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 ip_rt_fix_tos(struct flowi4 *fl4)
 510 {
 511         __u8 tos = RT_FL_TOS(fl4);
 512
 513         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
 514         fl4->flowi4_scope = tos & RTO_ONLINK ?
 515                             RT_SCOPE_LINK : RT_SCOPE_UNIVERSE;
 516 }
 517
 518 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
 519                              const struct sock *sk,
 520                              const struct iphdr *iph,
 521                              int oif, u8 tos,
 522                              u8 prot, u32 mark, int flow_flags)
 523 {
 524         if (sk) {
 525                 const struct inet_sock *inet = inet_sk(sk);
 526
 527                 oif = sk->sk_bound_dev_if;
 528                 mark = sk->sk_mark;
 529                 tos = RT_CONN_FLAGS(sk);
 530                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 531         }
 532         flowi4_init_output(fl4, oif, mark, tos,
 533                            RT_SCOPE_UNIVERSE, prot,
 534                            flow_flags,
 535                            iph->daddr, iph->saddr, 0, 0,
 536                            sock_net_uid(net, sk));
 537 }
 538
 539 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 540                                const struct sock *sk)
 541 {
 542         const struct net *net = dev_net(skb->dev);
 543         const struct iphdr *iph = ip_hdr(skb);
 544         int oif = skb->dev->ifindex;
 545         u8 tos = RT_TOS(iph->tos);
 546         u8 prot = iph->protocol;
 547         u32 mark = skb->mark;
 548
 549         __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
 550 }
 551
 552 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 553 {
 554         const struct inet_sock *inet = inet_sk(sk);
 555         const struct ip_options_rcu *inet_opt;
 556         __be32 daddr = inet->inet_daddr;
 557
 558         rcu_read_lock();
 559         inet_opt = rcu_dereference(inet->inet_opt);
 560         if (inet_opt && inet_opt->opt.srr)
 561                 daddr = inet_opt->opt.faddr;
 562         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 563                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 564                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 565                            inet_sk_flowi_flags(sk),
 566                            daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
 567         rcu_read_unlock();
 568 }
 569
 570 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 571                                  const struct sk_buff *skb)
 572 {
 573         if (skb)
 574                 build_skb_flow_key(fl4, skb, sk);
 575         else
 576                 build_sk_flow_key(fl4, sk);
 577 }
 578
 579 static DEFINE_SPINLOCK(fnhe_lock);
 580
 581 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 582 {
 583         struct rtable *rt;
 584
 585         rt = rcu_dereference(fnhe->fnhe_rth_input);
 586         if (rt) {
 587                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 588                 dst_dev_put(&rt->dst);
 589                 dst_release(&rt->dst);
 590         }
 591         rt = rcu_dereference(fnhe->fnhe_rth_output);
 592         if (rt) {
 593                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 594                 dst_dev_put(&rt->dst);
 595                 dst_release(&rt->dst);
 596         }
 597 }
 598
 599 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash)
 600 {
 601         struct fib_nh_exception __rcu **fnhe_p, **oldest_p;
 602         struct fib_nh_exception *fnhe, *oldest = NULL;
 603
 604         for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) {
 605                 fnhe = rcu_dereference_protected(*fnhe_p,
 606                                                  lockdep_is_held(&fnhe_lock));
 607                 if (!fnhe)
 608                         break;
 609                 if (!oldest ||
 610                     time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) {
 611                         oldest = fnhe;
 612                         oldest_p = fnhe_p;
 613                 }
 614         }
 615         fnhe_flush_routes(oldest);
 616         *oldest_p = oldest->fnhe_next;
 617         kfree_rcu(oldest, rcu);
 618 }
 619
 620 static u32 fnhe_hashfun(__be32 daddr)
 621 {
 622         static siphash_key_t fnhe_hash_key __read_mostly;
 623         u64 hval;
 624
 625         net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key));
 626         hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key);
 627         return hash_64(hval, FNHE_HASH_SHIFT);
 628 }
 629
 630 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 631 {
 632         rt->rt_pmtu = fnhe->fnhe_pmtu;
 633         rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
 634         rt->dst.expires = fnhe->fnhe_expires;
 635
 636         if (fnhe->fnhe_gw) {
 637                 rt->rt_flags |= RTCF_REDIRECTED;
 638                 rt->rt_uses_gateway = 1;
 639                 rt->rt_gw_family = AF_INET;
 640                 rt->rt_gw4 = fnhe->fnhe_gw;
 641         }
 642 }
 643
 644 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
 645                                   __be32 gw, u32 pmtu, bool lock,
 646                                   unsigned long expires)
 647 {
 648         struct fnhe_hash_bucket *hash;
 649         struct fib_nh_exception *fnhe;
 650         struct rtable *rt;
 651         u32 genid, hval;
 652         unsigned int i;
 653         int depth;
 654
 655         genid = fnhe_genid(dev_net(nhc->nhc_dev));
 656         hval = fnhe_hashfun(daddr);
 657
 658         spin_lock_bh(&fnhe_lock);
 659
 660         hash = rcu_dereference(nhc->nhc_exceptions);
 661         if (!hash) {
 662                 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
 663                 if (!hash)
 664                         goto out_unlock;
 665                 rcu_assign_pointer(nhc->nhc_exceptions, hash);
 666         }
 667
 668         hash += hval;
 669
 670         depth = 0;
 671         for (fnhe = rcu_dereference(hash->chain); fnhe;
 672              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 673                 if (fnhe->fnhe_daddr == daddr)
 674                         break;
 675                 depth++;
 676         }
 677
 678         if (fnhe) {
 679                 if (fnhe->fnhe_genid != genid)
 680                         fnhe->fnhe_genid = genid;
 681                 if (gw)
 682                         fnhe->fnhe_gw = gw;
 683                 if (pmtu) {
 684                         fnhe->fnhe_pmtu = pmtu;
 685                         fnhe->fnhe_mtu_locked = lock;
 686                 }
 687                 fnhe->fnhe_expires = max(1UL, expires);
 688                 /* Update all cached dsts too */
 689                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 690                 if (rt)
 691                         fill_route_from_fnhe(rt, fnhe);
 692                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 693                 if (rt)
 694                         fill_route_from_fnhe(rt, fnhe);
 695         } else {
 696                 /* Randomize max depth to avoid some side channels attacks. */
 697                 int max_depth = FNHE_RECLAIM_DEPTH +
 698                                 prandom_u32_max(FNHE_RECLAIM_DEPTH);
 699
 700                 while (depth > max_depth) {
 701                         fnhe_remove_oldest(hash);
 702                         depth--;
 703                 }
 704
 705                 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 706                 if (!fnhe)
 707                         goto out_unlock;
 708
 709                 fnhe->fnhe_next = hash->chain;
 710
 711                 fnhe->fnhe_genid = genid;
 712                 fnhe->fnhe_daddr = daddr;
 713                 fnhe->fnhe_gw = gw;
 714                 fnhe->fnhe_pmtu = pmtu;
 715                 fnhe->fnhe_mtu_locked = lock;
 716                 fnhe->fnhe_expires = max(1UL, expires);
 717
 718                 rcu_assign_pointer(hash->chain, fnhe);
 719
 720                 /* Exception created; mark the cached routes for the nexthop
 721                  * stale, so anyone caching it rechecks if this exception
 722                  * applies to them.
 723                  */
 724                 rt = rcu_dereference(nhc->nhc_rth_input);
 725                 if (rt)
 726                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 727
 728                 for_each_possible_cpu(i) {
 729                         struct rtable __rcu **prt;
 730
 731                         prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
 732                         rt = rcu_dereference(*prt);
 733                         if (rt)
 734                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 735                 }
 736         }
 737
 738         fnhe->fnhe_stamp = jiffies;
 739
 740 out_unlock:
 741         spin_unlock_bh(&fnhe_lock);
 742 }
 743
 744 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 745                              bool kill_route)
 746 {
 747         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 748         __be32 old_gw = ip_hdr(skb)->saddr;
 749         struct net_device *dev = skb->dev;
 750         struct in_device *in_dev;
 751         struct fib_result res;
 752         struct neighbour *n;
 753         struct net *net;
 754
 755         switch (icmp_hdr(skb)->code & 7) {
 756         case ICMP_REDIR_NET:
 757         case ICMP_REDIR_NETTOS:
 758         case ICMP_REDIR_HOST:
 759         case ICMP_REDIR_HOSTTOS:
 760                 break;
 761
 762         default:
 763                 return;
 764         }
 765
 766         if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
 767                 return;
 768
 769         in_dev = __in_dev_get_rcu(dev);
 770         if (!in_dev)
 771                 return;
 772
 773         net = dev_net(dev);
 774         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 775             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 776             ipv4_is_zeronet(new_gw))
 777                 goto reject_redirect;
 778
 779         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 780                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 781                         goto reject_redirect;
 782                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 783                         goto reject_redirect;
 784         } else {
 785                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 786                         goto reject_redirect;
 787         }
 788
 789         n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
 790         if (!n)
 791                 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
 792         if (!IS_ERR(n)) {
 793                 if (!(n->nud_state & NUD_VALID)) {
 794                         neigh_event_send(n, NULL);
 795                 } else {
 796                         if (fib_lookup(net, fl4, &res, 0) == 0) {
 797                                 struct fib_nh_common *nhc;
 798
 799                                 fib_select_path(net, &res, fl4, skb);
 800                                 nhc = FIB_RES_NHC(res);
 801                                 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
 802                                                 0, false,
 803                                                 jiffies + ip_rt_gc_timeout);
 804                         }
 805                         if (kill_route)
 806                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 807                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 808                 }
 809                 neigh_release(n);
 810         }
 811         return;
 812
 813 reject_redirect:
 814 #ifdef CONFIG_IP_ROUTE_VERBOSE
 815         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 816                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 817                 __be32 daddr = iph->daddr;
 818                 __be32 saddr = iph->saddr;
 819
 820                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 821                                      "  Advised path = %pI4 -> %pI4\n",
 822                                      &old_gw, dev->name, &new_gw,
 823                                      &saddr, &daddr);
 824         }
 825 #endif
 826         ;
 827 }
 828
 829 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 830 {
 831         struct rtable *rt;
 832         struct flowi4 fl4;
 833         const struct iphdr *iph = (const struct iphdr *) skb->data;
 834         struct net *net = dev_net(skb->dev);
 835         int oif = skb->dev->ifindex;
 836         u8 tos = RT_TOS(iph->tos);
 837         u8 prot = iph->protocol;
 838         u32 mark = skb->mark;
 839
 840         rt = (struct rtable *) dst;
 841
 842         __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
 843         ip_rt_fix_tos(&fl4);
 844         __ip_do_redirect(rt, skb, &fl4, true);
 845 }
 846
 847 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 848 {
 849         struct rtable *rt = (struct rtable *)dst;
 850         struct dst_entry *ret = dst;
 851
 852         if (rt) {
 853                 if (dst->obsolete > 0) {
 854                         ip_rt_put(rt);
 855                         ret = NULL;
 856                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 857                            rt->dst.expires) {
 858                         ip_rt_put(rt);
 859                         ret = NULL;
 860                 }
 861         }
 862         return ret;
 863 }
 864
 865 /*
 866  * Algorithm:
 867  *      1. The first ip_rt_redirect_number redirects are sent
 868  *         with exponential backoff, then we stop sending them at all,
 869  *         assuming that the host ignores our redirects.
 870  *      2. If we did not see packets requiring redirects
 871  *         during ip_rt_redirect_silence, we assume that the host
 872  *         forgot redirected route and start to send redirects again.
 873  *
 874  * This algorithm is much cheaper and more intelligent than dumb load limiting
 875  * in icmp.c.
 876  *
 877  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 878  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 879  */
 880
 881 void ip_rt_send_redirect(struct sk_buff *skb)
 882 {
 883         struct rtable *rt = skb_rtable(skb);
 884         struct in_device *in_dev;
 885         struct inet_peer *peer;
 886         struct net *net;
 887         int log_martians;
 888         int vif;
 889
 890         rcu_read_lock();
 891         in_dev = __in_dev_get_rcu(rt->dst.dev);
 892         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 893                 rcu_read_unlock();
 894                 return;
 895         }
 896         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 897         vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
 898         rcu_read_unlock();
 899
 900         net = dev_net(rt->dst.dev);
 901         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
 902         if (!peer) {
 903                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 904                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 905                 return;
 906         }
 907
 908         /* No redirected packets during ip_rt_redirect_silence;
 909          * reset the algorithm.
 910          */
 911         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
 912                 peer->rate_tokens = 0;
 913                 peer->n_redirects = 0;
 914         }
 915
 916         /* Too many ignored redirects; do not send anything
 917          * set dst.rate_last to the last seen redirected packet.
 918          */
 919         if (peer->n_redirects >= ip_rt_redirect_number) {
 920                 peer->rate_last = jiffies;
 921                 goto out_put_peer;
 922         }
 923
 924         /* Check for load limit; set rate_last to the latest sent
 925          * redirect.
 926          */
 927         if (peer->n_redirects == 0 ||
 928             time_after(jiffies,
 929                        (peer->rate_last +
 930                         (ip_rt_redirect_load << peer->n_redirects)))) {
 931                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 932
 933                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 934                 peer->rate_last = jiffies;
 935                 ++peer->n_redirects;
 936 #ifdef CONFIG_IP_ROUTE_VERBOSE
 937                 if (log_martians &&
 938                     peer->n_redirects == ip_rt_redirect_number)
 939                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 940                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 941                                              &ip_hdr(skb)->daddr, &gw);
 942 #endif
 943         }
 944 out_put_peer:
 945         inet_putpeer(peer);
 946 }
 947
 948 static int ip_error(struct sk_buff *skb)
 949 {
 950         struct rtable *rt = skb_rtable(skb);
 951         struct net_device *dev = skb->dev;
 952         struct in_device *in_dev;
 953         struct inet_peer *peer;
 954         unsigned long now;
 955         struct net *net;
 956         bool send;
 957         int code;
 958
 959         if (netif_is_l3_master(skb->dev)) {
 960                 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
 961                 if (!dev)
 962                         goto out;
 963         }
 964
 965         in_dev = __in_dev_get_rcu(dev);
 966
 967         /* IP on this device is disabled. */
 968         if (!in_dev)
 969                 goto out;
 970
 971         net = dev_net(rt->dst.dev);
 972         if (!IN_DEV_FORWARD(in_dev)) {
 973                 switch (rt->dst.error) {
 974                 case EHOSTUNREACH:
 975                         __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
 976                         break;
 977
 978                 case ENETUNREACH:
 979                         __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 980                         break;
 981                 }
 982                 goto out;
 983         }
 984
 985         switch (rt->dst.error) {
 986         case EINVAL:
 987         default:
 988                 goto out;
 989         case EHOSTUNREACH:
 990                 code = ICMP_HOST_UNREACH;
 991                 break;
 992         case ENETUNREACH:
 993                 code = ICMP_NET_UNREACH;
 994                 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 995                 break;
 996         case EACCES:
 997                 code = ICMP_PKT_FILTERED;
 998                 break;
 999         }
1000
1001         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
1002                                l3mdev_master_ifindex(skb->dev), 1);
1003
1004         send = true;
1005         if (peer) {
1006                 now = jiffies;
1007                 peer->rate_tokens += now - peer->rate_last;
1008                 if (peer->rate_tokens > ip_rt_error_burst)
1009                         peer->rate_tokens = ip_rt_error_burst;
1010                 peer->rate_last = now;
1011                 if (peer->rate_tokens >= ip_rt_error_cost)
1012                         peer->rate_tokens -= ip_rt_error_cost;
1013                 else
1014                         send = false;
1015                 inet_putpeer(peer);
1016         }
1017         if (send)
1018                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1019
1020 out:    kfree_skb(skb);
1021         return 0;
1022 }
1023
1024 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1025 {
1026         struct dst_entry *dst = &rt->dst;
1027         struct net *net = dev_net(dst->dev);
1028         struct fib_result res;
1029         bool lock = false;
1030         u32 old_mtu;
1031
1032         if (ip_mtu_locked(dst))
1033                 return;
1034
1035         old_mtu = ipv4_mtu(dst);
1036         if (old_mtu < mtu)
1037                 return;
1038
1039         if (mtu < ip_rt_min_pmtu) {
1040                 lock = true;
1041                 mtu = min(old_mtu, ip_rt_min_pmtu);
1042         }
1043
1044         if (rt->rt_pmtu == mtu && !lock &&
1045             time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1046                 return;
1047
1048         rcu_read_lock();
1049         if (fib_lookup(net, fl4, &res, 0) == 0) {
1050                 struct fib_nh_common *nhc;
1051
1052                 fib_select_path(net, &res, fl4, NULL);
1053                 nhc = FIB_RES_NHC(res);
1054                 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1055                                       jiffies + ip_rt_mtu_expires);
1056         }
1057         rcu_read_unlock();
1058 }
1059
1060 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1061                               struct sk_buff *skb, u32 mtu,
1062                               bool confirm_neigh)
1063 {
1064         struct rtable *rt = (struct rtable *) dst;
1065         struct flowi4 fl4;
1066
1067         ip_rt_build_flow_key(&fl4, sk, skb);
1068         ip_rt_fix_tos(&fl4);
1069
1070         /* Don't make lookup fail for bridged encapsulations */
1071         if (skb && netif_is_any_bridge_port(skb->dev))
1072                 fl4.flowi4_oif = 0;
1073
1074         __ip_rt_update_pmtu(rt, &fl4, mtu);
1075 }
1076
1077 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1078                       int oif, u8 protocol)
1079 {
1080         const struct iphdr *iph = (const struct iphdr *)skb->data;
1081         struct flowi4 fl4;
1082         struct rtable *rt;
1083         u32 mark = IP4_REPLY_MARK(net, skb->mark);
1084
1085         __build_flow_key(net, &fl4, NULL, iph, oif,
1086                          RT_TOS(iph->tos), protocol, mark, 0);
1087         rt = __ip_route_output_key(net, &fl4);
1088         if (!IS_ERR(rt)) {
1089                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1090                 ip_rt_put(rt);
1091         }
1092 }
1093 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1094
1095 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1096 {
1097         const struct iphdr *iph = (const struct iphdr *)skb->data;
1098         struct flowi4 fl4;
1099         struct rtable *rt;
1100
1101         __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1102
1103         if (!fl4.flowi4_mark)
1104                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1105
1106         rt = __ip_route_output_key(sock_net(sk), &fl4);
1107         if (!IS_ERR(rt)) {
1108                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1109                 ip_rt_put(rt);
1110         }
1111 }
1112
1113 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1114 {
1115         const struct iphdr *iph = (const struct iphdr *)skb->data;
1116         struct flowi4 fl4;
1117         struct rtable *rt;
1118         struct dst_entry *odst = NULL;
1119         bool new = false;
1120         struct net *net = sock_net(sk);
1121
1122         bh_lock_sock(sk);
1123
1124         if (!ip_sk_accept_pmtu(sk))
1125                 goto out;
1126
1127         odst = sk_dst_get(sk);
1128
1129         if (sock_owned_by_user(sk) || !odst) {
1130                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1131                 goto out;
1132         }
1133
1134         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1135
1136         rt = (struct rtable *)odst;
1137         if (odst->obsolete && !odst->ops->check(odst, 0)) {
1138                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1139                 if (IS_ERR(rt))
1140                         goto out;
1141
1142                 new = true;
1143         } else {
1144                 ip_rt_fix_tos(&fl4);
1145         }
1146
1147         __ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu);
1148
1149         if (!dst_check(&rt->dst, 0)) {
1150                 if (new)
1151                         dst_release(&rt->dst);
1152
1153                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1154                 if (IS_ERR(rt))
1155                         goto out;
1156
1157                 new = true;
1158         }
1159
1160         if (new)
1161                 sk_dst_set(sk, &rt->dst);
1162
1163 out:
1164         bh_unlock_sock(sk);
1165         dst_release(odst);
1166 }
1167 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1168
1169 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1170                    int oif, u8 protocol)
1171 {
1172         const struct iphdr *iph = (const struct iphdr *)skb->data;
1173         struct flowi4 fl4;
1174         struct rtable *rt;
1175
1176         __build_flow_key(net, &fl4, NULL, iph, oif,
1177                          RT_TOS(iph->tos), protocol, 0, 0);
1178         rt = __ip_route_output_key(net, &fl4);
1179         if (!IS_ERR(rt)) {
1180                 __ip_do_redirect(rt, skb, &fl4, false);
1181                 ip_rt_put(rt);
1182         }
1183 }
1184 EXPORT_SYMBOL_GPL(ipv4_redirect);
1185
1186 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1187 {
1188         const struct iphdr *iph = (const struct iphdr *)skb->data;
1189         struct flowi4 fl4;
1190         struct rtable *rt;
1191         struct net *net = sock_net(sk);
1192
1193         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1194         rt = __ip_route_output_key(net, &fl4);
1195         if (!IS_ERR(rt)) {
1196                 __ip_do_redirect(rt, skb, &fl4, false);
1197                 ip_rt_put(rt);
1198         }
1199 }
1200 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1201
1202 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1203                                                          u32 cookie)
1204 {
1205         struct rtable *rt = (struct rtable *) dst;
1206
1207         /* All IPV4 dsts are created with ->obsolete set to the value
1208          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1209          * into this function always.
1210          *
1211          * When a PMTU/redirect information update invalidates a route,
1212          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1213          * DST_OBSOLETE_DEAD.
1214          */
1215         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1216                 return NULL;
1217         return dst;
1218 }
1219 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1220
1221 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1222 {
1223         struct net_device *dev;
1224         struct ip_options opt;
1225         int res;
1226
1227         /* Recompile ip options since IPCB may not be valid anymore.
1228          * Also check we have a reasonable ipv4 header.
1229          */
1230         if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1231             ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1232                 return;
1233
1234         memset(&opt, 0, sizeof(opt));
1235         if (ip_hdr(skb)->ihl > 5) {
1236                 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1237                         return;
1238                 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1239
1240                 rcu_read_lock();
1241                 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev;
1242                 res = __ip_options_compile(dev_net(dev), &opt, skb, NULL);
1243                 rcu_read_unlock();
1244
1245                 if (res)
1246                         return;
1247         }
1248         __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1249 }
1250
1251 static void ipv4_link_failure(struct sk_buff *skb)
1252 {
1253         struct rtable *rt;
1254
1255         ipv4_send_dest_unreach(skb);
1256
1257         rt = skb_rtable(skb);
1258         if (rt)
1259                 dst_set_expires(&rt->dst, 0);
1260 }
1261
1262 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1263 {
1264         pr_debug("%s: %pI4 -> %pI4, %s\n",
1265                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1266                  skb->dev ? skb->dev->name : "?");
1267         kfree_skb(skb);
1268         WARN_ON(1);
1269         return 0;
1270 }
1271
1272 /*
1273  * We do not cache source address of outgoing interface,
1274  * because it is used only by IP RR, TS and SRR options,
1275  * so that it out of fast path.
1276  *
1277  * BTW remember: "addr" is allowed to be not aligned
1278  * in IP options!
1279  */
1280
1281 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1282 {
1283         __be32 src;
1284
1285         if (rt_is_output_route(rt))
1286                 src = ip_hdr(skb)->saddr;
1287         else {
1288                 struct fib_result res;
1289                 struct iphdr *iph = ip_hdr(skb);
1290                 struct flowi4 fl4 = {
1291                         .daddr = iph->daddr,
1292                         .saddr = iph->saddr,
1293                         .flowi4_tos = RT_TOS(iph->tos),
1294                         .flowi4_oif = rt->dst.dev->ifindex,
1295                         .flowi4_iif = skb->dev->ifindex,
1296                         .flowi4_mark = skb->mark,
1297                 };
1298
1299                 rcu_read_lock();
1300                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1301                         src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1302                 else
1303                         src = inet_select_addr(rt->dst.dev,
1304                                                rt_nexthop(rt, iph->daddr),
1305                                                RT_SCOPE_UNIVERSE);
1306                 rcu_read_unlock();
1307         }
1308         memcpy(addr, &src, 4);
1309 }
1310
1311 #ifdef CONFIG_IP_ROUTE_CLASSID
1312 static void set_class_tag(struct rtable *rt, u32 tag)
1313 {
1314         if (!(rt->dst.tclassid & 0xFFFF))
1315                 rt->dst.tclassid |= tag & 0xFFFF;
1316         if (!(rt->dst.tclassid & 0xFFFF0000))
1317                 rt->dst.tclassid |= tag & 0xFFFF0000;
1318 }
1319 #endif
1320
1321 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1322 {
1323         unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1324         unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1325                                     ip_rt_min_advmss);
1326
1327         return min(advmss, IPV4_MAX_PMTU - header_size);
1328 }
1329
1330 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1331 {
1332         return ip_dst_mtu_maybe_forward(dst, false);
1333 }
1334 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1335
1336 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1337 {
1338         struct fnhe_hash_bucket *hash;
1339         struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1340         u32 hval = fnhe_hashfun(daddr);
1341
1342         spin_lock_bh(&fnhe_lock);
1343
1344         hash = rcu_dereference_protected(nhc->nhc_exceptions,
1345                                          lockdep_is_held(&fnhe_lock));
1346         hash += hval;
1347
1348         fnhe_p = &hash->chain;
1349         fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1350         while (fnhe) {
1351                 if (fnhe->fnhe_daddr == daddr) {
1352                         rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1353                                 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1354                         /* set fnhe_daddr to 0 to ensure it won't bind with
1355                          * new dsts in rt_bind_exception().
1356                          */
1357                         fnhe->fnhe_daddr = 0;
1358                         fnhe_flush_routes(fnhe);
1359                         kfree_rcu(fnhe, rcu);
1360                         break;
1361                 }
1362                 fnhe_p = &fnhe->fnhe_next;
1363                 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1364                                                  lockdep_is_held(&fnhe_lock));
1365         }
1366
1367         spin_unlock_bh(&fnhe_lock);
1368 }
1369
1370 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1371                                                __be32 daddr)
1372 {
1373         struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1374         struct fib_nh_exception *fnhe;
1375         u32 hval;
1376
1377         if (!hash)
1378                 return NULL;
1379
1380         hval = fnhe_hashfun(daddr);
1381
1382         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1383              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1384                 if (fnhe->fnhe_daddr == daddr) {
1385                         if (fnhe->fnhe_expires &&
1386                             time_after(jiffies, fnhe->fnhe_expires)) {
1387                                 ip_del_fnhe(nhc, daddr);
1388                                 break;
1389                         }
1390                         return fnhe;
1391                 }
1392         }
1393         return NULL;
1394 }
1395
1396 /* MTU selection:
1397  * 1. mtu on route is locked - use it
1398  * 2. mtu from nexthop exception
1399  * 3. mtu from egress device
1400  */
1401
1402 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1403 {
1404         struct fib_nh_common *nhc = res->nhc;
1405         struct net_device *dev = nhc->nhc_dev;
1406         struct fib_info *fi = res->fi;
1407         u32 mtu = 0;
1408
1409         if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) ||
1410             fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1411                 mtu = fi->fib_mtu;
1412
1413         if (likely(!mtu)) {
1414                 struct fib_nh_exception *fnhe;
1415
1416                 fnhe = find_exception(nhc, daddr);
1417                 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1418                         mtu = fnhe->fnhe_pmtu;
1419         }
1420
1421         if (likely(!mtu))
1422                 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1423
1424         return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1425 }
1426
1427 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1428                               __be32 daddr, const bool do_cache)
1429 {
1430         bool ret = false;
1431
1432         spin_lock_bh(&fnhe_lock);
1433
1434         if (daddr == fnhe->fnhe_daddr) {
1435                 struct rtable __rcu **porig;
1436                 struct rtable *orig;
1437                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1438
1439                 if (rt_is_input_route(rt))
1440                         porig = &fnhe->fnhe_rth_input;
1441                 else
1442                         porig = &fnhe->fnhe_rth_output;
1443                 orig = rcu_dereference(*porig);
1444
1445                 if (fnhe->fnhe_genid != genid) {
1446                         fnhe->fnhe_genid = genid;
1447                         fnhe->fnhe_gw = 0;
1448                         fnhe->fnhe_pmtu = 0;
1449                         fnhe->fnhe_expires = 0;
1450                         fnhe->fnhe_mtu_locked = false;
1451                         fnhe_flush_routes(fnhe);
1452                         orig = NULL;
1453                 }
1454                 fill_route_from_fnhe(rt, fnhe);
1455                 if (!rt->rt_gw4) {
1456                         rt->rt_gw4 = daddr;
1457                         rt->rt_gw_family = AF_INET;
1458                 }
1459
1460                 if (do_cache) {
1461                         dst_hold(&rt->dst);
1462                         rcu_assign_pointer(*porig, rt);
1463                         if (orig) {
1464                                 dst_dev_put(&orig->dst);
1465                                 dst_release(&orig->dst);
1466                         }
1467                         ret = true;
1468                 }
1469
1470                 fnhe->fnhe_stamp = jiffies;
1471         }
1472         spin_unlock_bh(&fnhe_lock);
1473
1474         return ret;
1475 }
1476
1477 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1478 {
1479         struct rtable *orig, *prev, **p;
1480         bool ret = true;
1481
1482         if (rt_is_input_route(rt)) {
1483                 p = (struct rtable **)&nhc->nhc_rth_input;
1484         } else {
1485                 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1486         }
1487         orig = *p;
1488
1489         /* hold dst before doing cmpxchg() to avoid race condition
1490          * on this dst
1491          */
1492         dst_hold(&rt->dst);
1493         prev = cmpxchg(p, orig, rt);
1494         if (prev == orig) {
1495                 if (orig) {
1496                         rt_add_uncached_list(orig);
1497                         dst_release(&orig->dst);
1498                 }
1499         } else {
1500                 dst_release(&rt->dst);
1501                 ret = false;
1502         }
1503
1504         return ret;
1505 }
1506
1507 struct uncached_list {
1508         spinlock_t              lock;
1509         struct list_head        head;
1510 };
1511
1512 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1513
1514 void rt_add_uncached_list(struct rtable *rt)
1515 {
1516         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1517
1518         rt->rt_uncached_list = ul;
1519
1520         spin_lock_bh(&ul->lock);
1521         list_add_tail(&rt->rt_uncached, &ul->head);
1522         spin_unlock_bh(&ul->lock);
1523 }
1524
1525 void rt_del_uncached_list(struct rtable *rt)
1526 {
1527         if (!list_empty(&rt->rt_uncached)) {
1528                 struct uncached_list *ul = rt->rt_uncached_list;
1529
1530                 spin_lock_bh(&ul->lock);
1531                 list_del(&rt->rt_uncached);
1532                 spin_unlock_bh(&ul->lock);
1533         }
1534 }
1535
1536 static void ipv4_dst_destroy(struct dst_entry *dst)
1537 {
1538         struct rtable *rt = (struct rtable *)dst;
1539
1540         ip_dst_metrics_put(dst);
1541         rt_del_uncached_list(rt);
1542 }
1543
1544 void rt_flush_dev(struct net_device *dev)
1545 {
1546         struct rtable *rt;
1547         int cpu;
1548
1549         for_each_possible_cpu(cpu) {
1550                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1551
1552                 spin_lock_bh(&ul->lock);
1553                 list_for_each_entry(rt, &ul->head, rt_uncached) {
1554                         if (rt->dst.dev != dev)
1555                                 continue;
1556                         rt->dst.dev = blackhole_netdev;
1557                         dev_hold(rt->dst.dev);
1558                         dev_put(dev);
1559                 }
1560                 spin_unlock_bh(&ul->lock);
1561         }
1562 }
1563
1564 static bool rt_cache_valid(const struct rtable *rt)
1565 {
1566         return  rt &&
1567                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1568                 !rt_is_expired(rt);
1569 }
1570
1571 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1572                            const struct fib_result *res,
1573                            struct fib_nh_exception *fnhe,
1574                            struct fib_info *fi, u16 type, u32 itag,
1575                            const bool do_cache)
1576 {
1577         bool cached = false;
1578
1579         if (fi) {
1580                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1581
1582                 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1583                         rt->rt_uses_gateway = 1;
1584                         rt->rt_gw_family = nhc->nhc_gw_family;
1585                         /* only INET and INET6 are supported */
1586                         if (likely(nhc->nhc_gw_family == AF_INET))
1587                                 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1588                         else
1589                                 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1590                 }
1591
1592                 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1593
1594 #ifdef CONFIG_IP_ROUTE_CLASSID
1595                 if (nhc->nhc_family == AF_INET) {
1596                         struct fib_nh *nh;
1597
1598                         nh = container_of(nhc, struct fib_nh, nh_common);
1599                         rt->dst.tclassid = nh->nh_tclassid;
1600                 }
1601 #endif
1602                 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1603                 if (unlikely(fnhe))
1604                         cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1605                 else if (do_cache)
1606                         cached = rt_cache_route(nhc, rt);
1607                 if (unlikely(!cached)) {
1608                         /* Routes we intend to cache in nexthop exception or
1609                          * FIB nexthop have the DST_NOCACHE bit clear.
1610                          * However, if we are unsuccessful at storing this
1611                          * route into the cache we really need to set it.
1612                          */
1613                         if (!rt->rt_gw4) {
1614                                 rt->rt_gw_family = AF_INET;
1615                                 rt->rt_gw4 = daddr;
1616                         }
1617                         rt_add_uncached_list(rt);
1618                 }
1619         } else
1620                 rt_add_uncached_list(rt);
1621
1622 #ifdef CONFIG_IP_ROUTE_CLASSID
1623 #ifdef CONFIG_IP_MULTIPLE_TABLES
1624         set_class_tag(rt, res->tclassid);
1625 #endif
1626         set_class_tag(rt, itag);
1627 #endif
1628 }
1629
1630 struct rtable *rt_dst_alloc(struct net_device *dev,
1631                             unsigned int flags, u16 type,
1632                             bool nopolicy, bool noxfrm)
1633 {
1634         struct rtable *rt;
1635
1636         rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1637                        (nopolicy ? DST_NOPOLICY : 0) |
1638                        (noxfrm ? DST_NOXFRM : 0));
1639
1640         if (rt) {
1641                 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1642                 rt->rt_flags = flags;
1643                 rt->rt_type = type;
1644                 rt->rt_is_input = 0;
1645                 rt->rt_iif = 0;
1646                 rt->rt_pmtu = 0;
1647                 rt->rt_mtu_locked = 0;
1648                 rt->rt_uses_gateway = 0;
1649                 rt->rt_gw_family = 0;
1650                 rt->rt_gw4 = 0;
1651                 INIT_LIST_HEAD(&rt->rt_uncached);
1652
1653                 rt->dst.output = ip_output;
1654                 if (flags & RTCF_LOCAL)
1655                         rt->dst.input = ip_local_deliver;
1656         }
1657
1658         return rt;
1659 }
1660 EXPORT_SYMBOL(rt_dst_alloc);
1661
1662 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1663 {
1664         struct rtable *new_rt;
1665
1666         new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1667                            rt->dst.flags);
1668
1669         if (new_rt) {
1670                 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1671                 new_rt->rt_flags = rt->rt_flags;
1672                 new_rt->rt_type = rt->rt_type;
1673                 new_rt->rt_is_input = rt->rt_is_input;
1674                 new_rt->rt_iif = rt->rt_iif;
1675                 new_rt->rt_pmtu = rt->rt_pmtu;
1676                 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1677                 new_rt->rt_gw_family = rt->rt_gw_family;
1678                 if (rt->rt_gw_family == AF_INET)
1679                         new_rt->rt_gw4 = rt->rt_gw4;
1680                 else if (rt->rt_gw_family == AF_INET6)
1681                         new_rt->rt_gw6 = rt->rt_gw6;
1682                 INIT_LIST_HEAD(&new_rt->rt_uncached);
1683
1684                 new_rt->dst.input = rt->dst.input;
1685                 new_rt->dst.output = rt->dst.output;
1686                 new_rt->dst.error = rt->dst.error;
1687                 new_rt->dst.lastuse = jiffies;
1688                 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1689         }
1690         return new_rt;
1691 }
1692 EXPORT_SYMBOL(rt_dst_clone);
1693
1694 /* called in rcu_read_lock() section */
1695 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1696                           u8 tos, struct net_device *dev,
1697                           struct in_device *in_dev, u32 *itag)
1698 {
1699         int err;
1700
1701         /* Primary sanity checks. */
1702         if (!in_dev)
1703                 return -EINVAL;
1704
1705         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1706             skb->protocol != htons(ETH_P_IP))
1707                 return -EINVAL;
1708
1709         if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1710                 return -EINVAL;
1711
1712         if (ipv4_is_zeronet(saddr)) {
1713                 if (!ipv4_is_local_multicast(daddr) &&
1714                     ip_hdr(skb)->protocol != IPPROTO_IGMP)
1715                         return -EINVAL;
1716         } else {
1717                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1718                                           in_dev, itag);
1719                 if (err < 0)
1720                         return err;
1721         }
1722         return 0;
1723 }
1724
1725 /* called in rcu_read_lock() section */
1726 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1727                              u8 tos, struct net_device *dev, int our)
1728 {
1729         struct in_device *in_dev = __in_dev_get_rcu(dev);
1730         unsigned int flags = RTCF_MULTICAST;
1731         struct rtable *rth;
1732         bool no_policy;
1733         u32 itag = 0;
1734         int err;
1735
1736         err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1737         if (err)
1738                 return err;
1739
1740         if (our)
1741                 flags |= RTCF_LOCAL;
1742
1743         no_policy = IN_DEV_ORCONF(in_dev, NOPOLICY);
1744         if (no_policy)
1745                 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1746
1747         rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1748                            no_policy, false);
1749         if (!rth)
1750                 return -ENOBUFS;
1751
1752 #ifdef CONFIG_IP_ROUTE_CLASSID
1753         rth->dst.tclassid = itag;
1754 #endif
1755         rth->dst.output = ip_rt_bug;
1756         rth->rt_is_input= 1;
1757
1758 #ifdef CONFIG_IP_MROUTE
1759         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1760                 rth->dst.input = ip_mr_input;
1761 #endif
1762         RT_CACHE_STAT_INC(in_slow_mc);
1763
1764         skb_dst_drop(skb);
1765         skb_dst_set(skb, &rth->dst);
1766         return 0;
1767 }
1768
1769
1770 static void ip_handle_martian_source(struct net_device *dev,
1771                                      struct in_device *in_dev,
1772                                      struct sk_buff *skb,
1773                                      __be32 daddr,
1774                                      __be32 saddr)
1775 {
1776         RT_CACHE_STAT_INC(in_martian_src);
1777 #ifdef CONFIG_IP_ROUTE_VERBOSE
1778         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1779                 /*
1780                  *      RFC1812 recommendation, if source is martian,
1781                  *      the only hint is MAC header.
1782                  */
1783                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1784                         &daddr, &saddr, dev->name);
1785                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1786                         print_hex_dump(KERN_WARNING, "ll header: ",
1787                                        DUMP_PREFIX_OFFSET, 16, 1,
1788                                        skb_mac_header(skb),
1789                                        dev->hard_header_len, false);
1790                 }
1791         }
1792 #endif
1793 }
1794
1795 /* called in rcu_read_lock() section */
1796 static int __mkroute_input(struct sk_buff *skb,
1797                            const struct fib_result *res,
1798                            struct in_device *in_dev,
1799                            __be32 daddr, __be32 saddr, u32 tos)
1800 {
1801         struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1802         struct net_device *dev = nhc->nhc_dev;
1803         struct fib_nh_exception *fnhe;
1804         struct rtable *rth;
1805         int err;
1806         struct in_device *out_dev;
1807         bool do_cache, no_policy;
1808         u32 itag = 0;
1809
1810         /* get a working reference to the output device */
1811         out_dev = __in_dev_get_rcu(dev);
1812         if (!out_dev) {
1813                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1814                 return -EINVAL;
1815         }
1816
1817         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1818                                   in_dev->dev, in_dev, &itag);
1819         if (err < 0) {
1820                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1821                                          saddr);
1822
1823                 goto cleanup;
1824         }
1825
1826         do_cache = res->fi && !itag;
1827         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1828             skb->protocol == htons(ETH_P_IP)) {
1829                 __be32 gw;
1830
1831                 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1832                 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1833                     inet_addr_onlink(out_dev, saddr, gw))
1834                         IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1835         }
1836
1837         if (skb->protocol != htons(ETH_P_IP)) {
1838                 /* Not IP (i.e. ARP). Do not create route, if it is
1839                  * invalid for proxy arp. DNAT routes are always valid.
1840                  *
1841                  * Proxy arp feature have been extended to allow, ARP
1842                  * replies back to the same interface, to support
1843                  * Private VLAN switch technologies. See arp.c.
1844                  */
1845                 if (out_dev == in_dev &&
1846                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1847                         err = -EINVAL;
1848                         goto cleanup;
1849                 }
1850         }
1851
1852         no_policy = IN_DEV_ORCONF(in_dev, NOPOLICY);
1853         if (no_policy)
1854                 IPCB(skb)->flags |= IPSKB_NOPOLICY;
1855
1856         fnhe = find_exception(nhc, daddr);
1857         if (do_cache) {
1858                 if (fnhe)
1859                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1860                 else
1861                         rth = rcu_dereference(nhc->nhc_rth_input);
1862                 if (rt_cache_valid(rth)) {
1863                         skb_dst_set_noref(skb, &rth->dst);
1864                         goto out;
1865                 }
1866         }
1867
1868         rth = rt_dst_alloc(out_dev->dev, 0, res->type, no_policy,
1869                            IN_DEV_ORCONF(out_dev, NOXFRM));
1870         if (!rth) {
1871                 err = -ENOBUFS;
1872                 goto cleanup;
1873         }
1874
1875         rth->rt_is_input = 1;
1876         RT_CACHE_STAT_INC(in_slow_tot);
1877
1878         rth->dst.input = ip_forward;
1879
1880         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1881                        do_cache);
1882         lwtunnel_set_redirect(&rth->dst);
1883         skb_dst_set(skb, &rth->dst);
1884 out:
1885         err = 0;
1886  cleanup:
1887         return err;
1888 }
1889
1890 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1891 /* To make ICMP packets follow the right flow, the multipath hash is
1892  * calculated from the inner IP addresses.
1893  */
1894 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1895                                  struct flow_keys *hash_keys)
1896 {
1897         const struct iphdr *outer_iph = ip_hdr(skb);
1898         const struct iphdr *key_iph = outer_iph;
1899         const struct iphdr *inner_iph;
1900         const struct icmphdr *icmph;
1901         struct iphdr _inner_iph;
1902         struct icmphdr _icmph;
1903
1904         if (likely(outer_iph->protocol != IPPROTO_ICMP))
1905                 goto out;
1906
1907         if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1908                 goto out;
1909
1910         icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1911                                    &_icmph);
1912         if (!icmph)
1913                 goto out;
1914
1915         if (!icmp_is_err(icmph->type))
1916                 goto out;
1917
1918         inner_iph = skb_header_pointer(skb,
1919                                        outer_iph->ihl * 4 + sizeof(_icmph),
1920                                        sizeof(_inner_iph), &_inner_iph);
1921         if (!inner_iph)
1922                 goto out;
1923
1924         key_iph = inner_iph;
1925 out:
1926         hash_keys->addrs.v4addrs.src = key_iph->saddr;
1927         hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1928 }
1929
1930 static u32 fib_multipath_custom_hash_outer(const struct net *net,
1931                                            const struct sk_buff *skb,
1932                                            bool *p_has_inner)
1933 {
1934         u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1935         struct flow_keys keys, hash_keys;
1936
1937         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
1938                 return 0;
1939
1940         memset(&hash_keys, 0, sizeof(hash_keys));
1941         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
1942
1943         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1944         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
1945                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1946         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
1947                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1948         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
1949                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1950         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
1951                 hash_keys.ports.src = keys.ports.src;
1952         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
1953                 hash_keys.ports.dst = keys.ports.dst;
1954
1955         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
1956         return flow_hash_from_keys(&hash_keys);
1957 }
1958
1959 static u32 fib_multipath_custom_hash_inner(const struct net *net,
1960                                            const struct sk_buff *skb,
1961                                            bool has_inner)
1962 {
1963         u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
1964         struct flow_keys keys, hash_keys;
1965
1966         /* We assume the packet carries an encapsulation, but if none was
1967          * encountered during dissection of the outer flow, then there is no
1968          * point in calling the flow dissector again.
1969          */
1970         if (!has_inner)
1971                 return 0;
1972
1973         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
1974                 return 0;
1975
1976         memset(&hash_keys, 0, sizeof(hash_keys));
1977         skb_flow_dissect_flow_keys(skb, &keys, 0);
1978
1979         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
1980                 return 0;
1981
1982         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1983                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1984                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1985                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1986                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1987                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1988         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1989                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1990                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
1991                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1992                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
1993                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1994                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
1995                         hash_keys.tags.flow_label = keys.tags.flow_label;
1996         }
1997
1998         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
1999                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2000         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2001                 hash_keys.ports.src = keys.ports.src;
2002         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2003                 hash_keys.ports.dst = keys.ports.dst;
2004
2005         return flow_hash_from_keys(&hash_keys);
2006 }
2007
2008 static u32 fib_multipath_custom_hash_skb(const struct net *net,
2009                                          const struct sk_buff *skb)
2010 {
2011         u32 mhash, mhash_inner;
2012         bool has_inner = true;
2013
2014         mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner);
2015         mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner);
2016
2017         return jhash_2words(mhash, mhash_inner, 0);
2018 }
2019
2020 static u32 fib_multipath_custom_hash_fl4(const struct net *net,
2021                                          const struct flowi4 *fl4)
2022 {
2023         u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields);
2024         struct flow_keys hash_keys;
2025
2026         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2027                 return 0;
2028
2029         memset(&hash_keys, 0, sizeof(hash_keys));
2030         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2031         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2032                 hash_keys.addrs.v4addrs.src = fl4->saddr;
2033         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2034                 hash_keys.addrs.v4addrs.dst = fl4->daddr;
2035         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2036                 hash_keys.basic.ip_proto = fl4->flowi4_proto;
2037         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2038                 hash_keys.ports.src = fl4->fl4_sport;
2039         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2040                 hash_keys.ports.dst = fl4->fl4_dport;
2041
2042         return flow_hash_from_keys(&hash_keys);
2043 }
2044
2045 /* if skb is set it will be used and fl4 can be NULL */
2046 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
2047                        const struct sk_buff *skb, struct flow_keys *flkeys)
2048 {
2049         u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
2050         struct flow_keys hash_keys;
2051         u32 mhash = 0;
2052
2053         switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) {
2054         case 0:
2055                 memset(&hash_keys, 0, sizeof(hash_keys));
2056                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2057                 if (skb) {
2058                         ip_multipath_l3_keys(skb, &hash_keys);
2059                 } else {
2060                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2061                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2062                 }
2063                 mhash = flow_hash_from_keys(&hash_keys);
2064                 break;
2065         case 1:
2066                 /* skb is currently provided only when forwarding */
2067                 if (skb) {
2068                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2069                         struct flow_keys keys;
2070
2071                         /* short-circuit if we already have L4 hash present */
2072                         if (skb->l4_hash)
2073                                 return skb_get_hash_raw(skb) >> 1;
2074
2075                         memset(&hash_keys, 0, sizeof(hash_keys));
2076
2077                         if (!flkeys) {
2078                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2079                                 flkeys = &keys;
2080                         }
2081
2082                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2083                         hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2084                         hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2085                         hash_keys.ports.src = flkeys->ports.src;
2086                         hash_keys.ports.dst = flkeys->ports.dst;
2087                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2088                 } else {
2089                         memset(&hash_keys, 0, sizeof(hash_keys));
2090                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2091                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2092                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2093                         hash_keys.ports.src = fl4->fl4_sport;
2094                         hash_keys.ports.dst = fl4->fl4_dport;
2095                         hash_keys.basic.ip_proto = fl4->flowi4_proto;
2096                 }
2097                 mhash = flow_hash_from_keys(&hash_keys);
2098                 break;
2099         case 2:
2100                 memset(&hash_keys, 0, sizeof(hash_keys));
2101                 /* skb is currently provided only when forwarding */
2102                 if (skb) {
2103                         struct flow_keys keys;
2104
2105                         skb_flow_dissect_flow_keys(skb, &keys, 0);
2106                         /* Inner can be v4 or v6 */
2107                         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2108                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2109                                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2110                                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2111                         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2112                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2113                                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2114                                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2115                                 hash_keys.tags.flow_label = keys.tags.flow_label;
2116                                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2117                         } else {
2118                                 /* Same as case 0 */
2119                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2120                                 ip_multipath_l3_keys(skb, &hash_keys);
2121                         }
2122                 } else {
2123                         /* Same as case 0 */
2124                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2125                         hash_keys.addrs.v4addrs.src = fl4->saddr;
2126                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
2127                 }
2128                 mhash = flow_hash_from_keys(&hash_keys);
2129                 break;
2130         case 3:
2131                 if (skb)
2132                         mhash = fib_multipath_custom_hash_skb(net, skb);
2133                 else
2134                         mhash = fib_multipath_custom_hash_fl4(net, fl4);
2135                 break;
2136         }
2137
2138         if (multipath_hash)
2139                 mhash = jhash_2words(mhash, multipath_hash, 0);
2140
2141         return mhash >> 1;
2142 }
2143 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2144
2145 static int ip_mkroute_input(struct sk_buff *skb,
2146                             struct fib_result *res,
2147                             struct in_device *in_dev,
2148                             __be32 daddr, __be32 saddr, u32 tos,
2149                             struct flow_keys *hkeys)
2150 {
2151 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2152         if (res->fi && fib_info_num_path(res->fi) > 1) {
2153                 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2154
2155                 fib_select_multipath(res, h);
2156                 IPCB(skb)->flags |= IPSKB_MULTIPATH;
2157         }
2158 #endif
2159
2160         /* create a routing cache entry */
2161         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2162 }
2163
2164 /* Implements all the saddr-related checks as ip_route_input_slow(),
2165  * assuming daddr is valid and the destination is not a local broadcast one.
2166  * Uses the provided hint instead of performing a route lookup.
2167  */
2168 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2169                       u8 tos, struct net_device *dev,
2170                       const struct sk_buff *hint)
2171 {
2172         struct in_device *in_dev = __in_dev_get_rcu(dev);
2173         struct rtable *rt = skb_rtable(hint);
2174         struct net *net = dev_net(dev);
2175         int err = -EINVAL;
2176         u32 tag = 0;
2177
2178         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2179                 goto martian_source;
2180
2181         if (ipv4_is_zeronet(saddr))
2182                 goto martian_source;
2183
2184         if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2185                 goto martian_source;
2186
2187         if (rt->rt_type != RTN_LOCAL)
2188                 goto skip_validate_source;
2189
2190         tos &= IPTOS_RT_MASK;
2191         err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2192         if (err < 0)
2193                 goto martian_source;
2194
2195 skip_validate_source:
2196         skb_dst_copy(skb, hint);
2197         return 0;
2198
2199 martian_source:
2200         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2201         return err;
2202 }
2203
2204 /* get device for dst_alloc with local routes */
2205 static struct net_device *ip_rt_get_dev(struct net *net,
2206                                         const struct fib_result *res)
2207 {
2208         struct fib_nh_common *nhc = res->fi ? res->nhc : NULL;
2209         struct net_device *dev = NULL;
2210
2211         if (nhc)
2212                 dev = l3mdev_master_dev_rcu(nhc->nhc_dev);
2213
2214         return dev ? : net->loopback_dev;
2215 }
2216
2217 /*
2218  *      NOTE. We drop all the packets that has local source
2219  *      addresses, because every properly looped back packet
2220  *      must have correct destination already attached by output routine.
2221  *      Changes in the enforced policies must be applied also to
2222  *      ip_route_use_hint().
2223  *
2224  *      Such approach solves two big problems:
2225  *      1. Not simplex devices are handled properly.
2226  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2227  *      called with rcu_read_lock()
2228  */
2229
2230 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2231                                u8 tos, struct net_device *dev,
2232                                struct fib_result *res)
2233 {
2234         struct in_device *in_dev = __in_dev_get_rcu(dev);
2235         struct flow_keys *flkeys = NULL, _flkeys;
2236         struct net    *net = dev_net(dev);
2237         struct ip_tunnel_info *tun_info;
2238         int             err = -EINVAL;
2239         unsigned int    flags = 0;
2240         u32             itag = 0;
2241         struct rtable   *rth;
2242         struct flowi4   fl4;
2243         bool do_cache = true;
2244         bool no_policy;
2245
2246         /* IP on this device is disabled. */
2247
2248         if (!in_dev)
2249                 goto out;
2250
2251         /* Check for the most weird martians, which can be not detected
2252          * by fib_lookup.
2253          */
2254
2255         tun_info = skb_tunnel_info(skb);
2256         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2257                 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2258         else
2259                 fl4.flowi4_tun_key.tun_id = 0;
2260         skb_dst_drop(skb);
2261
2262         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2263                 goto martian_source;
2264
2265         res->fi = NULL;
2266         res->table = NULL;
2267         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2268                 goto brd_input;
2269
2270         /* Accept zero addresses only to limited broadcast;
2271          * I even do not know to fix it or not. Waiting for complains :-)
2272          */
2273         if (ipv4_is_zeronet(saddr))
2274                 goto martian_source;
2275
2276         if (ipv4_is_zeronet(daddr))
2277                 goto martian_destination;
2278
2279         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2280          * and call it once if daddr or/and saddr are loopback addresses
2281          */
2282         if (ipv4_is_loopback(daddr)) {
2283                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2284                         goto martian_destination;
2285         } else if (ipv4_is_loopback(saddr)) {
2286                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2287                         goto martian_source;
2288         }
2289
2290         /*
2291          *      Now we are ready to route packet.
2292          */
2293         fl4.flowi4_oif = 0;
2294         fl4.flowi4_iif = dev->ifindex;
2295         fl4.flowi4_mark = skb->mark;
2296         fl4.flowi4_tos = tos;
2297         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2298         fl4.flowi4_flags = 0;
2299         fl4.daddr = daddr;
2300         fl4.saddr = saddr;
2301         fl4.flowi4_uid = sock_net_uid(net, NULL);
2302         fl4.flowi4_multipath_hash = 0;
2303
2304         if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2305                 flkeys = &_flkeys;
2306         } else {
2307                 fl4.flowi4_proto = 0;
2308                 fl4.fl4_sport = 0;
2309                 fl4.fl4_dport = 0;
2310         }
2311
2312         err = fib_lookup(net, &fl4, res, 0);
2313         if (err != 0) {
2314                 if (!IN_DEV_FORWARD(in_dev))
2315                         err = -EHOSTUNREACH;
2316                 goto no_route;
2317         }
2318
2319         if (res->type == RTN_BROADCAST) {
2320                 if (IN_DEV_BFORWARD(in_dev))
2321                         goto make_route;
2322                 /* not do cache if bc_forwarding is enabled */
2323                 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2324                         do_cache = false;
2325                 goto brd_input;
2326         }
2327
2328         if (res->type == RTN_LOCAL) {
2329                 err = fib_validate_source(skb, saddr, daddr, tos,
2330                                           0, dev, in_dev, &itag);
2331                 if (err < 0)
2332                         goto martian_source;
2333                 goto local_input;
2334         }
2335
2336         if (!IN_DEV_FORWARD(in_dev)) {
2337                 err = -EHOSTUNREACH;
2338                 goto no_route;
2339         }
2340         if (res->type != RTN_UNICAST)
2341                 goto martian_destination;
2342
2343 make_route:
2344         err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2345 out:    return err;
2346
2347 brd_input:
2348         if (skb->protocol != htons(ETH_P_IP))
2349                 goto e_inval;
2350
2351         if (!ipv4_is_zeronet(saddr)) {
2352                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2353                                           in_dev, &itag);
2354                 if (err < 0)
2355                         goto martian_source;
2356         }
2357         flags |= RTCF_BROADCAST;
2358         res->type = RTN_BROADCAST;
2359         RT_CACHE_STAT_INC(in_brd);
2360
2361 local_input:
2362         no_policy = IN_DEV_ORCONF(in_dev, NOPOLICY);
2363         if (no_policy)
2364                 IPCB(skb)->flags |= IPSKB_NOPOLICY;
2365
2366         do_cache &= res->fi && !itag;
2367         if (do_cache) {
2368                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2369
2370                 rth = rcu_dereference(nhc->nhc_rth_input);
2371                 if (rt_cache_valid(rth)) {
2372                         skb_dst_set_noref(skb, &rth->dst);
2373                         err = 0;
2374                         goto out;
2375                 }
2376         }
2377
2378         rth = rt_dst_alloc(ip_rt_get_dev(net, res),
2379                            flags | RTCF_LOCAL, res->type,
2380                            no_policy, false);
2381         if (!rth)
2382                 goto e_nobufs;
2383
2384         rth->dst.output= ip_rt_bug;
2385 #ifdef CONFIG_IP_ROUTE_CLASSID
2386         rth->dst.tclassid = itag;
2387 #endif
2388         rth->rt_is_input = 1;
2389
2390         RT_CACHE_STAT_INC(in_slow_tot);
2391         if (res->type == RTN_UNREACHABLE) {
2392                 rth->dst.input= ip_error;
2393                 rth->dst.error= -err;
2394                 rth->rt_flags   &= ~RTCF_LOCAL;
2395         }
2396
2397         if (do_cache) {
2398                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2399
2400                 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2401                 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2402                         WARN_ON(rth->dst.input == lwtunnel_input);
2403                         rth->dst.lwtstate->orig_input = rth->dst.input;
2404                         rth->dst.input = lwtunnel_input;
2405                 }
2406
2407                 if (unlikely(!rt_cache_route(nhc, rth)))
2408                         rt_add_uncached_list(rth);
2409         }
2410         skb_dst_set(skb, &rth->dst);
2411         err = 0;
2412         goto out;
2413
2414 no_route:
2415         RT_CACHE_STAT_INC(in_no_route);
2416         res->type = RTN_UNREACHABLE;
2417         res->fi = NULL;
2418         res->table = NULL;
2419         goto local_input;
2420
2421         /*
2422          *      Do not cache martian addresses: they should be logged (RFC1812)
2423          */
2424 martian_destination:
2425         RT_CACHE_STAT_INC(in_martian_dst);
2426 #ifdef CONFIG_IP_ROUTE_VERBOSE
2427         if (IN_DEV_LOG_MARTIANS(in_dev))
2428                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2429                                      &daddr, &saddr, dev->name);
2430 #endif
2431
2432 e_inval:
2433         err = -EINVAL;
2434         goto out;
2435
2436 e_nobufs:
2437         err = -ENOBUFS;
2438         goto out;
2439
2440 martian_source:
2441         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2442         goto out;
2443 }
2444
2445 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2446                          u8 tos, struct net_device *dev)
2447 {
2448         struct fib_result res;
2449         int err;
2450
2451         tos &= IPTOS_RT_MASK;
2452         rcu_read_lock();
2453         err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2454         rcu_read_unlock();
2455
2456         return err;
2457 }
2458 EXPORT_SYMBOL(ip_route_input_noref);
2459
2460 /* called with rcu_read_lock held */
2461 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2462                        u8 tos, struct net_device *dev, struct fib_result *res)
2463 {
2464         /* Multicast recognition logic is moved from route cache to here.
2465          * The problem was that too many Ethernet cards have broken/missing
2466          * hardware multicast filters :-( As result the host on multicasting
2467          * network acquires a lot of useless route cache entries, sort of
2468          * SDR messages from all the world. Now we try to get rid of them.
2469          * Really, provided software IP multicast filter is organized
2470          * reasonably (at least, hashed), it does not result in a slowdown
2471          * comparing with route cache reject entries.
2472          * Note, that multicast routers are not affected, because
2473          * route cache entry is created eventually.
2474          */
2475         if (ipv4_is_multicast(daddr)) {
2476                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2477                 int our = 0;
2478                 int err = -EINVAL;
2479
2480                 if (!in_dev)
2481                         return err;
2482                 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2483                                       ip_hdr(skb)->protocol);
2484
2485                 /* check l3 master if no match yet */
2486                 if (!our && netif_is_l3_slave(dev)) {
2487                         struct in_device *l3_in_dev;
2488
2489                         l3_in_dev = __in_dev_get_rcu(skb->dev);
2490                         if (l3_in_dev)
2491                                 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2492                                                       ip_hdr(skb)->protocol);
2493                 }
2494
2495                 if (our
2496 #ifdef CONFIG_IP_MROUTE
2497                         ||
2498                     (!ipv4_is_local_multicast(daddr) &&
2499                      IN_DEV_MFORWARD(in_dev))
2500 #endif
2501                    ) {
2502                         err = ip_route_input_mc(skb, daddr, saddr,
2503                                                 tos, dev, our);
2504                 }
2505                 return err;
2506         }
2507
2508         return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2509 }
2510
2511 /* called with rcu_read_lock() */
2512 static struct rtable *__mkroute_output(const struct fib_result *res,
2513                                        const struct flowi4 *fl4, int orig_oif,
2514                                        struct net_device *dev_out,
2515                                        unsigned int flags)
2516 {
2517         struct fib_info *fi = res->fi;
2518         struct fib_nh_exception *fnhe;
2519         struct in_device *in_dev;
2520         u16 type = res->type;
2521         struct rtable *rth;
2522         bool do_cache;
2523
2524         in_dev = __in_dev_get_rcu(dev_out);
2525         if (!in_dev)
2526                 return ERR_PTR(-EINVAL);
2527
2528         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2529                 if (ipv4_is_loopback(fl4->saddr) &&
2530                     !(dev_out->flags & IFF_LOOPBACK) &&
2531                     !netif_is_l3_master(dev_out))
2532                         return ERR_PTR(-EINVAL);
2533
2534         if (ipv4_is_lbcast(fl4->daddr))
2535                 type = RTN_BROADCAST;
2536         else if (ipv4_is_multicast(fl4->daddr))
2537                 type = RTN_MULTICAST;
2538         else if (ipv4_is_zeronet(fl4->daddr))
2539                 return ERR_PTR(-EINVAL);
2540
2541         if (dev_out->flags & IFF_LOOPBACK)
2542                 flags |= RTCF_LOCAL;
2543
2544         do_cache = true;
2545         if (type == RTN_BROADCAST) {
2546                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2547                 fi = NULL;
2548         } else if (type == RTN_MULTICAST) {
2549                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2550                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2551                                      fl4->flowi4_proto))
2552                         flags &= ~RTCF_LOCAL;
2553                 else
2554                         do_cache = false;
2555                 /* If multicast route do not exist use
2556                  * default one, but do not gateway in this case.
2557                  * Yes, it is hack.
2558                  */
2559                 if (fi && res->prefixlen < 4)
2560                         fi = NULL;
2561         } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2562                    (orig_oif != dev_out->ifindex)) {
2563                 /* For local routes that require a particular output interface
2564                  * we do not want to cache the result.  Caching the result
2565                  * causes incorrect behaviour when there are multiple source
2566                  * addresses on the interface, the end result being that if the
2567                  * intended recipient is waiting on that interface for the
2568                  * packet he won't receive it because it will be delivered on
2569                  * the loopback interface and the IP_PKTINFO ipi_ifindex will
2570                  * be set to the loopback interface as well.
2571                  */
2572                 do_cache = false;
2573         }
2574
2575         fnhe = NULL;
2576         do_cache &= fi != NULL;
2577         if (fi) {
2578                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2579                 struct rtable __rcu **prth;
2580
2581                 fnhe = find_exception(nhc, fl4->daddr);
2582                 if (!do_cache)
2583                         goto add;
2584                 if (fnhe) {
2585                         prth = &fnhe->fnhe_rth_output;
2586                 } else {
2587                         if (unlikely(fl4->flowi4_flags &
2588                                      FLOWI_FLAG_KNOWN_NH &&
2589                                      !(nhc->nhc_gw_family &&
2590                                        nhc->nhc_scope == RT_SCOPE_LINK))) {
2591                                 do_cache = false;
2592                                 goto add;
2593                         }
2594                         prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2595                 }
2596                 rth = rcu_dereference(*prth);
2597                 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2598                         return rth;
2599         }
2600
2601 add:
2602         rth = rt_dst_alloc(dev_out, flags, type,
2603                            IN_DEV_ORCONF(in_dev, NOPOLICY),
2604                            IN_DEV_ORCONF(in_dev, NOXFRM));
2605         if (!rth)
2606                 return ERR_PTR(-ENOBUFS);
2607
2608         rth->rt_iif = orig_oif;
2609
2610         RT_CACHE_STAT_INC(out_slow_tot);
2611
2612         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2613                 if (flags & RTCF_LOCAL &&
2614                     !(dev_out->flags & IFF_LOOPBACK)) {
2615                         rth->dst.output = ip_mc_output;
2616                         RT_CACHE_STAT_INC(out_slow_mc);
2617                 }
2618 #ifdef CONFIG_IP_MROUTE
2619                 if (type == RTN_MULTICAST) {
2620                         if (IN_DEV_MFORWARD(in_dev) &&
2621                             !ipv4_is_local_multicast(fl4->daddr)) {
2622                                 rth->dst.input = ip_mr_input;
2623                                 rth->dst.output = ip_mc_output;
2624                         }
2625                 }
2626 #endif
2627         }
2628
2629         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2630         lwtunnel_set_redirect(&rth->dst);
2631
2632         return rth;
2633 }
2634
2635 /*
2636  * Major route resolver routine.
2637  */
2638
2639 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2640                                         const struct sk_buff *skb)
2641 {
2642         struct fib_result res = {
2643                 .type           = RTN_UNSPEC,
2644                 .fi             = NULL,
2645                 .table          = NULL,
2646                 .tclassid       = 0,
2647         };
2648         struct rtable *rth;
2649
2650         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2651         ip_rt_fix_tos(fl4);
2652
2653         rcu_read_lock();
2654         rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2655         rcu_read_unlock();
2656
2657         return rth;
2658 }
2659 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2660
2661 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2662                                             struct fib_result *res,
2663                                             const struct sk_buff *skb)
2664 {
2665         struct net_device *dev_out = NULL;
2666         int orig_oif = fl4->flowi4_oif;
2667         unsigned int flags = 0;
2668         struct rtable *rth;
2669         int err;
2670
2671         if (fl4->saddr) {
2672                 if (ipv4_is_multicast(fl4->saddr) ||
2673                     ipv4_is_lbcast(fl4->saddr) ||
2674                     ipv4_is_zeronet(fl4->saddr)) {
2675                         rth = ERR_PTR(-EINVAL);
2676                         goto out;
2677                 }
2678
2679                 rth = ERR_PTR(-ENETUNREACH);
2680
2681                 /* I removed check for oif == dev_out->oif here.
2682                  * It was wrong for two reasons:
2683                  * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2684                  *    is assigned to multiple interfaces.
2685                  * 2. Moreover, we are allowed to send packets with saddr
2686                  *    of another iface. --ANK
2687                  */
2688
2689                 if (fl4->flowi4_oif == 0 &&
2690                     (ipv4_is_multicast(fl4->daddr) ||
2691                      ipv4_is_lbcast(fl4->daddr))) {
2692                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2693                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2694                         if (!dev_out)
2695                                 goto out;
2696
2697                         /* Special hack: user can direct multicasts
2698                          * and limited broadcast via necessary interface
2699                          * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2700                          * This hack is not just for fun, it allows
2701                          * vic,vat and friends to work.
2702                          * They bind socket to loopback, set ttl to zero
2703                          * and expect that it will work.
2704                          * From the viewpoint of routing cache they are broken,
2705                          * because we are not allowed to build multicast path
2706                          * with loopback source addr (look, routing cache
2707                          * cannot know, that ttl is zero, so that packet
2708                          * will not leave this host and route is valid).
2709                          * Luckily, this hack is good workaround.
2710                          */
2711
2712                         fl4->flowi4_oif = dev_out->ifindex;
2713                         goto make_route;
2714                 }
2715
2716                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2717                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2718                         if (!__ip_dev_find(net, fl4->saddr, false))
2719                                 goto out;
2720                 }
2721         }
2722
2723
2724         if (fl4->flowi4_oif) {
2725                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2726                 rth = ERR_PTR(-ENODEV);
2727                 if (!dev_out)
2728                         goto out;
2729
2730                 /* RACE: Check return value of inet_select_addr instead. */
2731                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2732                         rth = ERR_PTR(-ENETUNREACH);
2733                         goto out;
2734                 }
2735                 if (ipv4_is_local_multicast(fl4->daddr) ||
2736                     ipv4_is_lbcast(fl4->daddr) ||
2737                     fl4->flowi4_proto == IPPROTO_IGMP) {
2738                         if (!fl4->saddr)
2739                                 fl4->saddr = inet_select_addr(dev_out, 0,
2740                                                               RT_SCOPE_LINK);
2741                         goto make_route;
2742                 }
2743                 if (!fl4->saddr) {
2744                         if (ipv4_is_multicast(fl4->daddr))
2745                                 fl4->saddr = inet_select_addr(dev_out, 0,
2746                                                               fl4->flowi4_scope);
2747                         else if (!fl4->daddr)
2748                                 fl4->saddr = inet_select_addr(dev_out, 0,
2749                                                               RT_SCOPE_HOST);
2750                 }
2751         }
2752
2753         if (!fl4->daddr) {
2754                 fl4->daddr = fl4->saddr;
2755                 if (!fl4->daddr)
2756                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2757                 dev_out = net->loopback_dev;
2758                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2759                 res->type = RTN_LOCAL;
2760                 flags |= RTCF_LOCAL;
2761                 goto make_route;
2762         }
2763
2764         err = fib_lookup(net, fl4, res, 0);
2765         if (err) {
2766                 res->fi = NULL;
2767                 res->table = NULL;
2768                 if (fl4->flowi4_oif &&
2769                     (ipv4_is_multicast(fl4->daddr) ||
2770                     !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2771                         /* Apparently, routing tables are wrong. Assume,
2772                          * that the destination is on link.
2773                          *
2774                          * WHY? DW.
2775                          * Because we are allowed to send to iface
2776                          * even if it has NO routes and NO assigned
2777                          * addresses. When oif is specified, routing
2778                          * tables are looked up with only one purpose:
2779                          * to catch if destination is gatewayed, rather than
2780                          * direct. Moreover, if MSG_DONTROUTE is set,
2781                          * we send packet, ignoring both routing tables
2782                          * and ifaddr state. --ANK
2783                          *
2784                          *
2785                          * We could make it even if oif is unknown,
2786                          * likely IPv6, but we do not.
2787                          */
2788
2789                         if (fl4->saddr == 0)
2790                                 fl4->saddr = inet_select_addr(dev_out, 0,
2791                                                               RT_SCOPE_LINK);
2792                         res->type = RTN_UNICAST;
2793                         goto make_route;
2794                 }
2795                 rth = ERR_PTR(err);
2796                 goto out;
2797         }
2798
2799         if (res->type == RTN_LOCAL) {
2800                 if (!fl4->saddr) {
2801                         if (res->fi->fib_prefsrc)
2802                                 fl4->saddr = res->fi->fib_prefsrc;
2803                         else
2804                                 fl4->saddr = fl4->daddr;
2805                 }
2806
2807                 /* L3 master device is the loopback for that domain */
2808                 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2809                         net->loopback_dev;
2810
2811                 /* make sure orig_oif points to fib result device even
2812                  * though packet rx/tx happens over loopback or l3mdev
2813                  */
2814                 orig_oif = FIB_RES_OIF(*res);
2815
2816                 fl4->flowi4_oif = dev_out->ifindex;
2817                 flags |= RTCF_LOCAL;
2818                 goto make_route;
2819         }
2820
2821         fib_select_path(net, res, fl4, skb);
2822
2823         dev_out = FIB_RES_DEV(*res);
2824
2825 make_route:
2826         rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2827
2828 out:
2829         return rth;
2830 }
2831
2832 static struct dst_ops ipv4_dst_blackhole_ops = {
2833         .family                 = AF_INET,
2834         .default_advmss         = ipv4_default_advmss,
2835         .neigh_lookup           = ipv4_neigh_lookup,
2836         .check                  = dst_blackhole_check,
2837         .cow_metrics            = dst_blackhole_cow_metrics,
2838         .update_pmtu            = dst_blackhole_update_pmtu,
2839         .redirect               = dst_blackhole_redirect,
2840         .mtu                    = dst_blackhole_mtu,
2841 };
2842
2843 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2844 {
2845         struct rtable *ort = (struct rtable *) dst_orig;
2846         struct rtable *rt;
2847
2848         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2849         if (rt) {
2850                 struct dst_entry *new = &rt->dst;
2851
2852                 new->__use = 1;
2853                 new->input = dst_discard;
2854                 new->output = dst_discard_out;
2855
2856                 new->dev = net->loopback_dev;
2857                 dev_hold(new->dev);
2858
2859                 rt->rt_is_input = ort->rt_is_input;
2860                 rt->rt_iif = ort->rt_iif;
2861                 rt->rt_pmtu = ort->rt_pmtu;
2862                 rt->rt_mtu_locked = ort->rt_mtu_locked;
2863
2864                 rt->rt_genid = rt_genid_ipv4(net);
2865                 rt->rt_flags = ort->rt_flags;
2866                 rt->rt_type = ort->rt_type;
2867                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2868                 rt->rt_gw_family = ort->rt_gw_family;
2869                 if (rt->rt_gw_family == AF_INET)
2870                         rt->rt_gw4 = ort->rt_gw4;
2871                 else if (rt->rt_gw_family == AF_INET6)
2872                         rt->rt_gw6 = ort->rt_gw6;
2873
2874                 INIT_LIST_HEAD(&rt->rt_uncached);
2875         }
2876
2877         dst_release(dst_orig);
2878
2879         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2880 }
2881
2882 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2883                                     const struct sock *sk)
2884 {
2885         struct rtable *rt = __ip_route_output_key(net, flp4);
2886
2887         if (IS_ERR(rt))
2888                 return rt;
2889
2890         if (flp4->flowi4_proto) {
2891                 flp4->flowi4_oif = rt->dst.dev->ifindex;
2892                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2893                                                         flowi4_to_flowi(flp4),
2894                                                         sk, 0);
2895         }
2896
2897         return rt;
2898 }
2899 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2900
2901 struct rtable *ip_route_output_tunnel(struct sk_buff *skb,
2902                                       struct net_device *dev,
2903                                       struct net *net, __be32 *saddr,
2904                                       const struct ip_tunnel_info *info,
2905                                       u8 protocol, bool use_cache)
2906 {
2907 #ifdef CONFIG_DST_CACHE
2908         struct dst_cache *dst_cache;
2909 #endif
2910         struct rtable *rt = NULL;
2911         struct flowi4 fl4;
2912         __u8 tos;
2913
2914 #ifdef CONFIG_DST_CACHE
2915         dst_cache = (struct dst_cache *)&info->dst_cache;
2916         if (use_cache) {
2917                 rt = dst_cache_get_ip4(dst_cache, saddr);
2918                 if (rt)
2919                         return rt;
2920         }
2921 #endif
2922         memset(&fl4, 0, sizeof(fl4));
2923         fl4.flowi4_mark = skb->mark;
2924         fl4.flowi4_proto = protocol;
2925         fl4.daddr = info->key.u.ipv4.dst;
2926         fl4.saddr = info->key.u.ipv4.src;
2927         tos = info->key.tos;
2928         fl4.flowi4_tos = RT_TOS(tos);
2929
2930         rt = ip_route_output_key(net, &fl4);
2931         if (IS_ERR(rt)) {
2932                 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
2933                 return ERR_PTR(-ENETUNREACH);
2934         }
2935         if (rt->dst.dev == dev) { /* is this necessary? */
2936                 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
2937                 ip_rt_put(rt);
2938                 return ERR_PTR(-ELOOP);
2939         }
2940 #ifdef CONFIG_DST_CACHE
2941         if (use_cache)
2942                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2943 #endif
2944         *saddr = fl4.saddr;
2945         return rt;
2946 }
2947 EXPORT_SYMBOL_GPL(ip_route_output_tunnel);
2948
2949 /* called with rcu_read_lock held */
2950 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2951                         struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2952                         struct sk_buff *skb, u32 portid, u32 seq,
2953                         unsigned int flags)
2954 {
2955         struct rtmsg *r;
2956         struct nlmsghdr *nlh;
2957         unsigned long expires = 0;
2958         u32 error;
2959         u32 metrics[RTAX_MAX];
2960
2961         nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2962         if (!nlh)
2963                 return -EMSGSIZE;
2964
2965         r = nlmsg_data(nlh);
2966         r->rtm_family    = AF_INET;
2967         r->rtm_dst_len  = 32;
2968         r->rtm_src_len  = 0;
2969         r->rtm_tos      = fl4 ? fl4->flowi4_tos : 0;
2970         r->rtm_table    = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2971         if (nla_put_u32(skb, RTA_TABLE, table_id))
2972                 goto nla_put_failure;
2973         r->rtm_type     = rt->rt_type;
2974         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2975         r->rtm_protocol = RTPROT_UNSPEC;
2976         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2977         if (rt->rt_flags & RTCF_NOTIFY)
2978                 r->rtm_flags |= RTM_F_NOTIFY;
2979         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2980                 r->rtm_flags |= RTCF_DOREDIRECT;
2981
2982         if (nla_put_in_addr(skb, RTA_DST, dst))
2983                 goto nla_put_failure;
2984         if (src) {
2985                 r->rtm_src_len = 32;
2986                 if (nla_put_in_addr(skb, RTA_SRC, src))
2987                         goto nla_put_failure;
2988         }
2989         if (rt->dst.dev &&
2990             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2991                 goto nla_put_failure;
2992         if (rt->dst.lwtstate &&
2993             lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2994                 goto nla_put_failure;
2995 #ifdef CONFIG_IP_ROUTE_CLASSID
2996         if (rt->dst.tclassid &&
2997             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2998                 goto nla_put_failure;
2999 #endif
3000         if (fl4 && !rt_is_input_route(rt) &&
3001             fl4->saddr != src) {
3002                 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
3003                         goto nla_put_failure;
3004         }
3005         if (rt->rt_uses_gateway) {
3006                 if (rt->rt_gw_family == AF_INET &&
3007                     nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
3008                         goto nla_put_failure;
3009                 } else if (rt->rt_gw_family == AF_INET6) {
3010                         int alen = sizeof(struct in6_addr);
3011                         struct nlattr *nla;
3012                         struct rtvia *via;
3013
3014                         nla = nla_reserve(skb, RTA_VIA, alen + 2);
3015                         if (!nla)
3016                                 goto nla_put_failure;
3017
3018                         via = nla_data(nla);
3019                         via->rtvia_family = AF_INET6;
3020                         memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
3021                 }
3022         }
3023
3024         expires = rt->dst.expires;
3025         if (expires) {
3026                 unsigned long now = jiffies;
3027
3028                 if (time_before(now, expires))
3029                         expires -= now;
3030                 else
3031                         expires = 0;
3032         }
3033
3034         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
3035         if (rt->rt_pmtu && expires)
3036                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
3037         if (rt->rt_mtu_locked && expires)
3038                 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
3039         if (rtnetlink_put_metrics(skb, metrics) < 0)
3040                 goto nla_put_failure;
3041
3042         if (fl4) {
3043                 if (fl4->flowi4_mark &&
3044                     nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
3045                         goto nla_put_failure;
3046
3047                 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
3048                     nla_put_u32(skb, RTA_UID,
3049                                 from_kuid_munged(current_user_ns(),
3050                                                  fl4->flowi4_uid)))
3051                         goto nla_put_failure;
3052
3053                 if (rt_is_input_route(rt)) {
3054 #ifdef CONFIG_IP_MROUTE
3055                         if (ipv4_is_multicast(dst) &&
3056                             !ipv4_is_local_multicast(dst) &&
3057                             IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
3058                                 int err = ipmr_get_route(net, skb,
3059                                                          fl4->saddr, fl4->daddr,
3060                                                          r, portid);
3061
3062                                 if (err <= 0) {
3063                                         if (err == 0)
3064                                                 return 0;
3065                                         goto nla_put_failure;
3066                                 }
3067                         } else
3068 #endif
3069                                 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
3070                                         goto nla_put_failure;
3071                 }
3072         }
3073
3074         error = rt->dst.error;
3075
3076         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
3077                 goto nla_put_failure;
3078
3079         nlmsg_end(skb, nlh);
3080         return 0;
3081
3082 nla_put_failure:
3083         nlmsg_cancel(skb, nlh);
3084         return -EMSGSIZE;
3085 }
3086
3087 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
3088                             struct netlink_callback *cb, u32 table_id,
3089                             struct fnhe_hash_bucket *bucket, int genid,
3090                             int *fa_index, int fa_start, unsigned int flags)
3091 {
3092         int i;
3093
3094         for (i = 0; i < FNHE_HASH_SIZE; i++) {
3095                 struct fib_nh_exception *fnhe;
3096
3097                 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
3098                      fnhe = rcu_dereference(fnhe->fnhe_next)) {
3099                         struct rtable *rt;
3100                         int err;
3101
3102                         if (*fa_index < fa_start)
3103                                 goto next;
3104
3105                         if (fnhe->fnhe_genid != genid)
3106                                 goto next;
3107
3108                         if (fnhe->fnhe_expires &&
3109                             time_after(jiffies, fnhe->fnhe_expires))
3110                                 goto next;
3111
3112                         rt = rcu_dereference(fnhe->fnhe_rth_input);
3113                         if (!rt)
3114                                 rt = rcu_dereference(fnhe->fnhe_rth_output);
3115                         if (!rt)
3116                                 goto next;
3117
3118                         err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
3119                                            table_id, NULL, skb,
3120                                            NETLINK_CB(cb->skb).portid,
3121                                            cb->nlh->nlmsg_seq, flags);
3122                         if (err)
3123                                 return err;
3124 next:
3125                         (*fa_index)++;
3126                 }
3127         }
3128
3129         return 0;
3130 }
3131
3132 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
3133                        u32 table_id, struct fib_info *fi,
3134                        int *fa_index, int fa_start, unsigned int flags)
3135 {
3136         struct net *net = sock_net(cb->skb->sk);
3137         int nhsel, genid = fnhe_genid(net);
3138
3139         for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
3140                 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
3141                 struct fnhe_hash_bucket *bucket;
3142                 int err;
3143
3144                 if (nhc->nhc_flags & RTNH_F_DEAD)
3145                         continue;
3146
3147                 rcu_read_lock();
3148                 bucket = rcu_dereference(nhc->nhc_exceptions);
3149                 err = 0;
3150                 if (bucket)
3151                         err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
3152                                                genid, fa_index, fa_start,
3153                                                flags);
3154                 rcu_read_unlock();
3155                 if (err)
3156                         return err;
3157         }
3158
3159         return 0;
3160 }
3161
3162 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3163                                                    u8 ip_proto, __be16 sport,
3164                                                    __be16 dport)
3165 {
3166         struct sk_buff *skb;
3167         struct iphdr *iph;
3168
3169         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3170         if (!skb)
3171                 return NULL;
3172
3173         /* Reserve room for dummy headers, this skb can pass
3174          * through good chunk of routing engine.
3175          */
3176         skb_reset_mac_header(skb);
3177         skb_reset_network_header(skb);
3178         skb->protocol = htons(ETH_P_IP);
3179         iph = skb_put(skb, sizeof(struct iphdr));
3180         iph->protocol = ip_proto;
3181         iph->saddr = src;
3182         iph->daddr = dst;
3183         iph->version = 0x4;
3184         iph->frag_off = 0;
3185         iph->ihl = 0x5;
3186         skb_set_transport_header(skb, skb->len);
3187
3188         switch (iph->protocol) {
3189         case IPPROTO_UDP: {
3190                 struct udphdr *udph;
3191
3192                 udph = skb_put_zero(skb, sizeof(struct udphdr));
3193                 udph->source = sport;
3194                 udph->dest = dport;
3195                 udph->len = htons(sizeof(struct udphdr));
3196                 udph->check = 0;
3197                 break;
3198         }
3199         case IPPROTO_TCP: {
3200                 struct tcphdr *tcph;
3201
3202                 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3203                 tcph->source    = sport;
3204                 tcph->dest      = dport;
3205                 tcph->doff      = sizeof(struct tcphdr) / 4;
3206                 tcph->rst = 1;
3207                 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3208                                             src, dst, 0);
3209                 break;
3210         }
3211         case IPPROTO_ICMP: {
3212                 struct icmphdr *icmph;
3213
3214                 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3215                 icmph->type = ICMP_ECHO;
3216                 icmph->code = 0;
3217         }
3218         }
3219
3220         return skb;
3221 }
3222
3223 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3224                                        const struct nlmsghdr *nlh,
3225                                        struct nlattr **tb,
3226                                        struct netlink_ext_ack *extack)
3227 {
3228         struct rtmsg *rtm;
3229         int i, err;
3230
3231         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3232                 NL_SET_ERR_MSG(extack,
3233                                "ipv4: Invalid header for route get request");
3234                 return -EINVAL;
3235         }
3236
3237         if (!netlink_strict_get_check(skb))
3238                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3239                                               rtm_ipv4_policy, extack);
3240
3241         rtm = nlmsg_data(nlh);
3242         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3243             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3244             rtm->rtm_table || rtm->rtm_protocol ||
3245             rtm->rtm_scope || rtm->rtm_type) {
3246                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3247                 return -EINVAL;
3248         }
3249
3250         if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3251                                RTM_F_LOOKUP_TABLE |
3252                                RTM_F_FIB_MATCH)) {
3253                 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3254                 return -EINVAL;
3255         }
3256
3257         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3258                                             rtm_ipv4_policy, extack);
3259         if (err)
3260                 return err;
3261
3262         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3263             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3264                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3265                 return -EINVAL;
3266         }
3267
3268         for (i = 0; i <= RTA_MAX; i++) {
3269                 if (!tb[i])
3270                         continue;
3271
3272                 switch (i) {
3273                 case RTA_IIF:
3274                 case RTA_OIF:
3275                 case RTA_SRC:
3276                 case RTA_DST:
3277                 case RTA_IP_PROTO:
3278                 case RTA_SPORT:
3279                 case RTA_DPORT:
3280                 case RTA_MARK:
3281                 case RTA_UID:
3282                         break;
3283                 default:
3284                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3285                         return -EINVAL;
3286                 }
3287         }
3288
3289         return 0;
3290 }
3291
3292 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3293                              struct netlink_ext_ack *extack)
3294 {
3295         struct net *net = sock_net(in_skb->sk);
3296         struct nlattr *tb[RTA_MAX+1];
3297         u32 table_id = RT_TABLE_MAIN;
3298         __be16 sport = 0, dport = 0;
3299         struct fib_result res = {};
3300         u8 ip_proto = IPPROTO_UDP;
3301         struct rtable *rt = NULL;
3302         struct sk_buff *skb;
3303         struct rtmsg *rtm;
3304         struct flowi4 fl4 = {};
3305         __be32 dst = 0;
3306         __be32 src = 0;
3307         kuid_t uid;
3308         u32 iif;
3309         int err;
3310         int mark;
3311
3312         err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3313         if (err < 0)
3314                 return err;
3315
3316         rtm = nlmsg_data(nlh);
3317         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3318         dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3319         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3320         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3321         if (tb[RTA_UID])
3322                 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3323         else
3324                 uid = (iif ? INVALID_UID : current_uid());
3325
3326         if (tb[RTA_IP_PROTO]) {
3327                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3328                                                   &ip_proto, AF_INET, extack);
3329                 if (err)
3330                         return err;
3331         }
3332
3333         if (tb[RTA_SPORT])
3334                 sport = nla_get_be16(tb[RTA_SPORT]);
3335
3336         if (tb[RTA_DPORT])
3337                 dport = nla_get_be16(tb[RTA_DPORT]);
3338
3339         skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3340         if (!skb)
3341                 return -ENOBUFS;
3342
3343         fl4.daddr = dst;
3344         fl4.saddr = src;
3345         fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3346         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3347         fl4.flowi4_mark = mark;
3348         fl4.flowi4_uid = uid;
3349         if (sport)
3350                 fl4.fl4_sport = sport;
3351         if (dport)
3352                 fl4.fl4_dport = dport;
3353         fl4.flowi4_proto = ip_proto;
3354
3355         rcu_read_lock();
3356
3357         if (iif) {
3358                 struct net_device *dev;
3359
3360                 dev = dev_get_by_index_rcu(net, iif);
3361                 if (!dev) {
3362                         err = -ENODEV;
3363                         goto errout_rcu;
3364                 }
3365
3366                 fl4.flowi4_iif = iif; /* for rt_fill_info */
3367                 skb->dev        = dev;
3368                 skb->mark       = mark;
3369                 err = ip_route_input_rcu(skb, dst, src,
3370                                          rtm->rtm_tos & IPTOS_RT_MASK, dev,
3371                                          &res);
3372
3373                 rt = skb_rtable(skb);
3374                 if (err == 0 && rt->dst.error)
3375                         err = -rt->dst.error;
3376         } else {
3377                 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3378                 skb->dev = net->loopback_dev;
3379                 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3380                 err = 0;
3381                 if (IS_ERR(rt))
3382                         err = PTR_ERR(rt);
3383                 else
3384                         skb_dst_set(skb, &rt->dst);
3385         }
3386
3387         if (err)
3388                 goto errout_rcu;
3389
3390         if (rtm->rtm_flags & RTM_F_NOTIFY)
3391                 rt->rt_flags |= RTCF_NOTIFY;
3392
3393         if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3394                 table_id = res.table ? res.table->tb_id : 0;
3395
3396         /* reset skb for netlink reply msg */
3397         skb_trim(skb, 0);
3398         skb_reset_network_header(skb);
3399         skb_reset_transport_header(skb);
3400         skb_reset_mac_header(skb);
3401
3402         if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3403                 struct fib_rt_info fri;
3404
3405                 if (!res.fi) {
3406                         err = fib_props[res.type].error;
3407                         if (!err)
3408                                 err = -EHOSTUNREACH;
3409                         goto errout_rcu;
3410                 }
3411                 fri.fi = res.fi;
3412                 fri.tb_id = table_id;
3413                 fri.dst = res.prefix;
3414                 fri.dst_len = res.prefixlen;
3415                 fri.tos = fl4.flowi4_tos;
3416                 fri.type = rt->rt_type;
3417                 fri.offload = 0;
3418                 fri.trap = 0;
3419                 fri.offload_failed = 0;
3420                 if (res.fa_head) {
3421                         struct fib_alias *fa;
3422
3423                         hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3424                                 u8 slen = 32 - fri.dst_len;
3425
3426                                 if (fa->fa_slen == slen &&
3427                                     fa->tb_id == fri.tb_id &&
3428                                     fa->fa_tos == fri.tos &&
3429                                     fa->fa_info == res.fi &&
3430                                     fa->fa_type == fri.type) {
3431                                         fri.offload = READ_ONCE(fa->offload);
3432                                         fri.trap = READ_ONCE(fa->trap);
3433                                         fri.offload_failed =
3434                                                 READ_ONCE(fa->offload_failed);
3435                                         break;
3436                                 }
3437                         }
3438                 }
3439                 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3440                                     nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3441         } else {
3442                 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3443                                    NETLINK_CB(in_skb).portid,
3444                                    nlh->nlmsg_seq, 0);
3445         }
3446         if (err < 0)
3447                 goto errout_rcu;
3448
3449         rcu_read_unlock();
3450
3451         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3452
3453 errout_free:
3454         return err;
3455 errout_rcu:
3456         rcu_read_unlock();
3457         kfree_skb(skb);
3458         goto errout_free;
3459 }
3460
3461 void ip_rt_multicast_event(struct in_device *in_dev)
3462 {
3463         rt_cache_flush(dev_net(in_dev->dev));
3464 }
3465
3466 #ifdef CONFIG_SYSCTL
3467 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3468 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
3469 static int ip_rt_gc_elasticity __read_mostly    = 8;
3470 static int ip_min_valid_pmtu __read_mostly      = IPV4_MIN_MTU;
3471
3472 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3473                 void *buffer, size_t *lenp, loff_t *ppos)
3474 {
3475         struct net *net = (struct net *)__ctl->extra1;
3476
3477         if (write) {
3478                 rt_cache_flush(net);
3479                 fnhe_genid_bump(net);
3480                 return 0;
3481         }
3482
3483         return -EINVAL;
3484 }
3485
3486 static struct ctl_table ipv4_route_table[] = {
3487         {
3488                 .procname       = "gc_thresh",
3489                 .data           = &ipv4_dst_ops.gc_thresh,
3490                 .maxlen         = sizeof(int),
3491                 .mode           = 0644,
3492                 .proc_handler   = proc_dointvec,
3493         },
3494         {
3495                 .procname       = "max_size",
3496                 .data           = &ip_rt_max_size,
3497                 .maxlen         = sizeof(int),
3498                 .mode           = 0644,
3499                 .proc_handler   = proc_dointvec,
3500         },
3501         {
3502                 /*  Deprecated. Use gc_min_interval_ms */
3503
3504                 .procname       = "gc_min_interval",
3505                 .data           = &ip_rt_gc_min_interval,
3506                 .maxlen         = sizeof(int),
3507                 .mode           = 0644,
3508                 .proc_handler   = proc_dointvec_jiffies,
3509         },
3510         {
3511                 .procname       = "gc_min_interval_ms",
3512                 .data           = &ip_rt_gc_min_interval,
3513                 .maxlen         = sizeof(int),
3514                 .mode           = 0644,
3515                 .proc_handler   = proc_dointvec_ms_jiffies,
3516         },
3517         {
3518                 .procname       = "gc_timeout",
3519                 .data           = &ip_rt_gc_timeout,
3520                 .maxlen         = sizeof(int),
3521                 .mode           = 0644,
3522                 .proc_handler   = proc_dointvec_jiffies,
3523         },
3524         {
3525                 .procname       = "gc_interval",
3526                 .data           = &ip_rt_gc_interval,
3527                 .maxlen         = sizeof(int),
3528                 .mode           = 0644,
3529                 .proc_handler   = proc_dointvec_jiffies,
3530         },
3531         {
3532                 .procname       = "redirect_load",
3533                 .data           = &ip_rt_redirect_load,
3534                 .maxlen         = sizeof(int),
3535                 .mode           = 0644,
3536                 .proc_handler   = proc_dointvec,
3537         },
3538         {
3539                 .procname       = "redirect_number",
3540                 .data           = &ip_rt_redirect_number,
3541                 .maxlen         = sizeof(int),
3542                 .mode           = 0644,
3543                 .proc_handler   = proc_dointvec,
3544         },
3545         {
3546                 .procname       = "redirect_silence",
3547                 .data           = &ip_rt_redirect_silence,
3548                 .maxlen         = sizeof(int),
3549                 .mode           = 0644,
3550                 .proc_handler   = proc_dointvec,
3551         },
3552         {
3553                 .procname       = "error_cost",
3554                 .data           = &ip_rt_error_cost,
3555                 .maxlen         = sizeof(int),
3556                 .mode           = 0644,
3557                 .proc_handler   = proc_dointvec,
3558         },
3559         {
3560                 .procname       = "error_burst",
3561                 .data           = &ip_rt_error_burst,
3562                 .maxlen         = sizeof(int),
3563                 .mode           = 0644,
3564                 .proc_handler   = proc_dointvec,
3565         },
3566         {
3567                 .procname       = "gc_elasticity",
3568                 .data           = &ip_rt_gc_elasticity,
3569                 .maxlen         = sizeof(int),
3570                 .mode           = 0644,
3571                 .proc_handler   = proc_dointvec,
3572         },
3573         {
3574                 .procname       = "mtu_expires",
3575                 .data           = &ip_rt_mtu_expires,
3576                 .maxlen         = sizeof(int),
3577                 .mode           = 0644,
3578                 .proc_handler   = proc_dointvec_jiffies,
3579         },
3580         {
3581                 .procname       = "min_pmtu",
3582                 .data           = &ip_rt_min_pmtu,
3583                 .maxlen         = sizeof(int),
3584                 .mode           = 0644,
3585                 .proc_handler   = proc_dointvec_minmax,
3586                 .extra1         = &ip_min_valid_pmtu,
3587         },
3588         {
3589                 .procname       = "min_adv_mss",
3590                 .data           = &ip_rt_min_advmss,
3591                 .maxlen         = sizeof(int),
3592                 .mode           = 0644,
3593                 .proc_handler   = proc_dointvec,
3594         },
3595         { }
3596 };
3597
3598 static const char ipv4_route_flush_procname[] = "flush";
3599
3600 static struct ctl_table ipv4_route_flush_table[] = {
3601         {
3602                 .procname       = ipv4_route_flush_procname,
3603                 .maxlen         = sizeof(int),
3604                 .mode           = 0200,
3605                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3606         },
3607         { },
3608 };
3609
3610 static __net_init int sysctl_route_net_init(struct net *net)
3611 {
3612         struct ctl_table *tbl;
3613
3614         tbl = ipv4_route_flush_table;
3615         if (!net_eq(net, &init_net)) {
3616                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3617                 if (!tbl)
3618                         goto err_dup;
3619
3620                 /* Don't export non-whitelisted sysctls to unprivileged users */
3621                 if (net->user_ns != &init_user_ns) {
3622                         if (tbl[0].procname != ipv4_route_flush_procname)
3623                                 tbl[0].procname = NULL;
3624                 }
3625         }
3626         tbl[0].extra1 = net;
3627
3628         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3629         if (!net->ipv4.route_hdr)
3630                 goto err_reg;
3631         return 0;
3632
3633 err_reg:
3634         if (tbl != ipv4_route_flush_table)
3635                 kfree(tbl);
3636 err_dup:
3637         return -ENOMEM;
3638 }
3639
3640 static __net_exit void sysctl_route_net_exit(struct net *net)
3641 {
3642         struct ctl_table *tbl;
3643
3644         tbl = net->ipv4.route_hdr->ctl_table_arg;
3645         unregister_net_sysctl_table(net->ipv4.route_hdr);
3646         BUG_ON(tbl == ipv4_route_flush_table);
3647         kfree(tbl);
3648 }
3649
3650 static __net_initdata struct pernet_operations sysctl_route_ops = {
3651         .init = sysctl_route_net_init,
3652         .exit = sysctl_route_net_exit,
3653 };
3654 #endif
3655
3656 static __net_init int rt_genid_init(struct net *net)
3657 {
3658         atomic_set(&net->ipv4.rt_genid, 0);
3659         atomic_set(&net->fnhe_genid, 0);
3660         atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3661         return 0;
3662 }
3663
3664 static __net_initdata struct pernet_operations rt_genid_ops = {
3665         .init = rt_genid_init,
3666 };
3667
3668 static int __net_init ipv4_inetpeer_init(struct net *net)
3669 {
3670         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3671
3672         if (!bp)
3673                 return -ENOMEM;
3674         inet_peer_base_init(bp);
3675         net->ipv4.peers = bp;
3676         return 0;
3677 }
3678
3679 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3680 {
3681         struct inet_peer_base *bp = net->ipv4.peers;
3682
3683         net->ipv4.peers = NULL;
3684         inetpeer_invalidate_tree(bp);
3685         kfree(bp);
3686 }
3687
3688 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3689         .init   =       ipv4_inetpeer_init,
3690         .exit   =       ipv4_inetpeer_exit,
3691 };
3692
3693 #ifdef CONFIG_IP_ROUTE_CLASSID
3694 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3695 #endif /* CONFIG_IP_ROUTE_CLASSID */
3696
3697 int __init ip_rt_init(void)
3698 {
3699         void *idents_hash;
3700         int cpu;
3701
3702         /* For modern hosts, this will use 2 MB of memory */
3703         idents_hash = alloc_large_system_hash("IP idents",
3704                                               sizeof(*ip_idents) + sizeof(*ip_tstamps),
3705                                               0,
3706                                               16, /* one bucket per 64 KB */
3707                                               HASH_ZERO,
3708                                               NULL,
3709                                               &ip_idents_mask,
3710                                               2048,
3711                                               256*1024);
3712
3713         ip_idents = idents_hash;
3714
3715         prandom_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3716
3717         ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3718
3719         for_each_possible_cpu(cpu) {
3720                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3721
3722                 INIT_LIST_HEAD(&ul->head);
3723                 spin_lock_init(&ul->lock);
3724         }
3725 #ifdef CONFIG_IP_ROUTE_CLASSID
3726         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3727         if (!ip_rt_acct)
3728                 panic("IP: failed to allocate ip_rt_acct\n");
3729 #endif
3730
3731         ipv4_dst_ops.kmem_cachep =
3732                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3733                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3734
3735         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3736
3737         if (dst_entries_init(&ipv4_dst_ops) < 0)
3738                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3739
3740         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3741                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3742
3743         ipv4_dst_ops.gc_thresh = ~0;
3744         ip_rt_max_size = INT_MAX;
3745
3746         devinet_init();
3747         ip_fib_init();
3748
3749         if (ip_rt_proc_init())
3750                 pr_err("Unable to create route proc files\n");
3751 #ifdef CONFIG_XFRM
3752         xfrm_init();
3753         xfrm4_init();
3754 #endif
3755         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3756                       RTNL_FLAG_DOIT_UNLOCKED);
3757
3758 #ifdef CONFIG_SYSCTL
3759         register_pernet_subsys(&sysctl_route_ops);
3760 #endif
3761         register_pernet_subsys(&rt_genid_ops);
3762         register_pernet_subsys(&ipv4_inetpeer_ops);
3763         return 0;
3764 }
3765
3766 #ifdef CONFIG_SYSCTL
3767 /*
3768  * We really need to sanitize the damn ipv4 init order, then all
3769  * this nonsense will go away.
3770  */
3771 void __init ip_static_sysctl_init(void)
3772 {
3773         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3774 }
3775 #endif