net/ipv4/route.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              ROUTE - implementation of the IP router.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *              Linus Torvalds, <Linus.Torvalds@helsinki.fi>
  12  *              Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  13  *
  14  * Fixes:
  15  *              Alan Cox        :       Verify area fixes.
  16  *              Alan Cox        :       cli() protects routing changes
  17  *              Rui Oliveira    :       ICMP routing table updates
  18  *              (rco@di.uminho.pt)      Routing table insertion and update
  19  *              Linus Torvalds  :       Rewrote bits to be sensible
  20  *              Alan Cox        :       Added BSD route gw semantics
  21  *              Alan Cox        :       Super /proc >4K
  22  *              Alan Cox        :       MTU in route table
  23  *              Alan Cox        :       MSS actually. Also added the window
  24  *                                      clamper.
  25  *              Sam Lantinga    :       Fixed route matching in rt_del()
  26  *              Alan Cox        :       Routing cache support.
  27  *              Alan Cox        :       Removed compatibility cruft.
  28  *              Alan Cox        :       RTF_REJECT support.
  29  *              Alan Cox        :       TCP irtt support.
  30  *              Jonathan Naylor :       Added Metric support.
  31  *      Miquel van Smoorenburg  :       BSD API fixes.
  32  *      Miquel van Smoorenburg  :       Metrics.
  33  *              Alan Cox        :       Use __u32 properly
  34  *              Alan Cox        :       Aligned routing errors more closely with BSD
  35  *                                      our system is still very different.
  36  *              Alan Cox        :       Faster /proc handling
  37  *      Alexey Kuznetsov        :       Massive rework to support tree based routing,
  38  *                                      routing caches and better behaviour.
  39  *
  40  *              Olaf Erb        :       irtt wasn't being copied right.
  41  *              Bjorn Ekwall    :       Kerneld route support.
  42  *              Alan Cox        :       Multicast fixed (I hope)
  43  *              Pavel Krauz     :       Limited broadcast fixed
  44  *              Mike McLagan    :       Routing by source
  45  *      Alexey Kuznetsov        :       End of old history. Split to fib.c and
  46  *                                      route.c and rewritten from scratch.
  47  *              Andi Kleen      :       Load-limit warning messages.
  48  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  49  *      Vitaly E. Lavrov        :       Race condition in ip_route_input_slow.
  50  *      Tobias Ringstrom        :       Uninitialized res.type in ip_route_output_slow.
  51  *      Vladimir V. Ivanov      :       IP rule info (flowid) is really useful.
  52  *              Marc Boucher    :       routing by fwmark
  53  *      Robert Olsson           :       Added rt_cache statistics
  54  *      Arnaldo C. Melo         :       Convert proc stuff to seq_file
  55  *      Eric Dumazet            :       hashed spinlocks and rt_check_expire() fixes.
  56  *      Ilia Sotnikov           :       Ignore TOS on PMTUD and Redirect
  57  *      Ilia Sotnikov           :       Removed TOS from hash calculations
  58  *
  59  *              This program is free software; you can redistribute it and/or
  60  *              modify it under the terms of the GNU General Public License
  61  *              as published by the Free Software Foundation; either version
  62  *              2 of the License, or (at your option) any later version.
  63  */
  64
  65 #define pr_fmt(fmt) "IPv4: " fmt
  66
  67 #include <linux/module.h>
  68 #include <linux/uaccess.h>
  69 #include <linux/bitops.h>
  70 #include <linux/types.h>
  71 #include <linux/kernel.h>
  72 #include <linux/mm.h>
  73 #include <linux/string.h>
  74 #include <linux/socket.h>
  75 #include <linux/sockios.h>
  76 #include <linux/errno.h>
  77 #include <linux/in.h>
  78 #include <linux/inet.h>
  79 #include <linux/netdevice.h>
  80 #include <linux/proc_fs.h>
  81 #include <linux/init.h>
  82 #include <linux/skbuff.h>
  83 #include <linux/inetdevice.h>
  84 #include <linux/igmp.h>
  85 #include <linux/pkt_sched.h>
  86 #include <linux/mroute.h>
  87 #include <linux/netfilter_ipv4.h>
  88 #include <linux/random.h>
  89 #include <linux/rcupdate.h>
  90 #include <linux/times.h>
  91 #include <linux/slab.h>
  92 #include <linux/jhash.h>
  93 #include <net/dst.h>
  94 #include <net/dst_metadata.h>
  95 #include <net/net_namespace.h>
  96 #include <net/protocol.h>
  97 #include <net/ip.h>
  98 #include <net/route.h>
  99 #include <net/inetpeer.h>
 100 #include <net/sock.h>
 101 #include <net/ip_fib.h>
 102 #include <net/arp.h>
 103 #include <net/tcp.h>
 104 #include <net/icmp.h>
 105 #include <net/xfrm.h>
 106 #include <net/lwtunnel.h>
 107 #include <net/netevent.h>
 108 #include <net/rtnetlink.h>
 109 #ifdef CONFIG_SYSCTL
 110 #include <linux/sysctl.h>
 111 #endif
 112 #include <net/secure_seq.h>
 113 #include <net/ip_tunnels.h>
 114 #include <net/l3mdev.h>
 115
 116 #include "fib_lookup.h"
 117
 118 #define RT_FL_TOS(oldflp4) \
 119         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
 120
 121 #define RT_GC_TIMEOUT (300*HZ)
 122
 123 static int ip_rt_max_size;
 124 static int ip_rt_redirect_number __read_mostly  = 9;
 125 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 126 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 127 static int ip_rt_error_cost __read_mostly       = HZ;
 128 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 129 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
 130 static u32 ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
 131 static int ip_rt_min_advmss __read_mostly       = 256;
 132
 133 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
 134
 135 /*
 136  *      Interface to generic destination cache.
 137  */
 138
 139 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 140 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 141 static 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 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 147                                         struct sk_buff *skb);
 148 static void             ipv4_dst_destroy(struct dst_entry *dst);
 149
 150 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 151 {
 152         WARN_ON(1);
 153         return NULL;
 154 }
 155
 156 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 157                                            struct sk_buff *skb,
 158                                            const void *daddr);
 159 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
 160
 161 static struct dst_ops ipv4_dst_ops = {
 162         .family =               AF_INET,
 163         .check =                ipv4_dst_check,
 164         .default_advmss =       ipv4_default_advmss,
 165         .mtu =                  ipv4_mtu,
 166         .cow_metrics =          ipv4_cow_metrics,
 167         .destroy =              ipv4_dst_destroy,
 168         .negative_advice =      ipv4_negative_advice,
 169         .link_failure =         ipv4_link_failure,
 170         .update_pmtu =          ip_rt_update_pmtu,
 171         .redirect =             ip_do_redirect,
 172         .local_out =            __ip_local_out,
 173         .neigh_lookup =         ipv4_neigh_lookup,
 174         .confirm_neigh =        ipv4_confirm_neigh,
 175 };
 176
 177 #define ECN_OR_COST(class)      TC_PRIO_##class
 178
 179 const __u8 ip_tos2prio[16] = {
 180         TC_PRIO_BESTEFFORT,
 181         ECN_OR_COST(BESTEFFORT),
 182         TC_PRIO_BESTEFFORT,
 183         ECN_OR_COST(BESTEFFORT),
 184         TC_PRIO_BULK,
 185         ECN_OR_COST(BULK),
 186         TC_PRIO_BULK,
 187         ECN_OR_COST(BULK),
 188         TC_PRIO_INTERACTIVE,
 189         ECN_OR_COST(INTERACTIVE),
 190         TC_PRIO_INTERACTIVE,
 191         ECN_OR_COST(INTERACTIVE),
 192         TC_PRIO_INTERACTIVE_BULK,
 193         ECN_OR_COST(INTERACTIVE_BULK),
 194         TC_PRIO_INTERACTIVE_BULK,
 195         ECN_OR_COST(INTERACTIVE_BULK)
 196 };
 197 EXPORT_SYMBOL(ip_tos2prio);
 198
 199 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 200 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
 201
 202 #ifdef CONFIG_PROC_FS
 203 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 204 {
 205         if (*pos)
 206                 return NULL;
 207         return SEQ_START_TOKEN;
 208 }
 209
 210 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 211 {
 212         ++*pos;
 213         return NULL;
 214 }
 215
 216 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 217 {
 218 }
 219
 220 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 221 {
 222         if (v == SEQ_START_TOKEN)
 223                 seq_printf(seq, "%-127s\n",
 224                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 225                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 226                            "HHUptod\tSpecDst");
 227         return 0;
 228 }
 229
 230 static const struct seq_operations rt_cache_seq_ops = {
 231         .start  = rt_cache_seq_start,
 232         .next   = rt_cache_seq_next,
 233         .stop   = rt_cache_seq_stop,
 234         .show   = rt_cache_seq_show,
 235 };
 236
 237 static int rt_cache_seq_open(struct inode *inode, struct file *file)
 238 {
 239         return seq_open(file, &rt_cache_seq_ops);
 240 }
 241
 242 static const struct file_operations rt_cache_seq_fops = {
 243         .open    = rt_cache_seq_open,
 244         .read    = seq_read,
 245         .llseek  = seq_lseek,
 246         .release = seq_release,
 247 };
 248
 249
 250 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 251 {
 252         int cpu;
 253
 254         if (*pos == 0)
 255                 return SEQ_START_TOKEN;
 256
 257         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 258                 if (!cpu_possible(cpu))
 259                         continue;
 260                 *pos = cpu+1;
 261                 return &per_cpu(rt_cache_stat, cpu);
 262         }
 263         return NULL;
 264 }
 265
 266 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 267 {
 268         int cpu;
 269
 270         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 271                 if (!cpu_possible(cpu))
 272                         continue;
 273                 *pos = cpu+1;
 274                 return &per_cpu(rt_cache_stat, cpu);
 275         }
 276         return NULL;
 277
 278 }
 279
 280 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 281 {
 282
 283 }
 284
 285 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 286 {
 287         struct rt_cache_stat *st = v;
 288
 289         if (v == SEQ_START_TOKEN) {
 290                 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");
 291                 return 0;
 292         }
 293
 294         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 295                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 296                    dst_entries_get_slow(&ipv4_dst_ops),
 297                    0, /* st->in_hit */
 298                    st->in_slow_tot,
 299                    st->in_slow_mc,
 300                    st->in_no_route,
 301                    st->in_brd,
 302                    st->in_martian_dst,
 303                    st->in_martian_src,
 304
 305                    0, /* st->out_hit */
 306                    st->out_slow_tot,
 307                    st->out_slow_mc,
 308
 309                    0, /* st->gc_total */
 310                    0, /* st->gc_ignored */
 311                    0, /* st->gc_goal_miss */
 312                    0, /* st->gc_dst_overflow */
 313                    0, /* st->in_hlist_search */
 314                    0  /* st->out_hlist_search */
 315                 );
 316         return 0;
 317 }
 318
 319 static const struct seq_operations rt_cpu_seq_ops = {
 320         .start  = rt_cpu_seq_start,
 321         .next   = rt_cpu_seq_next,
 322         .stop   = rt_cpu_seq_stop,
 323         .show   = rt_cpu_seq_show,
 324 };
 325
 326
 327 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
 328 {
 329         return seq_open(file, &rt_cpu_seq_ops);
 330 }
 331
 332 static const struct file_operations rt_cpu_seq_fops = {
 333         .open    = rt_cpu_seq_open,
 334         .read    = seq_read,
 335         .llseek  = seq_lseek,
 336         .release = seq_release,
 337 };
 338
 339 #ifdef CONFIG_IP_ROUTE_CLASSID
 340 static int rt_acct_proc_show(struct seq_file *m, void *v)
 341 {
 342         struct ip_rt_acct *dst, *src;
 343         unsigned int i, j;
 344
 345         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 346         if (!dst)
 347                 return -ENOMEM;
 348
 349         for_each_possible_cpu(i) {
 350                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 351                 for (j = 0; j < 256; j++) {
 352                         dst[j].o_bytes   += src[j].o_bytes;
 353                         dst[j].o_packets += src[j].o_packets;
 354                         dst[j].i_bytes   += src[j].i_bytes;
 355                         dst[j].i_packets += src[j].i_packets;
 356                 }
 357         }
 358
 359         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 360         kfree(dst);
 361         return 0;
 362 }
 363 #endif
 364
 365 static int __net_init ip_rt_do_proc_init(struct net *net)
 366 {
 367         struct proc_dir_entry *pde;
 368
 369         pde = proc_create("rt_cache", 0444, net->proc_net,
 370                           &rt_cache_seq_fops);
 371         if (!pde)
 372                 goto err1;
 373
 374         pde = proc_create("rt_cache", 0444,
 375                           net->proc_net_stat, &rt_cpu_seq_fops);
 376         if (!pde)
 377                 goto err2;
 378
 379 #ifdef CONFIG_IP_ROUTE_CLASSID
 380         pde = proc_create_single("rt_acct", 0, net->proc_net,
 381                         rt_acct_proc_show);
 382         if (!pde)
 383                 goto err3;
 384 #endif
 385         return 0;
 386
 387 #ifdef CONFIG_IP_ROUTE_CLASSID
 388 err3:
 389         remove_proc_entry("rt_cache", net->proc_net_stat);
 390 #endif
 391 err2:
 392         remove_proc_entry("rt_cache", net->proc_net);
 393 err1:
 394         return -ENOMEM;
 395 }
 396
 397 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 398 {
 399         remove_proc_entry("rt_cache", net->proc_net_stat);
 400         remove_proc_entry("rt_cache", net->proc_net);
 401 #ifdef CONFIG_IP_ROUTE_CLASSID
 402         remove_proc_entry("rt_acct", net->proc_net);
 403 #endif
 404 }
 405
 406 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 407         .init = ip_rt_do_proc_init,
 408         .exit = ip_rt_do_proc_exit,
 409 };
 410
 411 static int __init ip_rt_proc_init(void)
 412 {
 413         return register_pernet_subsys(&ip_rt_proc_ops);
 414 }
 415
 416 #else
 417 static inline int ip_rt_proc_init(void)
 418 {
 419         return 0;
 420 }
 421 #endif /* CONFIG_PROC_FS */
 422
 423 static inline bool rt_is_expired(const struct rtable *rth)
 424 {
 425         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 426 }
 427
 428 void rt_cache_flush(struct net *net)
 429 {
 430         rt_genid_bump_ipv4(net);
 431 }
 432
 433 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 434                                            struct sk_buff *skb,
 435                                            const void *daddr)
 436 {
 437         const struct rtable *rt = container_of(dst, struct rtable, dst);
 438         struct net_device *dev = dst->dev;
 439         struct neighbour *n;
 440
 441         rcu_read_lock_bh();
 442
 443         if (likely(rt->rt_gw_family == AF_INET)) {
 444                 n = ip_neigh_gw4(dev, rt->rt_gw4);
 445         } else if (rt->rt_gw_family == AF_INET6) {
 446                 n = ip_neigh_gw6(dev, &rt->rt_gw6);
 447         } else {
 448                 __be32 pkey;
 449
 450                 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
 451                 n = ip_neigh_gw4(dev, pkey);
 452         }
 453
 454         if (n && !refcount_inc_not_zero(&n->refcnt))
 455                 n = NULL;
 456
 457         rcu_read_unlock_bh();
 458
 459         return n;
 460 }
 461
 462 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 463 {
 464         const struct rtable *rt = container_of(dst, struct rtable, dst);
 465         struct net_device *dev = dst->dev;
 466         const __be32 *pkey = daddr;
 467
 468         if (rt->rt_gw_family == AF_INET) {
 469                 pkey = (const __be32 *)&rt->rt_gw4;
 470         } else if (rt->rt_gw_family == AF_INET6) {
 471                 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
 472         } else if (!daddr ||
 473                  (rt->rt_flags &
 474                   (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
 475                 return;
 476         }
 477         __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
 478 }
 479
 480 #define IP_IDENTS_SZ 2048u
 481
 482 static atomic_t *ip_idents __read_mostly;
 483 static u32 *ip_tstamps __read_mostly;
 484
 485 /* In order to protect privacy, we add a perturbation to identifiers
 486  * if one generator is seldom used. This makes hard for an attacker
 487  * to infer how many packets were sent between two points in time.
 488  */
 489 u32 ip_idents_reserve(u32 hash, int segs)
 490 {
 491         u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
 492         atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
 493         u32 old = READ_ONCE(*p_tstamp);
 494         u32 now = (u32)jiffies;
 495         u32 new, delta = 0;
 496
 497         if (old != now && cmpxchg(p_tstamp, old, now) == old)
 498                 delta = prandom_u32_max(now - old);
 499
 500         /* Do not use atomic_add_return() as it makes UBSAN unhappy */
 501         do {
 502                 old = (u32)atomic_read(p_id);
 503                 new = old + delta + segs;
 504         } while (atomic_cmpxchg(p_id, old, new) != old);
 505
 506         return new - segs;
 507 }
 508 EXPORT_SYMBOL(ip_idents_reserve);
 509
 510 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
 511 {
 512         u32 hash, id;
 513
 514         /* Note the following code is not safe, but this is okay. */
 515         if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
 516                 get_random_bytes(&net->ipv4.ip_id_key,
 517                                  sizeof(net->ipv4.ip_id_key));
 518
 519         hash = siphash_3u32((__force u32)iph->daddr,
 520                             (__force u32)iph->saddr,
 521                             iph->protocol,
 522                             &net->ipv4.ip_id_key);
 523         id = ip_idents_reserve(hash, segs);
 524         iph->id = htons(id);
 525 }
 526 EXPORT_SYMBOL(__ip_select_ident);
 527
 528 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
 529                              const struct sock *sk,
 530                              const struct iphdr *iph,
 531                              int oif, u8 tos,
 532                              u8 prot, u32 mark, int flow_flags)
 533 {
 534         if (sk) {
 535                 const struct inet_sock *inet = inet_sk(sk);
 536
 537                 oif = sk->sk_bound_dev_if;
 538                 mark = sk->sk_mark;
 539                 tos = RT_CONN_FLAGS(sk);
 540                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 541         }
 542         flowi4_init_output(fl4, oif, mark, tos,
 543                            RT_SCOPE_UNIVERSE, prot,
 544                            flow_flags,
 545                            iph->daddr, iph->saddr, 0, 0,
 546                            sock_net_uid(net, sk));
 547 }
 548
 549 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 550                                const struct sock *sk)
 551 {
 552         const struct net *net = dev_net(skb->dev);
 553         const struct iphdr *iph = ip_hdr(skb);
 554         int oif = skb->dev->ifindex;
 555         u8 tos = RT_TOS(iph->tos);
 556         u8 prot = iph->protocol;
 557         u32 mark = skb->mark;
 558
 559         __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
 560 }
 561
 562 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 563 {
 564         const struct inet_sock *inet = inet_sk(sk);
 565         const struct ip_options_rcu *inet_opt;
 566         __be32 daddr = inet->inet_daddr;
 567
 568         rcu_read_lock();
 569         inet_opt = rcu_dereference(inet->inet_opt);
 570         if (inet_opt && inet_opt->opt.srr)
 571                 daddr = inet_opt->opt.faddr;
 572         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 573                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 574                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 575                            inet_sk_flowi_flags(sk),
 576                            daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
 577         rcu_read_unlock();
 578 }
 579
 580 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 581                                  const struct sk_buff *skb)
 582 {
 583         if (skb)
 584                 build_skb_flow_key(fl4, skb, sk);
 585         else
 586                 build_sk_flow_key(fl4, sk);
 587 }
 588
 589 static DEFINE_SPINLOCK(fnhe_lock);
 590
 591 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 592 {
 593         struct rtable *rt;
 594
 595         rt = rcu_dereference(fnhe->fnhe_rth_input);
 596         if (rt) {
 597                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 598                 dst_dev_put(&rt->dst);
 599                 dst_release(&rt->dst);
 600         }
 601         rt = rcu_dereference(fnhe->fnhe_rth_output);
 602         if (rt) {
 603                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 604                 dst_dev_put(&rt->dst);
 605                 dst_release(&rt->dst);
 606         }
 607 }
 608
 609 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
 610 {
 611         struct fib_nh_exception *fnhe, *oldest;
 612
 613         oldest = rcu_dereference(hash->chain);
 614         for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
 615              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 616                 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
 617                         oldest = fnhe;
 618         }
 619         fnhe_flush_routes(oldest);
 620         return oldest;
 621 }
 622
 623 static inline u32 fnhe_hashfun(__be32 daddr)
 624 {
 625         static u32 fnhe_hashrnd __read_mostly;
 626         u32 hval;
 627
 628         net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
 629         hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
 630         return hash_32(hval, FNHE_HASH_SHIFT);
 631 }
 632
 633 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 634 {
 635         rt->rt_pmtu = fnhe->fnhe_pmtu;
 636         rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
 637         rt->dst.expires = fnhe->fnhe_expires;
 638
 639         if (fnhe->fnhe_gw) {
 640                 rt->rt_flags |= RTCF_REDIRECTED;
 641                 rt->rt_gw_family = AF_INET;
 642                 rt->rt_gw4 = fnhe->fnhe_gw;
 643         }
 644 }
 645
 646 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
 647                                   u32 pmtu, bool lock, unsigned long expires)
 648 {
 649         struct fnhe_hash_bucket *hash;
 650         struct fib_nh_exception *fnhe;
 651         struct rtable *rt;
 652         u32 genid, hval;
 653         unsigned int i;
 654         int depth;
 655
 656         genid = fnhe_genid(dev_net(nh->fib_nh_dev));
 657         hval = fnhe_hashfun(daddr);
 658
 659         spin_lock_bh(&fnhe_lock);
 660
 661         hash = rcu_dereference(nh->nh_exceptions);
 662         if (!hash) {
 663                 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
 664                 if (!hash)
 665                         goto out_unlock;
 666                 rcu_assign_pointer(nh->nh_exceptions, hash);
 667         }
 668
 669         hash += hval;
 670
 671         depth = 0;
 672         for (fnhe = rcu_dereference(hash->chain); fnhe;
 673              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 674                 if (fnhe->fnhe_daddr == daddr)
 675                         break;
 676                 depth++;
 677         }
 678
 679         if (fnhe) {
 680                 if (fnhe->fnhe_genid != genid)
 681                         fnhe->fnhe_genid = genid;
 682                 if (gw)
 683                         fnhe->fnhe_gw = gw;
 684                 if (pmtu) {
 685                         fnhe->fnhe_pmtu = pmtu;
 686                         fnhe->fnhe_mtu_locked = lock;
 687                 }
 688                 fnhe->fnhe_expires = max(1UL, expires);
 689                 /* Update all cached dsts too */
 690                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 691                 if (rt)
 692                         fill_route_from_fnhe(rt, fnhe);
 693                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 694                 if (rt)
 695                         fill_route_from_fnhe(rt, fnhe);
 696         } else {
 697                 if (depth > FNHE_RECLAIM_DEPTH)
 698                         fnhe = fnhe_oldest(hash);
 699                 else {
 700                         fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 701                         if (!fnhe)
 702                                 goto out_unlock;
 703
 704                         fnhe->fnhe_next = hash->chain;
 705                         rcu_assign_pointer(hash->chain, fnhe);
 706                 }
 707                 fnhe->fnhe_genid = genid;
 708                 fnhe->fnhe_daddr = daddr;
 709                 fnhe->fnhe_gw = gw;
 710                 fnhe->fnhe_pmtu = pmtu;
 711                 fnhe->fnhe_mtu_locked = lock;
 712                 fnhe->fnhe_expires = max(1UL, expires);
 713
 714                 /* Exception created; mark the cached routes for the nexthop
 715                  * stale, so anyone caching it rechecks if this exception
 716                  * applies to them.
 717                  */
 718                 rt = rcu_dereference(nh->nh_rth_input);
 719                 if (rt)
 720                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 721
 722                 for_each_possible_cpu(i) {
 723                         struct rtable __rcu **prt;
 724                         prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
 725                         rt = rcu_dereference(*prt);
 726                         if (rt)
 727                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 728                 }
 729         }
 730
 731         fnhe->fnhe_stamp = jiffies;
 732
 733 out_unlock:
 734         spin_unlock_bh(&fnhe_lock);
 735 }
 736
 737 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 738                              bool kill_route)
 739 {
 740         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 741         __be32 old_gw = ip_hdr(skb)->saddr;
 742         struct net_device *dev = skb->dev;
 743         struct in_device *in_dev;
 744         struct fib_result res;
 745         struct neighbour *n;
 746         struct net *net;
 747
 748         switch (icmp_hdr(skb)->code & 7) {
 749         case ICMP_REDIR_NET:
 750         case ICMP_REDIR_NETTOS:
 751         case ICMP_REDIR_HOST:
 752         case ICMP_REDIR_HOSTTOS:
 753                 break;
 754
 755         default:
 756                 return;
 757         }
 758
 759         if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
 760                 return;
 761
 762         in_dev = __in_dev_get_rcu(dev);
 763         if (!in_dev)
 764                 return;
 765
 766         net = dev_net(dev);
 767         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 768             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 769             ipv4_is_zeronet(new_gw))
 770                 goto reject_redirect;
 771
 772         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 773                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 774                         goto reject_redirect;
 775                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 776                         goto reject_redirect;
 777         } else {
 778                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 779                         goto reject_redirect;
 780         }
 781
 782         n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
 783         if (!n)
 784                 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
 785         if (!IS_ERR(n)) {
 786                 if (!(n->nud_state & NUD_VALID)) {
 787                         neigh_event_send(n, NULL);
 788                 } else {
 789                         if (fib_lookup(net, fl4, &res, 0) == 0) {
 790                                 struct fib_nh_common *nhc = FIB_RES_NHC(res);
 791                                 struct fib_nh *nh;
 792
 793                                 nh = container_of(nhc, struct fib_nh, nh_common);
 794                                 update_or_create_fnhe(nh, fl4->daddr, new_gw,
 795                                                 0, false,
 796                                                 jiffies + ip_rt_gc_timeout);
 797                         }
 798                         if (kill_route)
 799                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 800                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 801                 }
 802                 neigh_release(n);
 803         }
 804         return;
 805
 806 reject_redirect:
 807 #ifdef CONFIG_IP_ROUTE_VERBOSE
 808         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 809                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 810                 __be32 daddr = iph->daddr;
 811                 __be32 saddr = iph->saddr;
 812
 813                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 814                                      "  Advised path = %pI4 -> %pI4\n",
 815                                      &old_gw, dev->name, &new_gw,
 816                                      &saddr, &daddr);
 817         }
 818 #endif
 819         ;
 820 }
 821
 822 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 823 {
 824         struct rtable *rt;
 825         struct flowi4 fl4;
 826         const struct iphdr *iph = (const struct iphdr *) skb->data;
 827         struct net *net = dev_net(skb->dev);
 828         int oif = skb->dev->ifindex;
 829         u8 tos = RT_TOS(iph->tos);
 830         u8 prot = iph->protocol;
 831         u32 mark = skb->mark;
 832
 833         rt = (struct rtable *) dst;
 834
 835         __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
 836         __ip_do_redirect(rt, skb, &fl4, true);
 837 }
 838
 839 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 840 {
 841         struct rtable *rt = (struct rtable *)dst;
 842         struct dst_entry *ret = dst;
 843
 844         if (rt) {
 845                 if (dst->obsolete > 0) {
 846                         ip_rt_put(rt);
 847                         ret = NULL;
 848                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 849                            rt->dst.expires) {
 850                         ip_rt_put(rt);
 851                         ret = NULL;
 852                 }
 853         }
 854         return ret;
 855 }
 856
 857 /*
 858  * Algorithm:
 859  *      1. The first ip_rt_redirect_number redirects are sent
 860  *         with exponential backoff, then we stop sending them at all,
 861  *         assuming that the host ignores our redirects.
 862  *      2. If we did not see packets requiring redirects
 863  *         during ip_rt_redirect_silence, we assume that the host
 864  *         forgot redirected route and start to send redirects again.
 865  *
 866  * This algorithm is much cheaper and more intelligent than dumb load limiting
 867  * in icmp.c.
 868  *
 869  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 870  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 871  */
 872
 873 void ip_rt_send_redirect(struct sk_buff *skb)
 874 {
 875         struct rtable *rt = skb_rtable(skb);
 876         struct in_device *in_dev;
 877         struct inet_peer *peer;
 878         struct net *net;
 879         int log_martians;
 880         int vif;
 881
 882         rcu_read_lock();
 883         in_dev = __in_dev_get_rcu(rt->dst.dev);
 884         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 885                 rcu_read_unlock();
 886                 return;
 887         }
 888         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 889         vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
 890         rcu_read_unlock();
 891
 892         net = dev_net(rt->dst.dev);
 893         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
 894         if (!peer) {
 895                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 896                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 897                 return;
 898         }
 899
 900         /* No redirected packets during ip_rt_redirect_silence;
 901          * reset the algorithm.
 902          */
 903         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
 904                 peer->rate_tokens = 0;
 905                 peer->n_redirects = 0;
 906         }
 907
 908         /* Too many ignored redirects; do not send anything
 909          * set dst.rate_last to the last seen redirected packet.
 910          */
 911         if (peer->n_redirects >= ip_rt_redirect_number) {
 912                 peer->rate_last = jiffies;
 913                 goto out_put_peer;
 914         }
 915
 916         /* Check for load limit; set rate_last to the latest sent
 917          * redirect.
 918          */
 919         if (peer->rate_tokens == 0 ||
 920             time_after(jiffies,
 921                        (peer->rate_last +
 922                         (ip_rt_redirect_load << peer->rate_tokens)))) {
 923                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 924
 925                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 926                 peer->rate_last = jiffies;
 927                 ++peer->rate_tokens;
 928                 ++peer->n_redirects;
 929 #ifdef CONFIG_IP_ROUTE_VERBOSE
 930                 if (log_martians &&
 931                     peer->rate_tokens == ip_rt_redirect_number)
 932                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 933                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 934                                              &ip_hdr(skb)->daddr, &gw);
 935 #endif
 936         }
 937 out_put_peer:
 938         inet_putpeer(peer);
 939 }
 940
 941 static int ip_error(struct sk_buff *skb)
 942 {
 943         struct rtable *rt = skb_rtable(skb);
 944         struct net_device *dev = skb->dev;
 945         struct in_device *in_dev;
 946         struct inet_peer *peer;
 947         unsigned long now;
 948         struct net *net;
 949         bool send;
 950         int code;
 951
 952         if (netif_is_l3_master(skb->dev)) {
 953                 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
 954                 if (!dev)
 955                         goto out;
 956         }
 957
 958         in_dev = __in_dev_get_rcu(dev);
 959
 960         /* IP on this device is disabled. */
 961         if (!in_dev)
 962                 goto out;
 963
 964         net = dev_net(rt->dst.dev);
 965         if (!IN_DEV_FORWARD(in_dev)) {
 966                 switch (rt->dst.error) {
 967                 case EHOSTUNREACH:
 968                         __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
 969                         break;
 970
 971                 case ENETUNREACH:
 972                         __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 973                         break;
 974                 }
 975                 goto out;
 976         }
 977
 978         switch (rt->dst.error) {
 979         case EINVAL:
 980         default:
 981                 goto out;
 982         case EHOSTUNREACH:
 983                 code = ICMP_HOST_UNREACH;
 984                 break;
 985         case ENETUNREACH:
 986                 code = ICMP_NET_UNREACH;
 987                 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 988                 break;
 989         case EACCES:
 990                 code = ICMP_PKT_FILTERED;
 991                 break;
 992         }
 993
 994         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
 995                                l3mdev_master_ifindex(skb->dev), 1);
 996
 997         send = true;
 998         if (peer) {
 999                 now = jiffies;
1000                 peer->rate_tokens += now - peer->rate_last;
1001                 if (peer->rate_tokens > ip_rt_error_burst)
1002                         peer->rate_tokens = ip_rt_error_burst;
1003                 peer->rate_last = now;
1004                 if (peer->rate_tokens >= ip_rt_error_cost)
1005                         peer->rate_tokens -= ip_rt_error_cost;
1006                 else
1007                         send = false;
1008                 inet_putpeer(peer);
1009         }
1010         if (send)
1011                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1012
1013 out:    kfree_skb(skb);
1014         return 0;
1015 }
1016
1017 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1018 {
1019         struct dst_entry *dst = &rt->dst;
1020         u32 old_mtu = ipv4_mtu(dst);
1021         struct fib_result res;
1022         bool lock = false;
1023
1024         if (ip_mtu_locked(dst))
1025                 return;
1026
1027         if (old_mtu < mtu)
1028                 return;
1029
1030         if (mtu < ip_rt_min_pmtu) {
1031                 lock = true;
1032                 mtu = min(old_mtu, ip_rt_min_pmtu);
1033         }
1034
1035         if (rt->rt_pmtu == mtu && !lock &&
1036             time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1037                 return;
1038
1039         rcu_read_lock();
1040         if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1041                 struct fib_nh_common *nhc = FIB_RES_NHC(res);
1042                 struct fib_nh *nh;
1043
1044                 nh = container_of(nhc, struct fib_nh, nh_common);
1045                 update_or_create_fnhe(nh, fl4->daddr, 0, mtu, lock,
1046                                       jiffies + ip_rt_mtu_expires);
1047         }
1048         rcu_read_unlock();
1049 }
1050
1051 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1052                               struct sk_buff *skb, u32 mtu)
1053 {
1054         struct rtable *rt = (struct rtable *) dst;
1055         struct flowi4 fl4;
1056
1057         ip_rt_build_flow_key(&fl4, sk, skb);
1058         __ip_rt_update_pmtu(rt, &fl4, mtu);
1059 }
1060
1061 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1062                       int oif, u8 protocol)
1063 {
1064         const struct iphdr *iph = (const struct iphdr *) skb->data;
1065         struct flowi4 fl4;
1066         struct rtable *rt;
1067         u32 mark = IP4_REPLY_MARK(net, skb->mark);
1068
1069         __build_flow_key(net, &fl4, NULL, iph, oif,
1070                          RT_TOS(iph->tos), protocol, mark, 0);
1071         rt = __ip_route_output_key(net, &fl4);
1072         if (!IS_ERR(rt)) {
1073                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1074                 ip_rt_put(rt);
1075         }
1076 }
1077 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1078
1079 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1080 {
1081         const struct iphdr *iph = (const struct iphdr *) skb->data;
1082         struct flowi4 fl4;
1083         struct rtable *rt;
1084
1085         __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1086
1087         if (!fl4.flowi4_mark)
1088                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1089
1090         rt = __ip_route_output_key(sock_net(sk), &fl4);
1091         if (!IS_ERR(rt)) {
1092                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1093                 ip_rt_put(rt);
1094         }
1095 }
1096
1097 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1098 {
1099         const struct iphdr *iph = (const struct iphdr *) skb->data;
1100         struct flowi4 fl4;
1101         struct rtable *rt;
1102         struct dst_entry *odst = NULL;
1103         bool new = false;
1104         struct net *net = sock_net(sk);
1105
1106         bh_lock_sock(sk);
1107
1108         if (!ip_sk_accept_pmtu(sk))
1109                 goto out;
1110
1111         odst = sk_dst_get(sk);
1112
1113         if (sock_owned_by_user(sk) || !odst) {
1114                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1115                 goto out;
1116         }
1117
1118         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1119
1120         rt = (struct rtable *)odst;
1121         if (odst->obsolete && !odst->ops->check(odst, 0)) {
1122                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1123                 if (IS_ERR(rt))
1124                         goto out;
1125
1126                 new = true;
1127         }
1128
1129         __ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu);
1130
1131         if (!dst_check(&rt->dst, 0)) {
1132                 if (new)
1133                         dst_release(&rt->dst);
1134
1135                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1136                 if (IS_ERR(rt))
1137                         goto out;
1138
1139                 new = true;
1140         }
1141
1142         if (new)
1143                 sk_dst_set(sk, &rt->dst);
1144
1145 out:
1146         bh_unlock_sock(sk);
1147         dst_release(odst);
1148 }
1149 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1150
1151 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1152                    int oif, u8 protocol)
1153 {
1154         const struct iphdr *iph = (const struct iphdr *) skb->data;
1155         struct flowi4 fl4;
1156         struct rtable *rt;
1157
1158         __build_flow_key(net, &fl4, NULL, iph, oif,
1159                          RT_TOS(iph->tos), protocol, 0, 0);
1160         rt = __ip_route_output_key(net, &fl4);
1161         if (!IS_ERR(rt)) {
1162                 __ip_do_redirect(rt, skb, &fl4, false);
1163                 ip_rt_put(rt);
1164         }
1165 }
1166 EXPORT_SYMBOL_GPL(ipv4_redirect);
1167
1168 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1169 {
1170         const struct iphdr *iph = (const struct iphdr *) skb->data;
1171         struct flowi4 fl4;
1172         struct rtable *rt;
1173         struct net *net = sock_net(sk);
1174
1175         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1176         rt = __ip_route_output_key(net, &fl4);
1177         if (!IS_ERR(rt)) {
1178                 __ip_do_redirect(rt, skb, &fl4, false);
1179                 ip_rt_put(rt);
1180         }
1181 }
1182 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1183
1184 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1185 {
1186         struct rtable *rt = (struct rtable *) dst;
1187
1188         /* All IPV4 dsts are created with ->obsolete set to the value
1189          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1190          * into this function always.
1191          *
1192          * When a PMTU/redirect information update invalidates a route,
1193          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1194          * DST_OBSOLETE_DEAD.
1195          */
1196         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1197                 return NULL;
1198         return dst;
1199 }
1200
1201 static void ipv4_link_failure(struct sk_buff *skb)
1202 {
1203         struct ip_options opt;
1204         struct rtable *rt;
1205         int res;
1206
1207         /* Recompile ip options since IPCB may not be valid anymore.
1208          */
1209         memset(&opt, 0, sizeof(opt));
1210         opt.optlen = ip_hdr(skb)->ihl*4 - sizeof(struct iphdr);
1211
1212         rcu_read_lock();
1213         res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1214         rcu_read_unlock();
1215
1216         if (res)
1217                 return;
1218
1219         __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1220
1221         rt = skb_rtable(skb);
1222         if (rt)
1223                 dst_set_expires(&rt->dst, 0);
1224 }
1225
1226 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1227 {
1228         pr_debug("%s: %pI4 -> %pI4, %s\n",
1229                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1230                  skb->dev ? skb->dev->name : "?");
1231         kfree_skb(skb);
1232         WARN_ON(1);
1233         return 0;
1234 }
1235
1236 /*
1237    We do not cache source address of outgoing interface,
1238    because it is used only by IP RR, TS and SRR options,
1239    so that it out of fast path.
1240
1241    BTW remember: "addr" is allowed to be not aligned
1242    in IP options!
1243  */
1244
1245 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1246 {
1247         __be32 src;
1248
1249         if (rt_is_output_route(rt))
1250                 src = ip_hdr(skb)->saddr;
1251         else {
1252                 struct fib_result res;
1253                 struct iphdr *iph = ip_hdr(skb);
1254                 struct flowi4 fl4 = {
1255                         .daddr = iph->daddr,
1256                         .saddr = iph->saddr,
1257                         .flowi4_tos = RT_TOS(iph->tos),
1258                         .flowi4_oif = rt->dst.dev->ifindex,
1259                         .flowi4_iif = skb->dev->ifindex,
1260                         .flowi4_mark = skb->mark,
1261                 };
1262
1263                 rcu_read_lock();
1264                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1265                         src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1266                 else
1267                         src = inet_select_addr(rt->dst.dev,
1268                                                rt_nexthop(rt, iph->daddr),
1269                                                RT_SCOPE_UNIVERSE);
1270                 rcu_read_unlock();
1271         }
1272         memcpy(addr, &src, 4);
1273 }
1274
1275 #ifdef CONFIG_IP_ROUTE_CLASSID
1276 static void set_class_tag(struct rtable *rt, u32 tag)
1277 {
1278         if (!(rt->dst.tclassid & 0xFFFF))
1279                 rt->dst.tclassid |= tag & 0xFFFF;
1280         if (!(rt->dst.tclassid & 0xFFFF0000))
1281                 rt->dst.tclassid |= tag & 0xFFFF0000;
1282 }
1283 #endif
1284
1285 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1286 {
1287         unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1288         unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1289                                     ip_rt_min_advmss);
1290
1291         return min(advmss, IPV4_MAX_PMTU - header_size);
1292 }
1293
1294 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1295 {
1296         const struct rtable *rt = (const struct rtable *) dst;
1297         unsigned int mtu = rt->rt_pmtu;
1298
1299         if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1300                 mtu = dst_metric_raw(dst, RTAX_MTU);
1301
1302         if (mtu)
1303                 return mtu;
1304
1305         mtu = READ_ONCE(dst->dev->mtu);
1306
1307         if (unlikely(ip_mtu_locked(dst))) {
1308                 if (rt->rt_gw_family && mtu > 576)
1309                         mtu = 576;
1310         }
1311
1312         mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1313
1314         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1315 }
1316
1317 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1318 {
1319         struct fnhe_hash_bucket *hash;
1320         struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1321         u32 hval = fnhe_hashfun(daddr);
1322
1323         spin_lock_bh(&fnhe_lock);
1324
1325         hash = rcu_dereference_protected(nh->nh_exceptions,
1326                                          lockdep_is_held(&fnhe_lock));
1327         hash += hval;
1328
1329         fnhe_p = &hash->chain;
1330         fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1331         while (fnhe) {
1332                 if (fnhe->fnhe_daddr == daddr) {
1333                         rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1334                                 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1335                         /* set fnhe_daddr to 0 to ensure it won't bind with
1336                          * new dsts in rt_bind_exception().
1337                          */
1338                         fnhe->fnhe_daddr = 0;
1339                         fnhe_flush_routes(fnhe);
1340                         kfree_rcu(fnhe, rcu);
1341                         break;
1342                 }
1343                 fnhe_p = &fnhe->fnhe_next;
1344                 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1345                                                  lockdep_is_held(&fnhe_lock));
1346         }
1347
1348         spin_unlock_bh(&fnhe_lock);
1349 }
1350
1351 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1352 {
1353         struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1354         struct fib_nh_exception *fnhe;
1355         u32 hval;
1356
1357         if (!hash)
1358                 return NULL;
1359
1360         hval = fnhe_hashfun(daddr);
1361
1362         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1363              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1364                 if (fnhe->fnhe_daddr == daddr) {
1365                         if (fnhe->fnhe_expires &&
1366                             time_after(jiffies, fnhe->fnhe_expires)) {
1367                                 ip_del_fnhe(nh, daddr);
1368                                 break;
1369                         }
1370                         return fnhe;
1371                 }
1372         }
1373         return NULL;
1374 }
1375
1376 /* MTU selection:
1377  * 1. mtu on route is locked - use it
1378  * 2. mtu from nexthop exception
1379  * 3. mtu from egress device
1380  */
1381
1382 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1383 {
1384         struct fib_nh_common *nhc = res->nhc;
1385         struct net_device *dev = nhc->nhc_dev;
1386         struct fib_info *fi = res->fi;
1387         u32 mtu = 0;
1388
1389         if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1390             fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1391                 mtu = fi->fib_mtu;
1392
1393         if (likely(!mtu)) {
1394                 struct fib_nh *nh = container_of(nhc, struct fib_nh, nh_common);
1395                 struct fib_nh_exception *fnhe;
1396
1397                 fnhe = find_exception(nh, daddr);
1398                 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1399                         mtu = fnhe->fnhe_pmtu;
1400         }
1401
1402         if (likely(!mtu))
1403                 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1404
1405         return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1406 }
1407
1408 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1409                               __be32 daddr, const bool do_cache)
1410 {
1411         bool ret = false;
1412
1413         spin_lock_bh(&fnhe_lock);
1414
1415         if (daddr == fnhe->fnhe_daddr) {
1416                 struct rtable __rcu **porig;
1417                 struct rtable *orig;
1418                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1419
1420                 if (rt_is_input_route(rt))
1421                         porig = &fnhe->fnhe_rth_input;
1422                 else
1423                         porig = &fnhe->fnhe_rth_output;
1424                 orig = rcu_dereference(*porig);
1425
1426                 if (fnhe->fnhe_genid != genid) {
1427                         fnhe->fnhe_genid = genid;
1428                         fnhe->fnhe_gw = 0;
1429                         fnhe->fnhe_pmtu = 0;
1430                         fnhe->fnhe_expires = 0;
1431                         fnhe->fnhe_mtu_locked = false;
1432                         fnhe_flush_routes(fnhe);
1433                         orig = NULL;
1434                 }
1435                 fill_route_from_fnhe(rt, fnhe);
1436                 if (!rt->rt_gw4) {
1437                         rt->rt_gw4 = daddr;
1438                         rt->rt_gw_family = AF_INET;
1439                 }
1440
1441                 if (do_cache) {
1442                         dst_hold(&rt->dst);
1443                         rcu_assign_pointer(*porig, rt);
1444                         if (orig) {
1445                                 dst_dev_put(&orig->dst);
1446                                 dst_release(&orig->dst);
1447                         }
1448                         ret = true;
1449                 }
1450
1451                 fnhe->fnhe_stamp = jiffies;
1452         }
1453         spin_unlock_bh(&fnhe_lock);
1454
1455         return ret;
1456 }
1457
1458 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1459 {
1460         struct rtable *orig, *prev, **p;
1461         bool ret = true;
1462
1463         if (rt_is_input_route(rt)) {
1464                 p = (struct rtable **)&nh->nh_rth_input;
1465         } else {
1466                 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1467         }
1468         orig = *p;
1469
1470         /* hold dst before doing cmpxchg() to avoid race condition
1471          * on this dst
1472          */
1473         dst_hold(&rt->dst);
1474         prev = cmpxchg(p, orig, rt);
1475         if (prev == orig) {
1476                 if (orig) {
1477                         dst_dev_put(&orig->dst);
1478                         dst_release(&orig->dst);
1479                 }
1480         } else {
1481                 dst_release(&rt->dst);
1482                 ret = false;
1483         }
1484
1485         return ret;
1486 }
1487
1488 struct uncached_list {
1489         spinlock_t              lock;
1490         struct list_head        head;
1491 };
1492
1493 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1494
1495 void rt_add_uncached_list(struct rtable *rt)
1496 {
1497         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1498
1499         rt->rt_uncached_list = ul;
1500
1501         spin_lock_bh(&ul->lock);
1502         list_add_tail(&rt->rt_uncached, &ul->head);
1503         spin_unlock_bh(&ul->lock);
1504 }
1505
1506 void rt_del_uncached_list(struct rtable *rt)
1507 {
1508         if (!list_empty(&rt->rt_uncached)) {
1509                 struct uncached_list *ul = rt->rt_uncached_list;
1510
1511                 spin_lock_bh(&ul->lock);
1512                 list_del(&rt->rt_uncached);
1513                 spin_unlock_bh(&ul->lock);
1514         }
1515 }
1516
1517 static void ipv4_dst_destroy(struct dst_entry *dst)
1518 {
1519         struct rtable *rt = (struct rtable *)dst;
1520
1521         ip_dst_metrics_put(dst);
1522         rt_del_uncached_list(rt);
1523 }
1524
1525 void rt_flush_dev(struct net_device *dev)
1526 {
1527         struct net *net = dev_net(dev);
1528         struct rtable *rt;
1529         int cpu;
1530
1531         for_each_possible_cpu(cpu) {
1532                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1533
1534                 spin_lock_bh(&ul->lock);
1535                 list_for_each_entry(rt, &ul->head, rt_uncached) {
1536                         if (rt->dst.dev != dev)
1537                                 continue;
1538                         rt->dst.dev = net->loopback_dev;
1539                         dev_hold(rt->dst.dev);
1540                         dev_put(dev);
1541                 }
1542                 spin_unlock_bh(&ul->lock);
1543         }
1544 }
1545
1546 static bool rt_cache_valid(const struct rtable *rt)
1547 {
1548         return  rt &&
1549                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1550                 !rt_is_expired(rt);
1551 }
1552
1553 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1554                            const struct fib_result *res,
1555                            struct fib_nh_exception *fnhe,
1556                            struct fib_info *fi, u16 type, u32 itag,
1557                            const bool do_cache)
1558 {
1559         bool cached = false;
1560
1561         if (fi) {
1562                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1563                 struct fib_nh *nh;
1564
1565                 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1566                         rt->rt_gw_family = nhc->nhc_gw_family;
1567                         /* only INET and INET6 are supported */
1568                         if (likely(nhc->nhc_gw_family == AF_INET))
1569                                 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1570                         else
1571                                 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1572                 }
1573
1574                 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1575
1576                 nh = container_of(nhc, struct fib_nh, nh_common);
1577 #ifdef CONFIG_IP_ROUTE_CLASSID
1578                 rt->dst.tclassid = nh->nh_tclassid;
1579 #endif
1580                 rt->dst.lwtstate = lwtstate_get(nh->fib_nh_lws);
1581                 if (unlikely(fnhe))
1582                         cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1583                 else if (do_cache)
1584                         cached = rt_cache_route(nh, rt);
1585                 if (unlikely(!cached)) {
1586                         /* Routes we intend to cache in nexthop exception or
1587                          * FIB nexthop have the DST_NOCACHE bit clear.
1588                          * However, if we are unsuccessful at storing this
1589                          * route into the cache we really need to set it.
1590                          */
1591                         if (!rt->rt_gw4) {
1592                                 rt->rt_gw_family = AF_INET;
1593                                 rt->rt_gw4 = daddr;
1594                         }
1595                         rt_add_uncached_list(rt);
1596                 }
1597         } else
1598                 rt_add_uncached_list(rt);
1599
1600 #ifdef CONFIG_IP_ROUTE_CLASSID
1601 #ifdef CONFIG_IP_MULTIPLE_TABLES
1602         set_class_tag(rt, res->tclassid);
1603 #endif
1604         set_class_tag(rt, itag);
1605 #endif
1606 }
1607
1608 struct rtable *rt_dst_alloc(struct net_device *dev,
1609                             unsigned int flags, u16 type,
1610                             bool nopolicy, bool noxfrm, bool will_cache)
1611 {
1612         struct rtable *rt;
1613
1614         rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1615                        (will_cache ? 0 : DST_HOST) |
1616                        (nopolicy ? DST_NOPOLICY : 0) |
1617                        (noxfrm ? DST_NOXFRM : 0));
1618
1619         if (rt) {
1620                 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1621                 rt->rt_flags = flags;
1622                 rt->rt_type = type;
1623                 rt->rt_is_input = 0;
1624                 rt->rt_iif = 0;
1625                 rt->rt_pmtu = 0;
1626                 rt->rt_mtu_locked = 0;
1627                 rt->rt_gw_family = 0;
1628                 rt->rt_gw4 = 0;
1629                 INIT_LIST_HEAD(&rt->rt_uncached);
1630
1631                 rt->dst.output = ip_output;
1632                 if (flags & RTCF_LOCAL)
1633                         rt->dst.input = ip_local_deliver;
1634         }
1635
1636         return rt;
1637 }
1638 EXPORT_SYMBOL(rt_dst_alloc);
1639
1640 /* called in rcu_read_lock() section */
1641 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1642                           u8 tos, struct net_device *dev,
1643                           struct in_device *in_dev, u32 *itag)
1644 {
1645         int err;
1646
1647         /* Primary sanity checks. */
1648         if (!in_dev)
1649                 return -EINVAL;
1650
1651         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1652             skb->protocol != htons(ETH_P_IP))
1653                 return -EINVAL;
1654
1655         if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1656                 return -EINVAL;
1657
1658         if (ipv4_is_zeronet(saddr)) {
1659                 if (!ipv4_is_local_multicast(daddr) &&
1660                     ip_hdr(skb)->protocol != IPPROTO_IGMP)
1661                         return -EINVAL;
1662         } else {
1663                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1664                                           in_dev, itag);
1665                 if (err < 0)
1666                         return err;
1667         }
1668         return 0;
1669 }
1670
1671 /* called in rcu_read_lock() section */
1672 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1673                              u8 tos, struct net_device *dev, int our)
1674 {
1675         struct in_device *in_dev = __in_dev_get_rcu(dev);
1676         unsigned int flags = RTCF_MULTICAST;
1677         struct rtable *rth;
1678         u32 itag = 0;
1679         int err;
1680
1681         err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1682         if (err)
1683                 return err;
1684
1685         if (our)
1686                 flags |= RTCF_LOCAL;
1687
1688         rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1689                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1690         if (!rth)
1691                 return -ENOBUFS;
1692
1693 #ifdef CONFIG_IP_ROUTE_CLASSID
1694         rth->dst.tclassid = itag;
1695 #endif
1696         rth->dst.output = ip_rt_bug;
1697         rth->rt_is_input= 1;
1698
1699 #ifdef CONFIG_IP_MROUTE
1700         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1701                 rth->dst.input = ip_mr_input;
1702 #endif
1703         RT_CACHE_STAT_INC(in_slow_mc);
1704
1705         skb_dst_set(skb, &rth->dst);
1706         return 0;
1707 }
1708
1709
1710 static void ip_handle_martian_source(struct net_device *dev,
1711                                      struct in_device *in_dev,
1712                                      struct sk_buff *skb,
1713                                      __be32 daddr,
1714                                      __be32 saddr)
1715 {
1716         RT_CACHE_STAT_INC(in_martian_src);
1717 #ifdef CONFIG_IP_ROUTE_VERBOSE
1718         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1719                 /*
1720                  *      RFC1812 recommendation, if source is martian,
1721                  *      the only hint is MAC header.
1722                  */
1723                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1724                         &daddr, &saddr, dev->name);
1725                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1726                         print_hex_dump(KERN_WARNING, "ll header: ",
1727                                        DUMP_PREFIX_OFFSET, 16, 1,
1728                                        skb_mac_header(skb),
1729                                        dev->hard_header_len, false);
1730                 }
1731         }
1732 #endif
1733 }
1734
1735 /* called in rcu_read_lock() section */
1736 static int __mkroute_input(struct sk_buff *skb,
1737                            const struct fib_result *res,
1738                            struct in_device *in_dev,
1739                            __be32 daddr, __be32 saddr, u32 tos)
1740 {
1741         struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1742         struct net_device *dev = nhc->nhc_dev;
1743         struct fib_nh_exception *fnhe;
1744         struct rtable *rth;
1745         struct fib_nh *nh;
1746         int err;
1747         struct in_device *out_dev;
1748         bool do_cache;
1749         u32 itag = 0;
1750
1751         /* get a working reference to the output device */
1752         out_dev = __in_dev_get_rcu(dev);
1753         if (!out_dev) {
1754                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1755                 return -EINVAL;
1756         }
1757
1758         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1759                                   in_dev->dev, in_dev, &itag);
1760         if (err < 0) {
1761                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1762                                          saddr);
1763
1764                 goto cleanup;
1765         }
1766
1767         do_cache = res->fi && !itag;
1768         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1769             skb->protocol == htons(ETH_P_IP)) {
1770                 __be32 gw;
1771
1772                 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1773                 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1774                     inet_addr_onlink(out_dev, saddr, gw))
1775                         IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1776         }
1777
1778         if (skb->protocol != htons(ETH_P_IP)) {
1779                 /* Not IP (i.e. ARP). Do not create route, if it is
1780                  * invalid for proxy arp. DNAT routes are always valid.
1781                  *
1782                  * Proxy arp feature have been extended to allow, ARP
1783                  * replies back to the same interface, to support
1784                  * Private VLAN switch technologies. See arp.c.
1785                  */
1786                 if (out_dev == in_dev &&
1787                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1788                         err = -EINVAL;
1789                         goto cleanup;
1790                 }
1791         }
1792
1793         nh = container_of(nhc, struct fib_nh, nh_common);
1794         fnhe = find_exception(nh, daddr);
1795         if (do_cache) {
1796                 if (fnhe)
1797                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1798                 else
1799                         rth = rcu_dereference(nh->nh_rth_input);
1800                 if (rt_cache_valid(rth)) {
1801                         skb_dst_set_noref(skb, &rth->dst);
1802                         goto out;
1803                 }
1804         }
1805
1806         rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1807                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1808                            IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1809         if (!rth) {
1810                 err = -ENOBUFS;
1811                 goto cleanup;
1812         }
1813
1814         rth->rt_is_input = 1;
1815         RT_CACHE_STAT_INC(in_slow_tot);
1816
1817         rth->dst.input = ip_forward;
1818
1819         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1820                        do_cache);
1821         lwtunnel_set_redirect(&rth->dst);
1822         skb_dst_set(skb, &rth->dst);
1823 out:
1824         err = 0;
1825  cleanup:
1826         return err;
1827 }
1828
1829 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1830 /* To make ICMP packets follow the right flow, the multipath hash is
1831  * calculated from the inner IP addresses.
1832  */
1833 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1834                                  struct flow_keys *hash_keys)
1835 {
1836         const struct iphdr *outer_iph = ip_hdr(skb);
1837         const struct iphdr *key_iph = outer_iph;
1838         const struct iphdr *inner_iph;
1839         const struct icmphdr *icmph;
1840         struct iphdr _inner_iph;
1841         struct icmphdr _icmph;
1842
1843         if (likely(outer_iph->protocol != IPPROTO_ICMP))
1844                 goto out;
1845
1846         if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1847                 goto out;
1848
1849         icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1850                                    &_icmph);
1851         if (!icmph)
1852                 goto out;
1853
1854         if (icmph->type != ICMP_DEST_UNREACH &&
1855             icmph->type != ICMP_REDIRECT &&
1856             icmph->type != ICMP_TIME_EXCEEDED &&
1857             icmph->type != ICMP_PARAMETERPROB)
1858                 goto out;
1859
1860         inner_iph = skb_header_pointer(skb,
1861                                        outer_iph->ihl * 4 + sizeof(_icmph),
1862                                        sizeof(_inner_iph), &_inner_iph);
1863         if (!inner_iph)
1864                 goto out;
1865
1866         key_iph = inner_iph;
1867 out:
1868         hash_keys->addrs.v4addrs.src = key_iph->saddr;
1869         hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1870 }
1871
1872 /* if skb is set it will be used and fl4 can be NULL */
1873 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
1874                        const struct sk_buff *skb, struct flow_keys *flkeys)
1875 {
1876         u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
1877         struct flow_keys hash_keys;
1878         u32 mhash;
1879
1880         switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1881         case 0:
1882                 memset(&hash_keys, 0, sizeof(hash_keys));
1883                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1884                 if (skb) {
1885                         ip_multipath_l3_keys(skb, &hash_keys);
1886                 } else {
1887                         hash_keys.addrs.v4addrs.src = fl4->saddr;
1888                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
1889                 }
1890                 break;
1891         case 1:
1892                 /* skb is currently provided only when forwarding */
1893                 if (skb) {
1894                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1895                         struct flow_keys keys;
1896
1897                         /* short-circuit if we already have L4 hash present */
1898                         if (skb->l4_hash)
1899                                 return skb_get_hash_raw(skb) >> 1;
1900
1901                         memset(&hash_keys, 0, sizeof(hash_keys));
1902
1903                         if (!flkeys) {
1904                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
1905                                 flkeys = &keys;
1906                         }
1907
1908                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1909                         hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
1910                         hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
1911                         hash_keys.ports.src = flkeys->ports.src;
1912                         hash_keys.ports.dst = flkeys->ports.dst;
1913                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
1914                 } else {
1915                         memset(&hash_keys, 0, sizeof(hash_keys));
1916                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1917                         hash_keys.addrs.v4addrs.src = fl4->saddr;
1918                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
1919                         hash_keys.ports.src = fl4->fl4_sport;
1920                         hash_keys.ports.dst = fl4->fl4_dport;
1921                         hash_keys.basic.ip_proto = fl4->flowi4_proto;
1922                 }
1923                 break;
1924         }
1925         mhash = flow_hash_from_keys(&hash_keys);
1926
1927         if (multipath_hash)
1928                 mhash = jhash_2words(mhash, multipath_hash, 0);
1929
1930         return mhash >> 1;
1931 }
1932 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1933
1934 static int ip_mkroute_input(struct sk_buff *skb,
1935                             struct fib_result *res,
1936                             struct in_device *in_dev,
1937                             __be32 daddr, __be32 saddr, u32 tos,
1938                             struct flow_keys *hkeys)
1939 {
1940 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1941         if (res->fi && res->fi->fib_nhs > 1) {
1942                 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
1943
1944                 fib_select_multipath(res, h);
1945         }
1946 #endif
1947
1948         /* create a routing cache entry */
1949         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1950 }
1951
1952 /*
1953  *      NOTE. We drop all the packets that has local source
1954  *      addresses, because every properly looped back packet
1955  *      must have correct destination already attached by output routine.
1956  *
1957  *      Such approach solves two big problems:
1958  *      1. Not simplex devices are handled properly.
1959  *      2. IP spoofing attempts are filtered with 100% of guarantee.
1960  *      called with rcu_read_lock()
1961  */
1962
1963 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1964                                u8 tos, struct net_device *dev,
1965                                struct fib_result *res)
1966 {
1967         struct in_device *in_dev = __in_dev_get_rcu(dev);
1968         struct flow_keys *flkeys = NULL, _flkeys;
1969         struct net    *net = dev_net(dev);
1970         struct ip_tunnel_info *tun_info;
1971         int             err = -EINVAL;
1972         unsigned int    flags = 0;
1973         u32             itag = 0;
1974         struct rtable   *rth;
1975         struct flowi4   fl4;
1976         bool do_cache;
1977
1978         /* IP on this device is disabled. */
1979
1980         if (!in_dev)
1981                 goto out;
1982
1983         /* Check for the most weird martians, which can be not detected
1984            by fib_lookup.
1985          */
1986
1987         tun_info = skb_tunnel_info(skb);
1988         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1989                 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1990         else
1991                 fl4.flowi4_tun_key.tun_id = 0;
1992         skb_dst_drop(skb);
1993
1994         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1995                 goto martian_source;
1996
1997         res->fi = NULL;
1998         res->table = NULL;
1999         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2000                 goto brd_input;
2001
2002         /* Accept zero addresses only to limited broadcast;
2003          * I even do not know to fix it or not. Waiting for complains :-)
2004          */
2005         if (ipv4_is_zeronet(saddr))
2006                 goto martian_source;
2007
2008         if (ipv4_is_zeronet(daddr))
2009                 goto martian_destination;
2010
2011         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2012          * and call it once if daddr or/and saddr are loopback addresses
2013          */
2014         if (ipv4_is_loopback(daddr)) {
2015                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2016                         goto martian_destination;
2017         } else if (ipv4_is_loopback(saddr)) {
2018                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2019                         goto martian_source;
2020         }
2021
2022         /*
2023          *      Now we are ready to route packet.
2024          */
2025         fl4.flowi4_oif = 0;
2026         fl4.flowi4_iif = dev->ifindex;
2027         fl4.flowi4_mark = skb->mark;
2028         fl4.flowi4_tos = tos;
2029         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2030         fl4.flowi4_flags = 0;
2031         fl4.daddr = daddr;
2032         fl4.saddr = saddr;
2033         fl4.flowi4_uid = sock_net_uid(net, NULL);
2034
2035         if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2036                 flkeys = &_flkeys;
2037         } else {
2038                 fl4.flowi4_proto = 0;
2039                 fl4.fl4_sport = 0;
2040                 fl4.fl4_dport = 0;
2041         }
2042
2043         err = fib_lookup(net, &fl4, res, 0);
2044         if (err != 0) {
2045                 if (!IN_DEV_FORWARD(in_dev))
2046                         err = -EHOSTUNREACH;
2047                 goto no_route;
2048         }
2049
2050         if (res->type == RTN_BROADCAST) {
2051                 if (IN_DEV_BFORWARD(in_dev))
2052                         goto make_route;
2053                 goto brd_input;
2054         }
2055
2056         if (res->type == RTN_LOCAL) {
2057                 err = fib_validate_source(skb, saddr, daddr, tos,
2058                                           0, dev, in_dev, &itag);
2059                 if (err < 0)
2060                         goto martian_source;
2061                 goto local_input;
2062         }
2063
2064         if (!IN_DEV_FORWARD(in_dev)) {
2065                 err = -EHOSTUNREACH;
2066                 goto no_route;
2067         }
2068         if (res->type != RTN_UNICAST)
2069                 goto martian_destination;
2070
2071 make_route:
2072         err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2073 out:    return err;
2074
2075 brd_input:
2076         if (skb->protocol != htons(ETH_P_IP))
2077                 goto e_inval;
2078
2079         if (!ipv4_is_zeronet(saddr)) {
2080                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2081                                           in_dev, &itag);
2082                 if (err < 0)
2083                         goto martian_source;
2084         }
2085         flags |= RTCF_BROADCAST;
2086         res->type = RTN_BROADCAST;
2087         RT_CACHE_STAT_INC(in_brd);
2088
2089 local_input:
2090         do_cache = false;
2091         if (res->fi) {
2092                 if (!itag) {
2093                         struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2094                         struct fib_nh *nh;
2095
2096                         nh = container_of(nhc, struct fib_nh, nh_common);
2097                         rth = rcu_dereference(nh->nh_rth_input);
2098                         if (rt_cache_valid(rth)) {
2099                                 skb_dst_set_noref(skb, &rth->dst);
2100                                 err = 0;
2101                                 goto out;
2102                         }
2103                         do_cache = true;
2104                 }
2105         }
2106
2107         rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2108                            flags | RTCF_LOCAL, res->type,
2109                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2110         if (!rth)
2111                 goto e_nobufs;
2112
2113         rth->dst.output= ip_rt_bug;
2114 #ifdef CONFIG_IP_ROUTE_CLASSID
2115         rth->dst.tclassid = itag;
2116 #endif
2117         rth->rt_is_input = 1;
2118
2119         RT_CACHE_STAT_INC(in_slow_tot);
2120         if (res->type == RTN_UNREACHABLE) {
2121                 rth->dst.input= ip_error;
2122                 rth->dst.error= -err;
2123                 rth->rt_flags   &= ~RTCF_LOCAL;
2124         }
2125
2126         if (do_cache) {
2127                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2128                 struct fib_nh *nh;
2129
2130                 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2131                 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2132                         WARN_ON(rth->dst.input == lwtunnel_input);
2133                         rth->dst.lwtstate->orig_input = rth->dst.input;
2134                         rth->dst.input = lwtunnel_input;
2135                 }
2136
2137                 nh = container_of(nhc, struct fib_nh, nh_common);
2138                 if (unlikely(!rt_cache_route(nh, rth)))
2139                         rt_add_uncached_list(rth);
2140         }
2141         skb_dst_set(skb, &rth->dst);
2142         err = 0;
2143         goto out;
2144
2145 no_route:
2146         RT_CACHE_STAT_INC(in_no_route);
2147         res->type = RTN_UNREACHABLE;
2148         res->fi = NULL;
2149         res->table = NULL;
2150         goto local_input;
2151
2152         /*
2153          *      Do not cache martian addresses: they should be logged (RFC1812)
2154          */
2155 martian_destination:
2156         RT_CACHE_STAT_INC(in_martian_dst);
2157 #ifdef CONFIG_IP_ROUTE_VERBOSE
2158         if (IN_DEV_LOG_MARTIANS(in_dev))
2159                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2160                                      &daddr, &saddr, dev->name);
2161 #endif
2162
2163 e_inval:
2164         err = -EINVAL;
2165         goto out;
2166
2167 e_nobufs:
2168         err = -ENOBUFS;
2169         goto out;
2170
2171 martian_source:
2172         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2173         goto out;
2174 }
2175
2176 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2177                          u8 tos, struct net_device *dev)
2178 {
2179         struct fib_result res;
2180         int err;
2181
2182         tos &= IPTOS_RT_MASK;
2183         rcu_read_lock();
2184         err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2185         rcu_read_unlock();
2186
2187         return err;
2188 }
2189 EXPORT_SYMBOL(ip_route_input_noref);
2190
2191 /* called with rcu_read_lock held */
2192 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2193                        u8 tos, struct net_device *dev, struct fib_result *res)
2194 {
2195         /* Multicast recognition logic is moved from route cache to here.
2196            The problem was that too many Ethernet cards have broken/missing
2197            hardware multicast filters :-( As result the host on multicasting
2198            network acquires a lot of useless route cache entries, sort of
2199            SDR messages from all the world. Now we try to get rid of them.
2200            Really, provided software IP multicast filter is organized
2201            reasonably (at least, hashed), it does not result in a slowdown
2202            comparing with route cache reject entries.
2203            Note, that multicast routers are not affected, because
2204            route cache entry is created eventually.
2205          */
2206         if (ipv4_is_multicast(daddr)) {
2207                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2208                 int our = 0;
2209                 int err = -EINVAL;
2210
2211                 if (!in_dev)
2212                         return err;
2213                 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2214                                       ip_hdr(skb)->protocol);
2215
2216                 /* check l3 master if no match yet */
2217                 if (!our && netif_is_l3_slave(dev)) {
2218                         struct in_device *l3_in_dev;
2219
2220                         l3_in_dev = __in_dev_get_rcu(skb->dev);
2221                         if (l3_in_dev)
2222                                 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2223                                                       ip_hdr(skb)->protocol);
2224                 }
2225
2226                 if (our
2227 #ifdef CONFIG_IP_MROUTE
2228                         ||
2229                     (!ipv4_is_local_multicast(daddr) &&
2230                      IN_DEV_MFORWARD(in_dev))
2231 #endif
2232                    ) {
2233                         err = ip_route_input_mc(skb, daddr, saddr,
2234                                                 tos, dev, our);
2235                 }
2236                 return err;
2237         }
2238
2239         return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2240 }
2241
2242 /* called with rcu_read_lock() */
2243 static struct rtable *__mkroute_output(const struct fib_result *res,
2244                                        const struct flowi4 *fl4, int orig_oif,
2245                                        struct net_device *dev_out,
2246                                        unsigned int flags)
2247 {
2248         struct fib_info *fi = res->fi;
2249         struct fib_nh_exception *fnhe;
2250         struct in_device *in_dev;
2251         u16 type = res->type;
2252         struct rtable *rth;
2253         bool do_cache;
2254
2255         in_dev = __in_dev_get_rcu(dev_out);
2256         if (!in_dev)
2257                 return ERR_PTR(-EINVAL);
2258
2259         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2260                 if (ipv4_is_loopback(fl4->saddr) &&
2261                     !(dev_out->flags & IFF_LOOPBACK) &&
2262                     !netif_is_l3_master(dev_out))
2263                         return ERR_PTR(-EINVAL);
2264
2265         if (ipv4_is_lbcast(fl4->daddr))
2266                 type = RTN_BROADCAST;
2267         else if (ipv4_is_multicast(fl4->daddr))
2268                 type = RTN_MULTICAST;
2269         else if (ipv4_is_zeronet(fl4->daddr))
2270                 return ERR_PTR(-EINVAL);
2271
2272         if (dev_out->flags & IFF_LOOPBACK)
2273                 flags |= RTCF_LOCAL;
2274
2275         do_cache = true;
2276         if (type == RTN_BROADCAST) {
2277                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2278                 fi = NULL;
2279         } else if (type == RTN_MULTICAST) {
2280                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2281                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2282                                      fl4->flowi4_proto))
2283                         flags &= ~RTCF_LOCAL;
2284                 else
2285                         do_cache = false;
2286                 /* If multicast route do not exist use
2287                  * default one, but do not gateway in this case.
2288                  * Yes, it is hack.
2289                  */
2290                 if (fi && res->prefixlen < 4)
2291                         fi = NULL;
2292         } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2293                    (orig_oif != dev_out->ifindex)) {
2294                 /* For local routes that require a particular output interface
2295                  * we do not want to cache the result.  Caching the result
2296                  * causes incorrect behaviour when there are multiple source
2297                  * addresses on the interface, the end result being that if the
2298                  * intended recipient is waiting on that interface for the
2299                  * packet he won't receive it because it will be delivered on
2300                  * the loopback interface and the IP_PKTINFO ipi_ifindex will
2301                  * be set to the loopback interface as well.
2302                  */
2303                 do_cache = false;
2304         }
2305
2306         fnhe = NULL;
2307         do_cache &= fi != NULL;
2308         if (fi) {
2309                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2310                 struct fib_nh *nh = container_of(nhc, struct fib_nh, nh_common);
2311                 struct rtable __rcu **prth;
2312
2313                 fnhe = find_exception(nh, fl4->daddr);
2314                 if (!do_cache)
2315                         goto add;
2316                 if (fnhe) {
2317                         prth = &fnhe->fnhe_rth_output;
2318                 } else {
2319                         if (unlikely(fl4->flowi4_flags &
2320                                      FLOWI_FLAG_KNOWN_NH &&
2321                                      !(nhc->nhc_gw_family &&
2322                                        nhc->nhc_scope == RT_SCOPE_LINK))) {
2323                                 do_cache = false;
2324                                 goto add;
2325                         }
2326                         prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2327                 }
2328                 rth = rcu_dereference(*prth);
2329                 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2330                         return rth;
2331         }
2332
2333 add:
2334         rth = rt_dst_alloc(dev_out, flags, type,
2335                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
2336                            IN_DEV_CONF_GET(in_dev, NOXFRM),
2337                            do_cache);
2338         if (!rth)
2339                 return ERR_PTR(-ENOBUFS);
2340
2341         rth->rt_iif = orig_oif;
2342
2343         RT_CACHE_STAT_INC(out_slow_tot);
2344
2345         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2346                 if (flags & RTCF_LOCAL &&
2347                     !(dev_out->flags & IFF_LOOPBACK)) {
2348                         rth->dst.output = ip_mc_output;
2349                         RT_CACHE_STAT_INC(out_slow_mc);
2350                 }
2351 #ifdef CONFIG_IP_MROUTE
2352                 if (type == RTN_MULTICAST) {
2353                         if (IN_DEV_MFORWARD(in_dev) &&
2354                             !ipv4_is_local_multicast(fl4->daddr)) {
2355                                 rth->dst.input = ip_mr_input;
2356                                 rth->dst.output = ip_mc_output;
2357                         }
2358                 }
2359 #endif
2360         }
2361
2362         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2363         lwtunnel_set_redirect(&rth->dst);
2364
2365         return rth;
2366 }
2367
2368 /*
2369  * Major route resolver routine.
2370  */
2371
2372 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2373                                         const struct sk_buff *skb)
2374 {
2375         __u8 tos = RT_FL_TOS(fl4);
2376         struct fib_result res = {
2377                 .type           = RTN_UNSPEC,
2378                 .fi             = NULL,
2379                 .table          = NULL,
2380                 .tclassid       = 0,
2381         };
2382         struct rtable *rth;
2383
2384         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2385         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2386         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2387                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2388
2389         rcu_read_lock();
2390         rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2391         rcu_read_unlock();
2392
2393         return rth;
2394 }
2395 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2396
2397 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2398                                             struct fib_result *res,
2399                                             const struct sk_buff *skb)
2400 {
2401         struct net_device *dev_out = NULL;
2402         int orig_oif = fl4->flowi4_oif;
2403         unsigned int flags = 0;
2404         struct rtable *rth;
2405         int err = -ENETUNREACH;
2406
2407         if (fl4->saddr) {
2408                 rth = ERR_PTR(-EINVAL);
2409                 if (ipv4_is_multicast(fl4->saddr) ||
2410                     ipv4_is_lbcast(fl4->saddr) ||
2411                     ipv4_is_zeronet(fl4->saddr))
2412                         goto out;
2413
2414                 /* I removed check for oif == dev_out->oif here.
2415                    It was wrong for two reasons:
2416                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2417                       is assigned to multiple interfaces.
2418                    2. Moreover, we are allowed to send packets with saddr
2419                       of another iface. --ANK
2420                  */
2421
2422                 if (fl4->flowi4_oif == 0 &&
2423                     (ipv4_is_multicast(fl4->daddr) ||
2424                      ipv4_is_lbcast(fl4->daddr))) {
2425                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2426                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2427                         if (!dev_out)
2428                                 goto out;
2429
2430                         /* Special hack: user can direct multicasts
2431                            and limited broadcast via necessary interface
2432                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2433                            This hack is not just for fun, it allows
2434                            vic,vat and friends to work.
2435                            They bind socket to loopback, set ttl to zero
2436                            and expect that it will work.
2437                            From the viewpoint of routing cache they are broken,
2438                            because we are not allowed to build multicast path
2439                            with loopback source addr (look, routing cache
2440                            cannot know, that ttl is zero, so that packet
2441                            will not leave this host and route is valid).
2442                            Luckily, this hack is good workaround.
2443                          */
2444
2445                         fl4->flowi4_oif = dev_out->ifindex;
2446                         goto make_route;
2447                 }
2448
2449                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2450                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2451                         if (!__ip_dev_find(net, fl4->saddr, false))
2452                                 goto out;
2453                 }
2454         }
2455
2456
2457         if (fl4->flowi4_oif) {
2458                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2459                 rth = ERR_PTR(-ENODEV);
2460                 if (!dev_out)
2461                         goto out;
2462
2463                 /* RACE: Check return value of inet_select_addr instead. */
2464                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2465                         rth = ERR_PTR(-ENETUNREACH);
2466                         goto out;
2467                 }
2468                 if (ipv4_is_local_multicast(fl4->daddr) ||
2469                     ipv4_is_lbcast(fl4->daddr) ||
2470                     fl4->flowi4_proto == IPPROTO_IGMP) {
2471                         if (!fl4->saddr)
2472                                 fl4->saddr = inet_select_addr(dev_out, 0,
2473                                                               RT_SCOPE_LINK);
2474                         goto make_route;
2475                 }
2476                 if (!fl4->saddr) {
2477                         if (ipv4_is_multicast(fl4->daddr))
2478                                 fl4->saddr = inet_select_addr(dev_out, 0,
2479                                                               fl4->flowi4_scope);
2480                         else if (!fl4->daddr)
2481                                 fl4->saddr = inet_select_addr(dev_out, 0,
2482                                                               RT_SCOPE_HOST);
2483                 }
2484         }
2485
2486         if (!fl4->daddr) {
2487                 fl4->daddr = fl4->saddr;
2488                 if (!fl4->daddr)
2489                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2490                 dev_out = net->loopback_dev;
2491                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2492                 res->type = RTN_LOCAL;
2493                 flags |= RTCF_LOCAL;
2494                 goto make_route;
2495         }
2496
2497         err = fib_lookup(net, fl4, res, 0);
2498         if (err) {
2499                 res->fi = NULL;
2500                 res->table = NULL;
2501                 if (fl4->flowi4_oif &&
2502                     (ipv4_is_multicast(fl4->daddr) ||
2503                     !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2504                         /* Apparently, routing tables are wrong. Assume,
2505                            that the destination is on link.
2506
2507                            WHY? DW.
2508                            Because we are allowed to send to iface
2509                            even if it has NO routes and NO assigned
2510                            addresses. When oif is specified, routing
2511                            tables are looked up with only one purpose:
2512                            to catch if destination is gatewayed, rather than
2513                            direct. Moreover, if MSG_DONTROUTE is set,
2514                            we send packet, ignoring both routing tables
2515                            and ifaddr state. --ANK
2516
2517
2518                            We could make it even if oif is unknown,
2519                            likely IPv6, but we do not.
2520                          */
2521
2522                         if (fl4->saddr == 0)
2523                                 fl4->saddr = inet_select_addr(dev_out, 0,
2524                                                               RT_SCOPE_LINK);
2525                         res->type = RTN_UNICAST;
2526                         goto make_route;
2527                 }
2528                 rth = ERR_PTR(err);
2529                 goto out;
2530         }
2531
2532         if (res->type == RTN_LOCAL) {
2533                 if (!fl4->saddr) {
2534                         if (res->fi->fib_prefsrc)
2535                                 fl4->saddr = res->fi->fib_prefsrc;
2536                         else
2537                                 fl4->saddr = fl4->daddr;
2538                 }
2539
2540                 /* L3 master device is the loopback for that domain */
2541                 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2542                         net->loopback_dev;
2543
2544                 /* make sure orig_oif points to fib result device even
2545                  * though packet rx/tx happens over loopback or l3mdev
2546                  */
2547                 orig_oif = FIB_RES_OIF(*res);
2548
2549                 fl4->flowi4_oif = dev_out->ifindex;
2550                 flags |= RTCF_LOCAL;
2551                 goto make_route;
2552         }
2553
2554         fib_select_path(net, res, fl4, skb);
2555
2556         dev_out = FIB_RES_DEV(*res);
2557         fl4->flowi4_oif = dev_out->ifindex;
2558
2559
2560 make_route:
2561         rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2562
2563 out:
2564         return rth;
2565 }
2566
2567 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2568 {
2569         return NULL;
2570 }
2571
2572 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2573 {
2574         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2575
2576         return mtu ? : dst->dev->mtu;
2577 }
2578
2579 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2580                                           struct sk_buff *skb, u32 mtu)
2581 {
2582 }
2583
2584 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2585                                        struct sk_buff *skb)
2586 {
2587 }
2588
2589 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2590                                           unsigned long old)
2591 {
2592         return NULL;
2593 }
2594
2595 static struct dst_ops ipv4_dst_blackhole_ops = {
2596         .family                 =       AF_INET,
2597         .check                  =       ipv4_blackhole_dst_check,
2598         .mtu                    =       ipv4_blackhole_mtu,
2599         .default_advmss         =       ipv4_default_advmss,
2600         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2601         .redirect               =       ipv4_rt_blackhole_redirect,
2602         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
2603         .neigh_lookup           =       ipv4_neigh_lookup,
2604 };
2605
2606 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2607 {
2608         struct rtable *ort = (struct rtable *) dst_orig;
2609         struct rtable *rt;
2610
2611         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2612         if (rt) {
2613                 struct dst_entry *new = &rt->dst;
2614
2615                 new->__use = 1;
2616                 new->input = dst_discard;
2617                 new->output = dst_discard_out;
2618
2619                 new->dev = net->loopback_dev;
2620                 if (new->dev)
2621                         dev_hold(new->dev);
2622
2623                 rt->rt_is_input = ort->rt_is_input;
2624                 rt->rt_iif = ort->rt_iif;
2625                 rt->rt_pmtu = ort->rt_pmtu;
2626                 rt->rt_mtu_locked = ort->rt_mtu_locked;
2627
2628                 rt->rt_genid = rt_genid_ipv4(net);
2629                 rt->rt_flags = ort->rt_flags;
2630                 rt->rt_type = ort->rt_type;
2631                 rt->rt_gw_family = ort->rt_gw_family;
2632                 if (rt->rt_gw_family == AF_INET)
2633                         rt->rt_gw4 = ort->rt_gw4;
2634                 else if (rt->rt_gw_family == AF_INET6)
2635                         rt->rt_gw6 = ort->rt_gw6;
2636
2637                 INIT_LIST_HEAD(&rt->rt_uncached);
2638         }
2639
2640         dst_release(dst_orig);
2641
2642         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2643 }
2644
2645 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2646                                     const struct sock *sk)
2647 {
2648         struct rtable *rt = __ip_route_output_key(net, flp4);
2649
2650         if (IS_ERR(rt))
2651                 return rt;
2652
2653         if (flp4->flowi4_proto)
2654                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2655                                                         flowi4_to_flowi(flp4),
2656                                                         sk, 0);
2657
2658         return rt;
2659 }
2660 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2661
2662 /* called with rcu_read_lock held */
2663 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2664                         struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2665                         struct sk_buff *skb, u32 portid, u32 seq)
2666 {
2667         struct rtmsg *r;
2668         struct nlmsghdr *nlh;
2669         unsigned long expires = 0;
2670         u32 error;
2671         u32 metrics[RTAX_MAX];
2672
2673         nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0);
2674         if (!nlh)
2675                 return -EMSGSIZE;
2676
2677         r = nlmsg_data(nlh);
2678         r->rtm_family    = AF_INET;
2679         r->rtm_dst_len  = 32;
2680         r->rtm_src_len  = 0;
2681         r->rtm_tos      = fl4->flowi4_tos;
2682         r->rtm_table    = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2683         if (nla_put_u32(skb, RTA_TABLE, table_id))
2684                 goto nla_put_failure;
2685         r->rtm_type     = rt->rt_type;
2686         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2687         r->rtm_protocol = RTPROT_UNSPEC;
2688         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2689         if (rt->rt_flags & RTCF_NOTIFY)
2690                 r->rtm_flags |= RTM_F_NOTIFY;
2691         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2692                 r->rtm_flags |= RTCF_DOREDIRECT;
2693
2694         if (nla_put_in_addr(skb, RTA_DST, dst))
2695                 goto nla_put_failure;
2696         if (src) {
2697                 r->rtm_src_len = 32;
2698                 if (nla_put_in_addr(skb, RTA_SRC, src))
2699                         goto nla_put_failure;
2700         }
2701         if (rt->dst.dev &&
2702             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2703                 goto nla_put_failure;
2704 #ifdef CONFIG_IP_ROUTE_CLASSID
2705         if (rt->dst.tclassid &&
2706             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2707                 goto nla_put_failure;
2708 #endif
2709         if (!rt_is_input_route(rt) &&
2710             fl4->saddr != src) {
2711                 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2712                         goto nla_put_failure;
2713         }
2714         if (rt->rt_gw_family == AF_INET &&
2715             nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2716                 goto nla_put_failure;
2717         } else if (rt->rt_gw_family == AF_INET6) {
2718                 int alen = sizeof(struct in6_addr);
2719                 struct nlattr *nla;
2720                 struct rtvia *via;
2721
2722                 nla = nla_reserve(skb, RTA_VIA, alen + 2);
2723                 if (!nla)
2724                         goto nla_put_failure;
2725
2726                 via = nla_data(nla);
2727                 via->rtvia_family = AF_INET6;
2728                 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2729         }
2730
2731         expires = rt->dst.expires;
2732         if (expires) {
2733                 unsigned long now = jiffies;
2734
2735                 if (time_before(now, expires))
2736                         expires -= now;
2737                 else
2738                         expires = 0;
2739         }
2740
2741         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2742         if (rt->rt_pmtu && expires)
2743                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2744         if (rt->rt_mtu_locked && expires)
2745                 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2746         if (rtnetlink_put_metrics(skb, metrics) < 0)
2747                 goto nla_put_failure;
2748
2749         if (fl4->flowi4_mark &&
2750             nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2751                 goto nla_put_failure;
2752
2753         if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2754             nla_put_u32(skb, RTA_UID,
2755                         from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2756                 goto nla_put_failure;
2757
2758         error = rt->dst.error;
2759
2760         if (rt_is_input_route(rt)) {
2761 #ifdef CONFIG_IP_MROUTE
2762                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2763                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2764                         int err = ipmr_get_route(net, skb,
2765                                                  fl4->saddr, fl4->daddr,
2766                                                  r, portid);
2767
2768                         if (err <= 0) {
2769                                 if (err == 0)
2770                                         return 0;
2771                                 goto nla_put_failure;
2772                         }
2773                 } else
2774 #endif
2775                         if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
2776                                 goto nla_put_failure;
2777         }
2778
2779         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2780                 goto nla_put_failure;
2781
2782         nlmsg_end(skb, nlh);
2783         return 0;
2784
2785 nla_put_failure:
2786         nlmsg_cancel(skb, nlh);
2787         return -EMSGSIZE;
2788 }
2789
2790 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
2791                                                    u8 ip_proto, __be16 sport,
2792                                                    __be16 dport)
2793 {
2794         struct sk_buff *skb;
2795         struct iphdr *iph;
2796
2797         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2798         if (!skb)
2799                 return NULL;
2800
2801         /* Reserve room for dummy headers, this skb can pass
2802          * through good chunk of routing engine.
2803          */
2804         skb_reset_mac_header(skb);
2805         skb_reset_network_header(skb);
2806         skb->protocol = htons(ETH_P_IP);
2807         iph = skb_put(skb, sizeof(struct iphdr));
2808         iph->protocol = ip_proto;
2809         iph->saddr = src;
2810         iph->daddr = dst;
2811         iph->version = 0x4;
2812         iph->frag_off = 0;
2813         iph->ihl = 0x5;
2814         skb_set_transport_header(skb, skb->len);
2815
2816         switch (iph->protocol) {
2817         case IPPROTO_UDP: {
2818                 struct udphdr *udph;
2819
2820                 udph = skb_put_zero(skb, sizeof(struct udphdr));
2821                 udph->source = sport;
2822                 udph->dest = dport;
2823                 udph->len = sizeof(struct udphdr);
2824                 udph->check = 0;
2825                 break;
2826         }
2827         case IPPROTO_TCP: {
2828                 struct tcphdr *tcph;
2829
2830                 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
2831                 tcph->source    = sport;
2832                 tcph->dest      = dport;
2833                 tcph->doff      = sizeof(struct tcphdr) / 4;
2834                 tcph->rst = 1;
2835                 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
2836                                             src, dst, 0);
2837                 break;
2838         }
2839         case IPPROTO_ICMP: {
2840                 struct icmphdr *icmph;
2841
2842                 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
2843                 icmph->type = ICMP_ECHO;
2844                 icmph->code = 0;
2845         }
2846         }
2847
2848         return skb;
2849 }
2850
2851 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
2852                                        const struct nlmsghdr *nlh,
2853                                        struct nlattr **tb,
2854                                        struct netlink_ext_ack *extack)
2855 {
2856         struct rtmsg *rtm;
2857         int i, err;
2858
2859         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2860                 NL_SET_ERR_MSG(extack,
2861                                "ipv4: Invalid header for route get request");
2862                 return -EINVAL;
2863         }
2864
2865         if (!netlink_strict_get_check(skb))
2866                 return nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX,
2867                                    rtm_ipv4_policy, extack);
2868
2869         rtm = nlmsg_data(nlh);
2870         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
2871             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
2872             rtm->rtm_table || rtm->rtm_protocol ||
2873             rtm->rtm_scope || rtm->rtm_type) {
2874                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
2875                 return -EINVAL;
2876         }
2877
2878         if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
2879                                RTM_F_LOOKUP_TABLE |
2880                                RTM_F_FIB_MATCH)) {
2881                 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
2882                 return -EINVAL;
2883         }
2884
2885         err = nlmsg_parse_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2886                                  rtm_ipv4_policy, extack);
2887         if (err)
2888                 return err;
2889
2890         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
2891             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
2892                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
2893                 return -EINVAL;
2894         }
2895
2896         for (i = 0; i <= RTA_MAX; i++) {
2897                 if (!tb[i])
2898                         continue;
2899
2900                 switch (i) {
2901                 case RTA_IIF:
2902                 case RTA_OIF:
2903                 case RTA_SRC:
2904                 case RTA_DST:
2905                 case RTA_IP_PROTO:
2906                 case RTA_SPORT:
2907                 case RTA_DPORT:
2908                 case RTA_MARK:
2909                 case RTA_UID:
2910                         break;
2911                 default:
2912                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
2913                         return -EINVAL;
2914                 }
2915         }
2916
2917         return 0;
2918 }
2919
2920 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2921                              struct netlink_ext_ack *extack)
2922 {
2923         struct net *net = sock_net(in_skb->sk);
2924         struct nlattr *tb[RTA_MAX+1];
2925         u32 table_id = RT_TABLE_MAIN;
2926         __be16 sport = 0, dport = 0;
2927         struct fib_result res = {};
2928         u8 ip_proto = IPPROTO_UDP;
2929         struct rtable *rt = NULL;
2930         struct sk_buff *skb;
2931         struct rtmsg *rtm;
2932         struct flowi4 fl4 = {};
2933         __be32 dst = 0;
2934         __be32 src = 0;
2935         kuid_t uid;
2936         u32 iif;
2937         int err;
2938         int mark;
2939
2940         err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
2941         if (err < 0)
2942                 return err;
2943
2944         rtm = nlmsg_data(nlh);
2945         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2946         dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2947         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2948         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2949         if (tb[RTA_UID])
2950                 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2951         else
2952                 uid = (iif ? INVALID_UID : current_uid());
2953
2954         if (tb[RTA_IP_PROTO]) {
2955                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
2956                                                   &ip_proto, AF_INET, extack);
2957                 if (err)
2958                         return err;
2959         }
2960
2961         if (tb[RTA_SPORT])
2962                 sport = nla_get_be16(tb[RTA_SPORT]);
2963
2964         if (tb[RTA_DPORT])
2965                 dport = nla_get_be16(tb[RTA_DPORT]);
2966
2967         skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
2968         if (!skb)
2969                 return -ENOBUFS;
2970
2971         fl4.daddr = dst;
2972         fl4.saddr = src;
2973         fl4.flowi4_tos = rtm->rtm_tos;
2974         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2975         fl4.flowi4_mark = mark;
2976         fl4.flowi4_uid = uid;
2977         if (sport)
2978                 fl4.fl4_sport = sport;
2979         if (dport)
2980                 fl4.fl4_dport = dport;
2981         fl4.flowi4_proto = ip_proto;
2982
2983         rcu_read_lock();
2984
2985         if (iif) {
2986                 struct net_device *dev;
2987
2988                 dev = dev_get_by_index_rcu(net, iif);
2989                 if (!dev) {
2990                         err = -ENODEV;
2991                         goto errout_rcu;
2992                 }
2993
2994                 fl4.flowi4_iif = iif; /* for rt_fill_info */
2995                 skb->dev        = dev;
2996                 skb->mark       = mark;
2997                 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
2998                                          dev, &res);
2999
3000                 rt = skb_rtable(skb);
3001                 if (err == 0 && rt->dst.error)
3002                         err = -rt->dst.error;
3003         } else {
3004                 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3005                 skb->dev = net->loopback_dev;
3006                 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3007                 err = 0;
3008                 if (IS_ERR(rt))
3009                         err = PTR_ERR(rt);
3010                 else
3011                         skb_dst_set(skb, &rt->dst);
3012         }
3013
3014         if (err)
3015                 goto errout_rcu;
3016
3017         if (rtm->rtm_flags & RTM_F_NOTIFY)
3018                 rt->rt_flags |= RTCF_NOTIFY;
3019
3020         if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3021                 table_id = res.table ? res.table->tb_id : 0;
3022
3023         /* reset skb for netlink reply msg */
3024         skb_trim(skb, 0);
3025         skb_reset_network_header(skb);
3026         skb_reset_transport_header(skb);
3027         skb_reset_mac_header(skb);
3028
3029         if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3030                 if (!res.fi) {
3031                         err = fib_props[res.type].error;
3032                         if (!err)
3033                                 err = -EHOSTUNREACH;
3034                         goto errout_rcu;
3035                 }
3036                 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3037                                     nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
3038                                     rt->rt_type, res.prefix, res.prefixlen,
3039                                     fl4.flowi4_tos, res.fi, 0);
3040         } else {
3041                 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3042                                    NETLINK_CB(in_skb).portid, nlh->nlmsg_seq);
3043         }
3044         if (err < 0)
3045                 goto errout_rcu;
3046
3047         rcu_read_unlock();
3048
3049         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3050
3051 errout_free:
3052         return err;
3053 errout_rcu:
3054         rcu_read_unlock();
3055         kfree_skb(skb);
3056         goto errout_free;
3057 }
3058
3059 void ip_rt_multicast_event(struct in_device *in_dev)
3060 {
3061         rt_cache_flush(dev_net(in_dev->dev));
3062 }
3063
3064 #ifdef CONFIG_SYSCTL
3065 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3066 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
3067 static int ip_rt_gc_elasticity __read_mostly    = 8;
3068 static int ip_min_valid_pmtu __read_mostly      = IPV4_MIN_MTU;
3069
3070 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3071                                         void __user *buffer,
3072                                         size_t *lenp, loff_t *ppos)
3073 {
3074         struct net *net = (struct net *)__ctl->extra1;
3075
3076         if (write) {
3077                 rt_cache_flush(net);
3078                 fnhe_genid_bump(net);
3079                 return 0;
3080         }
3081
3082         return -EINVAL;
3083 }
3084
3085 static struct ctl_table ipv4_route_table[] = {
3086         {
3087                 .procname       = "gc_thresh",
3088                 .data           = &ipv4_dst_ops.gc_thresh,
3089                 .maxlen         = sizeof(int),
3090                 .mode           = 0644,
3091                 .proc_handler   = proc_dointvec,
3092         },
3093         {
3094                 .procname       = "max_size",
3095                 .data           = &ip_rt_max_size,
3096                 .maxlen         = sizeof(int),
3097                 .mode           = 0644,
3098                 .proc_handler   = proc_dointvec,
3099         },
3100         {
3101                 /*  Deprecated. Use gc_min_interval_ms */
3102
3103                 .procname       = "gc_min_interval",
3104                 .data           = &ip_rt_gc_min_interval,
3105                 .maxlen         = sizeof(int),
3106                 .mode           = 0644,
3107                 .proc_handler   = proc_dointvec_jiffies,
3108         },
3109         {
3110                 .procname       = "gc_min_interval_ms",
3111                 .data           = &ip_rt_gc_min_interval,
3112                 .maxlen         = sizeof(int),
3113                 .mode           = 0644,
3114                 .proc_handler   = proc_dointvec_ms_jiffies,
3115         },
3116         {
3117                 .procname       = "gc_timeout",
3118                 .data           = &ip_rt_gc_timeout,
3119                 .maxlen         = sizeof(int),
3120                 .mode           = 0644,
3121                 .proc_handler   = proc_dointvec_jiffies,
3122         },
3123         {
3124                 .procname       = "gc_interval",
3125                 .data           = &ip_rt_gc_interval,
3126                 .maxlen         = sizeof(int),
3127                 .mode           = 0644,
3128                 .proc_handler   = proc_dointvec_jiffies,
3129         },
3130         {
3131                 .procname       = "redirect_load",
3132                 .data           = &ip_rt_redirect_load,
3133                 .maxlen         = sizeof(int),
3134                 .mode           = 0644,
3135                 .proc_handler   = proc_dointvec,
3136         },
3137         {
3138                 .procname       = "redirect_number",
3139                 .data           = &ip_rt_redirect_number,
3140                 .maxlen         = sizeof(int),
3141                 .mode           = 0644,
3142                 .proc_handler   = proc_dointvec,
3143         },
3144         {
3145                 .procname       = "redirect_silence",
3146                 .data           = &ip_rt_redirect_silence,
3147                 .maxlen         = sizeof(int),
3148                 .mode           = 0644,
3149                 .proc_handler   = proc_dointvec,
3150         },
3151         {
3152                 .procname       = "error_cost",
3153                 .data           = &ip_rt_error_cost,
3154                 .maxlen         = sizeof(int),
3155                 .mode           = 0644,
3156                 .proc_handler   = proc_dointvec,
3157         },
3158         {
3159                 .procname       = "error_burst",
3160                 .data           = &ip_rt_error_burst,
3161                 .maxlen         = sizeof(int),
3162                 .mode           = 0644,
3163                 .proc_handler   = proc_dointvec,
3164         },
3165         {
3166                 .procname       = "gc_elasticity",
3167                 .data           = &ip_rt_gc_elasticity,
3168                 .maxlen         = sizeof(int),
3169                 .mode           = 0644,
3170                 .proc_handler   = proc_dointvec,
3171         },
3172         {
3173                 .procname       = "mtu_expires",
3174                 .data           = &ip_rt_mtu_expires,
3175                 .maxlen         = sizeof(int),
3176                 .mode           = 0644,
3177                 .proc_handler   = proc_dointvec_jiffies,
3178         },
3179         {
3180                 .procname       = "min_pmtu",
3181                 .data           = &ip_rt_min_pmtu,
3182                 .maxlen         = sizeof(int),
3183                 .mode           = 0644,
3184                 .proc_handler   = proc_dointvec_minmax,
3185                 .extra1         = &ip_min_valid_pmtu,
3186         },
3187         {
3188                 .procname       = "min_adv_mss",
3189                 .data           = &ip_rt_min_advmss,
3190                 .maxlen         = sizeof(int),
3191                 .mode           = 0644,
3192                 .proc_handler   = proc_dointvec,
3193         },
3194         { }
3195 };
3196
3197 static struct ctl_table ipv4_route_flush_table[] = {
3198         {
3199                 .procname       = "flush",
3200                 .maxlen         = sizeof(int),
3201                 .mode           = 0200,
3202                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3203         },
3204         { },
3205 };
3206
3207 static __net_init int sysctl_route_net_init(struct net *net)
3208 {
3209         struct ctl_table *tbl;
3210
3211         tbl = ipv4_route_flush_table;
3212         if (!net_eq(net, &init_net)) {
3213                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3214                 if (!tbl)
3215                         goto err_dup;
3216
3217                 /* Don't export sysctls to unprivileged users */
3218                 if (net->user_ns != &init_user_ns)
3219                         tbl[0].procname = NULL;
3220         }
3221         tbl[0].extra1 = net;
3222
3223         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3224         if (!net->ipv4.route_hdr)
3225                 goto err_reg;
3226         return 0;
3227
3228 err_reg:
3229         if (tbl != ipv4_route_flush_table)
3230                 kfree(tbl);
3231 err_dup:
3232         return -ENOMEM;
3233 }
3234
3235 static __net_exit void sysctl_route_net_exit(struct net *net)
3236 {
3237         struct ctl_table *tbl;
3238
3239         tbl = net->ipv4.route_hdr->ctl_table_arg;
3240         unregister_net_sysctl_table(net->ipv4.route_hdr);
3241         BUG_ON(tbl == ipv4_route_flush_table);
3242         kfree(tbl);
3243 }
3244
3245 static __net_initdata struct pernet_operations sysctl_route_ops = {
3246         .init = sysctl_route_net_init,
3247         .exit = sysctl_route_net_exit,
3248 };
3249 #endif
3250
3251 static __net_init int rt_genid_init(struct net *net)
3252 {
3253         atomic_set(&net->ipv4.rt_genid, 0);
3254         atomic_set(&net->fnhe_genid, 0);
3255         atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3256         return 0;
3257 }
3258
3259 static __net_initdata struct pernet_operations rt_genid_ops = {
3260         .init = rt_genid_init,
3261 };
3262
3263 static int __net_init ipv4_inetpeer_init(struct net *net)
3264 {
3265         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3266
3267         if (!bp)
3268                 return -ENOMEM;
3269         inet_peer_base_init(bp);
3270         net->ipv4.peers = bp;
3271         return 0;
3272 }
3273
3274 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3275 {
3276         struct inet_peer_base *bp = net->ipv4.peers;
3277
3278         net->ipv4.peers = NULL;
3279         inetpeer_invalidate_tree(bp);
3280         kfree(bp);
3281 }
3282
3283 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3284         .init   =       ipv4_inetpeer_init,
3285         .exit   =       ipv4_inetpeer_exit,
3286 };
3287
3288 #ifdef CONFIG_IP_ROUTE_CLASSID
3289 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3290 #endif /* CONFIG_IP_ROUTE_CLASSID */
3291
3292 int __init ip_rt_init(void)
3293 {
3294         int cpu;
3295
3296         ip_idents = kmalloc_array(IP_IDENTS_SZ, sizeof(*ip_idents),
3297                                   GFP_KERNEL);
3298         if (!ip_idents)
3299                 panic("IP: failed to allocate ip_idents\n");
3300
3301         prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3302
3303         ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3304         if (!ip_tstamps)
3305                 panic("IP: failed to allocate ip_tstamps\n");
3306
3307         for_each_possible_cpu(cpu) {
3308                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3309
3310                 INIT_LIST_HEAD(&ul->head);
3311                 spin_lock_init(&ul->lock);
3312         }
3313 #ifdef CONFIG_IP_ROUTE_CLASSID
3314         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3315         if (!ip_rt_acct)
3316                 panic("IP: failed to allocate ip_rt_acct\n");
3317 #endif
3318
3319         ipv4_dst_ops.kmem_cachep =
3320                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3321                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3322
3323         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3324
3325         if (dst_entries_init(&ipv4_dst_ops) < 0)
3326                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3327
3328         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3329                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3330
3331         ipv4_dst_ops.gc_thresh = ~0;
3332         ip_rt_max_size = INT_MAX;
3333
3334         devinet_init();
3335         ip_fib_init();
3336
3337         if (ip_rt_proc_init())
3338                 pr_err("Unable to create route proc files\n");
3339 #ifdef CONFIG_XFRM
3340         xfrm_init();
3341         xfrm4_init();
3342 #endif
3343         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3344                       RTNL_FLAG_DOIT_UNLOCKED);
3345
3346 #ifdef CONFIG_SYSCTL
3347         register_pernet_subsys(&sysctl_route_ops);
3348 #endif
3349         register_pernet_subsys(&rt_genid_ops);
3350         register_pernet_subsys(&ipv4_inetpeer_ops);
3351         return 0;
3352 }
3353
3354 #ifdef CONFIG_SYSCTL
3355 /*
3356  * We really need to sanitize the damn ipv4 init order, then all
3357  * this nonsense will go away.
3358  */
3359 void __init ip_static_sysctl_init(void)
3360 {
3361         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3362 }
3363 #endif