net/ipv4/route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              ROUTE - implementation of the IP router.
   8  *
   9  * Authors:     Ross Biro
  10  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  12  *              Linus Torvalds, <Linus.Torvalds@helsinki.fi>
  13  *              Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  14  *
  15  * Fixes:
  16  *              Alan Cox        :       Verify area fixes.
  17  *              Alan Cox        :       cli() protects routing changes
  18  *              Rui Oliveira    :       ICMP routing table updates
  19  *              (rco@di.uminho.pt)      Routing table insertion and update
  20  *              Linus Torvalds  :       Rewrote bits to be sensible
  21  *              Alan Cox        :       Added BSD route gw semantics
  22  *              Alan Cox        :       Super /proc >4K
  23  *              Alan Cox        :       MTU in route table
  24  *              Alan Cox        :       MSS actually. Also added the window
  25  *                                      clamper.
  26  *              Sam Lantinga    :       Fixed route matching in rt_del()
  27  *              Alan Cox        :       Routing cache support.
  28  *              Alan Cox        :       Removed compatibility cruft.
  29  *              Alan Cox        :       RTF_REJECT support.
  30  *              Alan Cox        :       TCP irtt support.
  31  *              Jonathan Naylor :       Added Metric support.
  32  *      Miquel van Smoorenburg  :       BSD API fixes.
  33  *      Miquel van Smoorenburg  :       Metrics.
  34  *              Alan Cox        :       Use __u32 properly
  35  *              Alan Cox        :       Aligned routing errors more closely with BSD
  36  *                                      our system is still very different.
  37  *              Alan Cox        :       Faster /proc handling
  38  *      Alexey Kuznetsov        :       Massive rework to support tree based routing,
  39  *                                      routing caches and better behaviour.
  40  *
  41  *              Olaf Erb        :       irtt wasn't being copied right.
  42  *              Bjorn Ekwall    :       Kerneld route support.
  43  *              Alan Cox        :       Multicast fixed (I hope)
  44  *              Pavel Krauz     :       Limited broadcast fixed
  45  *              Mike McLagan    :       Routing by source
  46  *      Alexey Kuznetsov        :       End of old history. Split to fib.c and
  47  *                                      route.c and rewritten from scratch.
  48  *              Andi Kleen      :       Load-limit warning messages.
  49  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  50  *      Vitaly E. Lavrov        :       Race condition in ip_route_input_slow.
  51  *      Tobias Ringstrom        :       Uninitialized res.type in ip_route_output_slow.
  52  *      Vladimir V. Ivanov      :       IP rule info (flowid) is really useful.
  53  *              Marc Boucher    :       routing by fwmark
  54  *      Robert Olsson           :       Added rt_cache statistics
  55  *      Arnaldo C. Melo         :       Convert proc stuff to seq_file
  56  *      Eric Dumazet            :       hashed spinlocks and rt_check_expire() fixes.
  57  *      Ilia Sotnikov           :       Ignore TOS on PMTUD and Redirect
  58  *      Ilia Sotnikov           :       Removed TOS from hash calculations
  59  */
  60
  61 #define pr_fmt(fmt) "IPv4: " fmt
  62
  63 #include <linux/module.h>
  64 #include <linux/uaccess.h>
  65 #include <linux/bitops.h>
  66 #include <linux/types.h>
  67 #include <linux/kernel.h>
  68 #include <linux/mm.h>
  69 #include <linux/string.h>
  70 #include <linux/socket.h>
  71 #include <linux/sockios.h>
  72 #include <linux/errno.h>
  73 #include <linux/in.h>
  74 #include <linux/inet.h>
  75 #include <linux/netdevice.h>
  76 #include <linux/proc_fs.h>
  77 #include <linux/init.h>
  78 #include <linux/skbuff.h>
  79 #include <linux/inetdevice.h>
  80 #include <linux/igmp.h>
  81 #include <linux/pkt_sched.h>
  82 #include <linux/mroute.h>
  83 #include <linux/netfilter_ipv4.h>
  84 #include <linux/random.h>
  85 #include <linux/rcupdate.h>
  86 #include <linux/times.h>
  87 #include <linux/slab.h>
  88 #include <linux/jhash.h>
  89 #include <net/dst.h>
  90 #include <net/dst_metadata.h>
  91 #include <net/net_namespace.h>
  92 #include <net/protocol.h>
  93 #include <net/ip.h>
  94 #include <net/route.h>
  95 #include <net/inetpeer.h>
  96 #include <net/sock.h>
  97 #include <net/ip_fib.h>
  98 #include <net/nexthop.h>
  99 #include <net/arp.h>
 100 #include <net/tcp.h>
 101 #include <net/icmp.h>
 102 #include <net/xfrm.h>
 103 #include <net/lwtunnel.h>
 104 #include <net/netevent.h>
 105 #include <net/rtnetlink.h>
 106 #ifdef CONFIG_SYSCTL
 107 #include <linux/sysctl.h>
 108 #endif
 109 #include <net/secure_seq.h>
 110 #include <net/ip_tunnels.h>
 111 #include <net/l3mdev.h>
 112
 113 #include "fib_lookup.h"
 114
 115 #define RT_FL_TOS(oldflp4) \
 116         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
 117
 118 #define RT_GC_TIMEOUT (300*HZ)
 119
 120 static int ip_rt_max_size;
 121 static int ip_rt_redirect_number __read_mostly  = 9;
 122 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 123 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 124 static int ip_rt_error_cost __read_mostly       = HZ;
 125 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 126 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
 127 static u32 ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
 128 static int ip_rt_min_advmss __read_mostly       = 256;
 129
 130 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
 131
 132 /*
 133  *      Interface to generic destination cache.
 134  */
 135
 136 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 137 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 138 static unsigned int      ipv4_mtu(const struct dst_entry *dst);
 139 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 140 static void              ipv4_link_failure(struct sk_buff *skb);
 141 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 142                                            struct sk_buff *skb, u32 mtu);
 143 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
 144                                         struct sk_buff *skb);
 145 static void             ipv4_dst_destroy(struct dst_entry *dst);
 146
 147 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 148 {
 149         WARN_ON(1);
 150         return NULL;
 151 }
 152
 153 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 154                                            struct sk_buff *skb,
 155                                            const void *daddr);
 156 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
 157
 158 static struct dst_ops ipv4_dst_ops = {
 159         .family =               AF_INET,
 160         .check =                ipv4_dst_check,
 161         .default_advmss =       ipv4_default_advmss,
 162         .mtu =                  ipv4_mtu,
 163         .cow_metrics =          ipv4_cow_metrics,
 164         .destroy =              ipv4_dst_destroy,
 165         .negative_advice =      ipv4_negative_advice,
 166         .link_failure =         ipv4_link_failure,
 167         .update_pmtu =          ip_rt_update_pmtu,
 168         .redirect =             ip_do_redirect,
 169         .local_out =            __ip_local_out,
 170         .neigh_lookup =         ipv4_neigh_lookup,
 171         .confirm_neigh =        ipv4_confirm_neigh,
 172 };
 173
 174 #define ECN_OR_COST(class)      TC_PRIO_##class
 175
 176 const __u8 ip_tos2prio[16] = {
 177         TC_PRIO_BESTEFFORT,
 178         ECN_OR_COST(BESTEFFORT),
 179         TC_PRIO_BESTEFFORT,
 180         ECN_OR_COST(BESTEFFORT),
 181         TC_PRIO_BULK,
 182         ECN_OR_COST(BULK),
 183         TC_PRIO_BULK,
 184         ECN_OR_COST(BULK),
 185         TC_PRIO_INTERACTIVE,
 186         ECN_OR_COST(INTERACTIVE),
 187         TC_PRIO_INTERACTIVE,
 188         ECN_OR_COST(INTERACTIVE),
 189         TC_PRIO_INTERACTIVE_BULK,
 190         ECN_OR_COST(INTERACTIVE_BULK),
 191         TC_PRIO_INTERACTIVE_BULK,
 192         ECN_OR_COST(INTERACTIVE_BULK)
 193 };
 194 EXPORT_SYMBOL(ip_tos2prio);
 195
 196 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 197 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
 198
 199 #ifdef CONFIG_PROC_FS
 200 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 201 {
 202         if (*pos)
 203                 return NULL;
 204         return SEQ_START_TOKEN;
 205 }
 206
 207 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 208 {
 209         ++*pos;
 210         return NULL;
 211 }
 212
 213 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 214 {
 215 }
 216
 217 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 218 {
 219         if (v == SEQ_START_TOKEN)
 220                 seq_printf(seq, "%-127s\n",
 221                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 222                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 223                            "HHUptod\tSpecDst");
 224         return 0;
 225 }
 226
 227 static const struct seq_operations rt_cache_seq_ops = {
 228         .start  = rt_cache_seq_start,
 229         .next   = rt_cache_seq_next,
 230         .stop   = rt_cache_seq_stop,
 231         .show   = rt_cache_seq_show,
 232 };
 233
 234 static int rt_cache_seq_open(struct inode *inode, struct file *file)
 235 {
 236         return seq_open(file, &rt_cache_seq_ops);
 237 }
 238
 239 static const struct file_operations rt_cache_seq_fops = {
 240         .open    = rt_cache_seq_open,
 241         .read    = seq_read,
 242         .llseek  = seq_lseek,
 243         .release = seq_release,
 244 };
 245
 246
 247 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 248 {
 249         int cpu;
 250
 251         if (*pos == 0)
 252                 return SEQ_START_TOKEN;
 253
 254         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 255                 if (!cpu_possible(cpu))
 256                         continue;
 257                 *pos = cpu+1;
 258                 return &per_cpu(rt_cache_stat, cpu);
 259         }
 260         return NULL;
 261 }
 262
 263 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 264 {
 265         int cpu;
 266
 267         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 268                 if (!cpu_possible(cpu))
 269                         continue;
 270                 *pos = cpu+1;
 271                 return &per_cpu(rt_cache_stat, cpu);
 272         }
 273         return NULL;
 274
 275 }
 276
 277 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 278 {
 279
 280 }
 281
 282 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 283 {
 284         struct rt_cache_stat *st = v;
 285
 286         if (v == SEQ_START_TOKEN) {
 287                 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");
 288                 return 0;
 289         }
 290
 291         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 292                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 293                    dst_entries_get_slow(&ipv4_dst_ops),
 294                    0, /* st->in_hit */
 295                    st->in_slow_tot,
 296                    st->in_slow_mc,
 297                    st->in_no_route,
 298                    st->in_brd,
 299                    st->in_martian_dst,
 300                    st->in_martian_src,
 301
 302                    0, /* st->out_hit */
 303                    st->out_slow_tot,
 304                    st->out_slow_mc,
 305
 306                    0, /* st->gc_total */
 307                    0, /* st->gc_ignored */
 308                    0, /* st->gc_goal_miss */
 309                    0, /* st->gc_dst_overflow */
 310                    0, /* st->in_hlist_search */
 311                    0  /* st->out_hlist_search */
 312                 );
 313         return 0;
 314 }
 315
 316 static const struct seq_operations rt_cpu_seq_ops = {
 317         .start  = rt_cpu_seq_start,
 318         .next   = rt_cpu_seq_next,
 319         .stop   = rt_cpu_seq_stop,
 320         .show   = rt_cpu_seq_show,
 321 };
 322
 323
 324 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
 325 {
 326         return seq_open(file, &rt_cpu_seq_ops);
 327 }
 328
 329 static const struct file_operations rt_cpu_seq_fops = {
 330         .open    = rt_cpu_seq_open,
 331         .read    = seq_read,
 332         .llseek  = seq_lseek,
 333         .release = seq_release,
 334 };
 335
 336 #ifdef CONFIG_IP_ROUTE_CLASSID
 337 static int rt_acct_proc_show(struct seq_file *m, void *v)
 338 {
 339         struct ip_rt_acct *dst, *src;
 340         unsigned int i, j;
 341
 342         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 343         if (!dst)
 344                 return -ENOMEM;
 345
 346         for_each_possible_cpu(i) {
 347                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 348                 for (j = 0; j < 256; j++) {
 349                         dst[j].o_bytes   += src[j].o_bytes;
 350                         dst[j].o_packets += src[j].o_packets;
 351                         dst[j].i_bytes   += src[j].i_bytes;
 352                         dst[j].i_packets += src[j].i_packets;
 353                 }
 354         }
 355
 356         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 357         kfree(dst);
 358         return 0;
 359 }
 360 #endif
 361
 362 static int __net_init ip_rt_do_proc_init(struct net *net)
 363 {
 364         struct proc_dir_entry *pde;
 365
 366         pde = proc_create("rt_cache", 0444, net->proc_net,
 367                           &rt_cache_seq_fops);
 368         if (!pde)
 369                 goto err1;
 370
 371         pde = proc_create("rt_cache", 0444,
 372                           net->proc_net_stat, &rt_cpu_seq_fops);
 373         if (!pde)
 374                 goto err2;
 375
 376 #ifdef CONFIG_IP_ROUTE_CLASSID
 377         pde = proc_create_single("rt_acct", 0, net->proc_net,
 378                         rt_acct_proc_show);
 379         if (!pde)
 380                 goto err3;
 381 #endif
 382         return 0;
 383
 384 #ifdef CONFIG_IP_ROUTE_CLASSID
 385 err3:
 386         remove_proc_entry("rt_cache", net->proc_net_stat);
 387 #endif
 388 err2:
 389         remove_proc_entry("rt_cache", net->proc_net);
 390 err1:
 391         return -ENOMEM;
 392 }
 393
 394 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 395 {
 396         remove_proc_entry("rt_cache", net->proc_net_stat);
 397         remove_proc_entry("rt_cache", net->proc_net);
 398 #ifdef CONFIG_IP_ROUTE_CLASSID
 399         remove_proc_entry("rt_acct", net->proc_net);
 400 #endif
 401 }
 402
 403 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 404         .init = ip_rt_do_proc_init,
 405         .exit = ip_rt_do_proc_exit,
 406 };
 407
 408 static int __init ip_rt_proc_init(void)
 409 {
 410         return register_pernet_subsys(&ip_rt_proc_ops);
 411 }
 412
 413 #else
 414 static inline int ip_rt_proc_init(void)
 415 {
 416         return 0;
 417 }
 418 #endif /* CONFIG_PROC_FS */
 419
 420 static inline bool rt_is_expired(const struct rtable *rth)
 421 {
 422         return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
 423 }
 424
 425 void rt_cache_flush(struct net *net)
 426 {
 427         rt_genid_bump_ipv4(net);
 428 }
 429
 430 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
 431                                            struct sk_buff *skb,
 432                                            const void *daddr)
 433 {
 434         const struct rtable *rt = container_of(dst, struct rtable, dst);
 435         struct net_device *dev = dst->dev;
 436         struct neighbour *n;
 437
 438         rcu_read_lock_bh();
 439
 440         if (likely(rt->rt_gw_family == AF_INET)) {
 441                 n = ip_neigh_gw4(dev, rt->rt_gw4);
 442         } else if (rt->rt_gw_family == AF_INET6) {
 443                 n = ip_neigh_gw6(dev, &rt->rt_gw6);
 444         } else {
 445                 __be32 pkey;
 446
 447                 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr);
 448                 n = ip_neigh_gw4(dev, pkey);
 449         }
 450
 451         if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt))
 452                 n = NULL;
 453
 454         rcu_read_unlock_bh();
 455
 456         return n;
 457 }
 458
 459 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 460 {
 461         const struct rtable *rt = container_of(dst, struct rtable, dst);
 462         struct net_device *dev = dst->dev;
 463         const __be32 *pkey = daddr;
 464
 465         if (rt->rt_gw_family == AF_INET) {
 466                 pkey = (const __be32 *)&rt->rt_gw4;
 467         } else if (rt->rt_gw_family == AF_INET6) {
 468                 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6);
 469         } else if (!daddr ||
 470                  (rt->rt_flags &
 471                   (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) {
 472                 return;
 473         }
 474         __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
 475 }
 476
 477 #define IP_IDENTS_SZ 2048u
 478
 479 static atomic_t *ip_idents __read_mostly;
 480 static u32 *ip_tstamps __read_mostly;
 481
 482 /* In order to protect privacy, we add a perturbation to identifiers
 483  * if one generator is seldom used. This makes hard for an attacker
 484  * to infer how many packets were sent between two points in time.
 485  */
 486 u32 ip_idents_reserve(u32 hash, int segs)
 487 {
 488         u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
 489         atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
 490         u32 old = READ_ONCE(*p_tstamp);
 491         u32 now = (u32)jiffies;
 492         u32 new, delta = 0;
 493
 494         if (old != now && cmpxchg(p_tstamp, old, now) == old)
 495                 delta = prandom_u32_max(now - old);
 496
 497         /* Do not use atomic_add_return() as it makes UBSAN unhappy */
 498         do {
 499                 old = (u32)atomic_read(p_id);
 500                 new = old + delta + segs;
 501         } while (atomic_cmpxchg(p_id, old, new) != old);
 502
 503         return new - segs;
 504 }
 505 EXPORT_SYMBOL(ip_idents_reserve);
 506
 507 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
 508 {
 509         u32 hash, id;
 510
 511         /* Note the following code is not safe, but this is okay. */
 512         if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key)))
 513                 get_random_bytes(&net->ipv4.ip_id_key,
 514                                  sizeof(net->ipv4.ip_id_key));
 515
 516         hash = siphash_3u32((__force u32)iph->daddr,
 517                             (__force u32)iph->saddr,
 518                             iph->protocol,
 519                             &net->ipv4.ip_id_key);
 520         id = ip_idents_reserve(hash, segs);
 521         iph->id = htons(id);
 522 }
 523 EXPORT_SYMBOL(__ip_select_ident);
 524
 525 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
 526                              const struct sock *sk,
 527                              const struct iphdr *iph,
 528                              int oif, u8 tos,
 529                              u8 prot, u32 mark, int flow_flags)
 530 {
 531         if (sk) {
 532                 const struct inet_sock *inet = inet_sk(sk);
 533
 534                 oif = sk->sk_bound_dev_if;
 535                 mark = sk->sk_mark;
 536                 tos = RT_CONN_FLAGS(sk);
 537                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
 538         }
 539         flowi4_init_output(fl4, oif, mark, tos,
 540                            RT_SCOPE_UNIVERSE, prot,
 541                            flow_flags,
 542                            iph->daddr, iph->saddr, 0, 0,
 543                            sock_net_uid(net, sk));
 544 }
 545
 546 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
 547                                const struct sock *sk)
 548 {
 549         const struct net *net = dev_net(skb->dev);
 550         const struct iphdr *iph = ip_hdr(skb);
 551         int oif = skb->dev->ifindex;
 552         u8 tos = RT_TOS(iph->tos);
 553         u8 prot = iph->protocol;
 554         u32 mark = skb->mark;
 555
 556         __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
 557 }
 558
 559 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
 560 {
 561         const struct inet_sock *inet = inet_sk(sk);
 562         const struct ip_options_rcu *inet_opt;
 563         __be32 daddr = inet->inet_daddr;
 564
 565         rcu_read_lock();
 566         inet_opt = rcu_dereference(inet->inet_opt);
 567         if (inet_opt && inet_opt->opt.srr)
 568                 daddr = inet_opt->opt.faddr;
 569         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 570                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 571                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 572                            inet_sk_flowi_flags(sk),
 573                            daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
 574         rcu_read_unlock();
 575 }
 576
 577 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
 578                                  const struct sk_buff *skb)
 579 {
 580         if (skb)
 581                 build_skb_flow_key(fl4, skb, sk);
 582         else
 583                 build_sk_flow_key(fl4, sk);
 584 }
 585
 586 static DEFINE_SPINLOCK(fnhe_lock);
 587
 588 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
 589 {
 590         struct rtable *rt;
 591
 592         rt = rcu_dereference(fnhe->fnhe_rth_input);
 593         if (rt) {
 594                 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
 595                 dst_dev_put(&rt->dst);
 596                 dst_release(&rt->dst);
 597         }
 598         rt = rcu_dereference(fnhe->fnhe_rth_output);
 599         if (rt) {
 600                 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
 601                 dst_dev_put(&rt->dst);
 602                 dst_release(&rt->dst);
 603         }
 604 }
 605
 606 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
 607 {
 608         struct fib_nh_exception *fnhe, *oldest;
 609
 610         oldest = rcu_dereference(hash->chain);
 611         for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
 612              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 613                 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
 614                         oldest = fnhe;
 615         }
 616         fnhe_flush_routes(oldest);
 617         return oldest;
 618 }
 619
 620 static inline u32 fnhe_hashfun(__be32 daddr)
 621 {
 622         static u32 fnhe_hashrnd __read_mostly;
 623         u32 hval;
 624
 625         net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
 626         hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
 627         return hash_32(hval, FNHE_HASH_SHIFT);
 628 }
 629
 630 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
 631 {
 632         rt->rt_pmtu = fnhe->fnhe_pmtu;
 633         rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
 634         rt->dst.expires = fnhe->fnhe_expires;
 635
 636         if (fnhe->fnhe_gw) {
 637                 rt->rt_flags |= RTCF_REDIRECTED;
 638                 rt->rt_gw_family = AF_INET;
 639                 rt->rt_gw4 = fnhe->fnhe_gw;
 640         }
 641 }
 642
 643 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr,
 644                                   __be32 gw, u32 pmtu, bool lock,
 645                                   unsigned long expires)
 646 {
 647         struct fnhe_hash_bucket *hash;
 648         struct fib_nh_exception *fnhe;
 649         struct rtable *rt;
 650         u32 genid, hval;
 651         unsigned int i;
 652         int depth;
 653
 654         genid = fnhe_genid(dev_net(nhc->nhc_dev));
 655         hval = fnhe_hashfun(daddr);
 656
 657         spin_lock_bh(&fnhe_lock);
 658
 659         hash = rcu_dereference(nhc->nhc_exceptions);
 660         if (!hash) {
 661                 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC);
 662                 if (!hash)
 663                         goto out_unlock;
 664                 rcu_assign_pointer(nhc->nhc_exceptions, hash);
 665         }
 666
 667         hash += hval;
 668
 669         depth = 0;
 670         for (fnhe = rcu_dereference(hash->chain); fnhe;
 671              fnhe = rcu_dereference(fnhe->fnhe_next)) {
 672                 if (fnhe->fnhe_daddr == daddr)
 673                         break;
 674                 depth++;
 675         }
 676
 677         if (fnhe) {
 678                 if (fnhe->fnhe_genid != genid)
 679                         fnhe->fnhe_genid = genid;
 680                 if (gw)
 681                         fnhe->fnhe_gw = gw;
 682                 if (pmtu) {
 683                         fnhe->fnhe_pmtu = pmtu;
 684                         fnhe->fnhe_mtu_locked = lock;
 685                 }
 686                 fnhe->fnhe_expires = max(1UL, expires);
 687                 /* Update all cached dsts too */
 688                 rt = rcu_dereference(fnhe->fnhe_rth_input);
 689                 if (rt)
 690                         fill_route_from_fnhe(rt, fnhe);
 691                 rt = rcu_dereference(fnhe->fnhe_rth_output);
 692                 if (rt)
 693                         fill_route_from_fnhe(rt, fnhe);
 694         } else {
 695                 if (depth > FNHE_RECLAIM_DEPTH)
 696                         fnhe = fnhe_oldest(hash);
 697                 else {
 698                         fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
 699                         if (!fnhe)
 700                                 goto out_unlock;
 701
 702                         fnhe->fnhe_next = hash->chain;
 703                         rcu_assign_pointer(hash->chain, fnhe);
 704                 }
 705                 fnhe->fnhe_genid = genid;
 706                 fnhe->fnhe_daddr = daddr;
 707                 fnhe->fnhe_gw = gw;
 708                 fnhe->fnhe_pmtu = pmtu;
 709                 fnhe->fnhe_mtu_locked = lock;
 710                 fnhe->fnhe_expires = max(1UL, expires);
 711
 712                 /* Exception created; mark the cached routes for the nexthop
 713                  * stale, so anyone caching it rechecks if this exception
 714                  * applies to them.
 715                  */
 716                 rt = rcu_dereference(nhc->nhc_rth_input);
 717                 if (rt)
 718                         rt->dst.obsolete = DST_OBSOLETE_KILL;
 719
 720                 for_each_possible_cpu(i) {
 721                         struct rtable __rcu **prt;
 722                         prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i);
 723                         rt = rcu_dereference(*prt);
 724                         if (rt)
 725                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 726                 }
 727         }
 728
 729         fnhe->fnhe_stamp = jiffies;
 730
 731 out_unlock:
 732         spin_unlock_bh(&fnhe_lock);
 733 }
 734
 735 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
 736                              bool kill_route)
 737 {
 738         __be32 new_gw = icmp_hdr(skb)->un.gateway;
 739         __be32 old_gw = ip_hdr(skb)->saddr;
 740         struct net_device *dev = skb->dev;
 741         struct in_device *in_dev;
 742         struct fib_result res;
 743         struct neighbour *n;
 744         struct net *net;
 745
 746         switch (icmp_hdr(skb)->code & 7) {
 747         case ICMP_REDIR_NET:
 748         case ICMP_REDIR_NETTOS:
 749         case ICMP_REDIR_HOST:
 750         case ICMP_REDIR_HOSTTOS:
 751                 break;
 752
 753         default:
 754                 return;
 755         }
 756
 757         if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw)
 758                 return;
 759
 760         in_dev = __in_dev_get_rcu(dev);
 761         if (!in_dev)
 762                 return;
 763
 764         net = dev_net(dev);
 765         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
 766             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
 767             ipv4_is_zeronet(new_gw))
 768                 goto reject_redirect;
 769
 770         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
 771                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
 772                         goto reject_redirect;
 773                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
 774                         goto reject_redirect;
 775         } else {
 776                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
 777                         goto reject_redirect;
 778         }
 779
 780         n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
 781         if (!n)
 782                 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
 783         if (!IS_ERR(n)) {
 784                 if (!(n->nud_state & NUD_VALID)) {
 785                         neigh_event_send(n, NULL);
 786                 } else {
 787                         if (fib_lookup(net, fl4, &res, 0) == 0) {
 788                                 struct fib_nh_common *nhc = FIB_RES_NHC(res);
 789
 790                                 update_or_create_fnhe(nhc, fl4->daddr, new_gw,
 791                                                 0, false,
 792                                                 jiffies + ip_rt_gc_timeout);
 793                         }
 794                         if (kill_route)
 795                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
 796                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
 797                 }
 798                 neigh_release(n);
 799         }
 800         return;
 801
 802 reject_redirect:
 803 #ifdef CONFIG_IP_ROUTE_VERBOSE
 804         if (IN_DEV_LOG_MARTIANS(in_dev)) {
 805                 const struct iphdr *iph = (const struct iphdr *) skb->data;
 806                 __be32 daddr = iph->daddr;
 807                 __be32 saddr = iph->saddr;
 808
 809                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
 810                                      "  Advised path = %pI4 -> %pI4\n",
 811                                      &old_gw, dev->name, &new_gw,
 812                                      &saddr, &daddr);
 813         }
 814 #endif
 815         ;
 816 }
 817
 818 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
 819 {
 820         struct rtable *rt;
 821         struct flowi4 fl4;
 822         const struct iphdr *iph = (const struct iphdr *) skb->data;
 823         struct net *net = dev_net(skb->dev);
 824         int oif = skb->dev->ifindex;
 825         u8 tos = RT_TOS(iph->tos);
 826         u8 prot = iph->protocol;
 827         u32 mark = skb->mark;
 828
 829         rt = (struct rtable *) dst;
 830
 831         __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
 832         __ip_do_redirect(rt, skb, &fl4, true);
 833 }
 834
 835 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 836 {
 837         struct rtable *rt = (struct rtable *)dst;
 838         struct dst_entry *ret = dst;
 839
 840         if (rt) {
 841                 if (dst->obsolete > 0) {
 842                         ip_rt_put(rt);
 843                         ret = NULL;
 844                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
 845                            rt->dst.expires) {
 846                         ip_rt_put(rt);
 847                         ret = NULL;
 848                 }
 849         }
 850         return ret;
 851 }
 852
 853 /*
 854  * Algorithm:
 855  *      1. The first ip_rt_redirect_number redirects are sent
 856  *         with exponential backoff, then we stop sending them at all,
 857  *         assuming that the host ignores our redirects.
 858  *      2. If we did not see packets requiring redirects
 859  *         during ip_rt_redirect_silence, we assume that the host
 860  *         forgot redirected route and start to send redirects again.
 861  *
 862  * This algorithm is much cheaper and more intelligent than dumb load limiting
 863  * in icmp.c.
 864  *
 865  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
 866  * and "frag. need" (breaks PMTU discovery) in icmp.c.
 867  */
 868
 869 void ip_rt_send_redirect(struct sk_buff *skb)
 870 {
 871         struct rtable *rt = skb_rtable(skb);
 872         struct in_device *in_dev;
 873         struct inet_peer *peer;
 874         struct net *net;
 875         int log_martians;
 876         int vif;
 877
 878         rcu_read_lock();
 879         in_dev = __in_dev_get_rcu(rt->dst.dev);
 880         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
 881                 rcu_read_unlock();
 882                 return;
 883         }
 884         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
 885         vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
 886         rcu_read_unlock();
 887
 888         net = dev_net(rt->dst.dev);
 889         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
 890         if (!peer) {
 891                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
 892                           rt_nexthop(rt, ip_hdr(skb)->daddr));
 893                 return;
 894         }
 895
 896         /* No redirected packets during ip_rt_redirect_silence;
 897          * reset the algorithm.
 898          */
 899         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
 900                 peer->rate_tokens = 0;
 901                 peer->n_redirects = 0;
 902         }
 903
 904         /* Too many ignored redirects; do not send anything
 905          * set dst.rate_last to the last seen redirected packet.
 906          */
 907         if (peer->n_redirects >= ip_rt_redirect_number) {
 908                 peer->rate_last = jiffies;
 909                 goto out_put_peer;
 910         }
 911
 912         /* Check for load limit; set rate_last to the latest sent
 913          * redirect.
 914          */
 915         if (peer->rate_tokens == 0 ||
 916             time_after(jiffies,
 917                        (peer->rate_last +
 918                         (ip_rt_redirect_load << peer->rate_tokens)))) {
 919                 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
 920
 921                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
 922                 peer->rate_last = jiffies;
 923                 ++peer->rate_tokens;
 924                 ++peer->n_redirects;
 925 #ifdef CONFIG_IP_ROUTE_VERBOSE
 926                 if (log_martians &&
 927                     peer->rate_tokens == ip_rt_redirect_number)
 928                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
 929                                              &ip_hdr(skb)->saddr, inet_iif(skb),
 930                                              &ip_hdr(skb)->daddr, &gw);
 931 #endif
 932         }
 933 out_put_peer:
 934         inet_putpeer(peer);
 935 }
 936
 937 static int ip_error(struct sk_buff *skb)
 938 {
 939         struct rtable *rt = skb_rtable(skb);
 940         struct net_device *dev = skb->dev;
 941         struct in_device *in_dev;
 942         struct inet_peer *peer;
 943         unsigned long now;
 944         struct net *net;
 945         bool send;
 946         int code;
 947
 948         if (netif_is_l3_master(skb->dev)) {
 949                 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif);
 950                 if (!dev)
 951                         goto out;
 952         }
 953
 954         in_dev = __in_dev_get_rcu(dev);
 955
 956         /* IP on this device is disabled. */
 957         if (!in_dev)
 958                 goto out;
 959
 960         net = dev_net(rt->dst.dev);
 961         if (!IN_DEV_FORWARD(in_dev)) {
 962                 switch (rt->dst.error) {
 963                 case EHOSTUNREACH:
 964                         __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
 965                         break;
 966
 967                 case ENETUNREACH:
 968                         __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 969                         break;
 970                 }
 971                 goto out;
 972         }
 973
 974         switch (rt->dst.error) {
 975         case EINVAL:
 976         default:
 977                 goto out;
 978         case EHOSTUNREACH:
 979                 code = ICMP_HOST_UNREACH;
 980                 break;
 981         case ENETUNREACH:
 982                 code = ICMP_NET_UNREACH;
 983                 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
 984                 break;
 985         case EACCES:
 986                 code = ICMP_PKT_FILTERED;
 987                 break;
 988         }
 989
 990         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
 991                                l3mdev_master_ifindex(skb->dev), 1);
 992
 993         send = true;
 994         if (peer) {
 995                 now = jiffies;
 996                 peer->rate_tokens += now - peer->rate_last;
 997                 if (peer->rate_tokens > ip_rt_error_burst)
 998                         peer->rate_tokens = ip_rt_error_burst;
 999                 peer->rate_last = now;
1000                 if (peer->rate_tokens >= ip_rt_error_cost)
1001                         peer->rate_tokens -= ip_rt_error_cost;
1002                 else
1003                         send = false;
1004                 inet_putpeer(peer);
1005         }
1006         if (send)
1007                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1008
1009 out:    kfree_skb(skb);
1010         return 0;
1011 }
1012
1013 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1014 {
1015         struct dst_entry *dst = &rt->dst;
1016         u32 old_mtu = ipv4_mtu(dst);
1017         struct fib_result res;
1018         bool lock = false;
1019
1020         if (ip_mtu_locked(dst))
1021                 return;
1022
1023         if (old_mtu < mtu)
1024                 return;
1025
1026         if (mtu < ip_rt_min_pmtu) {
1027                 lock = true;
1028                 mtu = min(old_mtu, ip_rt_min_pmtu);
1029         }
1030
1031         if (rt->rt_pmtu == mtu && !lock &&
1032             time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1033                 return;
1034
1035         rcu_read_lock();
1036         if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1037                 struct fib_nh_common *nhc = FIB_RES_NHC(res);
1038
1039                 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock,
1040                                       jiffies + ip_rt_mtu_expires);
1041         }
1042         rcu_read_unlock();
1043 }
1044
1045 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1046                               struct sk_buff *skb, u32 mtu)
1047 {
1048         struct rtable *rt = (struct rtable *) dst;
1049         struct flowi4 fl4;
1050
1051         ip_rt_build_flow_key(&fl4, sk, skb);
1052         __ip_rt_update_pmtu(rt, &fl4, mtu);
1053 }
1054
1055 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1056                       int oif, u8 protocol)
1057 {
1058         const struct iphdr *iph = (const struct iphdr *) skb->data;
1059         struct flowi4 fl4;
1060         struct rtable *rt;
1061         u32 mark = IP4_REPLY_MARK(net, skb->mark);
1062
1063         __build_flow_key(net, &fl4, NULL, iph, oif,
1064                          RT_TOS(iph->tos), protocol, mark, 0);
1065         rt = __ip_route_output_key(net, &fl4);
1066         if (!IS_ERR(rt)) {
1067                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1068                 ip_rt_put(rt);
1069         }
1070 }
1071 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1072
1073 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1074 {
1075         const struct iphdr *iph = (const struct iphdr *) skb->data;
1076         struct flowi4 fl4;
1077         struct rtable *rt;
1078
1079         __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1080
1081         if (!fl4.flowi4_mark)
1082                 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1083
1084         rt = __ip_route_output_key(sock_net(sk), &fl4);
1085         if (!IS_ERR(rt)) {
1086                 __ip_rt_update_pmtu(rt, &fl4, mtu);
1087                 ip_rt_put(rt);
1088         }
1089 }
1090
1091 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1092 {
1093         const struct iphdr *iph = (const struct iphdr *) skb->data;
1094         struct flowi4 fl4;
1095         struct rtable *rt;
1096         struct dst_entry *odst = NULL;
1097         bool new = false;
1098         struct net *net = sock_net(sk);
1099
1100         bh_lock_sock(sk);
1101
1102         if (!ip_sk_accept_pmtu(sk))
1103                 goto out;
1104
1105         odst = sk_dst_get(sk);
1106
1107         if (sock_owned_by_user(sk) || !odst) {
1108                 __ipv4_sk_update_pmtu(skb, sk, mtu);
1109                 goto out;
1110         }
1111
1112         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1113
1114         rt = (struct rtable *)odst;
1115         if (odst->obsolete && !odst->ops->check(odst, 0)) {
1116                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1117                 if (IS_ERR(rt))
1118                         goto out;
1119
1120                 new = true;
1121         }
1122
1123         __ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu);
1124
1125         if (!dst_check(&rt->dst, 0)) {
1126                 if (new)
1127                         dst_release(&rt->dst);
1128
1129                 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1130                 if (IS_ERR(rt))
1131                         goto out;
1132
1133                 new = true;
1134         }
1135
1136         if (new)
1137                 sk_dst_set(sk, &rt->dst);
1138
1139 out:
1140         bh_unlock_sock(sk);
1141         dst_release(odst);
1142 }
1143 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1144
1145 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1146                    int oif, u8 protocol)
1147 {
1148         const struct iphdr *iph = (const struct iphdr *) skb->data;
1149         struct flowi4 fl4;
1150         struct rtable *rt;
1151
1152         __build_flow_key(net, &fl4, NULL, iph, oif,
1153                          RT_TOS(iph->tos), protocol, 0, 0);
1154         rt = __ip_route_output_key(net, &fl4);
1155         if (!IS_ERR(rt)) {
1156                 __ip_do_redirect(rt, skb, &fl4, false);
1157                 ip_rt_put(rt);
1158         }
1159 }
1160 EXPORT_SYMBOL_GPL(ipv4_redirect);
1161
1162 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1163 {
1164         const struct iphdr *iph = (const struct iphdr *) skb->data;
1165         struct flowi4 fl4;
1166         struct rtable *rt;
1167         struct net *net = sock_net(sk);
1168
1169         __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1170         rt = __ip_route_output_key(net, &fl4);
1171         if (!IS_ERR(rt)) {
1172                 __ip_do_redirect(rt, skb, &fl4, false);
1173                 ip_rt_put(rt);
1174         }
1175 }
1176 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1177
1178 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1179 {
1180         struct rtable *rt = (struct rtable *) dst;
1181
1182         /* All IPV4 dsts are created with ->obsolete set to the value
1183          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1184          * into this function always.
1185          *
1186          * When a PMTU/redirect information update invalidates a route,
1187          * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1188          * DST_OBSOLETE_DEAD.
1189          */
1190         if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1191                 return NULL;
1192         return dst;
1193 }
1194
1195 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1196 {
1197         struct ip_options opt;
1198         int res;
1199
1200         /* Recompile ip options since IPCB may not be valid anymore.
1201          * Also check we have a reasonable ipv4 header.
1202          */
1203         if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1204             ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1205                 return;
1206
1207         memset(&opt, 0, sizeof(opt));
1208         if (ip_hdr(skb)->ihl > 5) {
1209                 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1210                         return;
1211                 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1212
1213                 rcu_read_lock();
1214                 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1215                 rcu_read_unlock();
1216
1217                 if (res)
1218                         return;
1219         }
1220         __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1221 }
1222
1223 static void ipv4_link_failure(struct sk_buff *skb)
1224 {
1225         struct rtable *rt;
1226
1227         ipv4_send_dest_unreach(skb);
1228
1229         rt = skb_rtable(skb);
1230         if (rt)
1231                 dst_set_expires(&rt->dst, 0);
1232 }
1233
1234 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1235 {
1236         pr_debug("%s: %pI4 -> %pI4, %s\n",
1237                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1238                  skb->dev ? skb->dev->name : "?");
1239         kfree_skb(skb);
1240         WARN_ON(1);
1241         return 0;
1242 }
1243
1244 /*
1245    We do not cache source address of outgoing interface,
1246    because it is used only by IP RR, TS and SRR options,
1247    so that it out of fast path.
1248
1249    BTW remember: "addr" is allowed to be not aligned
1250    in IP options!
1251  */
1252
1253 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1254 {
1255         __be32 src;
1256
1257         if (rt_is_output_route(rt))
1258                 src = ip_hdr(skb)->saddr;
1259         else {
1260                 struct fib_result res;
1261                 struct iphdr *iph = ip_hdr(skb);
1262                 struct flowi4 fl4 = {
1263                         .daddr = iph->daddr,
1264                         .saddr = iph->saddr,
1265                         .flowi4_tos = RT_TOS(iph->tos),
1266                         .flowi4_oif = rt->dst.dev->ifindex,
1267                         .flowi4_iif = skb->dev->ifindex,
1268                         .flowi4_mark = skb->mark,
1269                 };
1270
1271                 rcu_read_lock();
1272                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1273                         src = fib_result_prefsrc(dev_net(rt->dst.dev), &res);
1274                 else
1275                         src = inet_select_addr(rt->dst.dev,
1276                                                rt_nexthop(rt, iph->daddr),
1277                                                RT_SCOPE_UNIVERSE);
1278                 rcu_read_unlock();
1279         }
1280         memcpy(addr, &src, 4);
1281 }
1282
1283 #ifdef CONFIG_IP_ROUTE_CLASSID
1284 static void set_class_tag(struct rtable *rt, u32 tag)
1285 {
1286         if (!(rt->dst.tclassid & 0xFFFF))
1287                 rt->dst.tclassid |= tag & 0xFFFF;
1288         if (!(rt->dst.tclassid & 0xFFFF0000))
1289                 rt->dst.tclassid |= tag & 0xFFFF0000;
1290 }
1291 #endif
1292
1293 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1294 {
1295         unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1296         unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1297                                     ip_rt_min_advmss);
1298
1299         return min(advmss, IPV4_MAX_PMTU - header_size);
1300 }
1301
1302 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1303 {
1304         const struct rtable *rt = (const struct rtable *) dst;
1305         unsigned int mtu = rt->rt_pmtu;
1306
1307         if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1308                 mtu = dst_metric_raw(dst, RTAX_MTU);
1309
1310         if (mtu)
1311                 return mtu;
1312
1313         mtu = READ_ONCE(dst->dev->mtu);
1314
1315         if (unlikely(ip_mtu_locked(dst))) {
1316                 if (rt->rt_gw_family && mtu > 576)
1317                         mtu = 576;
1318         }
1319
1320         mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1321
1322         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1323 }
1324
1325 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr)
1326 {
1327         struct fnhe_hash_bucket *hash;
1328         struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1329         u32 hval = fnhe_hashfun(daddr);
1330
1331         spin_lock_bh(&fnhe_lock);
1332
1333         hash = rcu_dereference_protected(nhc->nhc_exceptions,
1334                                          lockdep_is_held(&fnhe_lock));
1335         hash += hval;
1336
1337         fnhe_p = &hash->chain;
1338         fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1339         while (fnhe) {
1340                 if (fnhe->fnhe_daddr == daddr) {
1341                         rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1342                                 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1343                         /* set fnhe_daddr to 0 to ensure it won't bind with
1344                          * new dsts in rt_bind_exception().
1345                          */
1346                         fnhe->fnhe_daddr = 0;
1347                         fnhe_flush_routes(fnhe);
1348                         kfree_rcu(fnhe, rcu);
1349                         break;
1350                 }
1351                 fnhe_p = &fnhe->fnhe_next;
1352                 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1353                                                  lockdep_is_held(&fnhe_lock));
1354         }
1355
1356         spin_unlock_bh(&fnhe_lock);
1357 }
1358
1359 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc,
1360                                                __be32 daddr)
1361 {
1362         struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions);
1363         struct fib_nh_exception *fnhe;
1364         u32 hval;
1365
1366         if (!hash)
1367                 return NULL;
1368
1369         hval = fnhe_hashfun(daddr);
1370
1371         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1372              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1373                 if (fnhe->fnhe_daddr == daddr) {
1374                         if (fnhe->fnhe_expires &&
1375                             time_after(jiffies, fnhe->fnhe_expires)) {
1376                                 ip_del_fnhe(nhc, daddr);
1377                                 break;
1378                         }
1379                         return fnhe;
1380                 }
1381         }
1382         return NULL;
1383 }
1384
1385 /* MTU selection:
1386  * 1. mtu on route is locked - use it
1387  * 2. mtu from nexthop exception
1388  * 3. mtu from egress device
1389  */
1390
1391 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr)
1392 {
1393         struct fib_nh_common *nhc = res->nhc;
1394         struct net_device *dev = nhc->nhc_dev;
1395         struct fib_info *fi = res->fi;
1396         u32 mtu = 0;
1397
1398         if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu ||
1399             fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU))
1400                 mtu = fi->fib_mtu;
1401
1402         if (likely(!mtu)) {
1403                 struct fib_nh_exception *fnhe;
1404
1405                 fnhe = find_exception(nhc, daddr);
1406                 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires))
1407                         mtu = fnhe->fnhe_pmtu;
1408         }
1409
1410         if (likely(!mtu))
1411                 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU);
1412
1413         return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu);
1414 }
1415
1416 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1417                               __be32 daddr, const bool do_cache)
1418 {
1419         bool ret = false;
1420
1421         spin_lock_bh(&fnhe_lock);
1422
1423         if (daddr == fnhe->fnhe_daddr) {
1424                 struct rtable __rcu **porig;
1425                 struct rtable *orig;
1426                 int genid = fnhe_genid(dev_net(rt->dst.dev));
1427
1428                 if (rt_is_input_route(rt))
1429                         porig = &fnhe->fnhe_rth_input;
1430                 else
1431                         porig = &fnhe->fnhe_rth_output;
1432                 orig = rcu_dereference(*porig);
1433
1434                 if (fnhe->fnhe_genid != genid) {
1435                         fnhe->fnhe_genid = genid;
1436                         fnhe->fnhe_gw = 0;
1437                         fnhe->fnhe_pmtu = 0;
1438                         fnhe->fnhe_expires = 0;
1439                         fnhe->fnhe_mtu_locked = false;
1440                         fnhe_flush_routes(fnhe);
1441                         orig = NULL;
1442                 }
1443                 fill_route_from_fnhe(rt, fnhe);
1444                 if (!rt->rt_gw4) {
1445                         rt->rt_gw4 = daddr;
1446                         rt->rt_gw_family = AF_INET;
1447                 }
1448
1449                 if (do_cache) {
1450                         dst_hold(&rt->dst);
1451                         rcu_assign_pointer(*porig, rt);
1452                         if (orig) {
1453                                 dst_dev_put(&orig->dst);
1454                                 dst_release(&orig->dst);
1455                         }
1456                         ret = true;
1457                 }
1458
1459                 fnhe->fnhe_stamp = jiffies;
1460         }
1461         spin_unlock_bh(&fnhe_lock);
1462
1463         return ret;
1464 }
1465
1466 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt)
1467 {
1468         struct rtable *orig, *prev, **p;
1469         bool ret = true;
1470
1471         if (rt_is_input_route(rt)) {
1472                 p = (struct rtable **)&nhc->nhc_rth_input;
1473         } else {
1474                 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
1475         }
1476         orig = *p;
1477
1478         /* hold dst before doing cmpxchg() to avoid race condition
1479          * on this dst
1480          */
1481         dst_hold(&rt->dst);
1482         prev = cmpxchg(p, orig, rt);
1483         if (prev == orig) {
1484                 if (orig) {
1485                         dst_dev_put(&orig->dst);
1486                         dst_release(&orig->dst);
1487                 }
1488         } else {
1489                 dst_release(&rt->dst);
1490                 ret = false;
1491         }
1492
1493         return ret;
1494 }
1495
1496 struct uncached_list {
1497         spinlock_t              lock;
1498         struct list_head        head;
1499 };
1500
1501 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1502
1503 void rt_add_uncached_list(struct rtable *rt)
1504 {
1505         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1506
1507         rt->rt_uncached_list = ul;
1508
1509         spin_lock_bh(&ul->lock);
1510         list_add_tail(&rt->rt_uncached, &ul->head);
1511         spin_unlock_bh(&ul->lock);
1512 }
1513
1514 void rt_del_uncached_list(struct rtable *rt)
1515 {
1516         if (!list_empty(&rt->rt_uncached)) {
1517                 struct uncached_list *ul = rt->rt_uncached_list;
1518
1519                 spin_lock_bh(&ul->lock);
1520                 list_del(&rt->rt_uncached);
1521                 spin_unlock_bh(&ul->lock);
1522         }
1523 }
1524
1525 static void ipv4_dst_destroy(struct dst_entry *dst)
1526 {
1527         struct rtable *rt = (struct rtable *)dst;
1528
1529         ip_dst_metrics_put(dst);
1530         rt_del_uncached_list(rt);
1531 }
1532
1533 void rt_flush_dev(struct net_device *dev)
1534 {
1535         struct rtable *rt;
1536         int cpu;
1537
1538         for_each_possible_cpu(cpu) {
1539                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1540
1541                 spin_lock_bh(&ul->lock);
1542                 list_for_each_entry(rt, &ul->head, rt_uncached) {
1543                         if (rt->dst.dev != dev)
1544                                 continue;
1545                         rt->dst.dev = blackhole_netdev;
1546                         dev_hold(rt->dst.dev);
1547                         dev_put(dev);
1548                 }
1549                 spin_unlock_bh(&ul->lock);
1550         }
1551 }
1552
1553 static bool rt_cache_valid(const struct rtable *rt)
1554 {
1555         return  rt &&
1556                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1557                 !rt_is_expired(rt);
1558 }
1559
1560 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1561                            const struct fib_result *res,
1562                            struct fib_nh_exception *fnhe,
1563                            struct fib_info *fi, u16 type, u32 itag,
1564                            const bool do_cache)
1565 {
1566         bool cached = false;
1567
1568         if (fi) {
1569                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1570
1571                 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) {
1572                         rt->rt_gw_family = nhc->nhc_gw_family;
1573                         /* only INET and INET6 are supported */
1574                         if (likely(nhc->nhc_gw_family == AF_INET))
1575                                 rt->rt_gw4 = nhc->nhc_gw.ipv4;
1576                         else
1577                                 rt->rt_gw6 = nhc->nhc_gw.ipv6;
1578                 }
1579
1580                 ip_dst_init_metrics(&rt->dst, fi->fib_metrics);
1581
1582 #ifdef CONFIG_IP_ROUTE_CLASSID
1583                 if (nhc->nhc_family == AF_INET) {
1584                         struct fib_nh *nh;
1585
1586                         nh = container_of(nhc, struct fib_nh, nh_common);
1587                         rt->dst.tclassid = nh->nh_tclassid;
1588                 }
1589 #endif
1590                 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
1591                 if (unlikely(fnhe))
1592                         cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1593                 else if (do_cache)
1594                         cached = rt_cache_route(nhc, rt);
1595                 if (unlikely(!cached)) {
1596                         /* Routes we intend to cache in nexthop exception or
1597                          * FIB nexthop have the DST_NOCACHE bit clear.
1598                          * However, if we are unsuccessful at storing this
1599                          * route into the cache we really need to set it.
1600                          */
1601                         if (!rt->rt_gw4) {
1602                                 rt->rt_gw_family = AF_INET;
1603                                 rt->rt_gw4 = daddr;
1604                         }
1605                         rt_add_uncached_list(rt);
1606                 }
1607         } else
1608                 rt_add_uncached_list(rt);
1609
1610 #ifdef CONFIG_IP_ROUTE_CLASSID
1611 #ifdef CONFIG_IP_MULTIPLE_TABLES
1612         set_class_tag(rt, res->tclassid);
1613 #endif
1614         set_class_tag(rt, itag);
1615 #endif
1616 }
1617
1618 struct rtable *rt_dst_alloc(struct net_device *dev,
1619                             unsigned int flags, u16 type,
1620                             bool nopolicy, bool noxfrm, bool will_cache)
1621 {
1622         struct rtable *rt;
1623
1624         rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1625                        (will_cache ? 0 : DST_HOST) |
1626                        (nopolicy ? DST_NOPOLICY : 0) |
1627                        (noxfrm ? DST_NOXFRM : 0));
1628
1629         if (rt) {
1630                 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1631                 rt->rt_flags = flags;
1632                 rt->rt_type = type;
1633                 rt->rt_is_input = 0;
1634                 rt->rt_iif = 0;
1635                 rt->rt_pmtu = 0;
1636                 rt->rt_mtu_locked = 0;
1637                 rt->rt_gw_family = 0;
1638                 rt->rt_gw4 = 0;
1639                 INIT_LIST_HEAD(&rt->rt_uncached);
1640
1641                 rt->dst.output = ip_output;
1642                 if (flags & RTCF_LOCAL)
1643                         rt->dst.input = ip_local_deliver;
1644         }
1645
1646         return rt;
1647 }
1648 EXPORT_SYMBOL(rt_dst_alloc);
1649
1650 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt)
1651 {
1652         struct rtable *new_rt;
1653
1654         new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1655                            rt->dst.flags);
1656
1657         if (new_rt) {
1658                 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1659                 new_rt->rt_flags = rt->rt_flags;
1660                 new_rt->rt_type = rt->rt_type;
1661                 new_rt->rt_is_input = rt->rt_is_input;
1662                 new_rt->rt_iif = rt->rt_iif;
1663                 new_rt->rt_pmtu = rt->rt_pmtu;
1664                 new_rt->rt_mtu_locked = rt->rt_mtu_locked;
1665                 new_rt->rt_gw_family = rt->rt_gw_family;
1666                 if (rt->rt_gw_family == AF_INET)
1667                         new_rt->rt_gw4 = rt->rt_gw4;
1668                 else if (rt->rt_gw_family == AF_INET6)
1669                         new_rt->rt_gw6 = rt->rt_gw6;
1670                 INIT_LIST_HEAD(&new_rt->rt_uncached);
1671
1672                 new_rt->dst.flags |= DST_HOST;
1673                 new_rt->dst.input = rt->dst.input;
1674                 new_rt->dst.output = rt->dst.output;
1675                 new_rt->dst.error = rt->dst.error;
1676                 new_rt->dst.lastuse = jiffies;
1677                 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate);
1678         }
1679         return new_rt;
1680 }
1681 EXPORT_SYMBOL(rt_dst_clone);
1682
1683 /* called in rcu_read_lock() section */
1684 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1685                           u8 tos, struct net_device *dev,
1686                           struct in_device *in_dev, u32 *itag)
1687 {
1688         int err;
1689
1690         /* Primary sanity checks. */
1691         if (!in_dev)
1692                 return -EINVAL;
1693
1694         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1695             skb->protocol != htons(ETH_P_IP))
1696                 return -EINVAL;
1697
1698         if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1699                 return -EINVAL;
1700
1701         if (ipv4_is_zeronet(saddr)) {
1702                 if (!ipv4_is_local_multicast(daddr) &&
1703                     ip_hdr(skb)->protocol != IPPROTO_IGMP)
1704                         return -EINVAL;
1705         } else {
1706                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1707                                           in_dev, itag);
1708                 if (err < 0)
1709                         return err;
1710         }
1711         return 0;
1712 }
1713
1714 /* called in rcu_read_lock() section */
1715 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1716                              u8 tos, struct net_device *dev, int our)
1717 {
1718         struct in_device *in_dev = __in_dev_get_rcu(dev);
1719         unsigned int flags = RTCF_MULTICAST;
1720         struct rtable *rth;
1721         u32 itag = 0;
1722         int err;
1723
1724         err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1725         if (err)
1726                 return err;
1727
1728         if (our)
1729                 flags |= RTCF_LOCAL;
1730
1731         rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1732                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1733         if (!rth)
1734                 return -ENOBUFS;
1735
1736 #ifdef CONFIG_IP_ROUTE_CLASSID
1737         rth->dst.tclassid = itag;
1738 #endif
1739         rth->dst.output = ip_rt_bug;
1740         rth->rt_is_input= 1;
1741
1742 #ifdef CONFIG_IP_MROUTE
1743         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1744                 rth->dst.input = ip_mr_input;
1745 #endif
1746         RT_CACHE_STAT_INC(in_slow_mc);
1747
1748         skb_dst_set(skb, &rth->dst);
1749         return 0;
1750 }
1751
1752
1753 static void ip_handle_martian_source(struct net_device *dev,
1754                                      struct in_device *in_dev,
1755                                      struct sk_buff *skb,
1756                                      __be32 daddr,
1757                                      __be32 saddr)
1758 {
1759         RT_CACHE_STAT_INC(in_martian_src);
1760 #ifdef CONFIG_IP_ROUTE_VERBOSE
1761         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1762                 /*
1763                  *      RFC1812 recommendation, if source is martian,
1764                  *      the only hint is MAC header.
1765                  */
1766                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1767                         &daddr, &saddr, dev->name);
1768                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1769                         print_hex_dump(KERN_WARNING, "ll header: ",
1770                                        DUMP_PREFIX_OFFSET, 16, 1,
1771                                        skb_mac_header(skb),
1772                                        dev->hard_header_len, false);
1773                 }
1774         }
1775 #endif
1776 }
1777
1778 /* called in rcu_read_lock() section */
1779 static int __mkroute_input(struct sk_buff *skb,
1780                            const struct fib_result *res,
1781                            struct in_device *in_dev,
1782                            __be32 daddr, __be32 saddr, u32 tos)
1783 {
1784         struct fib_nh_common *nhc = FIB_RES_NHC(*res);
1785         struct net_device *dev = nhc->nhc_dev;
1786         struct fib_nh_exception *fnhe;
1787         struct rtable *rth;
1788         int err;
1789         struct in_device *out_dev;
1790         bool do_cache;
1791         u32 itag = 0;
1792
1793         /* get a working reference to the output device */
1794         out_dev = __in_dev_get_rcu(dev);
1795         if (!out_dev) {
1796                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1797                 return -EINVAL;
1798         }
1799
1800         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1801                                   in_dev->dev, in_dev, &itag);
1802         if (err < 0) {
1803                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1804                                          saddr);
1805
1806                 goto cleanup;
1807         }
1808
1809         do_cache = res->fi && !itag;
1810         if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1811             skb->protocol == htons(ETH_P_IP)) {
1812                 __be32 gw;
1813
1814                 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0;
1815                 if (IN_DEV_SHARED_MEDIA(out_dev) ||
1816                     inet_addr_onlink(out_dev, saddr, gw))
1817                         IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1818         }
1819
1820         if (skb->protocol != htons(ETH_P_IP)) {
1821                 /* Not IP (i.e. ARP). Do not create route, if it is
1822                  * invalid for proxy arp. DNAT routes are always valid.
1823                  *
1824                  * Proxy arp feature have been extended to allow, ARP
1825                  * replies back to the same interface, to support
1826                  * Private VLAN switch technologies. See arp.c.
1827                  */
1828                 if (out_dev == in_dev &&
1829                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1830                         err = -EINVAL;
1831                         goto cleanup;
1832                 }
1833         }
1834
1835         fnhe = find_exception(nhc, daddr);
1836         if (do_cache) {
1837                 if (fnhe)
1838                         rth = rcu_dereference(fnhe->fnhe_rth_input);
1839                 else
1840                         rth = rcu_dereference(nhc->nhc_rth_input);
1841                 if (rt_cache_valid(rth)) {
1842                         skb_dst_set_noref(skb, &rth->dst);
1843                         goto out;
1844                 }
1845         }
1846
1847         rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1848                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1849                            IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1850         if (!rth) {
1851                 err = -ENOBUFS;
1852                 goto cleanup;
1853         }
1854
1855         rth->rt_is_input = 1;
1856         RT_CACHE_STAT_INC(in_slow_tot);
1857
1858         rth->dst.input = ip_forward;
1859
1860         rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1861                        do_cache);
1862         lwtunnel_set_redirect(&rth->dst);
1863         skb_dst_set(skb, &rth->dst);
1864 out:
1865         err = 0;
1866  cleanup:
1867         return err;
1868 }
1869
1870 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1871 /* To make ICMP packets follow the right flow, the multipath hash is
1872  * calculated from the inner IP addresses.
1873  */
1874 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1875                                  struct flow_keys *hash_keys)
1876 {
1877         const struct iphdr *outer_iph = ip_hdr(skb);
1878         const struct iphdr *key_iph = outer_iph;
1879         const struct iphdr *inner_iph;
1880         const struct icmphdr *icmph;
1881         struct iphdr _inner_iph;
1882         struct icmphdr _icmph;
1883
1884         if (likely(outer_iph->protocol != IPPROTO_ICMP))
1885                 goto out;
1886
1887         if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1888                 goto out;
1889
1890         icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1891                                    &_icmph);
1892         if (!icmph)
1893                 goto out;
1894
1895         if (icmph->type != ICMP_DEST_UNREACH &&
1896             icmph->type != ICMP_REDIRECT &&
1897             icmph->type != ICMP_TIME_EXCEEDED &&
1898             icmph->type != ICMP_PARAMETERPROB)
1899                 goto out;
1900
1901         inner_iph = skb_header_pointer(skb,
1902                                        outer_iph->ihl * 4 + sizeof(_icmph),
1903                                        sizeof(_inner_iph), &_inner_iph);
1904         if (!inner_iph)
1905                 goto out;
1906
1907         key_iph = inner_iph;
1908 out:
1909         hash_keys->addrs.v4addrs.src = key_iph->saddr;
1910         hash_keys->addrs.v4addrs.dst = key_iph->daddr;
1911 }
1912
1913 /* if skb is set it will be used and fl4 can be NULL */
1914 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
1915                        const struct sk_buff *skb, struct flow_keys *flkeys)
1916 {
1917         u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
1918         struct flow_keys hash_keys;
1919         u32 mhash;
1920
1921         switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1922         case 0:
1923                 memset(&hash_keys, 0, sizeof(hash_keys));
1924                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1925                 if (skb) {
1926                         ip_multipath_l3_keys(skb, &hash_keys);
1927                 } else {
1928                         hash_keys.addrs.v4addrs.src = fl4->saddr;
1929                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
1930                 }
1931                 break;
1932         case 1:
1933                 /* skb is currently provided only when forwarding */
1934                 if (skb) {
1935                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1936                         struct flow_keys keys;
1937
1938                         /* short-circuit if we already have L4 hash present */
1939                         if (skb->l4_hash)
1940                                 return skb_get_hash_raw(skb) >> 1;
1941
1942                         memset(&hash_keys, 0, sizeof(hash_keys));
1943
1944                         if (!flkeys) {
1945                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
1946                                 flkeys = &keys;
1947                         }
1948
1949                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1950                         hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
1951                         hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
1952                         hash_keys.ports.src = flkeys->ports.src;
1953                         hash_keys.ports.dst = flkeys->ports.dst;
1954                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
1955                 } else {
1956                         memset(&hash_keys, 0, sizeof(hash_keys));
1957                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1958                         hash_keys.addrs.v4addrs.src = fl4->saddr;
1959                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
1960                         hash_keys.ports.src = fl4->fl4_sport;
1961                         hash_keys.ports.dst = fl4->fl4_dport;
1962                         hash_keys.basic.ip_proto = fl4->flowi4_proto;
1963                 }
1964                 break;
1965         case 2:
1966                 memset(&hash_keys, 0, sizeof(hash_keys));
1967                 /* skb is currently provided only when forwarding */
1968                 if (skb) {
1969                         struct flow_keys keys;
1970
1971                         skb_flow_dissect_flow_keys(skb, &keys, 0);
1972                         /* Inner can be v4 or v6 */
1973                         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
1974                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1975                                 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1976                                 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1977                         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
1978                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1979                                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
1980                                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
1981                                 hash_keys.tags.flow_label = keys.tags.flow_label;
1982                                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1983                         } else {
1984                                 /* Same as case 0 */
1985                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1986                                 ip_multipath_l3_keys(skb, &hash_keys);
1987                         }
1988                 } else {
1989                         /* Same as case 0 */
1990                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1991                         hash_keys.addrs.v4addrs.src = fl4->saddr;
1992                         hash_keys.addrs.v4addrs.dst = fl4->daddr;
1993                 }
1994                 break;
1995         }
1996         mhash = flow_hash_from_keys(&hash_keys);
1997
1998         if (multipath_hash)
1999                 mhash = jhash_2words(mhash, multipath_hash, 0);
2000
2001         return mhash >> 1;
2002 }
2003 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
2004
2005 static int ip_mkroute_input(struct sk_buff *skb,
2006                             struct fib_result *res,
2007                             struct in_device *in_dev,
2008                             __be32 daddr, __be32 saddr, u32 tos,
2009                             struct flow_keys *hkeys)
2010 {
2011 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2012         if (res->fi && fib_info_num_path(res->fi) > 1) {
2013                 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys);
2014
2015                 fib_select_multipath(res, h);
2016         }
2017 #endif
2018
2019         /* create a routing cache entry */
2020         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2021 }
2022
2023 /*
2024  *      NOTE. We drop all the packets that has local source
2025  *      addresses, because every properly looped back packet
2026  *      must have correct destination already attached by output routine.
2027  *
2028  *      Such approach solves two big problems:
2029  *      1. Not simplex devices are handled properly.
2030  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2031  *      called with rcu_read_lock()
2032  */
2033
2034 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2035                                u8 tos, struct net_device *dev,
2036                                struct fib_result *res)
2037 {
2038         struct in_device *in_dev = __in_dev_get_rcu(dev);
2039         struct flow_keys *flkeys = NULL, _flkeys;
2040         struct net    *net = dev_net(dev);
2041         struct ip_tunnel_info *tun_info;
2042         int             err = -EINVAL;
2043         unsigned int    flags = 0;
2044         u32             itag = 0;
2045         struct rtable   *rth;
2046         struct flowi4   fl4;
2047         bool do_cache = true;
2048
2049         /* IP on this device is disabled. */
2050
2051         if (!in_dev)
2052                 goto out;
2053
2054         /* Check for the most weird martians, which can be not detected
2055            by fib_lookup.
2056          */
2057
2058         tun_info = skb_tunnel_info(skb);
2059         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2060                 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
2061         else
2062                 fl4.flowi4_tun_key.tun_id = 0;
2063         skb_dst_drop(skb);
2064
2065         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
2066                 goto martian_source;
2067
2068         res->fi = NULL;
2069         res->table = NULL;
2070         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2071                 goto brd_input;
2072
2073         /* Accept zero addresses only to limited broadcast;
2074          * I even do not know to fix it or not. Waiting for complains :-)
2075          */
2076         if (ipv4_is_zeronet(saddr))
2077                 goto martian_source;
2078
2079         if (ipv4_is_zeronet(daddr))
2080                 goto martian_destination;
2081
2082         /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
2083          * and call it once if daddr or/and saddr are loopback addresses
2084          */
2085         if (ipv4_is_loopback(daddr)) {
2086                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2087                         goto martian_destination;
2088         } else if (ipv4_is_loopback(saddr)) {
2089                 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
2090                         goto martian_source;
2091         }
2092
2093         /*
2094          *      Now we are ready to route packet.
2095          */
2096         fl4.flowi4_oif = 0;
2097         fl4.flowi4_iif = dev->ifindex;
2098         fl4.flowi4_mark = skb->mark;
2099         fl4.flowi4_tos = tos;
2100         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2101         fl4.flowi4_flags = 0;
2102         fl4.daddr = daddr;
2103         fl4.saddr = saddr;
2104         fl4.flowi4_uid = sock_net_uid(net, NULL);
2105
2106         if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) {
2107                 flkeys = &_flkeys;
2108         } else {
2109                 fl4.flowi4_proto = 0;
2110                 fl4.fl4_sport = 0;
2111                 fl4.fl4_dport = 0;
2112         }
2113
2114         err = fib_lookup(net, &fl4, res, 0);
2115         if (err != 0) {
2116                 if (!IN_DEV_FORWARD(in_dev))
2117                         err = -EHOSTUNREACH;
2118                 goto no_route;
2119         }
2120
2121         if (res->type == RTN_BROADCAST) {
2122                 if (IN_DEV_BFORWARD(in_dev))
2123                         goto make_route;
2124                 /* not do cache if bc_forwarding is enabled */
2125                 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
2126                         do_cache = false;
2127                 goto brd_input;
2128         }
2129
2130         if (res->type == RTN_LOCAL) {
2131                 err = fib_validate_source(skb, saddr, daddr, tos,
2132                                           0, dev, in_dev, &itag);
2133                 if (err < 0)
2134                         goto martian_source;
2135                 goto local_input;
2136         }
2137
2138         if (!IN_DEV_FORWARD(in_dev)) {
2139                 err = -EHOSTUNREACH;
2140                 goto no_route;
2141         }
2142         if (res->type != RTN_UNICAST)
2143                 goto martian_destination;
2144
2145 make_route:
2146         err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys);
2147 out:    return err;
2148
2149 brd_input:
2150         if (skb->protocol != htons(ETH_P_IP))
2151                 goto e_inval;
2152
2153         if (!ipv4_is_zeronet(saddr)) {
2154                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2155                                           in_dev, &itag);
2156                 if (err < 0)
2157                         goto martian_source;
2158         }
2159         flags |= RTCF_BROADCAST;
2160         res->type = RTN_BROADCAST;
2161         RT_CACHE_STAT_INC(in_brd);
2162
2163 local_input:
2164         do_cache &= res->fi && !itag;
2165         if (do_cache) {
2166                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2167
2168                 rth = rcu_dereference(nhc->nhc_rth_input);
2169                 if (rt_cache_valid(rth)) {
2170                         skb_dst_set_noref(skb, &rth->dst);
2171                         err = 0;
2172                         goto out;
2173                 }
2174         }
2175
2176         rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2177                            flags | RTCF_LOCAL, res->type,
2178                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2179         if (!rth)
2180                 goto e_nobufs;
2181
2182         rth->dst.output= ip_rt_bug;
2183 #ifdef CONFIG_IP_ROUTE_CLASSID
2184         rth->dst.tclassid = itag;
2185 #endif
2186         rth->rt_is_input = 1;
2187
2188         RT_CACHE_STAT_INC(in_slow_tot);
2189         if (res->type == RTN_UNREACHABLE) {
2190                 rth->dst.input= ip_error;
2191                 rth->dst.error= -err;
2192                 rth->rt_flags   &= ~RTCF_LOCAL;
2193         }
2194
2195         if (do_cache) {
2196                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2197
2198                 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate);
2199                 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2200                         WARN_ON(rth->dst.input == lwtunnel_input);
2201                         rth->dst.lwtstate->orig_input = rth->dst.input;
2202                         rth->dst.input = lwtunnel_input;
2203                 }
2204
2205                 if (unlikely(!rt_cache_route(nhc, rth)))
2206                         rt_add_uncached_list(rth);
2207         }
2208         skb_dst_set(skb, &rth->dst);
2209         err = 0;
2210         goto out;
2211
2212 no_route:
2213         RT_CACHE_STAT_INC(in_no_route);
2214         res->type = RTN_UNREACHABLE;
2215         res->fi = NULL;
2216         res->table = NULL;
2217         goto local_input;
2218
2219         /*
2220          *      Do not cache martian addresses: they should be logged (RFC1812)
2221          */
2222 martian_destination:
2223         RT_CACHE_STAT_INC(in_martian_dst);
2224 #ifdef CONFIG_IP_ROUTE_VERBOSE
2225         if (IN_DEV_LOG_MARTIANS(in_dev))
2226                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2227                                      &daddr, &saddr, dev->name);
2228 #endif
2229
2230 e_inval:
2231         err = -EINVAL;
2232         goto out;
2233
2234 e_nobufs:
2235         err = -ENOBUFS;
2236         goto out;
2237
2238 martian_source:
2239         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2240         goto out;
2241 }
2242
2243 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2244                          u8 tos, struct net_device *dev)
2245 {
2246         struct fib_result res;
2247         int err;
2248
2249         tos &= IPTOS_RT_MASK;
2250         rcu_read_lock();
2251         err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2252         rcu_read_unlock();
2253
2254         return err;
2255 }
2256 EXPORT_SYMBOL(ip_route_input_noref);
2257
2258 /* called with rcu_read_lock held */
2259 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2260                        u8 tos, struct net_device *dev, struct fib_result *res)
2261 {
2262         /* Multicast recognition logic is moved from route cache to here.
2263            The problem was that too many Ethernet cards have broken/missing
2264            hardware multicast filters :-( As result the host on multicasting
2265            network acquires a lot of useless route cache entries, sort of
2266            SDR messages from all the world. Now we try to get rid of them.
2267            Really, provided software IP multicast filter is organized
2268            reasonably (at least, hashed), it does not result in a slowdown
2269            comparing with route cache reject entries.
2270            Note, that multicast routers are not affected, because
2271            route cache entry is created eventually.
2272          */
2273         if (ipv4_is_multicast(daddr)) {
2274                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2275                 int our = 0;
2276                 int err = -EINVAL;
2277
2278                 if (!in_dev)
2279                         return err;
2280                 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2281                                       ip_hdr(skb)->protocol);
2282
2283                 /* check l3 master if no match yet */
2284                 if (!our && netif_is_l3_slave(dev)) {
2285                         struct in_device *l3_in_dev;
2286
2287                         l3_in_dev = __in_dev_get_rcu(skb->dev);
2288                         if (l3_in_dev)
2289                                 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2290                                                       ip_hdr(skb)->protocol);
2291                 }
2292
2293                 if (our
2294 #ifdef CONFIG_IP_MROUTE
2295                         ||
2296                     (!ipv4_is_local_multicast(daddr) &&
2297                      IN_DEV_MFORWARD(in_dev))
2298 #endif
2299                    ) {
2300                         err = ip_route_input_mc(skb, daddr, saddr,
2301                                                 tos, dev, our);
2302                 }
2303                 return err;
2304         }
2305
2306         return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2307 }
2308
2309 /* called with rcu_read_lock() */
2310 static struct rtable *__mkroute_output(const struct fib_result *res,
2311                                        const struct flowi4 *fl4, int orig_oif,
2312                                        struct net_device *dev_out,
2313                                        unsigned int flags)
2314 {
2315         struct fib_info *fi = res->fi;
2316         struct fib_nh_exception *fnhe;
2317         struct in_device *in_dev;
2318         u16 type = res->type;
2319         struct rtable *rth;
2320         bool do_cache;
2321
2322         in_dev = __in_dev_get_rcu(dev_out);
2323         if (!in_dev)
2324                 return ERR_PTR(-EINVAL);
2325
2326         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2327                 if (ipv4_is_loopback(fl4->saddr) &&
2328                     !(dev_out->flags & IFF_LOOPBACK) &&
2329                     !netif_is_l3_master(dev_out))
2330                         return ERR_PTR(-EINVAL);
2331
2332         if (ipv4_is_lbcast(fl4->daddr))
2333                 type = RTN_BROADCAST;
2334         else if (ipv4_is_multicast(fl4->daddr))
2335                 type = RTN_MULTICAST;
2336         else if (ipv4_is_zeronet(fl4->daddr))
2337                 return ERR_PTR(-EINVAL);
2338
2339         if (dev_out->flags & IFF_LOOPBACK)
2340                 flags |= RTCF_LOCAL;
2341
2342         do_cache = true;
2343         if (type == RTN_BROADCAST) {
2344                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2345                 fi = NULL;
2346         } else if (type == RTN_MULTICAST) {
2347                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2348                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2349                                      fl4->flowi4_proto))
2350                         flags &= ~RTCF_LOCAL;
2351                 else
2352                         do_cache = false;
2353                 /* If multicast route do not exist use
2354                  * default one, but do not gateway in this case.
2355                  * Yes, it is hack.
2356                  */
2357                 if (fi && res->prefixlen < 4)
2358                         fi = NULL;
2359         } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2360                    (orig_oif != dev_out->ifindex)) {
2361                 /* For local routes that require a particular output interface
2362                  * we do not want to cache the result.  Caching the result
2363                  * causes incorrect behaviour when there are multiple source
2364                  * addresses on the interface, the end result being that if the
2365                  * intended recipient is waiting on that interface for the
2366                  * packet he won't receive it because it will be delivered on
2367                  * the loopback interface and the IP_PKTINFO ipi_ifindex will
2368                  * be set to the loopback interface as well.
2369                  */
2370                 do_cache = false;
2371         }
2372
2373         fnhe = NULL;
2374         do_cache &= fi != NULL;
2375         if (fi) {
2376                 struct fib_nh_common *nhc = FIB_RES_NHC(*res);
2377                 struct rtable __rcu **prth;
2378
2379                 fnhe = find_exception(nhc, fl4->daddr);
2380                 if (!do_cache)
2381                         goto add;
2382                 if (fnhe) {
2383                         prth = &fnhe->fnhe_rth_output;
2384                 } else {
2385                         if (unlikely(fl4->flowi4_flags &
2386                                      FLOWI_FLAG_KNOWN_NH &&
2387                                      !(nhc->nhc_gw_family &&
2388                                        nhc->nhc_scope == RT_SCOPE_LINK))) {
2389                                 do_cache = false;
2390                                 goto add;
2391                         }
2392                         prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output);
2393                 }
2394                 rth = rcu_dereference(*prth);
2395                 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2396                         return rth;
2397         }
2398
2399 add:
2400         rth = rt_dst_alloc(dev_out, flags, type,
2401                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
2402                            IN_DEV_CONF_GET(in_dev, NOXFRM),
2403                            do_cache);
2404         if (!rth)
2405                 return ERR_PTR(-ENOBUFS);
2406
2407         rth->rt_iif = orig_oif;
2408
2409         RT_CACHE_STAT_INC(out_slow_tot);
2410
2411         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2412                 if (flags & RTCF_LOCAL &&
2413                     !(dev_out->flags & IFF_LOOPBACK)) {
2414                         rth->dst.output = ip_mc_output;
2415                         RT_CACHE_STAT_INC(out_slow_mc);
2416                 }
2417 #ifdef CONFIG_IP_MROUTE
2418                 if (type == RTN_MULTICAST) {
2419                         if (IN_DEV_MFORWARD(in_dev) &&
2420                             !ipv4_is_local_multicast(fl4->daddr)) {
2421                                 rth->dst.input = ip_mr_input;
2422                                 rth->dst.output = ip_mc_output;
2423                         }
2424                 }
2425 #endif
2426         }
2427
2428         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2429         lwtunnel_set_redirect(&rth->dst);
2430
2431         return rth;
2432 }
2433
2434 /*
2435  * Major route resolver routine.
2436  */
2437
2438 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2439                                         const struct sk_buff *skb)
2440 {
2441         __u8 tos = RT_FL_TOS(fl4);
2442         struct fib_result res = {
2443                 .type           = RTN_UNSPEC,
2444                 .fi             = NULL,
2445                 .table          = NULL,
2446                 .tclassid       = 0,
2447         };
2448         struct rtable *rth;
2449
2450         fl4->flowi4_iif = LOOPBACK_IFINDEX;
2451         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2452         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2453                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2454
2455         rcu_read_lock();
2456         rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2457         rcu_read_unlock();
2458
2459         return rth;
2460 }
2461 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2462
2463 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2464                                             struct fib_result *res,
2465                                             const struct sk_buff *skb)
2466 {
2467         struct net_device *dev_out = NULL;
2468         int orig_oif = fl4->flowi4_oif;
2469         unsigned int flags = 0;
2470         struct rtable *rth;
2471         int err = -ENETUNREACH;
2472
2473         if (fl4->saddr) {
2474                 rth = ERR_PTR(-EINVAL);
2475                 if (ipv4_is_multicast(fl4->saddr) ||
2476                     ipv4_is_lbcast(fl4->saddr) ||
2477                     ipv4_is_zeronet(fl4->saddr))
2478                         goto out;
2479
2480                 /* I removed check for oif == dev_out->oif here.
2481                    It was wrong for two reasons:
2482                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2483                       is assigned to multiple interfaces.
2484                    2. Moreover, we are allowed to send packets with saddr
2485                       of another iface. --ANK
2486                  */
2487
2488                 if (fl4->flowi4_oif == 0 &&
2489                     (ipv4_is_multicast(fl4->daddr) ||
2490                      ipv4_is_lbcast(fl4->daddr))) {
2491                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2492                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2493                         if (!dev_out)
2494                                 goto out;
2495
2496                         /* Special hack: user can direct multicasts
2497                            and limited broadcast via necessary interface
2498                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2499                            This hack is not just for fun, it allows
2500                            vic,vat and friends to work.
2501                            They bind socket to loopback, set ttl to zero
2502                            and expect that it will work.
2503                            From the viewpoint of routing cache they are broken,
2504                            because we are not allowed to build multicast path
2505                            with loopback source addr (look, routing cache
2506                            cannot know, that ttl is zero, so that packet
2507                            will not leave this host and route is valid).
2508                            Luckily, this hack is good workaround.
2509                          */
2510
2511                         fl4->flowi4_oif = dev_out->ifindex;
2512                         goto make_route;
2513                 }
2514
2515                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2516                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2517                         if (!__ip_dev_find(net, fl4->saddr, false))
2518                                 goto out;
2519                 }
2520         }
2521
2522
2523         if (fl4->flowi4_oif) {
2524                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2525                 rth = ERR_PTR(-ENODEV);
2526                 if (!dev_out)
2527                         goto out;
2528
2529                 /* RACE: Check return value of inet_select_addr instead. */
2530                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2531                         rth = ERR_PTR(-ENETUNREACH);
2532                         goto out;
2533                 }
2534                 if (ipv4_is_local_multicast(fl4->daddr) ||
2535                     ipv4_is_lbcast(fl4->daddr) ||
2536                     fl4->flowi4_proto == IPPROTO_IGMP) {
2537                         if (!fl4->saddr)
2538                                 fl4->saddr = inet_select_addr(dev_out, 0,
2539                                                               RT_SCOPE_LINK);
2540                         goto make_route;
2541                 }
2542                 if (!fl4->saddr) {
2543                         if (ipv4_is_multicast(fl4->daddr))
2544                                 fl4->saddr = inet_select_addr(dev_out, 0,
2545                                                               fl4->flowi4_scope);
2546                         else if (!fl4->daddr)
2547                                 fl4->saddr = inet_select_addr(dev_out, 0,
2548                                                               RT_SCOPE_HOST);
2549                 }
2550         }
2551
2552         if (!fl4->daddr) {
2553                 fl4->daddr = fl4->saddr;
2554                 if (!fl4->daddr)
2555                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2556                 dev_out = net->loopback_dev;
2557                 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2558                 res->type = RTN_LOCAL;
2559                 flags |= RTCF_LOCAL;
2560                 goto make_route;
2561         }
2562
2563         err = fib_lookup(net, fl4, res, 0);
2564         if (err) {
2565                 res->fi = NULL;
2566                 res->table = NULL;
2567                 if (fl4->flowi4_oif &&
2568                     (ipv4_is_multicast(fl4->daddr) ||
2569                     !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2570                         /* Apparently, routing tables are wrong. Assume,
2571                            that the destination is on link.
2572
2573                            WHY? DW.
2574                            Because we are allowed to send to iface
2575                            even if it has NO routes and NO assigned
2576                            addresses. When oif is specified, routing
2577                            tables are looked up with only one purpose:
2578                            to catch if destination is gatewayed, rather than
2579                            direct. Moreover, if MSG_DONTROUTE is set,
2580                            we send packet, ignoring both routing tables
2581                            and ifaddr state. --ANK
2582
2583
2584                            We could make it even if oif is unknown,
2585                            likely IPv6, but we do not.
2586                          */
2587
2588                         if (fl4->saddr == 0)
2589                                 fl4->saddr = inet_select_addr(dev_out, 0,
2590                                                               RT_SCOPE_LINK);
2591                         res->type = RTN_UNICAST;
2592                         goto make_route;
2593                 }
2594                 rth = ERR_PTR(err);
2595                 goto out;
2596         }
2597
2598         if (res->type == RTN_LOCAL) {
2599                 if (!fl4->saddr) {
2600                         if (res->fi->fib_prefsrc)
2601                                 fl4->saddr = res->fi->fib_prefsrc;
2602                         else
2603                                 fl4->saddr = fl4->daddr;
2604                 }
2605
2606                 /* L3 master device is the loopback for that domain */
2607                 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2608                         net->loopback_dev;
2609
2610                 /* make sure orig_oif points to fib result device even
2611                  * though packet rx/tx happens over loopback or l3mdev
2612                  */
2613                 orig_oif = FIB_RES_OIF(*res);
2614
2615                 fl4->flowi4_oif = dev_out->ifindex;
2616                 flags |= RTCF_LOCAL;
2617                 goto make_route;
2618         }
2619
2620         fib_select_path(net, res, fl4, skb);
2621
2622         dev_out = FIB_RES_DEV(*res);
2623         fl4->flowi4_oif = dev_out->ifindex;
2624
2625
2626 make_route:
2627         rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2628
2629 out:
2630         return rth;
2631 }
2632
2633 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2634 {
2635         return NULL;
2636 }
2637
2638 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2639 {
2640         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2641
2642         return mtu ? : dst->dev->mtu;
2643 }
2644
2645 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2646                                           struct sk_buff *skb, u32 mtu)
2647 {
2648 }
2649
2650 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2651                                        struct sk_buff *skb)
2652 {
2653 }
2654
2655 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2656                                           unsigned long old)
2657 {
2658         return NULL;
2659 }
2660
2661 static struct dst_ops ipv4_dst_blackhole_ops = {
2662         .family                 =       AF_INET,
2663         .check                  =       ipv4_blackhole_dst_check,
2664         .mtu                    =       ipv4_blackhole_mtu,
2665         .default_advmss         =       ipv4_default_advmss,
2666         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2667         .redirect               =       ipv4_rt_blackhole_redirect,
2668         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
2669         .neigh_lookup           =       ipv4_neigh_lookup,
2670 };
2671
2672 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2673 {
2674         struct rtable *ort = (struct rtable *) dst_orig;
2675         struct rtable *rt;
2676
2677         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2678         if (rt) {
2679                 struct dst_entry *new = &rt->dst;
2680
2681                 new->__use = 1;
2682                 new->input = dst_discard;
2683                 new->output = dst_discard_out;
2684
2685                 new->dev = net->loopback_dev;
2686                 if (new->dev)
2687                         dev_hold(new->dev);
2688
2689                 rt->rt_is_input = ort->rt_is_input;
2690                 rt->rt_iif = ort->rt_iif;
2691                 rt->rt_pmtu = ort->rt_pmtu;
2692                 rt->rt_mtu_locked = ort->rt_mtu_locked;
2693
2694                 rt->rt_genid = rt_genid_ipv4(net);
2695                 rt->rt_flags = ort->rt_flags;
2696                 rt->rt_type = ort->rt_type;
2697                 rt->rt_gw_family = ort->rt_gw_family;
2698                 if (rt->rt_gw_family == AF_INET)
2699                         rt->rt_gw4 = ort->rt_gw4;
2700                 else if (rt->rt_gw_family == AF_INET6)
2701                         rt->rt_gw6 = ort->rt_gw6;
2702
2703                 INIT_LIST_HEAD(&rt->rt_uncached);
2704         }
2705
2706         dst_release(dst_orig);
2707
2708         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2709 }
2710
2711 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2712                                     const struct sock *sk)
2713 {
2714         struct rtable *rt = __ip_route_output_key(net, flp4);
2715
2716         if (IS_ERR(rt))
2717                 return rt;
2718
2719         if (flp4->flowi4_proto)
2720                 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2721                                                         flowi4_to_flowi(flp4),
2722                                                         sk, 0);
2723
2724         return rt;
2725 }
2726 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2727
2728 /* called with rcu_read_lock held */
2729 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2730                         struct rtable *rt, u32 table_id, struct flowi4 *fl4,
2731                         struct sk_buff *skb, u32 portid, u32 seq)
2732 {
2733         struct rtmsg *r;
2734         struct nlmsghdr *nlh;
2735         unsigned long expires = 0;
2736         u32 error;
2737         u32 metrics[RTAX_MAX];
2738
2739         nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0);
2740         if (!nlh)
2741                 return -EMSGSIZE;
2742
2743         r = nlmsg_data(nlh);
2744         r->rtm_family    = AF_INET;
2745         r->rtm_dst_len  = 32;
2746         r->rtm_src_len  = 0;
2747         r->rtm_tos      = fl4 ? fl4->flowi4_tos : 0;
2748         r->rtm_table    = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2749         if (nla_put_u32(skb, RTA_TABLE, table_id))
2750                 goto nla_put_failure;
2751         r->rtm_type     = rt->rt_type;
2752         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2753         r->rtm_protocol = RTPROT_UNSPEC;
2754         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2755         if (rt->rt_flags & RTCF_NOTIFY)
2756                 r->rtm_flags |= RTM_F_NOTIFY;
2757         if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2758                 r->rtm_flags |= RTCF_DOREDIRECT;
2759
2760         if (nla_put_in_addr(skb, RTA_DST, dst))
2761                 goto nla_put_failure;
2762         if (src) {
2763                 r->rtm_src_len = 32;
2764                 if (nla_put_in_addr(skb, RTA_SRC, src))
2765                         goto nla_put_failure;
2766         }
2767         if (rt->dst.dev &&
2768             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2769                 goto nla_put_failure;
2770 #ifdef CONFIG_IP_ROUTE_CLASSID
2771         if (rt->dst.tclassid &&
2772             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2773                 goto nla_put_failure;
2774 #endif
2775         if (fl4 && !rt_is_input_route(rt) &&
2776             fl4->saddr != src) {
2777                 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2778                         goto nla_put_failure;
2779         }
2780         if (rt->rt_gw_family == AF_INET &&
2781             nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) {
2782                 goto nla_put_failure;
2783         } else if (rt->rt_gw_family == AF_INET6) {
2784                 int alen = sizeof(struct in6_addr);
2785                 struct nlattr *nla;
2786                 struct rtvia *via;
2787
2788                 nla = nla_reserve(skb, RTA_VIA, alen + 2);
2789                 if (!nla)
2790                         goto nla_put_failure;
2791
2792                 via = nla_data(nla);
2793                 via->rtvia_family = AF_INET6;
2794                 memcpy(via->rtvia_addr, &rt->rt_gw6, alen);
2795         }
2796
2797         expires = rt->dst.expires;
2798         if (expires) {
2799                 unsigned long now = jiffies;
2800
2801                 if (time_before(now, expires))
2802                         expires -= now;
2803                 else
2804                         expires = 0;
2805         }
2806
2807         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2808         if (rt->rt_pmtu && expires)
2809                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2810         if (rt->rt_mtu_locked && expires)
2811                 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2812         if (rtnetlink_put_metrics(skb, metrics) < 0)
2813                 goto nla_put_failure;
2814
2815         if (fl4) {
2816                 if (fl4->flowi4_mark &&
2817                     nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2818                         goto nla_put_failure;
2819
2820                 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2821                     nla_put_u32(skb, RTA_UID,
2822                                 from_kuid_munged(current_user_ns(),
2823                                                  fl4->flowi4_uid)))
2824                         goto nla_put_failure;
2825
2826                 if (rt_is_input_route(rt)) {
2827 #ifdef CONFIG_IP_MROUTE
2828                         if (ipv4_is_multicast(dst) &&
2829                             !ipv4_is_local_multicast(dst) &&
2830                             IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2831                                 int err = ipmr_get_route(net, skb,
2832                                                          fl4->saddr, fl4->daddr,
2833                                                          r, portid);
2834
2835                                 if (err <= 0) {
2836                                         if (err == 0)
2837                                                 return 0;
2838                                         goto nla_put_failure;
2839                                 }
2840                         } else
2841 #endif
2842                                 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif))
2843                                         goto nla_put_failure;
2844                 }
2845         }
2846
2847         error = rt->dst.error;
2848
2849         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2850                 goto nla_put_failure;
2851
2852         nlmsg_end(skb, nlh);
2853         return 0;
2854
2855 nla_put_failure:
2856         nlmsg_cancel(skb, nlh);
2857         return -EMSGSIZE;
2858 }
2859
2860 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb,
2861                             struct netlink_callback *cb, u32 table_id,
2862                             struct fnhe_hash_bucket *bucket, int genid,
2863                             int *fa_index, int fa_start)
2864 {
2865         int i;
2866
2867         for (i = 0; i < FNHE_HASH_SIZE; i++) {
2868                 struct fib_nh_exception *fnhe;
2869
2870                 for (fnhe = rcu_dereference(bucket[i].chain); fnhe;
2871                      fnhe = rcu_dereference(fnhe->fnhe_next)) {
2872                         struct rtable *rt;
2873                         int err;
2874
2875                         if (*fa_index < fa_start)
2876                                 goto next;
2877
2878                         if (fnhe->fnhe_genid != genid)
2879                                 goto next;
2880
2881                         if (fnhe->fnhe_expires &&
2882                             time_after(jiffies, fnhe->fnhe_expires))
2883                                 goto next;
2884
2885                         rt = rcu_dereference(fnhe->fnhe_rth_input);
2886                         if (!rt)
2887                                 rt = rcu_dereference(fnhe->fnhe_rth_output);
2888                         if (!rt)
2889                                 goto next;
2890
2891                         err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt,
2892                                            table_id, NULL, skb,
2893                                            NETLINK_CB(cb->skb).portid,
2894                                            cb->nlh->nlmsg_seq);
2895                         if (err)
2896                                 return err;
2897 next:
2898                         (*fa_index)++;
2899                 }
2900         }
2901
2902         return 0;
2903 }
2904
2905 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb,
2906                        u32 table_id, struct fib_info *fi,
2907                        int *fa_index, int fa_start)
2908 {
2909         struct net *net = sock_net(cb->skb->sk);
2910         int nhsel, genid = fnhe_genid(net);
2911
2912         for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
2913                 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
2914                 struct fnhe_hash_bucket *bucket;
2915                 int err;
2916
2917                 if (nhc->nhc_flags & RTNH_F_DEAD)
2918                         continue;
2919
2920                 rcu_read_lock();
2921                 bucket = rcu_dereference(nhc->nhc_exceptions);
2922                 err = 0;
2923                 if (bucket)
2924                         err = fnhe_dump_bucket(net, skb, cb, table_id, bucket,
2925                                                genid, fa_index, fa_start);
2926                 rcu_read_unlock();
2927                 if (err)
2928                         return err;
2929         }
2930
2931         return 0;
2932 }
2933
2934 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst,
2935                                                    u8 ip_proto, __be16 sport,
2936                                                    __be16 dport)
2937 {
2938         struct sk_buff *skb;
2939         struct iphdr *iph;
2940
2941         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2942         if (!skb)
2943                 return NULL;
2944
2945         /* Reserve room for dummy headers, this skb can pass
2946          * through good chunk of routing engine.
2947          */
2948         skb_reset_mac_header(skb);
2949         skb_reset_network_header(skb);
2950         skb->protocol = htons(ETH_P_IP);
2951         iph = skb_put(skb, sizeof(struct iphdr));
2952         iph->protocol = ip_proto;
2953         iph->saddr = src;
2954         iph->daddr = dst;
2955         iph->version = 0x4;
2956         iph->frag_off = 0;
2957         iph->ihl = 0x5;
2958         skb_set_transport_header(skb, skb->len);
2959
2960         switch (iph->protocol) {
2961         case IPPROTO_UDP: {
2962                 struct udphdr *udph;
2963
2964                 udph = skb_put_zero(skb, sizeof(struct udphdr));
2965                 udph->source = sport;
2966                 udph->dest = dport;
2967                 udph->len = sizeof(struct udphdr);
2968                 udph->check = 0;
2969                 break;
2970         }
2971         case IPPROTO_TCP: {
2972                 struct tcphdr *tcph;
2973
2974                 tcph = skb_put_zero(skb, sizeof(struct tcphdr));
2975                 tcph->source    = sport;
2976                 tcph->dest      = dport;
2977                 tcph->doff      = sizeof(struct tcphdr) / 4;
2978                 tcph->rst = 1;
2979                 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr),
2980                                             src, dst, 0);
2981                 break;
2982         }
2983         case IPPROTO_ICMP: {
2984                 struct icmphdr *icmph;
2985
2986                 icmph = skb_put_zero(skb, sizeof(struct icmphdr));
2987                 icmph->type = ICMP_ECHO;
2988                 icmph->code = 0;
2989         }
2990         }
2991
2992         return skb;
2993 }
2994
2995 static int inet_rtm_valid_getroute_req(struct sk_buff *skb,
2996                                        const struct nlmsghdr *nlh,
2997                                        struct nlattr **tb,
2998                                        struct netlink_ext_ack *extack)
2999 {
3000         struct rtmsg *rtm;
3001         int i, err;
3002
3003         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
3004                 NL_SET_ERR_MSG(extack,
3005                                "ipv4: Invalid header for route get request");
3006                 return -EINVAL;
3007         }
3008
3009         if (!netlink_strict_get_check(skb))
3010                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
3011                                               rtm_ipv4_policy, extack);
3012
3013         rtm = nlmsg_data(nlh);
3014         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
3015             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
3016             rtm->rtm_table || rtm->rtm_protocol ||
3017             rtm->rtm_scope || rtm->rtm_type) {
3018                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request");
3019                 return -EINVAL;
3020         }
3021
3022         if (rtm->rtm_flags & ~(RTM_F_NOTIFY |
3023                                RTM_F_LOOKUP_TABLE |
3024                                RTM_F_FIB_MATCH)) {
3025                 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request");
3026                 return -EINVAL;
3027         }
3028
3029         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
3030                                             rtm_ipv4_policy, extack);
3031         if (err)
3032                 return err;
3033
3034         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
3035             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
3036                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
3037                 return -EINVAL;
3038         }
3039
3040         for (i = 0; i <= RTA_MAX; i++) {
3041                 if (!tb[i])
3042                         continue;
3043
3044                 switch (i) {
3045                 case RTA_IIF:
3046                 case RTA_OIF:
3047                 case RTA_SRC:
3048                 case RTA_DST:
3049                 case RTA_IP_PROTO:
3050                 case RTA_SPORT:
3051                 case RTA_DPORT:
3052                 case RTA_MARK:
3053                 case RTA_UID:
3054                         break;
3055                 default:
3056                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request");
3057                         return -EINVAL;
3058                 }
3059         }
3060
3061         return 0;
3062 }
3063
3064 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3065                              struct netlink_ext_ack *extack)
3066 {
3067         struct net *net = sock_net(in_skb->sk);
3068         struct nlattr *tb[RTA_MAX+1];
3069         u32 table_id = RT_TABLE_MAIN;
3070         __be16 sport = 0, dport = 0;
3071         struct fib_result res = {};
3072         u8 ip_proto = IPPROTO_UDP;
3073         struct rtable *rt = NULL;
3074         struct sk_buff *skb;
3075         struct rtmsg *rtm;
3076         struct flowi4 fl4 = {};
3077         __be32 dst = 0;
3078         __be32 src = 0;
3079         kuid_t uid;
3080         u32 iif;
3081         int err;
3082         int mark;
3083
3084         err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
3085         if (err < 0)
3086                 return err;
3087
3088         rtm = nlmsg_data(nlh);
3089         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
3090         dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
3091         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
3092         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
3093         if (tb[RTA_UID])
3094                 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
3095         else
3096                 uid = (iif ? INVALID_UID : current_uid());
3097
3098         if (tb[RTA_IP_PROTO]) {
3099                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
3100                                                   &ip_proto, AF_INET, extack);
3101                 if (err)
3102                         return err;
3103         }
3104
3105         if (tb[RTA_SPORT])
3106                 sport = nla_get_be16(tb[RTA_SPORT]);
3107
3108         if (tb[RTA_DPORT])
3109                 dport = nla_get_be16(tb[RTA_DPORT]);
3110
3111         skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport);
3112         if (!skb)
3113                 return -ENOBUFS;
3114
3115         fl4.daddr = dst;
3116         fl4.saddr = src;
3117         fl4.flowi4_tos = rtm->rtm_tos;
3118         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
3119         fl4.flowi4_mark = mark;
3120         fl4.flowi4_uid = uid;
3121         if (sport)
3122                 fl4.fl4_sport = sport;
3123         if (dport)
3124                 fl4.fl4_dport = dport;
3125         fl4.flowi4_proto = ip_proto;
3126
3127         rcu_read_lock();
3128
3129         if (iif) {
3130                 struct net_device *dev;
3131
3132                 dev = dev_get_by_index_rcu(net, iif);
3133                 if (!dev) {
3134                         err = -ENODEV;
3135                         goto errout_rcu;
3136                 }
3137
3138                 fl4.flowi4_iif = iif; /* for rt_fill_info */
3139                 skb->dev        = dev;
3140                 skb->mark       = mark;
3141                 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
3142                                          dev, &res);
3143
3144                 rt = skb_rtable(skb);
3145                 if (err == 0 && rt->dst.error)
3146                         err = -rt->dst.error;
3147         } else {
3148                 fl4.flowi4_iif = LOOPBACK_IFINDEX;
3149                 skb->dev = net->loopback_dev;
3150                 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
3151                 err = 0;
3152                 if (IS_ERR(rt))
3153                         err = PTR_ERR(rt);
3154                 else
3155                         skb_dst_set(skb, &rt->dst);
3156         }
3157
3158         if (err)
3159                 goto errout_rcu;
3160
3161         if (rtm->rtm_flags & RTM_F_NOTIFY)
3162                 rt->rt_flags |= RTCF_NOTIFY;
3163
3164         if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
3165                 table_id = res.table ? res.table->tb_id : 0;
3166
3167         /* reset skb for netlink reply msg */
3168         skb_trim(skb, 0);
3169         skb_reset_network_header(skb);
3170         skb_reset_transport_header(skb);
3171         skb_reset_mac_header(skb);
3172
3173         if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
3174                 if (!res.fi) {
3175                         err = fib_props[res.type].error;
3176                         if (!err)
3177                                 err = -EHOSTUNREACH;
3178                         goto errout_rcu;
3179                 }
3180                 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
3181                                     nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
3182                                     rt->rt_type, res.prefix, res.prefixlen,
3183                                     fl4.flowi4_tos, res.fi, 0);
3184         } else {
3185                 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb,
3186                                    NETLINK_CB(in_skb).portid, nlh->nlmsg_seq);
3187         }
3188         if (err < 0)
3189                 goto errout_rcu;
3190
3191         rcu_read_unlock();
3192
3193         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3194
3195 errout_free:
3196         return err;
3197 errout_rcu:
3198         rcu_read_unlock();
3199         kfree_skb(skb);
3200         goto errout_free;
3201 }
3202
3203 void ip_rt_multicast_event(struct in_device *in_dev)
3204 {
3205         rt_cache_flush(dev_net(in_dev->dev));
3206 }
3207
3208 #ifdef CONFIG_SYSCTL
3209 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
3210 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
3211 static int ip_rt_gc_elasticity __read_mostly    = 8;
3212 static int ip_min_valid_pmtu __read_mostly      = IPV4_MIN_MTU;
3213
3214 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
3215                                         void __user *buffer,
3216                                         size_t *lenp, loff_t *ppos)
3217 {
3218         struct net *net = (struct net *)__ctl->extra1;
3219
3220         if (write) {
3221                 rt_cache_flush(net);
3222                 fnhe_genid_bump(net);
3223                 return 0;
3224         }
3225
3226         return -EINVAL;
3227 }
3228
3229 static struct ctl_table ipv4_route_table[] = {
3230         {
3231                 .procname       = "gc_thresh",
3232                 .data           = &ipv4_dst_ops.gc_thresh,
3233                 .maxlen         = sizeof(int),
3234                 .mode           = 0644,
3235                 .proc_handler   = proc_dointvec,
3236         },
3237         {
3238                 .procname       = "max_size",
3239                 .data           = &ip_rt_max_size,
3240                 .maxlen         = sizeof(int),
3241                 .mode           = 0644,
3242                 .proc_handler   = proc_dointvec,
3243         },
3244         {
3245                 /*  Deprecated. Use gc_min_interval_ms */
3246
3247                 .procname       = "gc_min_interval",
3248                 .data           = &ip_rt_gc_min_interval,
3249                 .maxlen         = sizeof(int),
3250                 .mode           = 0644,
3251                 .proc_handler   = proc_dointvec_jiffies,
3252         },
3253         {
3254                 .procname       = "gc_min_interval_ms",
3255                 .data           = &ip_rt_gc_min_interval,
3256                 .maxlen         = sizeof(int),
3257                 .mode           = 0644,
3258                 .proc_handler   = proc_dointvec_ms_jiffies,
3259         },
3260         {
3261                 .procname       = "gc_timeout",
3262                 .data           = &ip_rt_gc_timeout,
3263                 .maxlen         = sizeof(int),
3264                 .mode           = 0644,
3265                 .proc_handler   = proc_dointvec_jiffies,
3266         },
3267         {
3268                 .procname       = "gc_interval",
3269                 .data           = &ip_rt_gc_interval,
3270                 .maxlen         = sizeof(int),
3271                 .mode           = 0644,
3272                 .proc_handler   = proc_dointvec_jiffies,
3273         },
3274         {
3275                 .procname       = "redirect_load",
3276                 .data           = &ip_rt_redirect_load,
3277                 .maxlen         = sizeof(int),
3278                 .mode           = 0644,
3279                 .proc_handler   = proc_dointvec,
3280         },
3281         {
3282                 .procname       = "redirect_number",
3283                 .data           = &ip_rt_redirect_number,
3284                 .maxlen         = sizeof(int),
3285                 .mode           = 0644,
3286                 .proc_handler   = proc_dointvec,
3287         },
3288         {
3289                 .procname       = "redirect_silence",
3290                 .data           = &ip_rt_redirect_silence,
3291                 .maxlen         = sizeof(int),
3292                 .mode           = 0644,
3293                 .proc_handler   = proc_dointvec,
3294         },
3295         {
3296                 .procname       = "error_cost",
3297                 .data           = &ip_rt_error_cost,
3298                 .maxlen         = sizeof(int),
3299                 .mode           = 0644,
3300                 .proc_handler   = proc_dointvec,
3301         },
3302         {
3303                 .procname       = "error_burst",
3304                 .data           = &ip_rt_error_burst,
3305                 .maxlen         = sizeof(int),
3306                 .mode           = 0644,
3307                 .proc_handler   = proc_dointvec,
3308         },
3309         {
3310                 .procname       = "gc_elasticity",
3311                 .data           = &ip_rt_gc_elasticity,
3312                 .maxlen         = sizeof(int),
3313                 .mode           = 0644,
3314                 .proc_handler   = proc_dointvec,
3315         },
3316         {
3317                 .procname       = "mtu_expires",
3318                 .data           = &ip_rt_mtu_expires,
3319                 .maxlen         = sizeof(int),
3320                 .mode           = 0644,
3321                 .proc_handler   = proc_dointvec_jiffies,
3322         },
3323         {
3324                 .procname       = "min_pmtu",
3325                 .data           = &ip_rt_min_pmtu,
3326                 .maxlen         = sizeof(int),
3327                 .mode           = 0644,
3328                 .proc_handler   = proc_dointvec_minmax,
3329                 .extra1         = &ip_min_valid_pmtu,
3330         },
3331         {
3332                 .procname       = "min_adv_mss",
3333                 .data           = &ip_rt_min_advmss,
3334                 .maxlen         = sizeof(int),
3335                 .mode           = 0644,
3336                 .proc_handler   = proc_dointvec,
3337         },
3338         { }
3339 };
3340
3341 static const char ipv4_route_flush_procname[] = "flush";
3342
3343 static struct ctl_table ipv4_route_flush_table[] = {
3344         {
3345                 .procname       = ipv4_route_flush_procname,
3346                 .maxlen         = sizeof(int),
3347                 .mode           = 0200,
3348                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3349         },
3350         { },
3351 };
3352
3353 static __net_init int sysctl_route_net_init(struct net *net)
3354 {
3355         struct ctl_table *tbl;
3356
3357         tbl = ipv4_route_flush_table;
3358         if (!net_eq(net, &init_net)) {
3359                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3360                 if (!tbl)
3361                         goto err_dup;
3362
3363                 /* Don't export non-whitelisted sysctls to unprivileged users */
3364                 if (net->user_ns != &init_user_ns) {
3365                         if (tbl[0].procname != ipv4_route_flush_procname)
3366                                 tbl[0].procname = NULL;
3367                 }
3368         }
3369         tbl[0].extra1 = net;
3370
3371         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3372         if (!net->ipv4.route_hdr)
3373                 goto err_reg;
3374         return 0;
3375
3376 err_reg:
3377         if (tbl != ipv4_route_flush_table)
3378                 kfree(tbl);
3379 err_dup:
3380         return -ENOMEM;
3381 }
3382
3383 static __net_exit void sysctl_route_net_exit(struct net *net)
3384 {
3385         struct ctl_table *tbl;
3386
3387         tbl = net->ipv4.route_hdr->ctl_table_arg;
3388         unregister_net_sysctl_table(net->ipv4.route_hdr);
3389         BUG_ON(tbl == ipv4_route_flush_table);
3390         kfree(tbl);
3391 }
3392
3393 static __net_initdata struct pernet_operations sysctl_route_ops = {
3394         .init = sysctl_route_net_init,
3395         .exit = sysctl_route_net_exit,
3396 };
3397 #endif
3398
3399 static __net_init int rt_genid_init(struct net *net)
3400 {
3401         atomic_set(&net->ipv4.rt_genid, 0);
3402         atomic_set(&net->fnhe_genid, 0);
3403         atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3404         return 0;
3405 }
3406
3407 static __net_initdata struct pernet_operations rt_genid_ops = {
3408         .init = rt_genid_init,
3409 };
3410
3411 static int __net_init ipv4_inetpeer_init(struct net *net)
3412 {
3413         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3414
3415         if (!bp)
3416                 return -ENOMEM;
3417         inet_peer_base_init(bp);
3418         net->ipv4.peers = bp;
3419         return 0;
3420 }
3421
3422 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3423 {
3424         struct inet_peer_base *bp = net->ipv4.peers;
3425
3426         net->ipv4.peers = NULL;
3427         inetpeer_invalidate_tree(bp);
3428         kfree(bp);
3429 }
3430
3431 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3432         .init   =       ipv4_inetpeer_init,
3433         .exit   =       ipv4_inetpeer_exit,
3434 };
3435
3436 #ifdef CONFIG_IP_ROUTE_CLASSID
3437 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3438 #endif /* CONFIG_IP_ROUTE_CLASSID */
3439
3440 int __init ip_rt_init(void)
3441 {
3442         int cpu;
3443
3444         ip_idents = kmalloc_array(IP_IDENTS_SZ, sizeof(*ip_idents),
3445                                   GFP_KERNEL);
3446         if (!ip_idents)
3447                 panic("IP: failed to allocate ip_idents\n");
3448
3449         prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3450
3451         ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3452         if (!ip_tstamps)
3453                 panic("IP: failed to allocate ip_tstamps\n");
3454
3455         for_each_possible_cpu(cpu) {
3456                 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3457
3458                 INIT_LIST_HEAD(&ul->head);
3459                 spin_lock_init(&ul->lock);
3460         }
3461 #ifdef CONFIG_IP_ROUTE_CLASSID
3462         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3463         if (!ip_rt_acct)
3464                 panic("IP: failed to allocate ip_rt_acct\n");
3465 #endif
3466
3467         ipv4_dst_ops.kmem_cachep =
3468                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3469                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3470
3471         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3472
3473         if (dst_entries_init(&ipv4_dst_ops) < 0)
3474                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3475
3476         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3477                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3478
3479         ipv4_dst_ops.gc_thresh = ~0;
3480         ip_rt_max_size = INT_MAX;
3481
3482         devinet_init();
3483         ip_fib_init();
3484
3485         if (ip_rt_proc_init())
3486                 pr_err("Unable to create route proc files\n");
3487 #ifdef CONFIG_XFRM
3488         xfrm_init();
3489         xfrm4_init();
3490 #endif
3491         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3492                       RTNL_FLAG_DOIT_UNLOCKED);
3493
3494 #ifdef CONFIG_SYSCTL
3495         register_pernet_subsys(&sysctl_route_ops);
3496 #endif
3497         register_pernet_subsys(&rt_genid_ops);
3498         register_pernet_subsys(&ipv4_inetpeer_ops);
3499         return 0;
3500 }
3501
3502 #ifdef CONFIG_SYSCTL
3503 /*
3504  * We really need to sanitize the damn ipv4 init order, then all
3505  * this nonsense will go away.
3506  */
3507 void __init ip_static_sysctl_init(void)
3508 {
3509         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3510 }
3511 #endif