net/ipv4/route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              ROUTE - implementation of the IP router.
   8  *
   9  * Authors:     Ross Biro
  10  *              Fred N. van Kempen, <[email protected]>
  11  *              Alan Cox, <[email protected]>
  12  *              Linus Torvalds, <[email protected]>
  13  *              Alexey Kuznetsov, <[email protected]>
  14  *
  15  * Fixes:
  16  *              Alan Cox        :       Verify area fixes.
  17  *              Alan Cox        :       cli() protects routing changes
  18  *              Rui Oliveira    :       ICMP routing table updates
  19  *              ([email protected])      Routing table insertion and update
  20  *              Linus Torvalds  :       Rewrote bits to be sensible
  21  *              Alan Cox        :       Added BSD route gw semantics
  22  *              Alan Cox        :       Super /proc >4K
  23  *              Alan Cox        :       MTU in route table
  24  *              Alan Cox        :       MSS actually. Also added the window
  25  *                                      clamper.
  26  *              Sam Lantinga    :       Fixed route matching in rt_del()
  27  *              Alan Cox        :       Routing cache support.
  28  *              Alan Cox        :       Removed compatibility cruft.
  29  *              Alan Cox        :       RTF_REJECT support.
  30  *              Alan Cox        :       TCP irtt support.
  31  *              Jonathan Naylor :       Added Metric support.
  32  *      Miquel van Smoorenburg  :       BSD API fixes.
  33  *      Miquel van Smoorenburg  :       Metrics.
  34  *              Alan Cox        :       Use __u32 properly
  35  *              Alan Cox        :       Aligned routing errors more closely with BSD
  36  *                                      our system is still very different.
  37  *              Alan Cox        :       Faster /proc handling
  38  *      Alexey Kuznetsov        :       Massive rework to support tree based routing,
  39  *                                      routing caches and better behaviour.
  40  *
  41  *              Olaf Erb        :       irtt wasn't being copied right.
  42  *              Bjorn Ekwall    :       Kerneld route support.
  43  *              Alan Cox        :       Multicast fixed (I hope)
  44  *              Pavel Krauz     :       Limited broadcast fixed
  45  *              Mike McLagan    :       Routing by source
  46  *      Alexey Kuznetsov        :       End of old history. Split to fib.c and
  47  *                                      route.c and rewritten from scratch.
  48  *              Andi Kleen      :       Load-limit warning messages.
  49  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  50  *      Vitaly E. Lavrov        :       Race condition in ip_route_input_slow.
  51  *      Tobias Ringstrom        :       Uninitialized res.type in ip_route_output_slow.
  52  *      Vladimir V. Ivanov      :       IP rule info (flowid) is really useful.
  53  *              Marc Boucher    :       routing by fwmark
  54  *      Robert Olsson           :       Added rt_cache statistics
  55  *      Arnaldo C. Melo         :       Convert proc stuff to seq_file
  56  *      Eric Dumazet            :       hashed spinlocks and rt_check_expire() fixes.
  57  *      Ilia Sotnikov           :       Ignore TOS on PMTUD and Redirect
  58  *      Ilia Sotnikov           :       Removed TOS from hash calculations
  59  */
  60
  61 #define pr_fmt(fmt) "IPv4: " fmt
  62
  63 #include <linux/module.h>
  64 #include <linux/uaccess.h>
  65 #include <linux/bitops.h>
  66 #include <linux/types.h>
  67 #include <linux/kernel.h>
  68 #include <linux/mm.h>
  69 #include <linux/memblock.h>
  70 #include <linux/string.h>
  71 #include <linux/socket.h>
  72 #include <linux/sockios.h>
  73 #include <linux/errno.h>
  74 #include <linux/in.h>
  75 #include <linux/inet.h>
  76 #include <linux/netdevice.h>
  77 #include <linux/proc_fs.h>
  78 #include <linux/init.h>
  79 #include <linux/skbuff.h>
  80 #include <linux/inetdevice.h>
  81 #include <linux/igmp.h>
  82 #include <linux/pkt_sched.h>
  83 #include <linux/mroute.h>
  84 #include <linux/netfilter_ipv4.h>
  85 #include <linux/random.h>
  86 #include <linux/rcupdate.h>
  87 #include <linux/times.h>
  88 #include <linux/slab.h>
  89 #include <linux/jhash.h>
  90 #include <net/dst.h>
  91 #include <net/dst_metadata.h>
  92 #include <net/net_namespace.h>
  93 #include <net/protocol.h>
  94 #include <net/ip.h>
  95 #include <net/route.h>
  96 #include <net/inetpeer.h>
  97 #include <net/sock.h>
  98 #include <net/ip_fib.h>
  99 #include <net/nexthop.h>
 100 #include <net/arp.h>
 101 #include <net/tcp.h>
 102 #include <net/icmp.h>
 103 #include <net/xfrm.h>
 104 #include <net/lwtunnel.h>
 105 #include <net/netevent.h>
 106 #include <net/rtnetlink.h>
 107 #ifdef CONFIG_SYSCTL
 108 #include <linux/sysctl.h>
 109 #endif
 110 #include <net/secure_seq.h>
 111 #include <net/ip_tunnels.h>
 112 #include <net/l3mdev.h>
 113
 114 #include "fib_lookup.h"
 115
 116 #define RT_FL_TOS(oldflp4) \
 117         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
 118
 119 #define RT_GC_TIMEOUT (300*HZ)
 120
 121 static int ip_rt_max_size;
 122 static int ip_rt_redirect_number __read_mostly  = 9;
 123 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 124 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 125 static int ip_rt_error_cost __read_mostly       = HZ;
 126 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 127 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
 128 static u32 ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
 129 static int ip_rt_min_advmss __read_mostly       = 256;
 130
 131 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
 132
 133 /*
 134  *      Interface to generic destination cache.
 135  */
 136
 137 INDIRECT_CALLABLE_SCOPE
 138 struct dst_entry        *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 139 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 140 INDIRECT_CALLABLE_SCOPE
 141 unsigned int            ipv4_mtu(const struct dst_entry *dst);
 142 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 143 static void              ipv4_link_failure(struct sk_buff *skb);
 144 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 145                                            struct sk_buff *skb, u32 mtu,
 146                                            bool confirm_neigh);
 147 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 148                                         struct sk_buff *skb);
 149 static void             ipv4_dst_destroy(struct dst_entry *dst);
 150
 151 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 152 {
 153         WARN_ON(1);
 154         return NULL;
 155 }
 156
 157 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 158                                            struct sk_buff *skb,
 159                                            const void *daddr);
 160 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
 161
 162 static struct dst_ops ipv4_dst_ops = {
 163         .family =               AF_INET,
 164         .check =                ipv4_dst_check,
 165         .default_advmss =       ipv4_default_advmss,
 166         .mtu =                  ipv4_mtu,
 167         .cow_metrics =          ipv4_cow_metrics,
 168         .destroy =              ipv4_dst_destroy,
 169         .negative_advice =      ipv4_negative_advice,
 170         .link_failure =         ipv4_link_failure,
 171         .update_pmtu =          ip_rt_update_pmtu,
 172         .redirect =             ip_do_redirect,
 173         .local_out =            __ip_local_out,
 174         .neigh_lookup =         ipv4_neigh_lookup,
 175         .confirm_neigh =        ipv4_confirm_neigh,
 176 };
 177
 178 #define ECN_OR_COST(class)      TC_PRIO_##class
 179
 180 const __u8 ip_tos2prio[16] = {
 181         TC_PRIO_BESTEFFORT,
 182         ECN_OR_COST(BESTEFFORT),
 183         TC_PRIO_BESTEFFORT,
 184         ECN_OR_COST(BESTEFFORT),
 185         TC_PRIO_BULK,
 186         ECN_OR_COST(BULK),
 187         TC_PRIO_BULK,
 188         ECN_OR_COST(BULK),
 189         TC_PRIO_INTERACTIVE,
 190         ECN_OR_COST(INTERACTIVE),
 191         TC_PRIO_INTERACTIVE,
 192         ECN_OR_COST(INTERACTIVE),
 193         TC_PRIO_INTERACTIVE_BULK,
 194         ECN_OR_COST(INTERACTIVE_BULK),
 195         TC_PRIO_INTERACTIVE_BULK,
 196         ECN_OR_COST(INTERACTIVE_BULK)
 197 };
 198 EXPORT_SYMBOL(ip_tos2prio);
 199
 200 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 201 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
 202
 203 #ifdef CONFIG_PROC_FS
 204 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 205 {
 206         if (*pos)
 207                 return NULL;
 208         return SEQ_START_TOKEN;
 209 }
 210
 211 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 212 {
 213         ++*pos;
 214         return NULL;
 215 }
 216
 217 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 218 {
 219 }
 220
 221 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 222 {
 223         if (v == SEQ_START_TOKEN)
 224                 seq_printf(seq, "%-127s\n",
 225                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 226                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 227                            "HHUptod\tSpecDst");
 228         return 0;
 229 }
 230
 231 static const struct seq_operations rt_cache_seq_ops = {
 232         .start  = rt_cache_seq_start,
 233         .next   = rt_cache_seq_next,
 234         .stop   = rt_cache_seq_stop,
 235         .show   = rt_cache_seq_show,
 236 };
 237
 238 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 239 {
 240         int cpu;
 241
 242         if (*pos == 0)
 243                 return SEQ_START_TOKEN;
 244
 245         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 246                 if (!cpu_possible(cpu))
 247                         continue;
 248                 *pos = cpu+1;
 249                 return &per_cpu(rt_cache_stat, cpu);
 250         }
 251         return NULL;
 252 }
 253
 254 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 255 {
 256         int cpu;
 257
 258         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 259                 if (!cpu_possible(cpu))
 260                         continue;
 261                 *pos = cpu+1;
 262                 return &per_cpu(rt_cache_stat, cpu);
 263         }
 264         (*pos)++;
 265         return NULL;
 266
 267 }
 268
 269 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 270 {
 271
 272 }
 273
 274 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 275 {
 276         struct rt_cache_stat *st = v;
 277
 278         if (v == SEQ_START_TOKEN) {
 279                 seq_printf(seq, "entries  in_hit in_slow_tot in_slow_mc in_no_route in_brd in_martian_dst in_martian_src  out_hit out_slow_tot out_slow_mc  gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
 280                 return 0;
 281         }
 282
 283         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 284                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 285                    dst_entries_get_slow(&ipv4_dst_ops),
 286                    0, /* st->in_hit */
 287                    st->in_slow_tot,
 288                    st->in_slow_mc,
 289                    st->in_no_route,
 290                    st->in_brd,
 291                    st->in_martian_dst,
 292                    st->in_martian_src,
 293
 294                    0, /* st->out_hit */
 295                    st->out_slow_tot,
 296                    st->out_slow_mc,
 297
 298                    0, /* st->gc_total */
 299                    0, /* st->gc_ignored */
 300                    0, /* st->gc_goal_miss */
 301                    0, /* st->gc_dst_overflow */
 302                    0, /* st->in_hlist_search */
 303                    0  /* st->out_hlist_search */
 304                 );
 305         return 0;
 306 }
 307
 308 static const struct seq_operations rt_cpu_seq_ops = {
 309         .start  = rt_cpu_seq_start,
 310         .next   = rt_cpu_seq_next,
 311         .stop   = rt_cpu_seq_stop,
 312         .show   = rt_cpu_seq_show,
 313 };
 314
 315 #ifdef CONFIG_IP_ROUTE_CLASSID
 316 static int rt_acct_proc_show(struct seq_file *m, void *v)
 317 {
 318         struct ip_rt_acct *dst, *src;
 319         unsigned int i, j;
 320
 321         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 322         if (!dst)
 323                 return -ENOMEM;
 324
 325         for_each_possible_cpu(i) {
 326                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 327                 for (j = 0; j < 256; j++) {
 328                         dst[j].o_bytes   += src[j].o_bytes;
 329                         dst[j].o_packets += src[j].o_packets;
 330                         dst[j].i_bytes   += src[j].i_bytes;
 331                         dst[j].i_packets += src[j].i_packets;
 332                 }
 333         }
 334
 335         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 336         kfree(dst);
 337         return 0;
 338 }
 339 #endif
 340
 341 static int __net_init ip_rt_do_proc_init(struct net *net)
 342 {
 343         struct proc_dir_entry *pde;
 344
 345         pde = proc_create_seq("rt_cache", 0444, net->proc_net,
 346                               &rt_cache_seq_ops);
 347         if (!pde)
 348                 goto err1;
 349
 350         pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat,
 351                               &rt_cpu_seq_ops);
 352         if (!pde)
 353                 goto err2;
 354
 355 #ifdef CONFIG_IP_ROUTE_CLASSID
 356         pde = proc_create_single("rt_acct", 0, net->proc_net,
 357                         rt_acct_proc_show);
 358         if (!pde)
 359                 goto err3;
 360 #endif
 361         return 0;
 362
 363 #ifdef CONFIG_IP_ROUTE_CLASSID
 364 err3:
 365         remove_proc_entry("rt_cache", net->proc_net_stat);
 366 #endif
 367 err2:
 368         remove_proc_entry("rt_cache", net->proc_net);
 369 err1:
 370         return -ENOMEM;
 371 }
 372
 373 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 374 {
 375         remove_proc_entry("rt_cache", net->proc_net_stat);
 376         remove_proc_entry("rt_cache", net->proc_net);
 377 #ifdef CONFIG_IP_ROUTE_CLASSID
 378         remove_proc_entry("rt_acct", net->proc_net);
 379 #endif
 380 }
 381
 382 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 383         .init = ip_rt_do_proc_init,
 384         .exit = ip_rt_do_proc_exit,
 385 };
 386
 387 static int __init ip_rt_proc_init(void)
 388 {
 389         return register_pernet_subsys(&ip_rt_proc_ops);
 390 }
 391
 392 #else
 393 static inline int ip_rt_proc_init(void)
 394 {
 395         return 0;
 396 }
 397 #endif /* CONFIG_PROC_FS */
 398
 399 static inline bool rt_is_expired(const struct rtable *rth)
 400 {
 401         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 402 }
 403
 404 void rt_cache_flush(struct net *net)
 405 {
 406         rt_genid_bump_ipv4(net);
 407 }
 408
 409 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 410                                            struct sk_buff *skb,
 411                                            const void *daddr)
 412 {
 413         const struct rtable *rt = container_of(dst, struct rtable, dst);
 414         struct net_device *dev = dst->dev;
 415         struct neighbour *n;
 416
 417         rcu_read_lock_bh();
 418
 419         if (likely(rt->rt_gw_family == AF_INET)) {
 420                 n = ip_neigh_gw4(dev, rt->rt_gw4);
 421         } else if (rt->rt_gw_family == AF_INET6) {
 422                 n = ip_neigh_gw6(dev, &rt->rt_gw6);
 423         } else {
 424                 __be32 pkey;
 425
 426                 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
 427                 n = ip_neigh_gw4(dev, pkey);
 428         }
 429
 430         if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
 431                 n = NULL;
 432
 433         rcu_read_unlock_bh();
 434
 435         return n;
 436 }
 437
 438 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 439 {
 440         const struct rtable *rt = container_of(dst, struct rtable, dst);
 441         struct net_device *dev = dst->dev;
 442         const __be32 *pkey = daddr;
 443
 444         if (rt->rt_gw_family == AF_INET) {
 445                 pkey = (const __be32 *)&rt->rt_gw4;
 446         } else if (rt->rt_gw_family == AF_INET6) {
 447                 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
 448         } else if (!daddr ||
 449                  (rt->rt_flags &
 450                   (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
 451                 return;
 452         }
 453         __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
 454 }
 455
 456 /* Hash tables of size 2048..262144 depending on RAM size.
 457  * Each bucket uses 8 bytes.
 458  */
 459 static u32 ip_idents_mask __read_mostly;
 460 static atomic_t *ip_idents __read_mostly;
 461 static u32 *ip_tstamps __read_mostly;
 462
 463 /* In order to protect privacy, we add a perturbation to identifiers
 464  * if one generator is seldom used. This makes hard for an attacker
 465  * to infer how many packets were sent between two points in time.
 466  */
 467 u32 ip_idents_reserve(u32 hash, int segs)
 468 {
 469         u32 bucket, old, now = (u32)jiffies;
 470         atomic_t *p_id;
 471         u32 *p_tstamp;
 472         u32 delta = 0;
 473
 474         bucket = hash & ip_idents_mask;
 475         p_tstamp = ip_tstamps + bucket;
 476         p_id = ip_idents + bucket;
 477         old = READ_ONCE(*p_tstamp);
 478
 479         if (old != now && cmpxchg(p_tstamp, old, now) == old)
 480                 delta = prandom_u32_max(now - old);
 481
 482         /* If UBSAN reports an error there, please make sure your compiler
 483          * supports -fno-strict-overflow before reporting it that was a bug
 484          * in UBSAN, and it has been fixed in GCC-8.
 485          */
 486         return atomic_add_return(segs + delta, p_id) - segs;
 487 }
 488 EXPORT_SYMBOL(ip_idents_reserve);
 489
 490 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
 491 {
 492         u32 hash, id;
 493
 494         /* Note the following code is not safe, but this is okay. */
 495         if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
 496                 get_random_bytes(&net->ipv4.ip_id_key,
 497                                  sizeof(net->ipv4.ip_id_key));
 498
 499         hash = siphash_3u32((__force u32)iph->daddr,
 500                             (__force u32)iph->saddr,
 501                             iph->protocol,
 502                             &net->ipv4.ip_id_key);
 503         id = ip_idents_reserve(hash, segs);
 504         iph->id = htons(id);
 505 }
 506 EXPORT_SYMBOL(__ip_select_ident);
 507
 508 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
 509                              const struct sock *sk,
 510                              const struct iphdr *iph,
 511                              int oif, u8 tos,
 512                              u8 prot, u32 mark, int flow_flags)
 513 {
 514         if (sk) {
 515                 const struct inet_sock *inet = inet_sk(sk);
 516
 517                 oif = sk->sk_bound_dev_if;
 518                 mark = sk->sk_mark;
 519                 tos = RT_CONN_FLAGS(sk);
 520                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 521         }
 522         flowi4_init_output(fl4, oif, mark, tos,
 523                            RT_SCOPE_UNIVERSE, prot,
 524                            flow_flags,
 525                            iph->daddr, iph->saddr, 0, 0,
 526                            sock_net_uid(net, sk));
 527 }
 528
 529 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 530                                const struct sock *sk)
 531 {
 532         const struct net *net = dev_net(skb->dev);
 533         const struct iphdr *iph = ip_hdr(skb);
 534         int oif = skb->dev->ifindex;
 535         u8 tos = RT_TOS(iph->tos);
 536         u8 prot = iph->protocol;
 537         u32 mark = skb->mark;
 538
 539         __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
 540 }
 541
 542 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 543 {
 544         const struct inet_sock *inet = inet_sk(sk);
 545         const struct ip_options_rcu *inet_opt;
 546         __be32 daddr = inet->inet_daddr;
 547
 548         rcu_read_lock();
 549         inet_opt = rcu_dereference(inet->inet_opt);
 550         if (inet_opt && inet_opt->opt.srr)
 551                 daddr = inet_opt->opt.faddr;
 552         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 553                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 554                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 555                            inet_sk_flowi_flags(sk),
 556                            daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
 557         rcu_read_unlock();
 558 }
 559
 560 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 561                                  const struct sk_buff *skb)
 562 {
 563         if (skb)
 564                 build_skb_flow_key(fl4, skb, sk);
 565         else
 566                 build_sk_flow_key(fl4, sk);
 567 }
 568
 569 static DEFINE_SPINLOCK(fnhe_lock);
 570
 571 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 572 {
 573         struct rtable *rt;
 574
 575         rt = rcu_dereference(fnhe->fnhe_rth_input);
 576         if (rt) {
 577                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 578                 dst_dev_put(&rt->dst);
 579                 dst_release(&rt->dst);
 580         }
 581         rt = rcu_dereference(fnhe->fnhe_rth_output);
 582         if (rt) {
 583                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 584                 dst_dev_put(&rt->dst);
 585                 dst_release(&rt->dst);
 586         }
 587 }
 588
 589 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
 590 {
 591         struct fib_nh_exception *fnhe, *oldest;
 592
 593         oldest = rcu_dereference(hash->chain);
 594         for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
 595              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 596                 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
 597                         oldest = fnhe;
 598         }
 599         fnhe_flush_routes(oldest);
 600         return oldest;
 601 }
 602
 603 static inline u32 fnhe_hashfun(__be32 daddr)
 604 {
 605         static u32 fnhe_hashrnd __read_mostly;
 606         u32 hval;
 607
 608         net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
 609         hval = jhash_1word((__force u32)daddr, fnhe_hashrnd);
 610         return hash_32(hval, FNHE_HASH_SHIFT);
 611 }
 612
 613 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 614 {
 615         rt->rt_pmtu = fnhe->fnhe_pmtu;
 616         rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
 617         rt->dst.expires = fnhe->fnhe_expires;
 618
 619         if (fnhe->fnhe_gw) {
 620                 rt->rt_flags |= RTCF_REDIRECTED;
 621                 rt->rt_uses_gateway = 1;
 622                 rt->rt_gw_family = AF_INET;
 623                 rt->rt_gw4 = fnhe->fnhe_gw;
 624         }
 625 }
 626
 627 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
 628                                   __be32 gw, u32 pmtu, bool lock,
 629                                   unsigned long expires)
 630 {
 631         struct fnhe_hash_bucket *hash;
 632         struct fib_nh_exception *fnhe;
 633         struct rtable *rt;
 634         u32 genid, hval;
 635         unsigned int i;
 636         int depth;
 637
 638         genid = fnhe_genid(dev_net(nhc->nhc_dev));
 639         hval = fnhe_hashfun(daddr);
 640
 641         spin_lock_bh(&fnhe_lock);
 642
 643         hash = rcu_dereference(nhc->nhc_exceptions);
 644         if (!hash) {
 645                 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
 646                 if (!hash)
 647                         goto out_unlock;
 648                 rcu_assign_pointer(nhc->nhc_exceptions, hash);
 649         }
 650
 651         hash += hval;
 652
 653         depth = 0;
 654         for (fnhe = rcu_dereference(hash->chain); fnhe;
 655              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 656                 if (fnhe->fnhe_daddr == daddr)
 657                         break;
 658                 depth++;
 659         }
 660
 661         if (fnhe) {
 662                 if (fnhe->fnhe_genid != genid)
 663                         fnhe->fnhe_genid = genid;
 664                 if (gw)
 665                         fnhe->fnhe_gw = gw;
 666                 if (pmtu) {
 667                         fnhe->fnhe_pmtu = pmtu;
 668                         fnhe->fnhe_mtu_locked = lock;
 669                 }
 670                 fnhe->fnhe_expires = max(1UL, expires);
 671                 /* Update all cached dsts too */
 672                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 673                 if (rt)
 674                         fill_route_from_fnhe(rt, fnhe);
 675                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 676                 if (rt)
 677                         fill_route_from_fnhe(rt, fnhe);
 678         } else {
 679                 if (depth > FNHE_RECLAIM_DEPTH)
 680                         fnhe = fnhe_oldest(hash);
 681                 else {
 682                         fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 683                         if (!fnhe)
 684                                 goto out_unlock;
 685
 686                         fnhe->fnhe_next = hash->chain;
 687                         rcu_assign_pointer(hash->chain, fnhe);
 688                 }
 689                 fnhe->fnhe_genid = genid;
 690                 fnhe->fnhe_daddr = daddr;
 691                 fnhe->fnhe_gw = gw;
 692                 fnhe->fnhe_pmtu = pmtu;
 693                 fnhe->fnhe_mtu_locked = lock;
 694                 fnhe->fnhe_expires = max(1UL, expires);
 695
 696                 /* Exception created; mark the cached routes for the nexthop
 697                  * stale, so anyone caching it rechecks if this exception
 698                  * applies to them.
 699                  */
 700                 rt = rcu_dereference(nhc->nhc_rth_input);
 701                 if (rt)
 702                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 703
 704                 for_each_possible_cpu(i) {
 705                         struct rtable __rcu **prt;
 706
 707                         prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
 708                         rt = rcu_dereference(*prt);
 709                         if (rt)
 710                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 711                 }
 712         }
 713
 714         fnhe->fnhe_stamp = jiffies;
 715
 716 out_unlock:
 717         spin_unlock_bh(&fnhe_lock);
 718 }
 719
 720 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 721                              bool kill_route)
 722 {
 723         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 724         __be32 old_gw = ip_hdr(skb)->saddr;
 725         struct net_device *dev = skb->dev;
 726         struct in_device *in_dev;
 727         struct fib_result res;
 728         struct neighbour *n;
 729         struct net *net;
 730
 731         switch (icmp_hdr(skb)->code & 7) {
 732         case ICMP_REDIR_NET:
 733         case ICMP_REDIR_NETTOS:
 734         case ICMP_REDIR_HOST:
 735         case ICMP_REDIR_HOSTTOS:
 736                 break;
 737
 738         default:
 739                 return;
 740         }
 741
 742         if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
 743                 return;
 744
 745         in_dev = __in_dev_get_rcu(dev);
 746         if (!in_dev)
 747                 return;
 748
 749         net = dev_net(dev);
 750         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 751             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 752             ipv4_is_zeronet(new_gw))
 753                 goto reject_redirect;
 754
 755         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 756                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 757                         goto reject_redirect;
 758                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 759                         goto reject_redirect;
 760         } else {
 761                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 762                         goto reject_redirect;
 763         }
 764
 765         n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
 766         if (!n)
 767                 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
 768         if (!IS_ERR(n)) {
 769                 if (!(n->nud_state & NUD_VALID)) {
 770                         neigh_event_send(n, NULL);
 771                 } else {
 772                         if (fib_lookup(net, fl4, &res, 0) == 0) {
 773                                 struct fib_nh_common *nhc;
 774
 775                                 fib_select_path(net, &res, fl4, skb);
 776                                 nhc = FIB_RES_NHC(res);
 777                                 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
 778                                                 0, false,
 779                                                 jiffies + ip_rt_gc_timeout);
 780                         }
 781                         if (kill_route)
 782                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 783                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 784                 }
 785                 neigh_release(n);
 786         }
 787         return;
 788
 789 reject_redirect:
 790 #ifdef CONFIG_IP_ROUTE_VERBOSE
 791         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 792                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 793                 __be32 daddr = iph->daddr;
 794                 __be32 saddr = iph->saddr;
 795
 796                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 797                                      "  Advised path = %pI4 -> %pI4\n",
 798                                      &old_gw, dev->name, &new_gw,
 799                                      &saddr, &daddr);
 800         }
 801 #endif
 802         ;
 803 }
 804
 805 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 806 {
 807         struct rtable *rt;
 808         struct flowi4 fl4;
 809         const struct iphdr *iph = (const struct iphdr *) skb->data;
 810         struct net *net = dev_net(skb->dev);
 811         int oif = skb->dev->ifindex;
 812         u8 tos = RT_TOS(iph->tos);
 813         u8 prot = iph->protocol;
 814         u32 mark = skb->mark;
 815
 816         rt = (struct rtable *) dst;
 817
 818         __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
 819         __ip_do_redirect(rt, skb, &fl4, true);
 820 }
 821
 822 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 823 {
 824         struct rtable *rt = (struct rtable *)dst;
 825         struct dst_entry *ret = dst;
 826
 827         if (rt) {
 828                 if (dst->obsolete > 0) {
 829                         ip_rt_put(rt);
 830                         ret = NULL;
 831                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 832                            rt->dst.expires) {
 833                         ip_rt_put(rt);
 834                         ret = NULL;
 835                 }
 836         }
 837         return ret;
 838 }
 839
 840 /*
 841  * Algorithm:
 842  *      1. The first ip_rt_redirect_number redirects are sent
 843  *         with exponential backoff, then we stop sending them at all,
 844  *         assuming that the host ignores our redirects.
 845  *      2. If we did not see packets requiring redirects
 846  *         during ip_rt_redirect_silence, we assume that the host
 847  *         forgot redirected route and start to send redirects again.
 848  *
 849  * This algorithm is much cheaper and more intelligent than dumb load limiting
 850  * in icmp.c.
 851  *
 852  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 853  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 854  */
 855
 856 void ip_rt_send_redirect(struct sk_buff *skb)
 857 {
 858         struct rtable *rt = skb_rtable(skb);
 859         struct in_device *in_dev;
 860         struct inet_peer *peer;
 861         struct net *net;
 862         int log_martians;
 863         int vif;
 864
 865         rcu_read_lock();
 866         in_dev = __in_dev_get_rcu(rt->dst.dev);
 867         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 868                 rcu_read_unlock();
 869                 return;
 870         }
 871         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 872         vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
 873         rcu_read_unlock();
 874
 875         net = dev_net(rt->dst.dev);
 876         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
 877         if (!peer) {
 878                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 879                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 880                 return;
 881         }
 882
 883         /* No redirected packets during ip_rt_redirect_silence;
 884          * reset the algorithm.
 885          */
 886         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
 887                 peer->rate_tokens = 0;
 888                 peer->n_redirects = 0;
 889         }
 890
 891         /* Too many ignored redirects; do not send anything
 892          * set dst.rate_last to the last seen redirected packet.
 893          */
 894         if (peer->n_redirects >= ip_rt_redirect_number) {
 895                 peer->rate_last = jiffies;
 896                 goto out_put_peer;
 897         }
 898
 899         /* Check for load limit; set rate_last to the latest sent
 900          * redirect.
 901          */
 902         if (peer->n_redirects == 0 ||
 903             time_after(jiffies,
 904                        (peer->rate_last +
 905                         (ip_rt_redirect_load << peer->n_redirects)))) {
 906                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 907
 908                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 909                 peer->rate_last = jiffies;
 910                 ++peer->n_redirects;
 911 #ifdef CONFIG_IP_ROUTE_VERBOSE
 912                 if (log_martians &&
 913                     peer->n_redirects == ip_rt_redirect_number)
 914                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 915                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 916                                              &ip_hdr(skb)->daddr, &gw);
 917 #endif
 918         }
 919 out_put_peer:
 920         inet_putpeer(peer);
 921 }
 922
 923 static int ip_error(struct sk_buff *skb)
 924 {
 925         struct rtable *rt = skb_rtable(skb);
 926         struct net_device *dev = skb->dev;
 927         struct in_device *in_dev;
 928         struct inet_peer *peer;
 929         unsigned long now;
 930         struct net *net;
 931         bool send;
 932         int code;
 933
 934         if (netif_is_l3_master(skb->dev)) {
 935                 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
 936                 if (!dev)
 937                         goto out;
 938         }
 939
 940         in_dev = __in_dev_get_rcu(dev);
 941
 942         /* IP on this device is disabled. */
 943         if (!in_dev)
 944                 goto out;
 945
 946         net = dev_net(rt->dst.dev);
 947         if (!IN_DEV_FORWARD(in_dev)) {
 948                 switch (rt->dst.error) {
 949                 case EHOSTUNREACH:
 950                         __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
 951                         break;
 952
 953                 case ENETUNREACH:
 954                         __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 955                         break;
 956                 }
 957                 goto out;
 958         }
 959
 960         switch (rt->dst.error) {
 961         case EINVAL:
 962         default:
 963                 goto out;
 964         case EHOSTUNREACH:
 965                 code = ICMP_HOST_UNREACH;
 966                 break;
 967         case ENETUNREACH:
 968                 code = ICMP_NET_UNREACH;
 969                 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 970                 break;
 971         case EACCES:
 972                 code = ICMP_PKT_FILTERED;
 973                 break;
 974         }
 975
 976         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
 977                                l3mdev_master_ifindex(skb->dev), 1);
 978
 979         send = true;
 980         if (peer) {
 981                 now = jiffies;
 982                 peer->rate_tokens += now - peer->rate_last;
 983                 if (peer->rate_tokens > ip_rt_error_burst)
 984                         peer->rate_tokens = ip_rt_error_burst;
 985                 peer->rate_last = now;
 986                 if (peer->rate_tokens >= ip_rt_error_cost)
 987                         peer->rate_tokens -= ip_rt_error_cost;
 988                 else
 989                         send = false;
 990                 inet_putpeer(peer);
 991         }
 992         if (send)
 993                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
 994
 995 out:    kfree_skb(skb);
 996         return 0;
 997 }
 998
 999 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1000 {
1001         struct dst_entry *dst = &rt->dst;
1002         struct net *net = dev_net(dst->dev);
1003         struct fib_result res;
1004         bool lock = false;
1005         u32 old_mtu;
1006
1007         if (ip_mtu_locked(dst))
1008                 return;
1009
1010         old_mtu = ipv4_mtu(dst);
1011         if (old_mtu < mtu)
1012                 return;
1013
1014         if (mtu < ip_rt_min_pmtu) {
1015                 lock = true;
1016                 mtu = min(old_mtu, ip_rt_min_pmtu);
1017         }
1018
1019         if (rt->rt_pmtu == mtu && !lock &&
1020             time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1021                 return;
1022
1023         rcu_read_lock();
1024         if (fib_lookup(net, fl4, &res, 0) == 0) {
1025                 struct fib_nh_common *nhc;
1026
1027                 fib_select_path(net, &res, fl4, NULL);
1028                 nhc = FIB_RES_NHC(res);
1029                 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1030                                       jiffies + ip_rt_mtu_expires);
1031         }
1032         rcu_read_unlock();
1033 }
1034
1035 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1036                               struct sk_buff *skb, u32 mtu,
1037                               bool confirm_neigh)
1038 {
1039         struct rtable *rt = (struct rtable *) dst;
1040         struct flowi4 fl4;
1041
1042         ip_rt_build_flow_key(&fl4, sk, skb);
1043
1044         /* Don't make lookup fail for bridged encapsulations */
1045         if (skb && netif_is_any_bridge_port(skb->dev))
1046                 fl4.flowi4_oif = 0;
1047
1048         __ip_rt_update_pmtu(rt, &fl4, mtu);
1049 }
1050
1051 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1052                       int oif, u8 protocol)
1053 {
1054         const struct iphdr *iph = (const struct iphdr *)skb->data;
1055         struct flowi4 fl4;
1056         struct rtable *rt;
1057         u32 mark = IP4_REPLY_MARK(net, skb->mark);
1058
1059         __build_flow_key(net, &fl4, NULL, iph, oif,
1060                          RT_TOS(iph->tos), protocol, mark, 0);
1061         rt = __ip_route_output_key(net, &fl4);
1062         if (!IS_ERR(rt)) {
1063                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1064                 ip_rt_put(rt);
1065         }
1066 }
1067 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1068
1069 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1070 {
1071         const struct iphdr *iph = (const struct iphdr *)skb->data;
1072         struct flowi4 fl4;
1073         struct rtable *rt;
1074
1075         __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1076
1077         if (!fl4.flowi4_mark)
1078                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1079
1080         rt = __ip_route_output_key(sock_net(sk), &fl4);
1081         if (!IS_ERR(rt)) {
1082                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1083                 ip_rt_put(rt);
1084         }
1085 }
1086
1087 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1088 {
1089         const struct iphdr *iph = (const struct iphdr *)skb->data;
1090         struct flowi4 fl4;
1091         struct rtable *rt;
1092         struct dst_entry *odst = NULL;
1093         bool new = false;
1094         struct net *net = sock_net(sk);
1095
1096         bh_lock_sock(sk);
1097
1098         if (!ip_sk_accept_pmtu(sk))
1099                 goto out;
1100
1101         odst = sk_dst_get(sk);
1102
1103         if (sock_owned_by_user(sk) || !odst) {
1104                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1105                 goto out;
1106         }
1107
1108         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1109
1110         rt = (struct rtable *)odst;
1111         if (odst->obsolete && !odst->ops->check(odst, 0)) {
1112                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1113                 if (IS_ERR(rt))
1114                         goto out;
1115
1116                 new = true;
1117         }
1118
1119         __ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu);
1120
1121         if (!dst_check(&rt->dst, 0)) {
1122                 if (new)
1123                         dst_release(&rt->dst);
1124
1125                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1126                 if (IS_ERR(rt))
1127                         goto out;
1128
1129                 new = true;
1130         }
1131
1132         if (new)
1133                 sk_dst_set(sk, &rt->dst);
1134
1135 out:
1136         bh_unlock_sock(sk);
1137         dst_release(odst);
1138 }
1139 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1140
1141 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1142                    int oif, u8 protocol)
1143 {
1144         const struct iphdr *iph = (const struct iphdr *)skb->data;
1145         struct flowi4 fl4;
1146         struct rtable *rt;
1147
1148         __build_flow_key(net, &fl4, NULL, iph, oif,
1149                          RT_TOS(iph->tos), protocol, 0, 0);
1150         rt = __ip_route_output_key(net, &fl4);
1151         if (!IS_ERR(rt)) {
1152                 __ip_do_redirect(rt, skb, &fl4, false);
1153                 ip_rt_put(rt);
1154         }
1155 }
1156 EXPORT_SYMBOL_GPL(ipv4_redirect);
1157
1158 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1159 {
1160         const struct iphdr *iph = (const struct iphdr *)skb->data;
1161         struct flowi4 fl4;
1162         struct rtable *rt;
1163         struct net *net = sock_net(sk);
1164
1165         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1166         rt = __ip_route_output_key(net, &fl4);
1167         if (!IS_ERR(rt)) {
1168                 __ip_do_redirect(rt, skb, &fl4, false);
1169                 ip_rt_put(rt);
1170         }
1171 }
1172 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1173
1174 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst,
1175                                                          u32 cookie)
1176 {
1177         struct rtable *rt = (struct rtable *) dst;
1178
1179         /* All IPV4 dsts are created with ->obsolete set to the value
1180          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1181          * into this function always.
1182          *
1183          * When a PMTU/redirect information update invalidates a route,
1184          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1185          * DST_OBSOLETE_DEAD.
1186          */
1187         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1188                 return NULL;
1189         return dst;
1190 }
1191 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check);
1192
1193 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1194 {
1195         struct ip_options opt;
1196         int res;
1197
1198         /* Recompile ip options since IPCB may not be valid anymore.
1199          * Also check we have a reasonable ipv4 header.
1200          */
1201         if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1202             ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1203                 return;
1204
1205         memset(&opt, 0, sizeof(opt));
1206         if (ip_hdr(skb)->ihl > 5) {
1207                 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1208                         return;
1209                 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1210
1211                 rcu_read_lock();
1212                 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1213                 rcu_read_unlock();
1214
1215                 if (res)
1216                         return;
1217         }
1218         __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1219 }
1220
1221 static void ipv4_link_failure(struct sk_buff *skb)
1222 {
1223         struct rtable *rt;
1224
1225         ipv4_send_dest_unreach(skb);
1226
1227         rt = skb_rtable(skb);
1228         if (rt)
1229                 dst_set_expires(&rt->dst, 0);
1230 }
1231
1232 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1233 {
1234         pr_debug("%s: %pI4 -> %pI4, %s\n",
1235                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1236                  skb->dev ? skb->dev->name : "?");
1237         kfree_skb(skb);
1238         WARN_ON(1);
1239         return 0;
1240 }
1241
1242 /*
1243  * We do not cache source address of outgoing interface,
1244  * because it is used only by IP RR, TS and SRR options,
1245  * so that it out of fast path.
1246  *
1247  * BTW remember: "addr" is allowed to be not aligned
1248  * in IP options!
1249  */
1250
1251 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1252 {
1253         __be32 src;
1254
1255         if (rt_is_output_route(rt))
1256                 src = ip_hdr(skb)->saddr;
1257         else {
1258                 struct fib_result res;
1259                 struct iphdr *iph = ip_hdr(skb);
1260                 struct flowi4 fl4 = {
1261                         .daddr = iph->daddr,
1262                         .saddr = iph->saddr,
1263                         .flowi4_tos = RT_TOS(iph->tos),
1264                         .flowi4_oif = rt->dst.dev->ifindex,
1265                         .flowi4_iif = skb->dev->ifindex,
1266                         .flowi4_mark = skb->mark,
1267                 };
1268
1269                 rcu_read_lock();
1270                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1271                         src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1272                 else
1273                         src = inet_select_addr(rt->dst.dev,
1274                                                rt_nexthop(rt, iph->daddr),
1275                                                RT_SCOPE_UNIVERSE);
1276                 rcu_read_unlock();
1277         }
1278         memcpy(addr, &src, 4);
1279 }
1280
1281 #ifdef CONFIG_IP_ROUTE_CLASSID
1282 static void set_class_tag(struct rtable *rt, u32 tag)
1283 {
1284         if (!(rt->dst.tclassid & 0xFFFF))
1285                 rt->dst.tclassid |= tag & 0xFFFF;
1286         if (!(rt->dst.tclassid & 0xFFFF0000))
1287                 rt->dst.tclassid |= tag & 0xFFFF0000;
1288 }
1289 #endif
1290
1291 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1292 {
1293         unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1294         unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1295                                     ip_rt_min_advmss);
1296
1297         return min(advmss, IPV4_MAX_PMTU - header_size);
1298 }
1299
1300 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst)
1301 {
1302         const struct rtable *rt = (const struct rtable *)dst;
1303         unsigned int mtu = rt->rt_pmtu;
1304
1305         if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1306                 mtu = dst_metric_raw(dst, RTAX_MTU);
1307
1308         if (mtu)
1309                 return mtu;
1310
1311         mtu = READ_ONCE(dst->dev->mtu);
1312
1313         if (unlikely(ip_mtu_locked(dst))) {
1314                 if (rt->rt_uses_gateway && mtu > 576)
1315                         mtu = 576;
1316         }
1317
1318         mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1319
1320         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1321 }
1322 EXPORT_INDIRECT_CALLABLE(ipv4_mtu);
1323
1324 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1325 {
1326         struct fnhe_hash_bucket *hash;
1327         struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1328         u32 hval = fnhe_hashfun(daddr);
1329
1330         spin_lock_bh(&fnhe_lock);
1331
1332         hash = rcu_dereference_protected(nhc->nhc_exceptions,
1333                                          lockdep_is_held(&fnhe_lock));
1334         hash += hval;
1335
1336         fnhe_p = &hash->chain;
1337         fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1338         while (fnhe) {
1339                 if (fnhe->fnhe_daddr == daddr) {
1340                         rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1341                                 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1342                         /* set fnhe_daddr to 0 to ensure it won't bind with
1343                          * new dsts in rt_bind_exception().
1344                          */
1345                         fnhe->fnhe_daddr = 0;
1346                         fnhe_flush_routes(fnhe);
1347                         kfree_rcu(fnhe, rcu);
1348                         break;
1349                 }
1350                 fnhe_p = &fnhe->fnhe_next;
1351                 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1352                                                  lockdep_is_held(&fnhe_lock));
1353         }
1354
1355         spin_unlock_bh(&fnhe_lock);
1356 }
1357
1358 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1359                                                __be32 daddr)
1360 {
1361         struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1362         struct fib_nh_exception *fnhe;
1363         u32 hval;
1364
1365         if (!hash)
1366                 return NULL;
1367
1368         hval = fnhe_hashfun(daddr);
1369
1370         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1371              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1372                 if (fnhe->fnhe_daddr == daddr) {
1373                         if (fnhe->fnhe_expires &&
1374                             time_after(jiffies, fnhe->fnhe_expires)) {
1375                                 ip_del_fnhe(nhc, daddr);
1376                                 break;
1377                         }
1378                         return fnhe;
1379                 }
1380         }
1381         return NULL;
1382 }
1383
1384 /* MTU selection:
1385  * 1. mtu on route is locked - use it
1386  * 2. mtu from nexthop exception
1387  * 3. mtu from egress device
1388  */
1389
1390 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1391 {
1392         struct fib_nh_common *nhc = res->nhc;
1393         struct net_device *dev = nhc->nhc_dev;
1394         struct fib_info *fi = res->fi;
1395         u32 mtu = 0;
1396
1397         if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1398             fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1399                 mtu = fi->fib_mtu;
1400
1401         if (likely(!mtu)) {
1402                 struct fib_nh_exception *fnhe;
1403
1404                 fnhe = find_exception(nhc, daddr);
1405                 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1406                         mtu = fnhe->fnhe_pmtu;
1407         }
1408
1409         if (likely(!mtu))
1410                 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1411
1412         return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1413 }
1414
1415 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1416                               __be32 daddr, const bool do_cache)
1417 {
1418         bool ret = false;
1419
1420         spin_lock_bh(&fnhe_lock);
1421
1422         if (daddr == fnhe->fnhe_daddr) {
1423                 struct rtable __rcu **porig;
1424                 struct rtable *orig;
1425                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1426
1427                 if (rt_is_input_route(rt))
1428                         porig = &fnhe->fnhe_rth_input;
1429                 else
1430                         porig = &fnhe->fnhe_rth_output;
1431                 orig = rcu_dereference(*porig);
1432
1433                 if (fnhe->fnhe_genid != genid) {
1434                         fnhe->fnhe_genid = genid;
1435                         fnhe->fnhe_gw = 0;
1436                         fnhe->fnhe_pmtu = 0;
1437                         fnhe->fnhe_expires = 0;
1438                         fnhe->fnhe_mtu_locked = false;
1439                         fnhe_flush_routes(fnhe);
1440                         orig = NULL;
1441                 }
1442                 fill_route_from_fnhe(rt, fnhe);
1443                 if (!rt->rt_gw4) {
1444                         rt->rt_gw4 = daddr;
1445                         rt->rt_gw_family = AF_INET;
1446                 }
1447
1448                 if (do_cache) {
1449                         dst_hold(&rt->dst);
1450                         rcu_assign_pointer(*porig, rt);
1451                         if (orig) {
1452                                 dst_dev_put(&orig->dst);
1453                                 dst_release(&orig->dst);
1454                         }
1455                         ret = true;
1456                 }
1457
1458                 fnhe->fnhe_stamp = jiffies;
1459         }
1460         spin_unlock_bh(&fnhe_lock);
1461
1462         return ret;
1463 }
1464
1465 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1466 {
1467         struct rtable *orig, *prev, **p;
1468         bool ret = true;
1469
1470         if (rt_is_input_route(rt)) {
1471                 p = (struct rtable **)&nhc->nhc_rth_input;
1472         } else {
1473                 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1474         }
1475         orig = *p;
1476
1477         /* hold dst before doing cmpxchg() to avoid race condition
1478          * on this dst
1479          */
1480         dst_hold(&rt->dst);
1481         prev = cmpxchg(p, orig, rt);
1482         if (prev == orig) {
1483                 if (orig) {
1484                         rt_add_uncached_list(orig);
1485                         dst_release(&orig->dst);
1486                 }
1487         } else {
1488                 dst_release(&rt->dst);
1489                 ret = false;
1490         }
1491
1492         return ret;
1493 }
1494
1495 struct uncached_list {
1496         spinlock_t              lock;
1497         struct list_head        head;
1498 };
1499
1500 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1501
1502 void rt_add_uncached_list(struct rtable *rt)
1503 {
1504         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1505
1506         rt->rt_uncached_list = ul;
1507
1508         spin_lock_bh(&ul->lock);
1509         list_add_tail(&rt->rt_uncached, &ul->head);
1510         spin_unlock_bh(&ul->lock);
1511 }
1512
1513 void rt_del_uncached_list(struct rtable *rt)
1514 {
1515         if (!list_empty(&rt->rt_uncached)) {
1516                 struct uncached_list *ul = rt->rt_uncached_list;
1517
1518                 spin_lock_bh(&ul->lock);
1519                 list_del(&rt->rt_uncached);
1520                 spin_unlock_bh(&ul->lock);
1521         }
1522 }
1523
1524 static void ipv4_dst_destroy(struct dst_entry *dst)
1525 {
1526         struct rtable *rt = (struct rtable *)dst;
1527
1528         ip_dst_metrics_put(dst);
1529         rt_del_uncached_list(rt);
1530 }
1531
1532 void rt_flush_dev(struct net_device *dev)
1533 {
1534         struct rtable *rt;
1535         int cpu;
1536
1537         for_each_possible_cpu(cpu) {
1538                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1539
1540                 spin_lock_bh(&ul->lock);
1541                 list_for_each_entry(rt, &ul->head, rt_uncached) {
1542                         if (rt->dst.dev != dev)
1543                                 continue;
1544                         rt->dst.dev = blackhole_netdev;
1545                         dev_hold(rt->dst.dev);
1546                         dev_put(dev);
1547                 }
1548                 spin_unlock_bh(&ul->lock);
1549         }
1550 }
1551
1552 static bool rt_cache_valid(const struct rtable *rt)
1553 {
1554         return  rt &&
1555                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1556                 !rt_is_expired(rt);
1557 }
1558
1559 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1560                            const struct fib_result *res,
1561                            struct fib_nh_exception *fnhe,
1562                            struct fib_info *fi, u16 type, u32 itag,
1563                            const bool do_cache)
1564 {
1565         bool cached = false;
1566
1567         if (fi) {
1568                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1569
1570                 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1571                         rt->rt_uses_gateway = 1;
1572                         rt->rt_gw_family = nhc->nhc_gw_family;
1573                         /* only INET and INET6 are supported */
1574                         if (likely(nhc->nhc_gw_family == AF_INET))
1575                                 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1576                         else
1577                                 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1578                 }
1579
1580                 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1581
1582 #ifdef CONFIG_IP_ROUTE_CLASSID
1583                 if (nhc->nhc_family == AF_INET) {
1584                         struct fib_nh *nh;
1585
1586                         nh = container_of(nhc, struct fib_nh, nh_common);
1587                         rt->dst.tclassid = nh->nh_tclassid;
1588                 }
1589 #endif
1590                 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1591                 if (unlikely(fnhe))
1592                         cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1593                 else if (do_cache)
1594                         cached = rt_cache_route(nhc, rt);
1595                 if (unlikely(!cached)) {
1596                         /* Routes we intend to cache in nexthop exception or
1597                          * FIB nexthop have the DST_NOCACHE bit clear.
1598                          * However, if we are unsuccessful at storing this
1599                          * route into the cache we really need to set it.
1600                          */
1601                         if (!rt->rt_gw4) {
1602                                 rt->rt_gw_family = AF_INET;
1603                                 rt->rt_gw4 = daddr;
1604                         }
1605                         rt_add_uncached_list(rt);
1606                 }
1607         } else
1608                 rt_add_uncached_list(rt);
1609
1610 #ifdef CONFIG_IP_ROUTE_CLASSID
1611 #ifdef CONFIG_IP_MULTIPLE_TABLES
1612         set_class_tag(rt, res->tclassid);
1613 #endif
1614         set_class_tag(rt, itag);
1615 #endif
1616 }
1617
1618 struct rtable *rt_dst_alloc(struct net_device *dev,
1619                             unsigned int flags, u16 type,
1620                             bool nopolicy, bool noxfrm)
1621 {
1622         struct rtable *rt;
1623
1624         rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1625                        (nopolicy ? DST_NOPOLICY : 0) |
1626                        (noxfrm ? DST_NOXFRM : 0));
1627
1628         if (rt) {
1629                 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1630                 rt->rt_flags = flags;
1631                 rt->rt_type = type;
1632                 rt->rt_is_input = 0;
1633                 rt->rt_iif = 0;
1634                 rt->rt_pmtu = 0;
1635                 rt->rt_mtu_locked = 0;
1636                 rt->rt_uses_gateway = 0;
1637                 rt->rt_gw_family = 0;
1638                 rt->rt_gw4 = 0;
1639                 INIT_LIST_HEAD(&rt->rt_uncached);
1640
1641                 rt->dst.output = ip_output;
1642                 if (flags & RTCF_LOCAL)
1643                         rt->dst.input = ip_local_deliver;
1644         }
1645
1646         return rt;
1647 }
1648 EXPORT_SYMBOL(rt_dst_alloc);
1649
1650 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1651 {
1652         struct rtable *new_rt;
1653
1654         new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1655                            rt->dst.flags);
1656
1657         if (new_rt) {
1658                 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1659                 new_rt->rt_flags = rt->rt_flags;
1660                 new_rt->rt_type = rt->rt_type;
1661                 new_rt->rt_is_input = rt->rt_is_input;
1662                 new_rt->rt_iif = rt->rt_iif;
1663                 new_rt->rt_pmtu = rt->rt_pmtu;
1664                 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1665                 new_rt->rt_gw_family = rt->rt_gw_family;
1666                 if (rt->rt_gw_family == AF_INET)
1667                         new_rt->rt_gw4 = rt->rt_gw4;
1668                 else if (rt->rt_gw_family == AF_INET6)
1669                         new_rt->rt_gw6 = rt->rt_gw6;
1670                 INIT_LIST_HEAD(&new_rt->rt_uncached);
1671
1672                 new_rt->dst.input = rt->dst.input;
1673                 new_rt->dst.output = rt->dst.output;
1674                 new_rt->dst.error = rt->dst.error;
1675                 new_rt->dst.lastuse = jiffies;
1676                 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1677         }
1678         return new_rt;
1679 }
1680 EXPORT_SYMBOL(rt_dst_clone);
1681
1682 /* called in rcu_read_lock() section */
1683 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1684                           u8 tos, struct net_device *dev,
1685                           struct in_device *in_dev, u32 *itag)
1686 {
1687         int err;
1688
1689         /* Primary sanity checks. */
1690         if (!in_dev)
1691                 return -EINVAL;
1692
1693         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1694             skb->protocol != htons(ETH_P_IP))
1695                 return -EINVAL;
1696
1697         if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1698                 return -EINVAL;
1699
1700         if (ipv4_is_zeronet(saddr)) {
1701                 if (!ipv4_is_local_multicast(daddr) &&
1702                     ip_hdr(skb)->protocol != IPPROTO_IGMP)
1703                         return -EINVAL;
1704         } else {
1705                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1706                                           in_dev, itag);
1707                 if (err < 0)
1708                         return err;
1709         }
1710         return 0;
1711 }
1712
1713 /* called in rcu_read_lock() section */
1714 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1715                              u8 tos, struct net_device *dev, int our)
1716 {
1717         struct in_device *in_dev = __in_dev_get_rcu(dev);
1718         unsigned int flags = RTCF_MULTICAST;
1719         struct rtable *rth;
1720         u32 itag = 0;
1721         int err;
1722
1723         err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1724         if (err)
1725                 return err;
1726
1727         if (our)
1728                 flags |= RTCF_LOCAL;
1729
1730         rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1731                            IN_DEV_ORCONF(in_dev, NOPOLICY), false);
1732         if (!rth)
1733                 return -ENOBUFS;
1734
1735 #ifdef CONFIG_IP_ROUTE_CLASSID
1736         rth->dst.tclassid = itag;
1737 #endif
1738         rth->dst.output = ip_rt_bug;
1739         rth->rt_is_input= 1;
1740
1741 #ifdef CONFIG_IP_MROUTE
1742         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1743                 rth->dst.input = ip_mr_input;
1744 #endif
1745         RT_CACHE_STAT_INC(in_slow_mc);
1746
1747         skb_dst_set(skb, &rth->dst);
1748         return 0;
1749 }
1750
1751
1752 static void ip_handle_martian_source(struct net_device *dev,
1753                                      struct in_device *in_dev,
1754                                      struct sk_buff *skb,
1755                                      __be32 daddr,
1756                                      __be32 saddr)
1757 {
1758         RT_CACHE_STAT_INC(in_martian_src);
1759 #ifdef CONFIG_IP_ROUTE_VERBOSE
1760         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1761                 /*
1762                  *      RFC1812 recommendation, if source is martian,
1763                  *      the only hint is MAC header.
1764                  */
1765                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1766                         &daddr, &saddr, dev->name);
1767                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1768                         print_hex_dump(KERN_WARNING, "ll header: ",
1769                                        DUMP_PREFIX_OFFSET, 16, 1,
1770                                        skb_mac_header(skb),
1771                                        dev->hard_header_len, false);
1772                 }
1773         }
1774 #endif
1775 }
1776
1777 /* called in rcu_read_lock() section */
1778 static int __mkroute_input(struct sk_buff *skb,
1779                            const struct fib_result *res,
1780                            struct in_device *in_dev,
1781                            __be32 daddr, __be32 saddr, u32 tos)
1782 {
1783         struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1784         struct net_device *dev = nhc->nhc_dev;
1785         struct fib_nh_exception *fnhe;
1786         struct rtable *rth;
1787         int err;
1788         struct in_device *out_dev;
1789         bool do_cache;
1790         u32 itag = 0;
1791
1792         /* get a working reference to the output device */
1793         out_dev = __in_dev_get_rcu(dev);
1794         if (!out_dev) {
1795                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1796                 return -EINVAL;
1797         }
1798
1799         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1800                                   in_dev->dev, in_dev, &itag);
1801         if (err < 0) {
1802                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1803                                          saddr);
1804
1805                 goto cleanup;
1806         }
1807
1808         do_cache = res->fi && !itag;
1809         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1810             skb->protocol == htons(ETH_P_IP)) {
1811                 __be32 gw;
1812
1813                 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1814                 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1815                     inet_addr_onlink(out_dev, saddr, gw))
1816                         IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1817         }
1818
1819         if (skb->protocol != htons(ETH_P_IP)) {
1820                 /* Not IP (i.e. ARP). Do not create route, if it is
1821                  * invalid for proxy arp. DNAT routes are always valid.
1822                  *
1823                  * Proxy arp feature have been extended to allow, ARP
1824                  * replies back to the same interface, to support
1825                  * Private VLAN switch technologies. See arp.c.
1826                  */
1827                 if (out_dev == in_dev &&
1828                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1829                         err = -EINVAL;
1830                         goto cleanup;
1831                 }
1832         }
1833
1834         fnhe = find_exception(nhc, daddr);
1835         if (do_cache) {
1836                 if (fnhe)
1837                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1838                 else
1839                         rth = rcu_dereference(nhc->nhc_rth_input);
1840                 if (rt_cache_valid(rth)) {
1841                         skb_dst_set_noref(skb, &rth->dst);
1842                         goto out;
1843                 }
1844         }
1845
1846         rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1847                            IN_DEV_ORCONF(in_dev, NOPOLICY),
1848                            IN_DEV_ORCONF(out_dev, NOXFRM));
1849         if (!rth) {
1850                 err = -ENOBUFS;
1851                 goto cleanup;
1852         }
1853
1854         rth->rt_is_input = 1;
1855         RT_CACHE_STAT_INC(in_slow_tot);
1856
1857         rth->dst.input = ip_forward;
1858
1859         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1860                        do_cache);
1861         lwtunnel_set_redirect(&rth->dst);
1862         skb_dst_set(skb, &rth->dst);
1863 out:
1864         err = 0;
1865  cleanup:
1866         return err;
1867 }
1868
1869 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1870 /* To make ICMP packets follow the right flow, the multipath hash is
1871  * calculated from the inner IP addresses.
1872  */
1873 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1874                                  struct flow_keys *hash_keys)
1875 {
1876         const struct iphdr *outer_iph = ip_hdr(skb);
1877         const struct iphdr *key_iph = outer_iph;
1878         const struct iphdr *inner_iph;
1879         const struct icmphdr *icmph;
1880         struct iphdr _inner_iph;
1881         struct icmphdr _icmph;
1882
1883         if (likely(outer_iph->protocol != IPPROTO_ICMP))
1884                 goto out;
1885
1886         if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1887                 goto out;
1888
1889         icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1890                                    &_icmph);
1891         if (!icmph)
1892                 goto out;
1893
1894         if (!icmp_is_err(icmph->type))
1895                 goto out;
1896
1897         inner_iph = skb_header_pointer(skb,
1898                                        outer_iph->ihl * 4 + sizeof(_icmph),
1899                                        sizeof(_inner_iph), &_inner_iph);
1900         if (!inner_iph)
1901                 goto out;
1902
1903         key_iph = inner_iph;
1904 out:
1905         hash_keys->addrs.v4addrs.src = key_iph->saddr;
1906         hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1907 }
1908
1909 /* if skb is set it will be used and fl4 can be NULL */
1910 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
1911                        const struct sk_buff *skb, struct flow_keys *flkeys)
1912 {
1913         u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
1914         struct flow_keys hash_keys;
1915         u32 mhash;
1916
1917         switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1918         case 0:
1919                 memset(&hash_keys, 0, sizeof(hash_keys));
1920                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1921                 if (skb) {
1922                         ip_multipath_l3_keys(skb, &hash_keys);
1923                 } else {
1924                         hash_keys.addrs.v4addrs.src = fl4->saddr;
1925                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
1926                 }
1927                 break;
1928         case 1:
1929                 /* skb is currently provided only when forwarding */
1930                 if (skb) {
1931                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1932                         struct flow_keys keys;
1933
1934                         /* short-circuit if we already have L4 hash present */
1935                         if (skb->l4_hash)
1936                                 return skb_get_hash_raw(skb) >> 1;
1937
1938                         memset(&hash_keys, 0, sizeof(hash_keys));
1939
1940                         if (!flkeys) {
1941                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
1942                                 flkeys = &keys;
1943                         }
1944
1945                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1946                         hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
1947                         hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
1948                         hash_keys.ports.src = flkeys->ports.src;
1949                         hash_keys.ports.dst = flkeys->ports.dst;
1950                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
1951                 } else {
1952                         memset(&hash_keys, 0, sizeof(hash_keys));
1953                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1954                         hash_keys.addrs.v4addrs.src = fl4->saddr;
1955                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
1956                         hash_keys.ports.src = fl4->fl4_sport;
1957                         hash_keys.ports.dst = fl4->fl4_dport;
1958                         hash_keys.basic.ip_proto = fl4->flowi4_proto;
1959                 }
1960                 break;
1961         case 2:
1962                 memset(&hash_keys, 0, sizeof(hash_keys));
1963                 /* skb is currently provided only when forwarding */
1964                 if (skb) {
1965                         struct flow_keys keys;
1966
1967                         skb_flow_dissect_flow_keys(skb, &keys, 0);
1968                         /* Inner can be v4 or v6 */
1969                         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1970                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1971                                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1972                                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1973                         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1974                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1975                                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1976                                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1977                                 hash_keys.tags.flow_label = keys.tags.flow_label;
1978                                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1979                         } else {
1980                                 /* Same as case 0 */
1981                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1982                                 ip_multipath_l3_keys(skb, &hash_keys);
1983                         }
1984                 } else {
1985                         /* Same as case 0 */
1986                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1987                         hash_keys.addrs.v4addrs.src = fl4->saddr;
1988                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
1989                 }
1990                 break;
1991         }
1992         mhash = flow_hash_from_keys(&hash_keys);
1993
1994         if (multipath_hash)
1995                 mhash = jhash_2words(mhash, multipath_hash, 0);
1996
1997         return mhash >> 1;
1998 }
1999 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2000
2001 static int ip_mkroute_input(struct sk_buff *skb,
2002                             struct fib_result *res,
2003                             struct in_device *in_dev,
2004                             __be32 daddr, __be32 saddr, u32 tos,
2005                             struct flow_keys *hkeys)
2006 {
2007 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2008         if (res->fi && fib_info_num_path(res->fi) > 1) {
2009                 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2010
2011                 fib_select_multipath(res, h);
2012         }
2013 #endif
2014
2015         /* create a routing cache entry */
2016         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2017 }
2018
2019 /* Implements all the saddr-related checks as ip_route_input_slow(),
2020  * assuming daddr is valid and the destination is not a local broadcast one.
2021  * Uses the provided hint instead of performing a route lookup.
2022  */
2023 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2024                       u8 tos, struct net_device *dev,
2025                       const struct sk_buff *hint)
2026 {
2027         struct in_device *in_dev = __in_dev_get_rcu(dev);
2028         struct rtable *rt = skb_rtable(hint);
2029         struct net *net = dev_net(dev);
2030         int err = -EINVAL;
2031         u32 tag = 0;
2032
2033         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2034                 goto martian_source;
2035
2036         if (ipv4_is_zeronet(saddr))
2037                 goto martian_source;
2038
2039         if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2040                 goto martian_source;
2041
2042         if (rt->rt_type != RTN_LOCAL)
2043                 goto skip_validate_source;
2044
2045         tos &= IPTOS_RT_MASK;
2046         err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag);
2047         if (err < 0)
2048                 goto martian_source;
2049
2050 skip_validate_source:
2051         skb_dst_copy(skb, hint);
2052         return 0;
2053
2054 martian_source:
2055         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2056         return err;
2057 }
2058
2059 /*
2060  *      NOTE. We drop all the packets that has local source
2061  *      addresses, because every properly looped back packet
2062  *      must have correct destination already attached by output routine.
2063  *      Changes in the enforced policies must be applied also to
2064  *      ip_route_use_hint().
2065  *
2066  *      Such approach solves two big problems:
2067  *      1. Not simplex devices are handled properly.
2068  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2069  *      called with rcu_read_lock()
2070  */
2071
2072 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2073                                u8 tos, struct net_device *dev,
2074                                struct fib_result *res)
2075 {
2076         struct in_device *in_dev = __in_dev_get_rcu(dev);
2077         struct flow_keys *flkeys = NULL, _flkeys;
2078         struct net    *net = dev_net(dev);
2079         struct ip_tunnel_info *tun_info;
2080         int             err = -EINVAL;
2081         unsigned int    flags = 0;
2082         u32             itag = 0;
2083         struct rtable   *rth;
2084         struct flowi4   fl4;
2085         bool do_cache = true;
2086
2087         /* IP on this device is disabled. */
2088
2089         if (!in_dev)
2090                 goto out;
2091
2092         /* Check for the most weird martians, which can be not detected
2093          * by fib_lookup.
2094          */
2095
2096         tun_info = skb_tunnel_info(skb);
2097         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2098                 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2099         else
2100                 fl4.flowi4_tun_key.tun_id = 0;
2101         skb_dst_drop(skb);
2102
2103         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2104                 goto martian_source;
2105
2106         res->fi = NULL;
2107         res->table = NULL;
2108         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2109                 goto brd_input;
2110
2111         /* Accept zero addresses only to limited broadcast;
2112          * I even do not know to fix it or not. Waiting for complains :-)
2113          */
2114         if (ipv4_is_zeronet(saddr))
2115                 goto martian_source;
2116
2117         if (ipv4_is_zeronet(daddr))
2118                 goto martian_destination;
2119
2120         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2121          * and call it once if daddr or/and saddr are loopback addresses
2122          */
2123         if (ipv4_is_loopback(daddr)) {
2124                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2125                         goto martian_destination;
2126         } else if (ipv4_is_loopback(saddr)) {
2127                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2128                         goto martian_source;
2129         }
2130
2131         /*
2132          *      Now we are ready to route packet.
2133          */
2134         fl4.flowi4_oif = 0;
2135         fl4.flowi4_iif = dev->ifindex;
2136         fl4.flowi4_mark = skb->mark;
2137         fl4.flowi4_tos = tos;
2138         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2139         fl4.flowi4_flags = 0;
2140         fl4.daddr = daddr;
2141         fl4.saddr = saddr;
2142         fl4.flowi4_uid = sock_net_uid(net, NULL);
2143         fl4.flowi4_multipath_hash = 0;
2144
2145         if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2146                 flkeys = &_flkeys;
2147         } else {
2148                 fl4.flowi4_proto = 0;
2149                 fl4.fl4_sport = 0;
2150                 fl4.fl4_dport = 0;
2151         }
2152
2153         err = fib_lookup(net, &fl4, res, 0);
2154         if (err != 0) {
2155                 if (!IN_DEV_FORWARD(in_dev))
2156                         err = -EHOSTUNREACH;
2157                 goto no_route;
2158         }
2159
2160         if (res->type == RTN_BROADCAST) {
2161                 if (IN_DEV_BFORWARD(in_dev))
2162                         goto make_route;
2163                 /* not do cache if bc_forwarding is enabled */
2164                 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2165                         do_cache = false;
2166                 goto brd_input;
2167         }
2168
2169         if (res->type == RTN_LOCAL) {
2170                 err = fib_validate_source(skb, saddr, daddr, tos,
2171                                           0, dev, in_dev, &itag);
2172                 if (err < 0)
2173                         goto martian_source;
2174                 goto local_input;
2175         }
2176
2177         if (!IN_DEV_FORWARD(in_dev)) {
2178                 err = -EHOSTUNREACH;
2179                 goto no_route;
2180         }
2181         if (res->type != RTN_UNICAST)
2182                 goto martian_destination;
2183
2184 make_route:
2185         err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2186 out:    return err;
2187
2188 brd_input:
2189         if (skb->protocol != htons(ETH_P_IP))
2190                 goto e_inval;
2191
2192         if (!ipv4_is_zeronet(saddr)) {
2193                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2194                                           in_dev, &itag);
2195                 if (err < 0)
2196                         goto martian_source;
2197         }
2198         flags |= RTCF_BROADCAST;
2199         res->type = RTN_BROADCAST;
2200         RT_CACHE_STAT_INC(in_brd);
2201
2202 local_input:
2203         do_cache &= res->fi && !itag;
2204         if (do_cache) {
2205                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2206
2207                 rth = rcu_dereference(nhc->nhc_rth_input);
2208                 if (rt_cache_valid(rth)) {
2209                         skb_dst_set_noref(skb, &rth->dst);
2210                         err = 0;
2211                         goto out;
2212                 }
2213         }
2214
2215         rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2216                            flags | RTCF_LOCAL, res->type,
2217                            IN_DEV_ORCONF(in_dev, NOPOLICY), false);
2218         if (!rth)
2219                 goto e_nobufs;
2220
2221         rth->dst.output= ip_rt_bug;
2222 #ifdef CONFIG_IP_ROUTE_CLASSID
2223         rth->dst.tclassid = itag;
2224 #endif
2225         rth->rt_is_input = 1;
2226
2227         RT_CACHE_STAT_INC(in_slow_tot);
2228         if (res->type == RTN_UNREACHABLE) {
2229                 rth->dst.input= ip_error;
2230                 rth->dst.error= -err;
2231                 rth->rt_flags   &= ~RTCF_LOCAL;
2232         }
2233
2234         if (do_cache) {
2235                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2236
2237                 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2238                 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2239                         WARN_ON(rth->dst.input == lwtunnel_input);
2240                         rth->dst.lwtstate->orig_input = rth->dst.input;
2241                         rth->dst.input = lwtunnel_input;
2242                 }
2243
2244                 if (unlikely(!rt_cache_route(nhc, rth)))
2245                         rt_add_uncached_list(rth);
2246         }
2247         skb_dst_set(skb, &rth->dst);
2248         err = 0;
2249         goto out;
2250
2251 no_route:
2252         RT_CACHE_STAT_INC(in_no_route);
2253         res->type = RTN_UNREACHABLE;
2254         res->fi = NULL;
2255         res->table = NULL;
2256         goto local_input;
2257
2258         /*
2259          *      Do not cache martian addresses: they should be logged (RFC1812)
2260          */
2261 martian_destination:
2262         RT_CACHE_STAT_INC(in_martian_dst);
2263 #ifdef CONFIG_IP_ROUTE_VERBOSE
2264         if (IN_DEV_LOG_MARTIANS(in_dev))
2265                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2266                                      &daddr, &saddr, dev->name);
2267 #endif
2268
2269 e_inval:
2270         err = -EINVAL;
2271         goto out;
2272
2273 e_nobufs:
2274         err = -ENOBUFS;
2275         goto out;
2276
2277 martian_source:
2278         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2279         goto out;
2280 }
2281
2282 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2283                          u8 tos, struct net_device *dev)
2284 {
2285         struct fib_result res;
2286         int err;
2287
2288         tos &= IPTOS_RT_MASK;
2289         rcu_read_lock();
2290         err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2291         rcu_read_unlock();
2292
2293         return err;
2294 }
2295 EXPORT_SYMBOL(ip_route_input_noref);
2296
2297 /* called with rcu_read_lock held */
2298 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2299                        u8 tos, struct net_device *dev, struct fib_result *res)
2300 {
2301         /* Multicast recognition logic is moved from route cache to here.
2302          * The problem was that too many Ethernet cards have broken/missing
2303          * hardware multicast filters :-( As result the host on multicasting
2304          * network acquires a lot of useless route cache entries, sort of
2305          * SDR messages from all the world. Now we try to get rid of them.
2306          * Really, provided software IP multicast filter is organized
2307          * reasonably (at least, hashed), it does not result in a slowdown
2308          * comparing with route cache reject entries.
2309          * Note, that multicast routers are not affected, because
2310          * route cache entry is created eventually.
2311          */
2312         if (ipv4_is_multicast(daddr)) {
2313                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2314                 int our = 0;
2315                 int err = -EINVAL;
2316
2317                 if (!in_dev)
2318                         return err;
2319                 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2320                                       ip_hdr(skb)->protocol);
2321
2322                 /* check l3 master if no match yet */
2323                 if (!our && netif_is_l3_slave(dev)) {
2324                         struct in_device *l3_in_dev;
2325
2326                         l3_in_dev = __in_dev_get_rcu(skb->dev);
2327                         if (l3_in_dev)
2328                                 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2329                                                       ip_hdr(skb)->protocol);
2330                 }
2331
2332                 if (our
2333 #ifdef CONFIG_IP_MROUTE
2334                         ||
2335                     (!ipv4_is_local_multicast(daddr) &&
2336                      IN_DEV_MFORWARD(in_dev))
2337 #endif
2338                    ) {
2339                         err = ip_route_input_mc(skb, daddr, saddr,
2340                                                 tos, dev, our);
2341                 }
2342                 return err;
2343         }
2344
2345         return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2346 }
2347
2348 /* called with rcu_read_lock() */
2349 static struct rtable *__mkroute_output(const struct fib_result *res,
2350                                        const struct flowi4 *fl4, int orig_oif,
2351                                        struct net_device *dev_out,
2352                                        unsigned int flags)
2353 {
2354         struct fib_info *fi = res->fi;
2355         struct fib_nh_exception *fnhe;
2356         struct in_device *in_dev;
2357         u16 type = res->type;
2358         struct rtable *rth;
2359         bool do_cache;
2360
2361         in_dev = __in_dev_get_rcu(dev_out);
2362         if (!in_dev)
2363                 return ERR_PTR(-EINVAL);
2364
2365         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2366                 if (ipv4_is_loopback(fl4->saddr) &&
2367                     !(dev_out->flags & IFF_LOOPBACK) &&
2368                     !netif_is_l3_master(dev_out))
2369                         return ERR_PTR(-EINVAL);
2370
2371         if (ipv4_is_lbcast(fl4->daddr))
2372                 type = RTN_BROADCAST;
2373         else if (ipv4_is_multicast(fl4->daddr))
2374                 type = RTN_MULTICAST;
2375         else if (ipv4_is_zeronet(fl4->daddr))
2376                 return ERR_PTR(-EINVAL);
2377
2378         if (dev_out->flags & IFF_LOOPBACK)
2379                 flags |= RTCF_LOCAL;
2380
2381         do_cache = true;
2382         if (type == RTN_BROADCAST) {
2383                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2384                 fi = NULL;
2385         } else if (type == RTN_MULTICAST) {
2386                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2387                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2388                                      fl4->flowi4_proto))
2389                         flags &= ~RTCF_LOCAL;
2390                 else
2391                         do_cache = false;
2392                 /* If multicast route do not exist use
2393                  * default one, but do not gateway in this case.
2394                  * Yes, it is hack.
2395                  */
2396                 if (fi && res->prefixlen < 4)
2397                         fi = NULL;
2398         } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2399                    (orig_oif != dev_out->ifindex)) {
2400                 /* For local routes that require a particular output interface
2401                  * we do not want to cache the result.  Caching the result
2402                  * causes incorrect behaviour when there are multiple source
2403                  * addresses on the interface, the end result being that if the
2404                  * intended recipient is waiting on that interface for the
2405                  * packet he won't receive it because it will be delivered on
2406                  * the loopback interface and the IP_PKTINFO ipi_ifindex will
2407                  * be set to the loopback interface as well.
2408                  */
2409                 do_cache = false;
2410         }
2411
2412         fnhe = NULL;
2413         do_cache &= fi != NULL;
2414         if (fi) {
2415                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2416                 struct rtable __rcu **prth;
2417
2418                 fnhe = find_exception(nhc, fl4->daddr);
2419                 if (!do_cache)
2420                         goto add;
2421                 if (fnhe) {
2422                         prth = &fnhe->fnhe_rth_output;
2423                 } else {
2424                         if (unlikely(fl4->flowi4_flags &
2425                                      FLOWI_FLAG_KNOWN_NH &&
2426                                      !(nhc->nhc_gw_family &&
2427                                        nhc->nhc_scope == RT_SCOPE_LINK))) {
2428                                 do_cache = false;
2429                                 goto add;
2430                         }
2431                         prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2432                 }
2433                 rth = rcu_dereference(*prth);
2434                 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2435                         return rth;
2436         }
2437
2438 add:
2439         rth = rt_dst_alloc(dev_out, flags, type,
2440                            IN_DEV_ORCONF(in_dev, NOPOLICY),
2441                            IN_DEV_ORCONF(in_dev, NOXFRM));
2442         if (!rth)
2443                 return ERR_PTR(-ENOBUFS);
2444
2445         rth->rt_iif = orig_oif;
2446
2447         RT_CACHE_STAT_INC(out_slow_tot);
2448
2449         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2450                 if (flags & RTCF_LOCAL &&
2451                     !(dev_out->flags & IFF_LOOPBACK)) {
2452                         rth->dst.output = ip_mc_output;
2453                         RT_CACHE_STAT_INC(out_slow_mc);
2454                 }
2455 #ifdef CONFIG_IP_MROUTE
2456                 if (type == RTN_MULTICAST) {
2457                         if (IN_DEV_MFORWARD(in_dev) &&
2458                             !ipv4_is_local_multicast(fl4->daddr)) {
2459                                 rth->dst.input = ip_mr_input;
2460                                 rth->dst.output = ip_mc_output;
2461                         }
2462                 }
2463 #endif
2464         }
2465
2466         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2467         lwtunnel_set_redirect(&rth->dst);
2468
2469         return rth;
2470 }
2471
2472 /*
2473  * Major route resolver routine.
2474  */
2475
2476 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2477                                         const struct sk_buff *skb)
2478 {
2479         __u8 tos = RT_FL_TOS(fl4);
2480         struct fib_result res = {
2481                 .type           = RTN_UNSPEC,
2482                 .fi             = NULL,
2483                 .table          = NULL,
2484                 .tclassid       = 0,
2485         };
2486         struct rtable *rth;
2487
2488         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2489         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2490         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2491                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2492
2493         rcu_read_lock();
2494         rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2495         rcu_read_unlock();
2496
2497         return rth;
2498 }
2499 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2500
2501 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2502                                             struct fib_result *res,
2503                                             const struct sk_buff *skb)
2504 {
2505         struct net_device *dev_out = NULL;
2506         int orig_oif = fl4->flowi4_oif;
2507         unsigned int flags = 0;
2508         struct rtable *rth;
2509         int err;
2510
2511         if (fl4->saddr) {
2512                 if (ipv4_is_multicast(fl4->saddr) ||
2513                     ipv4_is_lbcast(fl4->saddr) ||
2514                     ipv4_is_zeronet(fl4->saddr)) {
2515                         rth = ERR_PTR(-EINVAL);
2516                         goto out;
2517                 }
2518
2519                 rth = ERR_PTR(-ENETUNREACH);
2520
2521                 /* I removed check for oif == dev_out->oif here.
2522                  * It was wrong for two reasons:
2523                  * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2524                  *    is assigned to multiple interfaces.
2525                  * 2. Moreover, we are allowed to send packets with saddr
2526                  *    of another iface. --ANK
2527                  */
2528
2529                 if (fl4->flowi4_oif == 0 &&
2530                     (ipv4_is_multicast(fl4->daddr) ||
2531                      ipv4_is_lbcast(fl4->daddr))) {
2532                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2533                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2534                         if (!dev_out)
2535                                 goto out;
2536
2537                         /* Special hack: user can direct multicasts
2538                          * and limited broadcast via necessary interface
2539                          * without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2540                          * This hack is not just for fun, it allows
2541                          * vic,vat and friends to work.
2542                          * They bind socket to loopback, set ttl to zero
2543                          * and expect that it will work.
2544                          * From the viewpoint of routing cache they are broken,
2545                          * because we are not allowed to build multicast path
2546                          * with loopback source addr (look, routing cache
2547                          * cannot know, that ttl is zero, so that packet
2548                          * will not leave this host and route is valid).
2549                          * Luckily, this hack is good workaround.
2550                          */
2551
2552                         fl4->flowi4_oif = dev_out->ifindex;
2553                         goto make_route;
2554                 }
2555
2556                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2557                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2558                         if (!__ip_dev_find(net, fl4->saddr, false))
2559                                 goto out;
2560                 }
2561         }
2562
2563
2564         if (fl4->flowi4_oif) {
2565                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2566                 rth = ERR_PTR(-ENODEV);
2567                 if (!dev_out)
2568                         goto out;
2569
2570                 /* RACE: Check return value of inet_select_addr instead. */
2571                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2572                         rth = ERR_PTR(-ENETUNREACH);
2573                         goto out;
2574                 }
2575                 if (ipv4_is_local_multicast(fl4->daddr) ||
2576                     ipv4_is_lbcast(fl4->daddr) ||
2577                     fl4->flowi4_proto == IPPROTO_IGMP) {
2578                         if (!fl4->saddr)
2579                                 fl4->saddr = inet_select_addr(dev_out, 0,
2580                                                               RT_SCOPE_LINK);
2581                         goto make_route;
2582                 }
2583                 if (!fl4->saddr) {
2584                         if (ipv4_is_multicast(fl4->daddr))
2585                                 fl4->saddr = inet_select_addr(dev_out, 0,
2586                                                               fl4->flowi4_scope);
2587                         else if (!fl4->daddr)
2588                                 fl4->saddr = inet_select_addr(dev_out, 0,
2589                                                               RT_SCOPE_HOST);
2590                 }
2591         }
2592
2593         if (!fl4->daddr) {
2594                 fl4->daddr = fl4->saddr;
2595                 if (!fl4->daddr)
2596                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2597                 dev_out = net->loopback_dev;
2598                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2599                 res->type = RTN_LOCAL;
2600                 flags |= RTCF_LOCAL;
2601                 goto make_route;
2602         }
2603
2604         err = fib_lookup(net, fl4, res, 0);
2605         if (err) {
2606                 res->fi = NULL;
2607                 res->table = NULL;
2608                 if (fl4->flowi4_oif &&
2609                     (ipv4_is_multicast(fl4->daddr) ||
2610                     !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2611                         /* Apparently, routing tables are wrong. Assume,
2612                          * that the destination is on link.
2613                          *
2614                          * WHY? DW.
2615                          * Because we are allowed to send to iface
2616                          * even if it has NO routes and NO assigned
2617                          * addresses. When oif is specified, routing
2618                          * tables are looked up with only one purpose:
2619                          * to catch if destination is gatewayed, rather than
2620                          * direct. Moreover, if MSG_DONTROUTE is set,
2621                          * we send packet, ignoring both routing tables
2622                          * and ifaddr state. --ANK
2623                          *
2624                          *
2625                          * We could make it even if oif is unknown,
2626                          * likely IPv6, but we do not.
2627                          */
2628
2629                         if (fl4->saddr == 0)
2630                                 fl4->saddr = inet_select_addr(dev_out, 0,
2631                                                               RT_SCOPE_LINK);
2632                         res->type = RTN_UNICAST;
2633                         goto make_route;
2634                 }
2635                 rth = ERR_PTR(err);
2636                 goto out;
2637         }
2638
2639         if (res->type == RTN_LOCAL) {
2640                 if (!fl4->saddr) {
2641                         if (res->fi->fib_prefsrc)
2642                                 fl4->saddr = res->fi->fib_prefsrc;
2643                         else
2644                                 fl4->saddr = fl4->daddr;
2645                 }
2646
2647                 /* L3 master device is the loopback for that domain */
2648                 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2649                         net->loopback_dev;
2650
2651                 /* make sure orig_oif points to fib result device even
2652                  * though packet rx/tx happens over loopback or l3mdev
2653                  */
2654                 orig_oif = FIB_RES_OIF(*res);
2655
2656                 fl4->flowi4_oif = dev_out->ifindex;
2657                 flags |= RTCF_LOCAL;
2658                 goto make_route;
2659         }
2660
2661         fib_select_path(net, res, fl4, skb);
2662
2663         dev_out = FIB_RES_DEV(*res);
2664
2665 make_route:
2666         rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2667
2668 out:
2669         return rth;
2670 }
2671
2672 static struct dst_ops ipv4_dst_blackhole_ops = {
2673         .family                 = AF_INET,
2674         .default_advmss         = ipv4_default_advmss,
2675         .neigh_lookup           = ipv4_neigh_lookup,
2676         .check                  = dst_blackhole_check,
2677         .cow_metrics            = dst_blackhole_cow_metrics,
2678         .update_pmtu            = dst_blackhole_update_pmtu,
2679         .redirect               = dst_blackhole_redirect,
2680         .mtu                    = dst_blackhole_mtu,
2681 };
2682
2683 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2684 {
2685         struct rtable *ort = (struct rtable *) dst_orig;
2686         struct rtable *rt;
2687
2688         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2689         if (rt) {
2690                 struct dst_entry *new = &rt->dst;
2691
2692                 new->__use = 1;
2693                 new->input = dst_discard;
2694                 new->output = dst_discard_out;
2695
2696                 new->dev = net->loopback_dev;
2697                 if (new->dev)
2698                         dev_hold(new->dev);
2699
2700                 rt->rt_is_input = ort->rt_is_input;
2701                 rt->rt_iif = ort->rt_iif;
2702                 rt->rt_pmtu = ort->rt_pmtu;
2703                 rt->rt_mtu_locked = ort->rt_mtu_locked;
2704
2705                 rt->rt_genid = rt_genid_ipv4(net);
2706                 rt->rt_flags = ort->rt_flags;
2707                 rt->rt_type = ort->rt_type;
2708                 rt->rt_uses_gateway = ort->rt_uses_gateway;
2709                 rt->rt_gw_family = ort->rt_gw_family;
2710                 if (rt->rt_gw_family == AF_INET)
2711                         rt->rt_gw4 = ort->rt_gw4;
2712                 else if (rt->rt_gw_family == AF_INET6)
2713                         rt->rt_gw6 = ort->rt_gw6;
2714
2715                 INIT_LIST_HEAD(&rt->rt_uncached);
2716         }
2717
2718         dst_release(dst_orig);
2719
2720         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2721 }
2722
2723 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2724                                     const struct sock *sk)
2725 {
2726         struct rtable *rt = __ip_route_output_key(net, flp4);
2727
2728         if (IS_ERR(rt))
2729                 return rt;
2730
2731         if (flp4->flowi4_proto) {
2732                 flp4->flowi4_oif = rt->dst.dev->ifindex;
2733                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2734                                                         flowi4_to_flowi(flp4),
2735                                                         sk, 0);
2736         }
2737
2738         return rt;
2739 }
2740 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2741
2742 struct rtable *ip_route_output_tunnel(struct sk_buff *skb,
2743                                       struct net_device *dev,
2744                                       struct net *net, __be32 *saddr,
2745                                       const struct ip_tunnel_info *info,
2746                                       u8 protocol, bool use_cache)
2747 {
2748 #ifdef CONFIG_DST_CACHE
2749         struct dst_cache *dst_cache;
2750 #endif
2751         struct rtable *rt = NULL;
2752         struct flowi4 fl4;
2753         __u8 tos;
2754
2755 #ifdef CONFIG_DST_CACHE
2756         dst_cache = (struct dst_cache *)&info->dst_cache;
2757         if (use_cache) {
2758                 rt = dst_cache_get_ip4(dst_cache, saddr);
2759                 if (rt)
2760                         return rt;
2761         }
2762 #endif
2763         memset(&fl4, 0, sizeof(fl4));
2764         fl4.flowi4_mark = skb->mark;
2765         fl4.flowi4_proto = protocol;
2766         fl4.daddr = info->key.u.ipv4.dst;
2767         fl4.saddr = info->key.u.ipv4.src;
2768         tos = info->key.tos;
2769         fl4.flowi4_tos = RT_TOS(tos);
2770
2771         rt = ip_route_output_key(net, &fl4);
2772         if (IS_ERR(rt)) {
2773                 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
2774                 return ERR_PTR(-ENETUNREACH);
2775         }
2776         if (rt->dst.dev == dev) { /* is this necessary? */
2777                 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
2778                 ip_rt_put(rt);
2779                 return ERR_PTR(-ELOOP);
2780         }
2781 #ifdef CONFIG_DST_CACHE
2782         if (use_cache)
2783                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2784 #endif
2785         *saddr = fl4.saddr;
2786         return rt;
2787 }
2788 EXPORT_SYMBOL_GPL(ip_route_output_tunnel);
2789
2790 /* called with rcu_read_lock held */
2791 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2792                         struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2793                         struct sk_buff *skb, u32 portid, u32 seq,
2794                         unsigned int flags)
2795 {
2796         struct rtmsg *r;
2797         struct nlmsghdr *nlh;
2798         unsigned long expires = 0;
2799         u32 error;
2800         u32 metrics[RTAX_MAX];
2801
2802         nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags);
2803         if (!nlh)
2804                 return -EMSGSIZE;
2805
2806         r = nlmsg_data(nlh);
2807         r->rtm_family    = AF_INET;
2808         r->rtm_dst_len  = 32;
2809         r->rtm_src_len  = 0;
2810         r->rtm_tos      = fl4 ? fl4->flowi4_tos : 0;
2811         r->rtm_table    = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2812         if (nla_put_u32(skb, RTA_TABLE, table_id))
2813                 goto nla_put_failure;
2814         r->rtm_type     = rt->rt_type;
2815         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2816         r->rtm_protocol = RTPROT_UNSPEC;
2817         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2818         if (rt->rt_flags & RTCF_NOTIFY)
2819                 r->rtm_flags |= RTM_F_NOTIFY;
2820         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2821                 r->rtm_flags |= RTCF_DOREDIRECT;
2822
2823         if (nla_put_in_addr(skb, RTA_DST, dst))
2824                 goto nla_put_failure;
2825         if (src) {
2826                 r->rtm_src_len = 32;
2827                 if (nla_put_in_addr(skb, RTA_SRC, src))
2828                         goto nla_put_failure;
2829         }
2830         if (rt->dst.dev &&
2831             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2832                 goto nla_put_failure;
2833         if (rt->dst.lwtstate &&
2834             lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
2835                 goto nla_put_failure;
2836 #ifdef CONFIG_IP_ROUTE_CLASSID
2837         if (rt->dst.tclassid &&
2838             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2839                 goto nla_put_failure;
2840 #endif
2841         if (fl4 && !rt_is_input_route(rt) &&
2842             fl4->saddr != src) {
2843                 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2844                         goto nla_put_failure;
2845         }
2846         if (rt->rt_uses_gateway) {
2847                 if (rt->rt_gw_family == AF_INET &&
2848                     nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2849                         goto nla_put_failure;
2850                 } else if (rt->rt_gw_family == AF_INET6) {
2851                         int alen = sizeof(struct in6_addr);
2852                         struct nlattr *nla;
2853                         struct rtvia *via;
2854
2855                         nla = nla_reserve(skb, RTA_VIA, alen + 2);
2856                         if (!nla)
2857                                 goto nla_put_failure;
2858
2859                         via = nla_data(nla);
2860                         via->rtvia_family = AF_INET6;
2861                         memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2862                 }
2863         }
2864
2865         expires = rt->dst.expires;
2866         if (expires) {
2867                 unsigned long now = jiffies;
2868
2869                 if (time_before(now, expires))
2870                         expires -= now;
2871                 else
2872                         expires = 0;
2873         }
2874
2875         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2876         if (rt->rt_pmtu && expires)
2877                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2878         if (rt->rt_mtu_locked && expires)
2879                 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2880         if (rtnetlink_put_metrics(skb, metrics) < 0)
2881                 goto nla_put_failure;
2882
2883         if (fl4) {
2884                 if (fl4->flowi4_mark &&
2885                     nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2886                         goto nla_put_failure;
2887
2888                 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2889                     nla_put_u32(skb, RTA_UID,
2890                                 from_kuid_munged(current_user_ns(),
2891                                                  fl4->flowi4_uid)))
2892                         goto nla_put_failure;
2893
2894                 if (rt_is_input_route(rt)) {
2895 #ifdef CONFIG_IP_MROUTE
2896                         if (ipv4_is_multicast(dst) &&
2897                             !ipv4_is_local_multicast(dst) &&
2898                             IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2899                                 int err = ipmr_get_route(net, skb,
2900                                                          fl4->saddr, fl4->daddr,
2901                                                          r, portid);
2902
2903                                 if (err <= 0) {
2904                                         if (err == 0)
2905                                                 return 0;
2906                                         goto nla_put_failure;
2907                                 }
2908                         } else
2909 #endif
2910                                 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
2911                                         goto nla_put_failure;
2912                 }
2913         }
2914
2915         error = rt->dst.error;
2916
2917         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2918                 goto nla_put_failure;
2919
2920         nlmsg_end(skb, nlh);
2921         return 0;
2922
2923 nla_put_failure:
2924         nlmsg_cancel(skb, nlh);
2925         return -EMSGSIZE;
2926 }
2927
2928 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
2929                             struct netlink_callback *cb, u32 table_id,
2930                             struct fnhe_hash_bucket *bucket, int genid,
2931                             int *fa_index, int fa_start, unsigned int flags)
2932 {
2933         int i;
2934
2935         for (i = 0; i < FNHE_HASH_SIZE; i++) {
2936                 struct fib_nh_exception *fnhe;
2937
2938                 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
2939                      fnhe = rcu_dereference(fnhe->fnhe_next)) {
2940                         struct rtable *rt;
2941                         int err;
2942
2943                         if (*fa_index < fa_start)
2944                                 goto next;
2945
2946                         if (fnhe->fnhe_genid != genid)
2947                                 goto next;
2948
2949                         if (fnhe->fnhe_expires &&
2950                             time_after(jiffies, fnhe->fnhe_expires))
2951                                 goto next;
2952
2953                         rt = rcu_dereference(fnhe->fnhe_rth_input);
2954                         if (!rt)
2955                                 rt = rcu_dereference(fnhe->fnhe_rth_output);
2956                         if (!rt)
2957                                 goto next;
2958
2959                         err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
2960                                            table_id, NULL, skb,
2961                                            NETLINK_CB(cb->skb).portid,
2962                                            cb->nlh->nlmsg_seq, flags);
2963                         if (err)
2964                                 return err;
2965 next:
2966                         (*fa_index)++;
2967                 }
2968         }
2969
2970         return 0;
2971 }
2972
2973 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
2974                        u32 table_id, struct fib_info *fi,
2975                        int *fa_index, int fa_start, unsigned int flags)
2976 {
2977         struct net *net = sock_net(cb->skb->sk);
2978         int nhsel, genid = fnhe_genid(net);
2979
2980         for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
2981                 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
2982                 struct fnhe_hash_bucket *bucket;
2983                 int err;
2984
2985                 if (nhc->nhc_flags & RTNH_F_DEAD)
2986                         continue;
2987
2988                 rcu_read_lock();
2989                 bucket = rcu_dereference(nhc->nhc_exceptions);
2990                 err = 0;
2991                 if (bucket)
2992                         err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
2993                                                genid, fa_index, fa_start,
2994                                                flags);
2995                 rcu_read_unlock();
2996                 if (err)
2997                         return err;
2998         }
2999
3000         return 0;
3001 }
3002
3003 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
3004                                                    u8 ip_proto, __be16 sport,
3005                                                    __be16 dport)
3006 {
3007         struct sk_buff *skb;
3008         struct iphdr *iph;
3009
3010         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3011         if (!skb)
3012                 return NULL;
3013
3014         /* Reserve room for dummy headers, this skb can pass
3015          * through good chunk of routing engine.
3016          */
3017         skb_reset_mac_header(skb);
3018         skb_reset_network_header(skb);
3019         skb->protocol = htons(ETH_P_IP);
3020         iph = skb_put(skb, sizeof(struct iphdr));
3021         iph->protocol = ip_proto;
3022         iph->saddr = src;
3023         iph->daddr = dst;
3024         iph->version = 0x4;
3025         iph->frag_off = 0;
3026         iph->ihl = 0x5;
3027         skb_set_transport_header(skb, skb->len);
3028
3029         switch (iph->protocol) {
3030         case IPPROTO_UDP: {
3031                 struct udphdr *udph;
3032
3033                 udph = skb_put_zero(skb, sizeof(struct udphdr));
3034                 udph->source = sport;
3035                 udph->dest = dport;
3036                 udph->len = sizeof(struct udphdr);
3037                 udph->check = 0;
3038                 break;
3039         }
3040         case IPPROTO_TCP: {
3041                 struct tcphdr *tcph;
3042
3043                 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
3044                 tcph->source    = sport;
3045                 tcph->dest      = dport;
3046                 tcph->doff      = sizeof(struct tcphdr) / 4;
3047                 tcph->rst = 1;
3048                 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
3049                                             src, dst, 0);
3050                 break;
3051         }
3052         case IPPROTO_ICMP: {
3053                 struct icmphdr *icmph;
3054
3055                 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
3056                 icmph->type = ICMP_ECHO;
3057                 icmph->code = 0;
3058         }
3059         }
3060
3061         return skb;
3062 }
3063
3064 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
3065                                        const struct nlmsghdr *nlh,
3066                                        struct nlattr **tb,
3067                                        struct netlink_ext_ack *extack)
3068 {
3069         struct rtmsg *rtm;
3070         int i, err;
3071
3072         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3073                 NL_SET_ERR_MSG(extack,
3074                                "ipv4: Invalid header for route get request");
3075                 return -EINVAL;
3076         }
3077
3078         if (!netlink_strict_get_check(skb))
3079                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3080                                               rtm_ipv4_policy, extack);
3081
3082         rtm = nlmsg_data(nlh);
3083         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3084             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3085             rtm->rtm_table || rtm->rtm_protocol ||
3086             rtm->rtm_scope || rtm->rtm_type) {
3087                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3088                 return -EINVAL;
3089         }
3090
3091         if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3092                                RTM_F_LOOKUP_TABLE |
3093                                RTM_F_FIB_MATCH)) {
3094                 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3095                 return -EINVAL;
3096         }
3097
3098         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3099                                             rtm_ipv4_policy, extack);
3100         if (err)
3101                 return err;
3102
3103         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3104             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3105                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3106                 return -EINVAL;
3107         }
3108
3109         for (i = 0; i <= RTA_MAX; i++) {
3110                 if (!tb[i])
3111                         continue;
3112
3113                 switch (i) {
3114                 case RTA_IIF:
3115                 case RTA_OIF:
3116                 case RTA_SRC:
3117                 case RTA_DST:
3118                 case RTA_IP_PROTO:
3119                 case RTA_SPORT:
3120                 case RTA_DPORT:
3121                 case RTA_MARK:
3122                 case RTA_UID:
3123                         break;
3124                 default:
3125                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3126                         return -EINVAL;
3127                 }
3128         }
3129
3130         return 0;
3131 }
3132
3133 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3134                              struct netlink_ext_ack *extack)
3135 {
3136         struct net *net = sock_net(in_skb->sk);
3137         struct nlattr *tb[RTA_MAX+1];
3138         u32 table_id = RT_TABLE_MAIN;
3139         __be16 sport = 0, dport = 0;
3140         struct fib_result res = {};
3141         u8 ip_proto = IPPROTO_UDP;
3142         struct rtable *rt = NULL;
3143         struct sk_buff *skb;
3144         struct rtmsg *rtm;
3145         struct flowi4 fl4 = {};
3146         __be32 dst = 0;
3147         __be32 src = 0;
3148         kuid_t uid;
3149         u32 iif;
3150         int err;
3151         int mark;
3152
3153         err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3154         if (err < 0)
3155                 return err;
3156
3157         rtm = nlmsg_data(nlh);
3158         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3159         dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3160         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3161         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3162         if (tb[RTA_UID])
3163                 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3164         else
3165                 uid = (iif ? INVALID_UID : current_uid());
3166
3167         if (tb[RTA_IP_PROTO]) {
3168                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3169                                                   &ip_proto, AF_INET, extack);
3170                 if (err)
3171                         return err;
3172         }
3173
3174         if (tb[RTA_SPORT])
3175                 sport = nla_get_be16(tb[RTA_SPORT]);
3176
3177         if (tb[RTA_DPORT])
3178                 dport = nla_get_be16(tb[RTA_DPORT]);
3179
3180         skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3181         if (!skb)
3182                 return -ENOBUFS;
3183
3184         fl4.daddr = dst;
3185         fl4.saddr = src;
3186         fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK;
3187         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3188         fl4.flowi4_mark = mark;
3189         fl4.flowi4_uid = uid;
3190         if (sport)
3191                 fl4.fl4_sport = sport;
3192         if (dport)
3193                 fl4.fl4_dport = dport;
3194         fl4.flowi4_proto = ip_proto;
3195
3196         rcu_read_lock();
3197
3198         if (iif) {
3199                 struct net_device *dev;
3200
3201                 dev = dev_get_by_index_rcu(net, iif);
3202                 if (!dev) {
3203                         err = -ENODEV;
3204                         goto errout_rcu;
3205                 }
3206
3207                 fl4.flowi4_iif = iif; /* for rt_fill_info */
3208                 skb->dev        = dev;
3209                 skb->mark       = mark;
3210                 err = ip_route_input_rcu(skb, dst, src,
3211                                          rtm->rtm_tos & IPTOS_RT_MASK, dev,
3212                                          &res);
3213
3214                 rt = skb_rtable(skb);
3215                 if (err == 0 && rt->dst.error)
3216                         err = -rt->dst.error;
3217         } else {
3218                 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3219                 skb->dev = net->loopback_dev;
3220                 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3221                 err = 0;
3222                 if (IS_ERR(rt))
3223                         err = PTR_ERR(rt);
3224                 else
3225                         skb_dst_set(skb, &rt->dst);
3226         }
3227
3228         if (err)
3229                 goto errout_rcu;
3230
3231         if (rtm->rtm_flags & RTM_F_NOTIFY)
3232                 rt->rt_flags |= RTCF_NOTIFY;
3233
3234         if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3235                 table_id = res.table ? res.table->tb_id : 0;
3236
3237         /* reset skb for netlink reply msg */
3238         skb_trim(skb, 0);
3239         skb_reset_network_header(skb);
3240         skb_reset_transport_header(skb);
3241         skb_reset_mac_header(skb);
3242
3243         if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3244                 struct fib_rt_info fri;
3245
3246                 if (!res.fi) {
3247                         err = fib_props[res.type].error;
3248                         if (!err)
3249                                 err = -EHOSTUNREACH;
3250                         goto errout_rcu;
3251                 }
3252                 fri.fi = res.fi;
3253                 fri.tb_id = table_id;
3254                 fri.dst = res.prefix;
3255                 fri.dst_len = res.prefixlen;
3256                 fri.tos = fl4.flowi4_tos;
3257                 fri.type = rt->rt_type;
3258                 fri.offload = 0;
3259                 fri.trap = 0;
3260                 fri.offload_failed = 0;
3261                 if (res.fa_head) {
3262                         struct fib_alias *fa;
3263
3264                         hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) {
3265                                 u8 slen = 32 - fri.dst_len;
3266
3267                                 if (fa->fa_slen == slen &&
3268                                     fa->tb_id == fri.tb_id &&
3269                                     fa->fa_tos == fri.tos &&
3270                                     fa->fa_info == res.fi &&
3271                                     fa->fa_type == fri.type) {
3272                                         fri.offload = fa->offload;
3273                                         fri.trap = fa->trap;
3274                                         break;
3275                                 }
3276                         }
3277                 }
3278                 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3279                                     nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0);
3280         } else {
3281                 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3282                                    NETLINK_CB(in_skb).portid,
3283                                    nlh->nlmsg_seq, 0);
3284         }
3285         if (err < 0)
3286                 goto errout_rcu;
3287
3288         rcu_read_unlock();
3289
3290         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3291
3292 errout_free:
3293         return err;
3294 errout_rcu:
3295         rcu_read_unlock();
3296         kfree_skb(skb);
3297         goto errout_free;
3298 }
3299
3300 void ip_rt_multicast_event(struct in_device *in_dev)
3301 {
3302         rt_cache_flush(dev_net(in_dev->dev));
3303 }
3304
3305 #ifdef CONFIG_SYSCTL
3306 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3307 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
3308 static int ip_rt_gc_elasticity __read_mostly    = 8;
3309 static int ip_min_valid_pmtu __read_mostly      = IPV4_MIN_MTU;
3310
3311 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3312                 void *buffer, size_t *lenp, loff_t *ppos)
3313 {
3314         struct net *net = (struct net *)__ctl->extra1;
3315
3316         if (write) {
3317                 rt_cache_flush(net);
3318                 fnhe_genid_bump(net);
3319                 return 0;
3320         }
3321
3322         return -EINVAL;
3323 }
3324
3325 static struct ctl_table ipv4_route_table[] = {
3326         {
3327                 .procname       = "gc_thresh",
3328                 .data           = &ipv4_dst_ops.gc_thresh,
3329                 .maxlen         = sizeof(int),
3330                 .mode           = 0644,
3331                 .proc_handler   = proc_dointvec,
3332         },
3333         {
3334                 .procname       = "max_size",
3335                 .data           = &ip_rt_max_size,
3336                 .maxlen         = sizeof(int),
3337                 .mode           = 0644,
3338                 .proc_handler   = proc_dointvec,
3339         },
3340         {
3341                 /*  Deprecated. Use gc_min_interval_ms */
3342
3343                 .procname       = "gc_min_interval",
3344                 .data           = &ip_rt_gc_min_interval,
3345                 .maxlen         = sizeof(int),
3346                 .mode           = 0644,
3347                 .proc_handler   = proc_dointvec_jiffies,
3348         },
3349         {
3350                 .procname       = "gc_min_interval_ms",
3351                 .data           = &ip_rt_gc_min_interval,
3352                 .maxlen         = sizeof(int),
3353                 .mode           = 0644,
3354                 .proc_handler   = proc_dointvec_ms_jiffies,
3355         },
3356         {
3357                 .procname       = "gc_timeout",
3358                 .data           = &ip_rt_gc_timeout,
3359                 .maxlen         = sizeof(int),
3360                 .mode           = 0644,
3361                 .proc_handler   = proc_dointvec_jiffies,
3362         },
3363         {
3364                 .procname       = "gc_interval",
3365                 .data           = &ip_rt_gc_interval,
3366                 .maxlen         = sizeof(int),
3367                 .mode           = 0644,
3368                 .proc_handler   = proc_dointvec_jiffies,
3369         },
3370         {
3371                 .procname       = "redirect_load",
3372                 .data           = &ip_rt_redirect_load,
3373                 .maxlen         = sizeof(int),
3374                 .mode           = 0644,
3375                 .proc_handler   = proc_dointvec,
3376         },
3377         {
3378                 .procname       = "redirect_number",
3379                 .data           = &ip_rt_redirect_number,
3380                 .maxlen         = sizeof(int),
3381                 .mode           = 0644,
3382                 .proc_handler   = proc_dointvec,
3383         },
3384         {
3385                 .procname       = "redirect_silence",
3386                 .data           = &ip_rt_redirect_silence,
3387                 .maxlen         = sizeof(int),
3388                 .mode           = 0644,
3389                 .proc_handler   = proc_dointvec,
3390         },
3391         {
3392                 .procname       = "error_cost",
3393                 .data           = &ip_rt_error_cost,
3394                 .maxlen         = sizeof(int),
3395                 .mode           = 0644,
3396                 .proc_handler   = proc_dointvec,
3397         },
3398         {
3399                 .procname       = "error_burst",
3400                 .data           = &ip_rt_error_burst,
3401                 .maxlen         = sizeof(int),
3402                 .mode           = 0644,
3403                 .proc_handler   = proc_dointvec,
3404         },
3405         {
3406                 .procname       = "gc_elasticity",
3407                 .data           = &ip_rt_gc_elasticity,
3408                 .maxlen         = sizeof(int),
3409                 .mode           = 0644,
3410                 .proc_handler   = proc_dointvec,
3411         },
3412         {
3413                 .procname       = "mtu_expires",
3414                 .data           = &ip_rt_mtu_expires,
3415                 .maxlen         = sizeof(int),
3416                 .mode           = 0644,
3417                 .proc_handler   = proc_dointvec_jiffies,
3418         },
3419         {
3420                 .procname       = "min_pmtu",
3421                 .data           = &ip_rt_min_pmtu,
3422                 .maxlen         = sizeof(int),
3423                 .mode           = 0644,
3424                 .proc_handler   = proc_dointvec_minmax,
3425                 .extra1         = &ip_min_valid_pmtu,
3426         },
3427         {
3428                 .procname       = "min_adv_mss",
3429                 .data           = &ip_rt_min_advmss,
3430                 .maxlen         = sizeof(int),
3431                 .mode           = 0644,
3432                 .proc_handler   = proc_dointvec,
3433         },
3434         { }
3435 };
3436
3437 static const char ipv4_route_flush_procname[] = "flush";
3438
3439 static struct ctl_table ipv4_route_flush_table[] = {
3440         {
3441                 .procname       = ipv4_route_flush_procname,
3442                 .maxlen         = sizeof(int),
3443                 .mode           = 0200,
3444                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3445         },
3446         { },
3447 };
3448
3449 static __net_init int sysctl_route_net_init(struct net *net)
3450 {
3451         struct ctl_table *tbl;
3452
3453         tbl = ipv4_route_flush_table;
3454         if (!net_eq(net, &init_net)) {
3455                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3456                 if (!tbl)
3457                         goto err_dup;
3458
3459                 /* Don't export non-whitelisted sysctls to unprivileged users */
3460                 if (net->user_ns != &init_user_ns) {
3461                         if (tbl[0].procname != ipv4_route_flush_procname)
3462                                 tbl[0].procname = NULL;
3463                 }
3464         }
3465         tbl[0].extra1 = net;
3466
3467         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3468         if (!net->ipv4.route_hdr)
3469                 goto err_reg;
3470         return 0;
3471
3472 err_reg:
3473         if (tbl != ipv4_route_flush_table)
3474                 kfree(tbl);
3475 err_dup:
3476         return -ENOMEM;
3477 }
3478
3479 static __net_exit void sysctl_route_net_exit(struct net *net)
3480 {
3481         struct ctl_table *tbl;
3482
3483         tbl = net->ipv4.route_hdr->ctl_table_arg;
3484         unregister_net_sysctl_table(net->ipv4.route_hdr);
3485         BUG_ON(tbl == ipv4_route_flush_table);
3486         kfree(tbl);
3487 }
3488
3489 static __net_initdata struct pernet_operations sysctl_route_ops = {
3490         .init = sysctl_route_net_init,
3491         .exit = sysctl_route_net_exit,
3492 };
3493 #endif
3494
3495 static __net_init int rt_genid_init(struct net *net)
3496 {
3497         atomic_set(&net->ipv4.rt_genid, 0);
3498         atomic_set(&net->fnhe_genid, 0);
3499         atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3500         return 0;
3501 }
3502
3503 static __net_initdata struct pernet_operations rt_genid_ops = {
3504         .init = rt_genid_init,
3505 };
3506
3507 static int __net_init ipv4_inetpeer_init(struct net *net)
3508 {
3509         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3510
3511         if (!bp)
3512                 return -ENOMEM;
3513         inet_peer_base_init(bp);
3514         net->ipv4.peers = bp;
3515         return 0;
3516 }
3517
3518 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3519 {
3520         struct inet_peer_base *bp = net->ipv4.peers;
3521
3522         net->ipv4.peers = NULL;
3523         inetpeer_invalidate_tree(bp);
3524         kfree(bp);
3525 }
3526
3527 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3528         .init   =       ipv4_inetpeer_init,
3529         .exit   =       ipv4_inetpeer_exit,
3530 };
3531
3532 #ifdef CONFIG_IP_ROUTE_CLASSID
3533 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3534 #endif /* CONFIG_IP_ROUTE_CLASSID */
3535
3536 int __init ip_rt_init(void)
3537 {
3538         void *idents_hash;
3539         int cpu;
3540
3541         /* For modern hosts, this will use 2 MB of memory */
3542         idents_hash = alloc_large_system_hash("IP idents",
3543                                               sizeof(*ip_idents) + sizeof(*ip_tstamps),
3544                                               0,
3545                                               16, /* one bucket per 64 KB */
3546                                               HASH_ZERO,
3547                                               NULL,
3548                                               &ip_idents_mask,
3549                                               2048,
3550                                               256*1024);
3551
3552         ip_idents = idents_hash;
3553
3554         prandom_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
3555
3556         ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
3557
3558         for_each_possible_cpu(cpu) {
3559                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3560
3561                 INIT_LIST_HEAD(&ul->head);
3562                 spin_lock_init(&ul->lock);
3563         }
3564 #ifdef CONFIG_IP_ROUTE_CLASSID
3565         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3566         if (!ip_rt_acct)
3567                 panic("IP: failed to allocate ip_rt_acct\n");
3568 #endif
3569
3570         ipv4_dst_ops.kmem_cachep =
3571                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3572                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3573
3574         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3575
3576         if (dst_entries_init(&ipv4_dst_ops) < 0)
3577                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3578
3579         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3580                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3581
3582         ipv4_dst_ops.gc_thresh = ~0;
3583         ip_rt_max_size = INT_MAX;
3584
3585         devinet_init();
3586         ip_fib_init();
3587
3588         if (ip_rt_proc_init())
3589                 pr_err("Unable to create route proc files\n");
3590 #ifdef CONFIG_XFRM
3591         xfrm_init();
3592         xfrm4_init();
3593 #endif
3594         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3595                       RTNL_FLAG_DOIT_UNLOCKED);
3596
3597 #ifdef CONFIG_SYSCTL
3598         register_pernet_subsys(&sysctl_route_ops);
3599 #endif
3600         register_pernet_subsys(&rt_genid_ops);
3601         register_pernet_subsys(&ipv4_inetpeer_ops);
3602         return 0;
3603 }
3604
3605 #ifdef CONFIG_SYSCTL
3606 /*
3607  * We really need to sanitize the damn ipv4 init order, then all
3608  * this nonsense will go away.
3609  */
3610 void __init ip_static_sysctl_init(void)
3611 {
3612         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3613 }
3614 #endif