 {
1910         u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
1911         struct flow_keys keys, hash_keys;
1912
1913         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1914                 return 0;
1915
1916         memset(&hash_keys, 0, sizeof(hash_keys));
1917         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1918
1919         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1920         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1921                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1922         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1923                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1924         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1925                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1926         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1927                 hash_keys.ports.src = keys.ports.src;
1928         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1929                 hash_keys.ports.dst = keys.ports.dst;
1930
1931         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1932         return flow_hash_from_keys(&hash_keys);
1933 }
1934
1935 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1936                                            const struct sk_buff *skb,
1937                                            bool has_inner)
1938 {
1939         u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
1940         struct flow_keys keys, hash_keys;
1941
1942         /* We assume the packet carries an encapsulation, but if none was
1943          * encountered during dissection of the outer flow, then there is no
1944          * point in calling the flow dissector again.
1945          */
1946         if (!has_inner)
1947                 return 0;
1948
1949         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1950                 return 0;
1951
1952         memset(&hash_keys, 0, sizeof(hash_keys));
1953         skb_flow_dissect_flow_keys(skb, &keys, 0);
1954
1955         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1956                 return 0;
1957
1958         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1959                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1960                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1961                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1962                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1963                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1964         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1965                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1966                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1967                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1968                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1969                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1970                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1971                         hash_keys.tags.flow_label = keys.tags.flow_label;
1972         }
1973
1974         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1975                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1976         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
1977                 hash_keys.ports.src = keys.ports.src;
1978         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
1979                 hash_keys.ports.dst = keys.ports.dst;
1980
1981         return flow_hash_from_keys(&hash_keys);
1982 }
1983
1984 static u32 fib_multipath_custom_hash_skb(const struct net *net,
1985                                          const struct sk_buff *skb)
1986 {
1987         u32 mhash, mhash_inner;
1988         bool has_inner = true;
1989
1990         mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
1991         mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
1992
1993         return jhash_2words(mhash, mhash_inner, 0);
1994 }
1995
1996 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
1997                                          const struct flowi4 *fl4)
1998 {
1999         u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields;
2000         struct flow_keys hash_keys;
2001
2002         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2003                 return 0;
2004
2005         memset(&hash_keys, 0, sizeof(hash_keys));
2006         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2007         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2008                 hash_keys.addrs.v4addrs.src = fl4->saddr;
2009         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2010                 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2011         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2012                 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2013         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2014                 hash_keys.ports.src = fl4->fl4_sport;
2015         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2016                 hash_keys.ports.dst = fl4->fl4_dport;
2017
2018         return flow_hash_from_keys(&hash_keys);
2019 }
2020
2021 /* if skb is set it will be used and fl4 can be NULL */
2022 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2023                        const struct sk_buff *skb, struct flow_keys *flkeys)
2024 {
2025         u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2026         struct flow_keys hash_keys;
2027         u32 mhash = 0;
2028
2029         switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
2030         case 0:
2031                 memset(&hash_keys, 0, sizeof(hash_keys));
2032                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2033                 if (skb) {
2034                         ip_multipath_l3_keys(skb, &hash_keys);
2035                 } else {
2036                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2037                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2038                 }
2039                 mhash = flow_hash_from_keys(&hash_keys);
2040                 break;
2041         case 1:
2042                 /* skb is currently provided only when forwarding */
2043                 if (skb) {
2044                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2045                         struct flow_keys keys;
2046
2047                         /* short-circuit if we already have L4 hash present */
2048                         if (skb->l4_hash)
2049                                 return skb_get_hash_raw(skb) >> 1;
2050
2051                         memset(&hash_keys, 0, sizeof(hash_keys));
2052
2053                         if (!flkeys) {
2054                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2055                                 flkeys = &keys;
2056                         }
2057
2058                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2059                         hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2060                         hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2061                         hash_keys.ports.src = flkeys->ports.src;
2062                         hash_keys.ports.dst = flkeys->ports.dst;
2063                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2064                 } else {
2065                         memset(&hash_keys, 0, sizeof(hash_keys));
2066                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2067                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2068                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2069                         hash_keys.ports.src = fl4->fl4_sport;
2070                         hash_keys.ports.dst = fl4->fl4_dport;
2071                         hash_keys.basic.ip_proto = fl4->flowi4_proto;
2072                 }
2073                 mhash = flow_hash_from_keys(&hash_keys);
2074                 break;
2075         case 2:
2076                 memset(&hash_keys, 0, sizeof(hash_keys));
2077                 /* skb is currently provided only when forwarding */
2078                 if (skb) {
2079                         struct flow_keys keys;
2080
2081                         skb_flow_dissect_flow_keys(skb, &keys, 0);
2082                         /* Inner can be v4 or v6 */
2083                         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2084                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2085                                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2086                                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2087                         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2088                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2089                                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2090                                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2091                                 hash_keys.tags.flow_label = keys.tags.flow_label;
2092                                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2093                         } else {
2094                                 /* Same as case 0 */
2095                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2096                                 ip_multipath_l3_keys(skb, &hash_keys);
2097                         }
2098                 } else {
2099                         /* Same as case 0 */
2100                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2101                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2102                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2103                 }
2104                 mhash = flow_hash_from_keys(&hash_keys);
2105                 break;
2106         case 3:
2107                 if (skb)
2108                         mhash = fib_multipath_custom_hash_skb(net, skb);
2109                 else
2110                         mhash = fib_multipath_custom_hash_fl4(net, fl4);
2111                 break;
2112         }
2113
2114         if (multipath_hash)
2115                 mhash = jhash_2words(mhash, multipath_hash, 0);
2116
2117         return mhash >> 1;
2118 }
2119 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2120
2121 static int ip_mkroute_input(struct sk_buff *skb,
2122                             struct fib_result *res,
2123                             struct in_device *in_dev,
2124                             __be32 daddr, __be32 saddr, u32 tos,
2125                             struct flow_keys *hkeys)
2126 {
2127 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2128         if (res->fi && fib_info_num_path(res->fi) > 1) {
2129                 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2130
2131                 fib_select_multipath(res, h);
2132         }
2133 #endif
2134
2135         /* create a routing cache entry */
2136         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2137 }
2138
2139 /* Implements all the saddr-related checks as ip_route_input_slow(),
2140  * assuming daddr is valid and the destination is not a local broadcast one.
2141  * Uses the provided hint instead of performing a route lookup.
2142  */
2143 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2144                       u8 tos, struct net_device *dev,
2145                       const struct sk_buff *hint)
2146 {
2147         struct in_device *in_dev = __in_dev_get_rcu(dev);
2148         struct rtable *rt = skb_rtable(hint);
2149         struct net *net = dev_net(dev);
2150         int err = -EINVAL;
2151         u32 tag = 0;
2152
2153         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2154                 goto martian_source;
2155
2156         if (ipv4_is_zeronet(saddr))
2157                 goto martian_source;
2158
2159         if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2160                 goto martian_source;
2161
2162         if (rt->rt_type != RTN_LOCAL)
2163                 goto skip_validate_source;
2164
2165         tos &= IPTOS_RT_MASK;
2166         err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2167         if (err < 0)
2168                 goto martian_source;
2169
2170 skip_validate_source:
2171         skb_dst_copy(skb, hint);
2172         return 0;
2173
2174 martian_source:
2175         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2176         return err;
2177 }
2178
2179 /* get device for dst_alloc with local routes */
2180 static struct net_device *ip_rt_get_dev(struct net *net,
2181                                         const struct fib_result *res)
2182 {
2183         struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2184         struct net_device *dev = NULL;
2185
2186         if (nhc)
2187                 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2188
2189         return dev ? : net->loopback_dev;
2190 }
2191
2192 /*
2193  *      NOTE. We drop all the packets that has local source
2194  *      addresses, because every properly looped back packet
2195  *      must have correct destination already attached by output routine.
2196  *      Changes in the enforced policies must be applied also to
2197  *      ip_route_use_hint().
2198  *
2199  *      Such approach solves two big problems:
2200  *      1. Not simplex devices are handled properly.
2201  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2202  *      called with rcu_read_lock()
2203  */
2204
2205 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2206                                u8 tos, struct net_device *dev,
2207                                struct fib_result *res)
2208 {
2209         struct in_device *in_dev = __in_dev_get_rcu(dev);
2210         struct flow_keys *flkeys = NULL, _flkeys;
2211         struct net    *net = dev_net(dev);
2212         struct ip_tunnel_info *tun_info;
2213         int             err = -EINVAL;
2214         unsigned int    flags = 0;
2215         u32             itag = 0;
2216         struct rtable   *rth;
2217         struct flowi4   fl4;
2218         bool do_cache = true;
2219
2220         /* IP on this device is disabled. */
2221
2222         if (!in_dev)
2223                 goto out;
2224
2225         /* Check for the most weird martians, which can be not detected
2226          * by fib_lookup.
2227          */
2228
2229         tun_info = skb_tunnel_info(skb);
2230         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2231                 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2232         else
2233                 fl4.flowi4_tun_key.tun_id = 0;
2234         skb_dst_drop(skb);
2235
2236         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2237                 goto martian_source;
2238
2239         res->fi = NULL;
2240         res->table = NULL;
2241         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2242                 goto brd_input;
2243
2244         /* Accept zero addresses only to limited broadcast;
2245          * I even do not know to fix it or not. Waiting for complains :-)
2246          */
2247         if (ipv4_is_zeronet(saddr))
2248                 goto martian_source;
2249
2250         if (ipv4_is_zeronet(daddr))
2251                 goto martian_destination;
2252
2253         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2254          * and call it once if daddr or/and saddr are loopback addresses
2255          */
2256         if (ipv4_is_loopback(daddr)) {
2257                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2258                         goto martian_destination;
2259         } else if (ipv4_is_loopback(saddr)) {
2260                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2261                         goto martian_source;
2262         }
2263
2264         /*
2265          *      Now we are ready to route packet.
2266          */
2267         fl4.flowi4_oif = 0;
2268         fl4.flowi4_iif = dev->ifindex;
2269         fl4.flowi4_mark = skb->mark;
2270         fl4.flowi4_tos = tos;
2271         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2272         fl4.flowi4_flags = 0;
2273         fl4.daddr = daddr;
2274         fl4.saddr = saddr;
2275         fl4.flowi4_uid = sock_net_uid(net, NULL);
2276         fl4.flowi4_multipath_hash = 0;
2277
2278         if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2279                 flkeys = &_flkeys;
2280         } else {
2281                 fl4.flowi4_proto = 0;
2282                 fl4.fl4_sport = 0;
2283                 fl4.fl4_dport = 0;
2284         }
2285
2286         err = fib_lookup(net, &fl4, res, 0);
2287         if (err != 0) {
2288                 if (!IN_DEV_FORWARD(in_dev))
2289                         err = -EHOSTUNREACH;
2290                 goto no_route;
2291         }
2292
2293         if (res->type == RTN_BROADCAST) {
2294                 if (IN_DEV_BFORWARD(in_dev))
2295                         goto make_route;
2296                 /* not do cache if bc_forwarding is enabled */
2297                 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2298                         do_cache = false;
2299                 goto brd_input;
2300         }
2301
2302         if (res->type == RTN_LOCAL) {
2303                 err = fib_validate_source(skb, saddr, daddr, tos,
2304                                           0, dev, in_dev, &itag);
2305                 if (err < 0)
2306                         goto martian_source;
2307                 goto local_input;
2308         }
2309
2310         if (!IN_DEV_FORWARD(in_dev)) {
2311                 err = -EHOSTUNREACH;
2312                 goto no_route;
2313         }
2314         if (res->type != RTN_UNICAST)
2315                 goto martian_destination;
2316
2317 make_route:
2318         err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2319 out:    return err;
2320
2321 brd_input:
2322         if (skb->protocol != htons(ETH_P_IP))
2323                 goto e_inval;
2324
2325         if (!ipv4_is_zeronet(saddr)) {
2326                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2327                                           in_dev, &itag);
2328                 if (err < 0)
2329                         goto martian_source;
2330         }
2331         flags |= RTCF_BROADCAST;
2332         res->type = RTN_BROADCAST;
2333         RT_CACHE_STAT_INC(in_brd);
2334
2335 local_input:
2336         do_cache &= res->fi && !itag;
2337         if (do_cache) {
2338                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2339
2340                 rth = rcu_dereference(nhc->nhc_rth_input);
2341                 if (rt_cache_valid(rth)) {
2342                         skb_dst_set_noref(skb, &rth->dst);
2343                         err = 0;
2344                         goto out;
2345                 }
2346         }
2347
2348         rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2349                            flags | RTCF_LOCAL, res->type,
2350                            IN_DEV_ORCONF(in_dev, NOPOLICY), false);
2351         if (!rth)
2352                 goto e_nobufs;
2353
2354         rth->dst.output= ip_rt_bug;
2355 #ifdef CONFIG_IP_ROUTE_CLASSID
2356         rth->dst.tclassid = itag;
2357 #endif
2358         rth->rt_is_input = 1;
2359
2360         RT_CACHE_STAT_INC(in_slow_tot);
2361         if (res->type == RTN_UNREACHABLE) {
2362                 rth->dst.input= ip_error;
2363                 rth->dst.error= -err;
2364                 rth->rt_flags   &= ~RTCF_LOCAL;
2365         }
2366
2367         if (do_cache) {
2368                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2369
2370                 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2371                 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2372                         WARN_ON(rth->dst.input == lwtunnel_input);
2373                         rth->dst.lwtstate->orig_input = rth->dst.input;
2374                         rth->dst.input = lwtunnel_input;
2375                 }
2376
2377                 if (unlikely(!rt_cache_route(nhc, rth)))
2378                         rt_add_uncached_list(rth);
2379         }
2380         skb_dst_set(skb, &rth->dst);
2381         err = 0;
2382         goto out;
2383
2384 no_route:
2385         RT_CACHE_STAT_INC(in_no_route);
2386         res->type = RTN_UNREACHABLE;
2387         res->fi = NULL;
2388         res->table = NULL;
2389         goto local_input;
2390
2391         /*
2392          *      Do not cache martian addresses: they should be logged (RFC1812)
2393          */
2394 martian_destination:
2395         RT_CACHE_STAT_INC(in_martian_dst);
2396 #ifdef CONFIG_IP_ROUTE_VERBOSE
2397         if (IN_DEV_LOG_MARTIANS(in_dev))
2398                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2399                                      &daddr, &saddr, dev->name);
2400 #endif
2401
2402 e_inval:
2403         err = -EINVAL;
2404         goto out;
2405
2406 e_nobufs:
2407         err = -ENOBUFS;
2408         goto out;
2409
2410 martian_source:
2411         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2412         goto out;
2413 }
2414
2415 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2416                          u8 tos, struct net_device *dev)
2417 {
2418         struct fib_result res;
2419         int err;
2420
2421         tos &= IPTOS_RT_MASK;
2422         rcu_read_lock();
2423         err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2424         rcu_read_unlock();
2425
2426         return err;
2427 }
2428 EXPORT_SYMBOL(ip_route_input_noref);
2429
2430 /* called with rcu_read_lock held */
2431 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2432                        u8 tos, struct net_device *dev, struct fib_result *res)
2433 {
2434         /* Multicast recognition logic is moved from route cache to here.
2435          * The problem was that too many Ethernet cards have broken/missing
2436          * hardware multicast filters :-( As result the host on multicasting
2437          * network acquires a lot of useless route cache entries, sort of
2438          * SDR messages from all the world. Now we try to get rid of them.
2439          * Really, provided software IP multicast filter is organized
2440          * reasonably (at least, hashed), it does not result in a slowdown
2441          * comparing with route cache reject entries.
2442          * Note, that multicast routers are not affected, because
2443          * route cache entry is created eventually.
2444          */
2445         if (ipv4_is_multicast(daddr)) {
2446                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2447                 int our = 0;
2448                 int err = -EINVAL;
2449
2450                 if (!in_dev)
2451                         return err;
2452                 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2453                                       ip_hdr(skb)->protocol);
2454
2455                 /* check l3 master if no match yet */
2456                 if (!our && netif_is_l3_slave(dev)) {
2457                         struct in_device *l3_in_dev;
2458
2459                         l3_in_dev = __in_dev_get_rcu(skb->dev);
2460                         if (l3_in_dev)
2461                                 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2462                                                       ip_hdr(skb)->protocol);
2463                 }
2464
2465                 if (our
2466 #ifdef CONFIG_IP_MROUTE
2467                         ||
2468                     (!ipv4_is_local_multicast(daddr) &&
2469                      IN_DEV_MFORWARD(in_dev))
2470 #endif
2471                    ) {
2472                         err = ip_route_input_mc(skb, daddr, saddr,
2473                                                 tos, dev, our);
2474                 }
2475                 return err;
2476         }
2477
2478         return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2479 }
2480
2481 /* called with rcu_read_lock() */
2482 static struct rtable *__mkroute_output(const struct fib_result *res,
2483                                        const struct flowi4 *fl4, int orig_oif,
2484                                        struct net_device *dev_out,
2485                                        unsigned int flags)
2486 {
2487         struct fib_info *fi = res->fi;
2488         struct fib_nh_exception *fnhe;
2489         struct in_device *in_dev;
2490         u16 type = res->type;
2491         struct rtable *rth;
2492         bool do_cache;
2493
2494         in_dev = __in_dev_get_rcu(dev_out);
2495         if (!in_dev)
2496                 return ERR_PTR(-EINVAL);
2497
2498         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2499                 if (ipv4_is_loopback(fl4->saddr) &&
2500                     !(dev_out->flags & IFF_LOOPBACK) &&
2501                     !netif_is_l3_master(dev_out))
2502                         return ERR_PTR(-EINVAL);
2503
2504         if (ipv4_is_lbcast(fl4->daddr))
2505                 type = RTN_BROADCAST;
2506         else if (ipv4_is_multicast(fl4->daddr))
2507                 type = RTN_MULTICAST;
2508         else if (ipv4_is_zeronet(fl4->daddr))
2509                 return ERR_PTR(-EINVAL);
2510
2511         if (dev_out->flags & IFF_LOOPBACK)
2512                 flags |= RTCF_LOCAL;
2513
2514         do_cache = true;
2515         if (type == RTN_BROADCAST) {
2516                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2517                 fi = NULL;
2518         } else if (type == RTN_MULTICAST) {
2519                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2520                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2521                                      fl4->flowi4_proto))
2522                         flags &= ~RTCF_LOCAL;
2523                 else
2524                         do_cache = false;
2525                 /* If multicast route do not exist use
2526                  * default one, but do not gateway in this case.
2527                  * Yes, it is hack.
2528                  */
2529                 if (fi && res->prefixlen < 4)
2530                         fi = NULL;
2531         } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2532                    (orig_oif != dev_out->ifindex)) {
2533                 /* For local routes that require a particular output interface
2534                  * we do not want to cache the result.  Caching the result
2535                  * causes incorrect behaviour when there are multiple source
2536                  * addresses on the interface, the end result being that if the
2537                  * intended recipient is waiting on that interface for the
2538                  * packet he won't receive it because it will be delivered on
2539                  * the loopback interface and the IP_PKTINFO ipi_ifindex will
2540                  * be set to the loopback interface as well.
2541                  */
2542                 do_cache = false;
2543         }
2544
2545         fnhe = NULL;
2546         do_cache &= fi != NULL;
2547         if (fi) {
2548                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2549                 struct rtable __rcu **prth;
2550
2551                 fnhe = find_exception(nhc, fl4->daddr);
2552                 if (!do_cache)
2553                         goto add;
2554                 if (fnhe) {
2555                         prth = &fnhe->fnhe_rth_output;
2556                 } else {
2557                         if (unlikely(fl4->flowi4_flags &
2558                                      FLOWI_FLAG_KNOWN_NH &&
2559                                      !(nhc->nhc_gw_family &&
2560                                        nhc->nhc_scope == RT_SCOPE_LINK))) {
2561                                 do_cache = false;
2562                                 goto add;
2563                         }
2564                         prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2565                 }
2566                 rth = rcu_dereference(*prth);
2567                 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2568                         return rth;
2569         }
2570
2571 add:
2572         rth = rt_dst_alloc(dev_out, flags, type,
2573                            IN_DEV_ORCONF(in_dev, NOPOLICY),
2574                            IN_DEV_ORCONF(in_dev, NOXFRM));
2575         if (!rth)
2576                 return ERR_PTR(-ENOBUFS);
2577
2578         rth->rt_iif = orig_oif;
2579
2580         RT_CACHE_STAT_INC(out_slow_tot);
2581
2582         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2583                 if (flags & RTCF_LOCAL &&
2584                     !(dev_out->flags & IFF_LOOPBACK)) {
2585                         rth->dst.output = ip_mc_output;
2586                         RT_CACHE_STAT_INC(out_slow_mc);
2587                 }
2588 #ifdef CONFIG_IP_MROUTE
2589                 if (type == RTN_MULTICAST) {
2590                         if (IN_DEV_MFORWARD(in_dev) &&
2591                             !ipv4_is_local_multicast(fl4->daddr)) {
2592                                 rth->dst.input = ip_mr_input;
2593                                 rth->dst.output = ip_mc_output;
2594                         }
2595                 }
2596 #endif
2597         }
2598
2599         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2600         lwtunnel_set_redirect(&rth->dst);
2601
2602         return rth;
2603 }
2604
2605 /*
2606  * Major route resolver routine.
2607  */
2608
2609 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2610                                         const struct sk_buff *skb)
2611 {
2612         __u8 tos = RT_FL_TOS(fl4);
2613         struct fib_result res = {
2614                 .type           = RTN_UNSPEC,
2615                 .fi             = NULL,
2616                 .table          = NULL,
2617                 .tclassid       = 0,
2618         };
2619         struct rtable *rth;
2620
2621         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2622         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2623         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2624                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2625
2626         rcu_read_lock();
2627         rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2628         rcu_read_unlock();
2629
2630         return rth;
2631 }
2632 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2633
2634 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2635                                             struct fib_result *res,
2636                                             const struct sk_buff *skb)
2637 {
2638         struct net_device *dev_out = NULL;
2639         int orig_oif = fl4->flowi4_oif;
2640         unsigned int flags = 0;
2641         struct rtable *rth;
2642         int err;
2643
2644         if (fl4->saddr) {
2645                 if (ipv4_is_multicast(fl4->saddr) ||
2646                     ipv4_is_lbcast(fl4->saddr) ||
2647                     ipv4_is_zeronet(fl4->saddr)) {
2648                         rth = ERR_PTR(-EINVAL);
2649                         goto out;
2650                 }
2651
2652                 rth = ERR_PTR(-ENETUNREACH);
2653
2654                 /* I removed check for oif == dev_out->oif here.
2655                  * It was wrong for two reasons:
2656                  * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2657                  *    is assigned to multiple interfaces.
2658                  * 2. Moreover, we are allowed to send packets with saddr
2659                  *    of another iface. --ANK
2660                  */
2661
2662                 if (fl4->flowi4_oif == 0 &&
2663                     (ipv4_is_multicast(fl4->daddr) ||
2664                      ipv4_is_lbcast(fl4->daddr))) {
2665                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2666                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2667                         if (!dev_out)
2668                                 goto out;
2669
2670                         /* Special hack: user can direct multicasts
2671                          * and limited broadcast via necessary interface
2672                          * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2673                          * This hack is not just for fun, it allows
2674                          * vic,vat and friends to work.
2675                          * They bind socket to loopback, set ttl to zero
2676                          * and expect that it will work.
2677                          * From the viewpoint of routing cache they are broken,
2678                          * because we are not allowed to build multicast path
2679                          * with loopback source addr (look, routing cache
2680                          * cannot know, that ttl is zero, so that packet
2681                          * will not leave this host and route is valid).
2682                          * Luckily, this hack is good workaround.
2683                          */
2684
2685                         fl4->flowi4_oif = dev_out->ifindex;
2686                         goto make_route;
2687                 }
2688
2689                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2690                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2691                         if (!__ip_dev_find(net, fl4->saddr, false))
2692                                 goto out;
2693                 }
2694         }
2695
2696
2697         if (fl4->flowi4_oif) {
2698                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2699                 rth = ERR_PTR(-ENODEV);
2700                 if (!dev_out)
2701                         goto out;
2702
2703                 /* RACE: Check return value of inet_select_addr instead. */
2704                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2705                         rth = ERR_PTR(-ENETUNREACH);
2706                         goto out;
2707                 }
2708                 if (ipv4_is_local_multicast(fl4->daddr) ||
2709                     ipv4_is_lbcast(fl4->daddr) ||
2710                     fl4->flowi4_proto == IPPROTO_IGMP) {
2711                         if (!fl4->saddr)
2712                                 fl4->saddr = inet_select_addr(dev_out, 0,
2713                                                               RT_SCOPE_LINK);
2714                         goto make_route;
2715                 }
2716                 if (!fl4->saddr) {
2717                         if (ipv4_is_multicast(fl4->daddr))
2718                                 fl4->saddr = inet_select_addr(dev_out, 0,
2719                                                               fl4->flowi4_scope);
2720                         else if (!fl4->daddr)
2721                                 fl4->saddr = inet_select_addr(dev_out, 0,
2722                                                               RT_SCOPE_HOST);
2723                 }
2724         }
2725
2726         if (!fl4->daddr) {
2727                 fl4->daddr = fl4->saddr;
2728                 if (!fl4->daddr)
2729                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2730                 dev_out = net->loopback_dev;
2731                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2732                 res->type = RTN_LOCAL;
2733                 flags |= RTCF_LOCAL;
2734                 goto make_route;
2735         }
2736
2737         err = fib_lookup(net, fl4, res, 0);
2738         if (err) {
2739                 res->fi = NULL;
2740                 res->table = NULL;
2741                 if (fl4->flowi4_oif &&
2742                     (ipv4_is_multicast(fl4->daddr) ||
2743                     !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2744                         /* Apparently, routing tables are wrong. Assume,
2745                          * that the destination is on link.
2746                          *
2747                          * WHY? DW.
2748                          * Because we are allowed to send to iface
2749                          * even if it has NO routes and NO assigned
2750                          * addresses. When oif is specified, routing
2751                          * tables are looked up with only one purpose:
2752                          * to catch if destination is gatewayed, rather than
2753                          * direct. Moreover, if MSG_DONTROUTE is set,
2754                          * we send packet, ignoring both routing tables
2755                          * and ifaddr state. --ANK
2756                          *
2757                          *
2758                          * We could make it even if oif is unknown,
2759                          * likely IPv6, but we do not.
2760                          */
2761
2762                         if (fl4->saddr == 0)
2763                                 fl4->saddr = inet_select_addr(dev_out, 0,
2764                                                               RT_SCOPE_LINK);
2765                         res->type = RTN_UNICAST;
2766                         goto make_route;
2767                 }
2768                 rth = ERR_PTR(err);
2769                 goto out;
2770         }
2771
2772         if (res->type == RTN_LOCAL) {
2773                 if (!fl4->saddr) {
2774                         if (res->fi->fib_prefsrc)
2775                                 fl4->saddr = res->fi->fib_prefsrc;
2776                         else
2777                                 fl4->saddr = fl4->daddr;
2778                 }
2779
2780                 /* L3 master device is the loopback for that domain */
2781                 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2782                         net->loopback_dev;
2783
2784                 /* make sure orig_oif points to fib result device even
2785                  * though packet rx/tx happens over loopback or l3mdev
2786                  */
2787                 orig_oif = FIB_RES_OIF(*res);
2788
2789                 fl4->flowi4_oif = dev_out->ifindex;
2790                 flags |= RTCF_LOCAL;
2791                 goto make_route;
2792         }
2793
2794         fib_select_path(net, res, fl4, skb);
2795
2796         dev_out = FIB_RES_DEV(*res);
2797
2798 make_route:
2799         rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2800
2801 out:
2802         return rth;
2803 }
2804
2805 static struct dst_ops ipv4_dst_blackhole_ops = {
2806         .family                 = AF_INET,
2807         .default_advmss         = ipv4_default_advmss,
2808         .neigh_lookup           = ipv4_neigh_lookup,
2809         .check                  = dst_blackhole_check,
2810         .cow_metrics            = dst_blackhole_cow_metrics,
2811         .update_pmtu            = dst_blackhole_update_pmtu,
2812         .redirect               = dst_blackhole_redirect,
2813         .mtu                    = dst_blackhole_mtu,
2814 };
2815
2816 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2817 {
2818         struct rtable *ort = (struct rtable *) dst_orig;
2819         struct rtable *rt;
2820
2821         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2822         if (rt) {
2823                 struct dst_entry *new = &rt->dst;
2824
2825                 new->__use = 1;
2826                 new->input = dst_discard;
2827                 new->output = dst_discard_out;
2828
2829                 new->dev = net->loopback_dev;
2830                 dev_hold(new->dev);
2831
2832                 rt->rt_is_input = ort->rt_is_input;
2833                 rt->rt_iif = ort->rt_iif;
2834                 rt->rt_pmtu = ort->rt_pmtu;
2835                 rt->rt_mtu_locked = ort->rt_mtu_locked;
2836
2837                 rt->rt_genid = rt_genid_ipv4(net);
2838                 rt->rt_flags = ort->rt_flags;
2839                 rt->rt_type = ort->rt_type;
2840                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2841                 rt->rt_gw_family = ort->rt_gw_family;
2842                 if (rt->rt_gw_family == AF_INET)
2843                         rt->rt_gw4 = ort->rt_gw4;
2844                 else if (rt->rt_gw_family == AF_INET6)
2845                         rt->rt_gw6 = ort->rt_gw6;
2846
2847                 INIT_LIST_HEAD(&rt->rt_uncached);
2848         }
2849
2850         dst_release(dst_orig);
2851
2852         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2853 }
2854
2855 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2856                                     const struct sock *sk)
2857 {
2858         struct rtable *rt = __ip_route_output_key(net, flp4);
2859
2860         if (IS_ERR(rt))
2861                 return rt;
2862
2863         if (flp4->flowi4_proto) {
2864                 flp4->flowi4_oif = rt->dst.dev->ifindex;
2865                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2866                                                         flowi4_to_flowi(flp4),
2867                                                         sk, 0);
2868         }
2869
2870         return rt;
2871 }
2872 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2873
2874 struct rtable *ip_route_output_tunnel(struct sk_buff *skb,
2875                                       struct net_device *dev,
2876                                       struct net *net, __be32 *saddr,
2877                                       const struct ip_tunnel_info *info,
2878                                       u8 protocol, bool use_cache)
2879 {
2880 #ifdef CONFIG_DST_CACHE
2881         struct dst_cache *dst_cache;
2882 #endif
2883         struct rtable *rt = NULL;
2884         struct flowi4 fl4;
2885         __u8 tos;
2886
2887 #ifdef CONFIG_DST_CACHE
2888         dst_cache = (struct dst_cache *)&info->dst_cache;
2889         if (use_cache) {
2890                 rt = dst_cache_get_ip4(dst_cache, saddr);
2891                 if (rt)
2892                         return rt;
2893         }
2894 #endif
2895         memset(&fl4, 0, sizeof(fl4));
2896         fl4.flowi4_mark = skb->mark;
2897         fl4.flowi4_proto = protocol;
2898         fl4.daddr = info->key.u.ipv4.dst;
2899         fl4.saddr = info->key.u.ipv4.src;
2900         tos = info->key.tos;
2901         fl4.flowi4_tos = RT_TOS(tos);
2902
2903         rt = ip_route_output_key(net, &fl4);
2904         if (IS_ERR(rt)) {
2905                 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
2906                 return ERR_PTR(-ENETUNREACH);
2907         }
2908         if (rt->dst.dev == dev) { /* is this necessary? */
2909                 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
2910                 ip_rt_put(rt);
2911                 return ERR_PTR(-ELOOP);
2912         }
2913 #ifdef CONFIG_DST_CACHE
2914         if (use_cache)
2915                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2916 #endif
2917         *saddr = fl4.saddr;
2918         return rt;
2919 }
2920 EXPORT_SYMBOL_GPL(ip_route_output_tunnel);
2921
2922 /* called with rcu_read_lock held */
2923 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2924                         struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2925                         struct sk_buff *skb, u32 portid, u32 seq,
2926                         unsigned int flags)
2927 {
2928         struct rtmsg *r;
2929         struct nlmsghdr *nlh;
2930         unsigned long expires = 0;
2931         u32 error;
2932         u32 metrics[RTAX_MAX];
2933
2934         nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2935         if (!nlh)
2936                 return -EMSGSIZE;
2937
2938         r = nlmsg_data(nlh);
2939         r->rtm_family    = AF_INET;
2940         r->rtm_dst_len  = 32;
2941         r->rtm_src_len  = 0;
2942         r->rtm_tos      = fl4 ? fl4->flowi4_tos : 0;
2943         r->rtm_table    = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2944         if (nla_put_u32(skb, RTA_TABLE, table_id))
2945                 goto nla_put_failure;
2946         r->rtm_type     = rt->rt_type;
2947         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2948         r->rtm_protocol = RTPROT_UNSPEC;
2949         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2950         if (rt->rt_flags & RTCF_NOTIFY)
2951                 r->rtm_flags |= RTM_F_NOTIFY;
2952         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2953                 r->rtm_flags |= RTCF_DOREDIRECT;
2954
2955         if (nla_put_in_addr(skb, RTA_DST, dst))
2956                 goto nla_put_failure;
2957         if (src) {
2958                 r->rtm_src_len = 32;
2959                 if (nla_put_in_addr(skb, RTA_SRC, src))
2960                         goto nla_put_failure;
2961         }
2962         if (rt->dst.dev &&
2963             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2964                 goto nla_put_failure;
2965         if (rt->dst.lwtstate &&
2966             lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2967                 goto nla_put_failure;
2968 #ifdef CONFIG_IP_ROUTE_CLASSID
2969         if (rt->dst.tclassid &&
2970             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2971                 goto nla_put_failure;
2972 #endif
2973         if (fl4 && !rt_is_input_route(rt) &&
2974             fl4->saddr != src) {
2975                 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2976                         goto nla_put_failure;
2977         }
2978         if (rt->rt_uses_gateway) {
2979                 if (rt->rt_gw_family == AF_INET &&
2980                     nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2981                         goto nla_put_failure;
2982                 } else if (rt->rt_gw_family == AF_INET6) {
2983                         int alen = sizeof(struct in6_addr);
2984                         struct nlattr *nla;
2985                         struct rtvia *via;
2986
2987                         nla = nla_reserve(skb, RTA_VIA, alen + 2);
2988                         if (!nla)
2989                                 goto nla_put_failure;
2990
2991                         via = nla_data(nla);
2992                         via->rtvia_family = AF_INET6;
2993                         memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2994                 }
2995         }
2996
2997         expires = rt->dst.expires;
2998         if (expires) {
2999                 unsigned long now = jiffies;
3000
3001                 if (time_before(now, expires))
3002                         expires -= now;
3003                 else
3004                         expires = 0;
3005         }
3006
3007         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3008         if (rt->rt_pmtu && expires)
3009                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3010         if (rt->rt_mtu_locked && expires)
3011                 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3012         if (rtnetlink_put_metrics(skb, metrics) < 0)
3013                 goto nla_put_failure;
3014
3015         if (fl4) {
3016                 if (fl4->flowi4_mark &&
3017                     nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3018                         goto nla_put_failure;
3019
3020                 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3021                     nla_put_u32(skb, RTA_UID,
3022                                 from_kuid_munged(current_user_ns(),
3023                                                  fl4->flowi4_uid)))
3024                         goto nla_put_failure;
3025
3026                 if (rt_is_input_route(rt)) {
3027 #ifdef CONFIG_IP_MROUTE
3028                         if (ipv4_is_multicast(dst) &&
3029                             !ipv4_is_local_multicast(dst) &&
3030                             IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
3031                                 int err = ipmr_get_route(net, skb,
3032                                                          fl4->saddr, fl4->daddr,
3033                                                          r, portid);
3034
3035                                 if (err <= 0) {
3036                                         if (err == 0)
3037                                                 return 0;
3038                                         goto nla_put_failure;
3039                                 }
3040                         } else
3041 #endif
3042                                 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3043                                         goto nla_put_failure;
3044                 }
3045         }
3046
3047         error = rt->dst.error;
3048
3049         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3050                 goto nla_put_failure;
3051
3052         nlmsg_end(skb, nlh);
3053         return 0;
3054
3055 nla_put_failure:
3056         nlmsg_cancel(skb, nlh);
3057         return -EMSGSIZE;
3058 }
3059
3060 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3061                             struct netlink_callback *cb, u32 table_id,
3062                             struct fnhe_hash_bucket *bucket, int genid,
3063                             int *fa_index, int fa_start, unsigned int flags)
3064 {
3065         int i;
3066
3067         for (i = 0; i < FNHE_HASH_SIZE; i++) {
3068                 struct fib_nh_exception *fnhe;
3069
3070                 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3071                      fnhe = rcu_dereference(fnhe->fnhe_next)) {
3072                         struct rtable *rt;
3073                         int err;
3074
3075                         if (*fa_index < fa_start)
3076                                 goto next;
3077
3078                         if (fnhe->fnhe_genid != genid)
3079                                 goto next;
3080
3081                         if (fnhe->fnhe_expires &&
3082                             time_after(jiffies, fnhe->fnhe_expires))
3083                                 goto next;
3084
3085                         rt = rcu_dereference(fnhe->fnhe_rth_input);
3086                         if (!rt)
3087                                 rt = rcu_dereference(fnhe->fnhe_rth_output);
3088                         if (!rt)
3089                                 goto next;
3090
3091                         err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3092                                            table_id, NULL, skb,
3093                                            NETLINK_CB(cb->skb).portid,
3094                                            cb->nlh->nlmsg_seq, flags);
3095                         if (err)
3096                                 return err;
3097 next:
3098                         (*fa_index)++;
3099                 }
3100         }
3101
3102         return 0;
3103 }
3104
3105 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3106                        u32 table_id, struct fib_info *fi,
3107                        int *fa_index, int fa_start, unsigned int flags)
3108 {
3109         struct net *net = sock_net(cb->skb->sk);
3110         int nhsel, genid = fnhe_genid(net);
3111
3112         for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3113                 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3114                 struct fnhe_hash_bucket *bucket;
3115                 int err;
3116
3117                 if (nhc->nhc_flags & RTNH_F_DEAD)
3118                         continue;
3119
3120                 rcu_read_lock();
3121                 bucket = rcu_dereference(nhc->nhc_exceptions);
3122                 err = 0;
3123                 if (bucket)
3124                         err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3125                                                genid, fa_index, fa_start,
3126                                                flags);
3127                 rcu_read_unlock();
3128                 if (err)
3129                         return err;
3130         }
3131
3132         return 0;
3133 }
3134
3135 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3136                                                    u8 ip_proto, __be16 sport,
3137                                                    __be16 dport)
3138 {
3139         struct sk_buff *skb;
3140         struct iphdr *iph;
3141
3142         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3143         if (!skb)
3144                 return NULL;
3145
3146         /* Reserve room for dummy headers, this skb can pass
3147          * through good chunk of routing engine.
3148          */
3149         skb_reset_mac_header(skb);
3150         skb_reset_network_header(skb);
3151         skb->protocol = htons(ETH_P_IP);
3152         iph = skb_put(skb, sizeof(struct iphdr));
3153         iph->protocol = ip_proto;
3154         iph->saddr = src;
3155         iph->daddr = dst;
3156         iph->version = 0x4;
3157         iph->frag_off = 0;
3158         iph->ihl = 0x5;
3159         skb_set_transport_header(skb, skb->len);
3160
3161         switch (iph->protocol) {
3162         case IPPROTO_UDP: {
3163                 struct udphdr *udph;
3164
3165                 udph = skb_put_zero(skb, sizeof(struct udphdr));
3166                 udph->source = sport;
3167                 udph->dest = dport;
3168                 udph->len = sizeof(struct udphdr);
3169                 udph->check = 0;
3170                 break;
3171         }
3172         case IPPROTO_TCP: {
3173                 struct tcphdr *tcph;
3174
3175                 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3176                 tcph->source    = sport;
3177                 tcph->dest      = dport;
3178                 tcph->doff      = sizeof(struct tcphdr) / 4;
3179                 tcph->rst = 1;
3180                 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3181                                             src, dst, 0);
3182                 break;
3183         }
3184         case IPPROTO_ICMP: {
3185                 struct icmphdr *icmph;
3186
3187                 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3188                 icmph->type = ICMP_ECHO;
3189                 icmph->code = 0;
3190         }
3191         }
3192
3193         return skb;
3194 }
3195
3196 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3197                                        const struct nlmsghdr *nlh,
3198                                        struct nlattr **tb,
3199                                        struct netlink_ext_ack *extack)
3200 {
3201         struct rtmsg *rtm;
3202         int i, err;
3203
3204         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3205                 NL_SET_ERR_MSG(extack,
3206                                "ipv4: Invalid header for route get request");
3207                 return -EINVAL;
3208         }
3209
3210         if (!netlink_strict_get_check(skb))
3211                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3212                                               rtm_ipv4_policy, extack);
3213
3214         rtm = nlmsg_data(nlh);
3215         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3216             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3217             rtm->rtm_table || rtm->rtm_protocol ||
3218             rtm->rtm_scope || rtm->rtm_type) {
3219                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3220                 return -EINVAL;
3221         }
3222
3223         if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3224                                RTM_F_LOOKUP_TABLE |
3225                                RTM_F_FIB_MATCH)) {
3226                 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3227                 return -EINVAL;
3228         }
3229
3230         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3231                                             rtm_ipv4_policy, extack);
3232         if (err)
3233                 return err;
3234
3235         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3236             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3237                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3238                 return -EINVAL;
3239         }
3240
3241         for (i = 0; i <= RTA_MAX; i++) {
3242                 if (!tb[i])
3243                         continue;
3244
3245                 switch (i) {
3246                 case RTA_IIF:
3247                 case RTA_OIF:
3248                 case RTA_SRC:
3249                 case RTA_DST:
3250                 case RTA_IP_PROTO:
3251                 case RTA_SPORT:
3252                 case RTA_DPORT:
3253                 case RTA_MARK:
3254                 case RTA_UID:
3255                         break;
3256                 default:
3257                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3258                         return -EINVAL;
3259                 }
3260         }
3261
3262         return 0;
3263 }
3264
3265 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3266                              struct netlink_ext_ack *extack)
3267 {
3268         struct net *net = sock_net(in_skb->sk);
3269         struct nlattr *tb[RTA_MAX+1];
3270         u32 table_id = RT_TABLE_MAIN;
3271         __be16 sport = 0, dport = 0;
3272         struct fib_result res = {};
3273         u8 ip_proto = IPPROTO_UDP;
3274         struct rtable *rt = NULL;
3275         struct sk_buff *skb;
3276         struct rtmsg *rtm;
3277         struct flowi4 fl4 = {};
3278         __be32 dst = 0;
3279         __be32 src = 0;
3280         kuid_t uid;
3281         u32 iif;
3282         int err;
3283         int mark;
3284
3285         err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3286         if (err < 0)
3287                 return err;
3288
3289         rtm = nlmsg_data(nlh);
3290         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3291         dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3292         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3293         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3294         if (tb[RTA_UID])
3295                 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3296         else
3297                 uid = (iif ? INVALID_UID : current_uid());
3298
3299         if (tb[RTA_IP_PROTO]) {
3300                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3301                                                   &ip_proto, AF_INET, extack);
3302                 if (err)
3303                         return err;
3304         }
3305
3306         if (tb[RTA_SPORT])
3307                 sport = nla_get_be16(tb[RTA_SPORT]);
3308
3309         if (tb[RTA_DPORT])
3310                 dport = nla_get_be16(tb[RTA_DPORT]);
3311
3312         skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3313         if (!skb)
3314                 return -ENOBUFS;
3315
3316         fl4.daddr = dst;
3317         fl4.saddr = src;
3318         fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3319         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3320         fl4.flowi4_mark = mark;
3321         fl4.flowi4_uid = uid;
3322         if (sport)
3323                 fl4.fl4_sport = sport;
3324         if (dport)
3325                 fl4.fl4_dport = dport;
3326         fl4.flowi4_proto = ip_proto;
3327
3328         rcu_read_lock();
3329
3330         if (iif) {
3331                 struct net_device *dev;
3332
3333                 dev = dev_get_by_index_rcu(net, iif);
3334                 if (!dev) {
3335                         err = -ENODEV;
3336                         goto errout_rcu;
3337                 }
3338
3339                 fl4.flowi4_iif = iif; /* for rt_fill_info */
3340                 skb->dev        = dev;
3341                 skb->mark       = mark;
3342                 err = ip_route_input_rcu(skb, dst, src,
3343                                          rtm->rtm_tos & IPTOS_RT_MASK, dev,
3344                                          &res);
3345
3346                 rt = skb_rtable(skb);
3347                 if (err == 0 && rt->dst.error)
3348                         err = -rt->dst.error;
3349         } else {
3350                 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3351                 skb->dev = net->loopback_dev;
3352                 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3353                 err = 0;
3354                 if (IS_ERR(rt))
3355                         err = PTR_ERR(rt);
3356                 else
3357                         skb_dst_set(skb, &rt->dst);
3358         }
3359
3360         if (err)
3361                 goto errout_rcu;
3362
3363         if (rtm->rtm_flags & RTM_F_NOTIFY)
3364                 rt->rt_flags |= RTCF_NOTIFY;
3365
3366         if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3367                 table_id = res.table ? res.table->tb_id : 0;
3368
3369         /* reset skb for netlink reply msg */
3370         skb_trim(skb, 0);
3371         skb_reset_network_header(skb);
3372         skb_reset_transport_header(skb);
3373         skb_reset_mac_header(skb);
3374
3375         if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3376                 struct fib_rt_info fri;
3377
3378                 if (!res.fi) {
3379                         err = fib_props[res.type].error;
3380                         if (!err)
3381                                 err = -EHOSTUNREACH;
3382                         goto errout_rcu;
3383                 }
3384                 fri.fi = res.fi;
3385                 fri.tb_id = table_id;
3386                 fri.dst = res.prefix;
3387                 fri.dst_len = res.prefixlen;
3388                 fri.tos = fl4.flowi4_tos;
3389                 fri.type = rt->rt_type;
3390                 fri.offload = 0;
3391                 fri.trap = 0;
3392                 fri.offload_failed = 0;
3393                 if (res.fa_head) {
3394                         struct fib_alias *fa;
3395
3396                         hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3397                                 u8 slen = 32 - fri.dst_len;
3398
3399                                 if (fa->fa_slen == slen &&
3400                                     fa->tb_id == fri.tb_id &&
3401                                     fa->fa_tos == fri.tos &&
3402                                     fa->fa_info == res.fi &&
3403                                     fa->fa_type == fri.type) {
3404                                         fri.offload = fa->offload;
3405                                         fri.trap = fa->trap;
3406                                         break;
3407                                 }
3408                         }
3409                 }
3410                 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3411                                     nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3412         } else {
3413                 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3414                                    NETLINK_CB(in_skb).portid,
3415                                    nlh->nlmsg_seq, 0);
3416         }
3417         if (err < 0)
3418                 goto errout_rcu;
3419
3420         rcu_read_unlock();
3421
3422         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3423
3424 errout_free:
3425         return err;
3426 errout_rcu:
3427         rcu_read_unlock();
3428         kfree_skb(skb);
3429         goto errout_free;
3430 }
3431
3432 void ip_rt_multicast_event(struct in_device *in_dev)
3433 {
3434         rt_cache_flush(dev_net(in_dev->dev));
3435 }
3436
3437 #ifdef CONFIG_SYSCTL
3438 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3439 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
3440 static int ip_rt_gc_elasticity __read_mostly    = 8;
3441 static int ip_min_valid_pmtu __read_mostly      = IPV4_MIN_MTU;
3442
3443 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3444                 void *buffer, size_t *lenp, loff_t *ppos)
3445 {
3446         struct net *net = (struct net *)__ctl->extra1;
3447
3448         if (write) {
3449                 rt_cache_flush(net);
3450                 fnhe_genid_bump(net);
3451                 return 0;
3452         }
3453
3454         return -EINVAL;
3455 }
3456
3457 static struct ctl_table ipv4_route_table[] = {
3458         {
3459                 .procname       = "gc_thresh",
3460                 .data           = &ipv4_dst_ops.gc_thresh,
3461                 .maxlen         = sizeof(int),
3462                 .mode           = 0644,
3463                 .proc_handler   = proc_dointvec,
3464         },
3465         {
3466                 .procname       = "max_size",
3467                 .data           = &ip_rt_max_size,
3468                 .maxlen         = sizeof(int),
3469                 .mode           = 0644,
3470                 .proc_handler   = proc_dointvec,
3471         },
3472         {
3473                 /*  Deprecated. Use gc_min_interval_ms */
3474
3475                 .procname       = "gc_min_interval",
3476                 .data           = &ip_rt_gc_min_interval,
3477                 .maxlen         = sizeof(int),
3478                 .mode           = 0644,
3479                 .proc_handler   = proc_dointvec_jiffies,
3480         },
3481         {
3482                 .procname       = "gc_min_interval_ms",
3483                 .data           = &ip_rt_gc_min_interval,
3484                 .maxlen         = sizeof(int),
3485                 .mode           = 0644,
3486                 .proc_handler   = proc_dointvec_ms_jiffies,
3487         },
3488         {
3489                 .procname       = "gc_timeout",
3490                 .data           = &ip_rt_gc_timeout,
3491                 .maxlen         = sizeof(int),
3492                 .mode           = 0644,
3493                 .proc_handler   = proc_dointvec_jiffies,
3494         },
3495         {
3496                 .procname       = "gc_interval",
3497                 .data           = &ip_rt_gc_interval,
3498                 .maxlen         = sizeof(int),
3499                 .mode           = 0644,
3500                 .proc_handler   = proc_dointvec_jiffies,
3501         },
3502         {
3503                 .procname       = "redirect_load",
3504                 .data           = &ip_rt_redirect_load,
3505                 .maxlen         = sizeof(int),
3506                 .mode           = 0644,
3507                 .proc_handler   = proc_dointvec,
3508         },
3509         {
3510                 .procname       = "redirect_number",
3511                 .data           = &ip_rt_redirect_number,
3512                 .maxlen         = sizeof(int),
3513                 .mode           = 0644,
3514                 .proc_handler   = proc_dointvec,
3515         },
3516         {
3517                 .procname       = "redirect_silence",
3518                 .data           = &ip_rt_redirect_silence,
3519                 .maxlen         = sizeof(int),
3520                 .mode           = 0644,
3521                 .proc_handler   = proc_dointvec,
3522         },
3523         {
3524                 .procname       = "error_cost",
3525                 .data           = &ip_rt_error_cost,
3526                 .maxlen         = sizeof(int),
3527                 .mode           = 0644,
3528                 .proc_handler   = proc_dointvec,
3529         },
3530         {
3531                 .procname       = "error_burst",
3532                 .data           = &ip_rt_error_burst,
3533                 .maxlen         = sizeof(int),
3534                 .mode           = 0644,
3535                 .proc_handler   = proc_dointvec,
3536         },
3537         {
3538                 .procname       = "gc_elasticity",
3539                 .data           = &ip_rt_gc_elasticity,
3540                 .maxlen         = sizeof(int),
3541                 .mode           = 0644,
3542                 .proc_handler   = proc_dointvec,
3543         },
3544         {
3545                 .procname       = "mtu_expires",
3546                 .data           = &ip_rt_mtu_expires,
3547                 .maxlen         = sizeof(int),
3548                 .mode           = 0644,
3549                 .proc_handler   = proc_dointvec_jiffies,
3550         },
3551         {
3552                 .procname       = "min_pmtu",
3553                 .data           = &ip_rt_min_pmtu,
3554                 .maxlen         = sizeof(int),
3555                 .mode           = 0644,
3556                 .proc_handler   = proc_dointvec_minmax,
3557                 .extra1         = &ip_min_valid_pmtu,
3558         },
3559         {
3560                 .procname       = "min_adv_mss",
3561                 .data           = &ip_rt_min_advmss,
3562                 .maxlen         = sizeof(int),
3563                 .mode           = 0644,
3564                 .proc_handler   = proc_dointvec,
3565         },
3566         { }
3567 };
3568
3569 static const char ipv4_route_flush_procname[] = "flush";
3570
3571 static struct ctl_table ipv4_route_flush_table[] = {
3572         {
3573                 .procname       = ipv4_route_flush_procname,
3574                 .maxlen         = sizeof(int),
3575                 .mode           = 0200,
3576                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3577         },
3578         { },
3579 };
3580
3581 static __net_init int sysctl_route_net_init(struct net *net)
3582 {
3583         struct ctl_table *tbl;
3584
3585         tbl = ipv4_route_flush_table;
3586         if (!net_eq(net, &init_net)) {
3587                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3588                 if (!tbl)
3589                         goto err_dup;
3590
3591                 /* Don't export non-whitelisted sysctls to unprivileged users */
3592                 if (net->user_ns != &init_user_ns) {
3593                         if (tbl[0].procname != ipv4_route_flush_procname)
3594                                 tbl[0].procname = NULL;
3595                 }
3596         }
3597         tbl[0].extra1 = net;
3598
3599         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3600         if (!net->ipv4.route_hdr)
3601                 goto err_reg;
3602         return 0;
3603
3604 err_reg:
3605         if (tbl != ipv4_route_flush_table)
3606                 kfree(tbl);
3607 err_dup:
3608         return -ENOMEM;
3609 }
3610
3611 static __net_exit void sysctl_route_net_exit(struct net *net)
3612 {
3613         struct ctl_table *tbl;
3614
3615         tbl = net->ipv4.route_hdr->ctl_table_arg;
3616         unregister_net_sysctl_table(net->ipv4.route_hdr);
3617         BUG_ON(tbl == ipv4_route_flush_table);
3618         kfree(tbl);
3619 }
3620
3621 static __net_initdata struct pernet_operations sysctl_route_ops = {
3622         .init = sysctl_route_net_init,
3623         .exit = sysctl_route_net_exit,
3624 };
3625 #endif
3626
3627 static __net_init int rt_genid_init(struct net *net)
3628 {
3629         atomic_set(&net->ipv4.rt_genid, 0);
3630         atomic_set(&net->fnhe_genid, 0);
3631         atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3632         return 0;
3633 }
3634
3635 static __net_initdata struct pernet_operations rt_genid_ops = {
3636         .init = rt_genid_init,
3637 };
3638
3639 static int __net_init ipv4_inetpeer_init(struct net *net)
3640 {
3641         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3642
3643         if (!bp)
3644                 return -ENOMEM;
3645         inet_peer_base_init(bp);
3646         net->ipv4.peers = bp;
3647         return 0;
3648 }
3649
3650 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3651 {
3652         struct inet_peer_base *bp = net->ipv4.peers;
3653
3654         net->ipv4.peers = NULL;
3655         inetpeer_invalidate_tree(bp);
3656         kfree(bp);
3657 }
3658
3659 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3660         .init   =       ipv4_inetpeer_init,
3661         .exit   =       ipv4_inetpeer_exit,
3662 };
3663
3664 #ifdef CONFIG_IP_ROUTE_CLASSID
3665 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3666 #endif /* CONFIG_IP_ROUTE_CLASSID */
3667
3668 int __init ip_rt_init(void)
3669 {
3670         void *idents_hash;
3671         int cpu;
3672
3673         /* For modern hosts, this will use 2 MB of memory */
3674         idents_hash = alloc_large_system_hash("IP idents",
3675                                               sizeof(*ip_idents) + sizeof(*ip_tstamps),
3676                                               0,
3677                                               16, /* one bucket per 64 KB */
3678                                               HASH_ZERO,
3679                                               NULL,
3680                                               &ip_idents_mask,
3681                                               2048,
3682                                               256*1024);
3683
3684         ip_idents = idents_hash;
3685
3686         prandom_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3687
3688         ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3689
3690         for_each_possible_cpu(cpu) {
3691                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3692
3693                 INIT_LIST_HEAD(&ul->head);
3694                 spin_lock_init(&ul->lock);
3695         }
3696 #ifdef CONFIG_IP_ROUTE_CLASSID
3697         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3698         if (!ip_rt_acct)
3699                 panic("IP: failed to allocate ip_rt_acct\n");
3700 #endif
3701
3702         ipv4_dst_ops.kmem_cachep =
3703                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3704                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3705
3706         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3707
3708         if (dst_entries_init(&ipv4_dst_ops) < 0)
3709                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3710
3711         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3712                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3713
3714         ipv4_dst_ops.gc_thresh = ~0;
3715         ip_rt_max_size = INT_MAX;
3716
3717         devinet_init();
3718         ip_fib_init();
3719
3720         if (ip_rt_proc_init())
3721                 pr_err("Unable to create route proc files\n");
3722 #ifdef CONFIG_XFRM
3723         xfrm_init();
3724         xfrm4_init();
3725 #endif
3726         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3727                       RTNL_FLAG_DOIT_UNLOCKED);
3728
3729 #ifdef CONFIG_SYSCTL
3730         register_pernet_subsys(&sysctl_route_ops);
3731 #endif
3732         register_pernet_subsys(&rt_genid_ops);
3733         register_pernet_subsys(&ipv4_inetpeer_ops);
3734         return 0;
3735 }
3736
3737 #ifdef CONFIG_SYSCTL
3738 /*
3739  * We really need to sanitize the damn ipv4 init order, then all
3740  * this nonsense will go away.
3741  */
3742 void __init ip_static_sysctl_init(void)
3743 {
3744         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3745 }
3746 #endif