net/ipv4/route.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              ROUTE - implementation of the IP router.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *              Linus Torvalds, <Linus.Torvalds@helsinki.fi>
  12  *              Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  13  *
  14  * Fixes:
  15  *              Alan Cox        :       Verify area fixes.
  16  *              Alan Cox        :       cli() protects routing changes
  17  *              Rui Oliveira    :       ICMP routing table updates
  18  *              (rco@di.uminho.pt)      Routing table insertion and update
  19  *              Linus Torvalds  :       Rewrote bits to be sensible
  20  *              Alan Cox        :       Added BSD route gw semantics
  21  *              Alan Cox        :       Super /proc >4K
  22  *              Alan Cox        :       MTU in route table
  23  *              Alan Cox        :       MSS actually. Also added the window
  24  *                                      clamper.
  25  *              Sam Lantinga    :       Fixed route matching in rt_del()
  26  *              Alan Cox        :       Routing cache support.
  27  *              Alan Cox        :       Removed compatibility cruft.
  28  *              Alan Cox        :       RTF_REJECT support.
  29  *              Alan Cox        :       TCP irtt support.
  30  *              Jonathan Naylor :       Added Metric support.
  31  *      Miquel van Smoorenburg  :       BSD API fixes.
  32  *      Miquel van Smoorenburg  :       Metrics.
  33  *              Alan Cox        :       Use __u32 properly
  34  *              Alan Cox        :       Aligned routing errors more closely with BSD
  35  *                                      our system is still very different.
  36  *              Alan Cox        :       Faster /proc handling
  37  *      Alexey Kuznetsov        :       Massive rework to support tree based routing,
  38  *                                      routing caches and better behaviour.
  39  *
  40  *              Olaf Erb        :       irtt wasn't being copied right.
  41  *              Bjorn Ekwall    :       Kerneld route support.
  42  *              Alan Cox        :       Multicast fixed (I hope)
  43  *              Pavel Krauz     :       Limited broadcast fixed
  44  *              Mike McLagan    :       Routing by source
  45  *      Alexey Kuznetsov        :       End of old history. Split to fib.c and
  46  *                                      route.c and rewritten from scratch.
  47  *              Andi Kleen      :       Load-limit warning messages.
  48  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  49  *      Vitaly E. Lavrov        :       Race condition in ip_route_input_slow.
  50  *      Tobias Ringstrom        :       Uninitialized res.type in ip_route_output_slow.
  51  *      Vladimir V. Ivanov      :       IP rule info (flowid) is really useful.
  52  *              Marc Boucher    :       routing by fwmark
  53  *      Robert Olsson           :       Added rt_cache statistics
  54  *      Arnaldo C. Melo         :       Convert proc stuff to seq_file
  55  *      Eric Dumazet            :       hashed spinlocks and rt_check_expire() fixes.
  56  *      Ilia Sotnikov           :       Ignore TOS on PMTUD and Redirect
  57  *      Ilia Sotnikov           :       Removed TOS from hash calculations
  58  *
  59  *              This program is free software; you can redistribute it and/or
  60  *              modify it under the terms of the GNU General Public License
  61  *              as published by the Free Software Foundation; either version
  62  *              2 of the License, or (at your option) any later version.
  63  */
  64
  65 #include <linux/module.h>
  66 #include <asm/uaccess.h>
  67 #include <asm/system.h>
  68 #include <linux/bitops.h>
  69 #include <linux/types.h>
  70 #include <linux/kernel.h>
  71 #include <linux/mm.h>
  72 #include <linux/bootmem.h>
  73 #include <linux/string.h>
  74 #include <linux/socket.h>
  75 #include <linux/sockios.h>
  76 #include <linux/errno.h>
  77 #include <linux/in.h>
  78 #include <linux/inet.h>
  79 #include <linux/netdevice.h>
  80 #include <linux/proc_fs.h>
  81 #include <linux/init.h>
  82 #include <linux/workqueue.h>
  83 #include <linux/skbuff.h>
  84 #include <linux/inetdevice.h>
  85 #include <linux/igmp.h>
  86 #include <linux/pkt_sched.h>
  87 #include <linux/mroute.h>
  88 #include <linux/netfilter_ipv4.h>
  89 #include <linux/random.h>
  90 #include <linux/jhash.h>
  91 #include <linux/rcupdate.h>
  92 #include <linux/times.h>
  93 #include <linux/slab.h>
  94 #include <net/dst.h>
  95 #include <net/net_namespace.h>
  96 #include <net/protocol.h>
  97 #include <net/ip.h>
  98 #include <net/route.h>
  99 #include <net/inetpeer.h>
 100 #include <net/sock.h>
 101 #include <net/ip_fib.h>
 102 #include <net/arp.h>
 103 #include <net/tcp.h>
 104 #include <net/icmp.h>
 105 #include <net/xfrm.h>
 106 #include <net/netevent.h>
 107 #include <net/rtnetlink.h>
 108 #ifdef CONFIG_SYSCTL
 109 #include <linux/sysctl.h>
 110 #endif
 111
 112 #define RT_FL_TOS(oldflp4) \
 113     ((u32)(oldflp4->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
 114
 115 #define IP_MAX_MTU      0xFFF0
 116
 117 #define RT_GC_TIMEOUT (300*HZ)
 118
 119 static int ip_rt_max_size;
 120 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
 121 static int ip_rt_gc_interval __read_mostly      = 60 * HZ;
 122 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
 123 static int ip_rt_redirect_number __read_mostly  = 9;
 124 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
 125 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
 126 static int ip_rt_error_cost __read_mostly       = HZ;
 127 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
 128 static int ip_rt_gc_elasticity __read_mostly    = 8;
 129 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
 130 static int ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
 131 static int ip_rt_min_advmss __read_mostly       = 256;
 132 static int rt_chain_length_max __read_mostly    = 20;
 133
 134 /*
 135  *      Interface to generic destination cache.
 136  */
 137
 138 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 139 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 140 static unsigned int      ipv4_default_mtu(const struct dst_entry *dst);
 141 static void              ipv4_dst_destroy(struct dst_entry *dst);
 142 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 143 static void              ipv4_link_failure(struct sk_buff *skb);
 144 static void              ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
 145 static int rt_garbage_collect(struct dst_ops *ops);
 146
 147 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 148                             int how)
 149 {
 150 }
 151
 152 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
 153 {
 154         struct rtable *rt = (struct rtable *) dst;
 155         struct inet_peer *peer;
 156         u32 *p = NULL;
 157
 158         if (!rt->peer)
 159                 rt_bind_peer(rt, 1);
 160
 161         peer = rt->peer;
 162         if (peer) {
 163                 u32 *old_p = __DST_METRICS_PTR(old);
 164                 unsigned long prev, new;
 165
 166                 p = peer->metrics;
 167                 if (inet_metrics_new(peer))
 168                         memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
 169
 170                 new = (unsigned long) p;
 171                 prev = cmpxchg(&dst->_metrics, old, new);
 172
 173                 if (prev != old) {
 174                         p = __DST_METRICS_PTR(prev);
 175                         if (prev & DST_METRICS_READ_ONLY)
 176                                 p = NULL;
 177                 } else {
 178                         if (rt->fi) {
 179                                 fib_info_put(rt->fi);
 180                                 rt->fi = NULL;
 181                         }
 182                 }
 183         }
 184         return p;
 185 }
 186
 187 static struct dst_ops ipv4_dst_ops = {
 188         .family =               AF_INET,
 189         .protocol =             cpu_to_be16(ETH_P_IP),
 190         .gc =                   rt_garbage_collect,
 191         .check =                ipv4_dst_check,
 192         .default_advmss =       ipv4_default_advmss,
 193         .default_mtu =          ipv4_default_mtu,
 194         .cow_metrics =          ipv4_cow_metrics,
 195         .destroy =              ipv4_dst_destroy,
 196         .ifdown =               ipv4_dst_ifdown,
 197         .negative_advice =      ipv4_negative_advice,
 198         .link_failure =         ipv4_link_failure,
 199         .update_pmtu =          ip_rt_update_pmtu,
 200         .local_out =            __ip_local_out,
 201 };
 202
 203 #define ECN_OR_COST(class)      TC_PRIO_##class
 204
 205 const __u8 ip_tos2prio[16] = {
 206         TC_PRIO_BESTEFFORT,
 207         ECN_OR_COST(FILLER),
 208         TC_PRIO_BESTEFFORT,
 209         ECN_OR_COST(BESTEFFORT),
 210         TC_PRIO_BULK,
 211         ECN_OR_COST(BULK),
 212         TC_PRIO_BULK,
 213         ECN_OR_COST(BULK),
 214         TC_PRIO_INTERACTIVE,
 215         ECN_OR_COST(INTERACTIVE),
 216         TC_PRIO_INTERACTIVE,
 217         ECN_OR_COST(INTERACTIVE),
 218         TC_PRIO_INTERACTIVE_BULK,
 219         ECN_OR_COST(INTERACTIVE_BULK),
 220         TC_PRIO_INTERACTIVE_BULK,
 221         ECN_OR_COST(INTERACTIVE_BULK)
 222 };
 223
 224
 225 /*
 226  * Route cache.
 227  */
 228
 229 /* The locking scheme is rather straight forward:
 230  *
 231  * 1) Read-Copy Update protects the buckets of the central route hash.
 232  * 2) Only writers remove entries, and they hold the lock
 233  *    as they look at rtable reference counts.
 234  * 3) Only readers acquire references to rtable entries,
 235  *    they do so with atomic increments and with the
 236  *    lock held.
 237  */
 238
 239 struct rt_hash_bucket {
 240         struct rtable __rcu     *chain;
 241 };
 242
 243 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
 244         defined(CONFIG_PROVE_LOCKING)
 245 /*
 246  * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
 247  * The size of this table is a power of two and depends on the number of CPUS.
 248  * (on lockdep we have a quite big spinlock_t, so keep the size down there)
 249  */
 250 #ifdef CONFIG_LOCKDEP
 251 # define RT_HASH_LOCK_SZ        256
 252 #else
 253 # if NR_CPUS >= 32
 254 #  define RT_HASH_LOCK_SZ       4096
 255 # elif NR_CPUS >= 16
 256 #  define RT_HASH_LOCK_SZ       2048
 257 # elif NR_CPUS >= 8
 258 #  define RT_HASH_LOCK_SZ       1024
 259 # elif NR_CPUS >= 4
 260 #  define RT_HASH_LOCK_SZ       512
 261 # else
 262 #  define RT_HASH_LOCK_SZ       256
 263 # endif
 264 #endif
 265
 266 static spinlock_t       *rt_hash_locks;
 267 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
 268
 269 static __init void rt_hash_lock_init(void)
 270 {
 271         int i;
 272
 273         rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
 274                         GFP_KERNEL);
 275         if (!rt_hash_locks)
 276                 panic("IP: failed to allocate rt_hash_locks\n");
 277
 278         for (i = 0; i < RT_HASH_LOCK_SZ; i++)
 279                 spin_lock_init(&rt_hash_locks[i]);
 280 }
 281 #else
 282 # define rt_hash_lock_addr(slot) NULL
 283
 284 static inline void rt_hash_lock_init(void)
 285 {
 286 }
 287 #endif
 288
 289 static struct rt_hash_bucket    *rt_hash_table __read_mostly;
 290 static unsigned                 rt_hash_mask __read_mostly;
 291 static unsigned int             rt_hash_log  __read_mostly;
 292
 293 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 294 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
 295
 296 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
 297                                    int genid)
 298 {
 299         return jhash_3words((__force u32)daddr, (__force u32)saddr,
 300                             idx, genid)
 301                 & rt_hash_mask;
 302 }
 303
 304 static inline int rt_genid(struct net *net)
 305 {
 306         return atomic_read(&net->ipv4.rt_genid);
 307 }
 308
 309 #ifdef CONFIG_PROC_FS
 310 struct rt_cache_iter_state {
 311         struct seq_net_private p;
 312         int bucket;
 313         int genid;
 314 };
 315
 316 static struct rtable *rt_cache_get_first(struct seq_file *seq)
 317 {
 318         struct rt_cache_iter_state *st = seq->private;
 319         struct rtable *r = NULL;
 320
 321         for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
 322                 if (!rcu_dereference_raw(rt_hash_table[st->bucket].chain))
 323                         continue;
 324                 rcu_read_lock_bh();
 325                 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
 326                 while (r) {
 327                         if (dev_net(r->dst.dev) == seq_file_net(seq) &&
 328                             r->rt_genid == st->genid)
 329                                 return r;
 330                         r = rcu_dereference_bh(r->dst.rt_next);
 331                 }
 332                 rcu_read_unlock_bh();
 333         }
 334         return r;
 335 }
 336
 337 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
 338                                           struct rtable *r)
 339 {
 340         struct rt_cache_iter_state *st = seq->private;
 341
 342         r = rcu_dereference_bh(r->dst.rt_next);
 343         while (!r) {
 344                 rcu_read_unlock_bh();
 345                 do {
 346                         if (--st->bucket < 0)
 347                                 return NULL;
 348                 } while (!rcu_dereference_raw(rt_hash_table[st->bucket].chain));
 349                 rcu_read_lock_bh();
 350                 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
 351         }
 352         return r;
 353 }
 354
 355 static struct rtable *rt_cache_get_next(struct seq_file *seq,
 356                                         struct rtable *r)
 357 {
 358         struct rt_cache_iter_state *st = seq->private;
 359         while ((r = __rt_cache_get_next(seq, r)) != NULL) {
 360                 if (dev_net(r->dst.dev) != seq_file_net(seq))
 361                         continue;
 362                 if (r->rt_genid == st->genid)
 363                         break;
 364         }
 365         return r;
 366 }
 367
 368 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
 369 {
 370         struct rtable *r = rt_cache_get_first(seq);
 371
 372         if (r)
 373                 while (pos && (r = rt_cache_get_next(seq, r)))
 374                         --pos;
 375         return pos ? NULL : r;
 376 }
 377
 378 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 379 {
 380         struct rt_cache_iter_state *st = seq->private;
 381         if (*pos)
 382                 return rt_cache_get_idx(seq, *pos - 1);
 383         st->genid = rt_genid(seq_file_net(seq));
 384         return SEQ_START_TOKEN;
 385 }
 386
 387 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 388 {
 389         struct rtable *r;
 390
 391         if (v == SEQ_START_TOKEN)
 392                 r = rt_cache_get_first(seq);
 393         else
 394                 r = rt_cache_get_next(seq, v);
 395         ++*pos;
 396         return r;
 397 }
 398
 399 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 400 {
 401         if (v && v != SEQ_START_TOKEN)
 402                 rcu_read_unlock_bh();
 403 }
 404
 405 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 406 {
 407         if (v == SEQ_START_TOKEN)
 408                 seq_printf(seq, "%-127s\n",
 409                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 410                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 411                            "HHUptod\tSpecDst");
 412         else {
 413                 struct rtable *r = v;
 414                 int len;
 415
 416                 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
 417                               "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
 418                         r->dst.dev ? r->dst.dev->name : "*",
 419                         (__force u32)r->rt_dst,
 420                         (__force u32)r->rt_gateway,
 421                         r->rt_flags, atomic_read(&r->dst.__refcnt),
 422                         r->dst.__use, 0, (__force u32)r->rt_src,
 423                         dst_metric_advmss(&r->dst) + 40,
 424                         dst_metric(&r->dst, RTAX_WINDOW),
 425                         (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
 426                               dst_metric(&r->dst, RTAX_RTTVAR)),
 427                         r->rt_tos,
 428                         r->dst.hh ? atomic_read(&r->dst.hh->hh_refcnt) : -1,
 429                         r->dst.hh ? (r->dst.hh->hh_output ==
 430                                        dev_queue_xmit) : 0,
 431                         r->rt_spec_dst, &len);
 432
 433                 seq_printf(seq, "%*s\n", 127 - len, "");
 434         }
 435         return 0;
 436 }
 437
 438 static const struct seq_operations rt_cache_seq_ops = {
 439         .start  = rt_cache_seq_start,
 440         .next   = rt_cache_seq_next,
 441         .stop   = rt_cache_seq_stop,
 442         .show   = rt_cache_seq_show,
 443 };
 444
 445 static int rt_cache_seq_open(struct inode *inode, struct file *file)
 446 {
 447         return seq_open_net(inode, file, &rt_cache_seq_ops,
 448                         sizeof(struct rt_cache_iter_state));
 449 }
 450
 451 static const struct file_operations rt_cache_seq_fops = {
 452         .owner   = THIS_MODULE,
 453         .open    = rt_cache_seq_open,
 454         .read    = seq_read,
 455         .llseek  = seq_lseek,
 456         .release = seq_release_net,
 457 };
 458
 459
 460 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 461 {
 462         int cpu;
 463
 464         if (*pos == 0)
 465                 return SEQ_START_TOKEN;
 466
 467         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 468                 if (!cpu_possible(cpu))
 469                         continue;
 470                 *pos = cpu+1;
 471                 return &per_cpu(rt_cache_stat, cpu);
 472         }
 473         return NULL;
 474 }
 475
 476 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 477 {
 478         int cpu;
 479
 480         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 481                 if (!cpu_possible(cpu))
 482                         continue;
 483                 *pos = cpu+1;
 484                 return &per_cpu(rt_cache_stat, cpu);
 485         }
 486         return NULL;
 487
 488 }
 489
 490 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 491 {
 492
 493 }
 494
 495 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 496 {
 497         struct rt_cache_stat *st = v;
 498
 499         if (v == SEQ_START_TOKEN) {
 500                 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");
 501                 return 0;
 502         }
 503
 504         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 505                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 506                    dst_entries_get_slow(&ipv4_dst_ops),
 507                    st->in_hit,
 508                    st->in_slow_tot,
 509                    st->in_slow_mc,
 510                    st->in_no_route,
 511                    st->in_brd,
 512                    st->in_martian_dst,
 513                    st->in_martian_src,
 514
 515                    st->out_hit,
 516                    st->out_slow_tot,
 517                    st->out_slow_mc,
 518
 519                    st->gc_total,
 520                    st->gc_ignored,
 521                    st->gc_goal_miss,
 522                    st->gc_dst_overflow,
 523                    st->in_hlist_search,
 524                    st->out_hlist_search
 525                 );
 526         return 0;
 527 }
 528
 529 static const struct seq_operations rt_cpu_seq_ops = {
 530         .start  = rt_cpu_seq_start,
 531         .next   = rt_cpu_seq_next,
 532         .stop   = rt_cpu_seq_stop,
 533         .show   = rt_cpu_seq_show,
 534 };
 535
 536
 537 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
 538 {
 539         return seq_open(file, &rt_cpu_seq_ops);
 540 }
 541
 542 static const struct file_operations rt_cpu_seq_fops = {
 543         .owner   = THIS_MODULE,
 544         .open    = rt_cpu_seq_open,
 545         .read    = seq_read,
 546         .llseek  = seq_lseek,
 547         .release = seq_release,
 548 };
 549
 550 #ifdef CONFIG_IP_ROUTE_CLASSID
 551 static int rt_acct_proc_show(struct seq_file *m, void *v)
 552 {
 553         struct ip_rt_acct *dst, *src;
 554         unsigned int i, j;
 555
 556         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 557         if (!dst)
 558                 return -ENOMEM;
 559
 560         for_each_possible_cpu(i) {
 561                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 562                 for (j = 0; j < 256; j++) {
 563                         dst[j].o_bytes   += src[j].o_bytes;
 564                         dst[j].o_packets += src[j].o_packets;
 565                         dst[j].i_bytes   += src[j].i_bytes;
 566                         dst[j].i_packets += src[j].i_packets;
 567                 }
 568         }
 569
 570         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 571         kfree(dst);
 572         return 0;
 573 }
 574
 575 static int rt_acct_proc_open(struct inode *inode, struct file *file)
 576 {
 577         return single_open(file, rt_acct_proc_show, NULL);
 578 }
 579
 580 static const struct file_operations rt_acct_proc_fops = {
 581         .owner          = THIS_MODULE,
 582         .open           = rt_acct_proc_open,
 583         .read           = seq_read,
 584         .llseek         = seq_lseek,
 585         .release        = single_release,
 586 };
 587 #endif
 588
 589 static int __net_init ip_rt_do_proc_init(struct net *net)
 590 {
 591         struct proc_dir_entry *pde;
 592
 593         pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
 594                         &rt_cache_seq_fops);
 595         if (!pde)
 596                 goto err1;
 597
 598         pde = proc_create("rt_cache", S_IRUGO,
 599                           net->proc_net_stat, &rt_cpu_seq_fops);
 600         if (!pde)
 601                 goto err2;
 602
 603 #ifdef CONFIG_IP_ROUTE_CLASSID
 604         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
 605         if (!pde)
 606                 goto err3;
 607 #endif
 608         return 0;
 609
 610 #ifdef CONFIG_IP_ROUTE_CLASSID
 611 err3:
 612         remove_proc_entry("rt_cache", net->proc_net_stat);
 613 #endif
 614 err2:
 615         remove_proc_entry("rt_cache", net->proc_net);
 616 err1:
 617         return -ENOMEM;
 618 }
 619
 620 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 621 {
 622         remove_proc_entry("rt_cache", net->proc_net_stat);
 623         remove_proc_entry("rt_cache", net->proc_net);
 624 #ifdef CONFIG_IP_ROUTE_CLASSID
 625         remove_proc_entry("rt_acct", net->proc_net);
 626 #endif
 627 }
 628
 629 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 630         .init = ip_rt_do_proc_init,
 631         .exit = ip_rt_do_proc_exit,
 632 };
 633
 634 static int __init ip_rt_proc_init(void)
 635 {
 636         return register_pernet_subsys(&ip_rt_proc_ops);
 637 }
 638
 639 #else
 640 static inline int ip_rt_proc_init(void)
 641 {
 642         return 0;
 643 }
 644 #endif /* CONFIG_PROC_FS */
 645
 646 static inline void rt_free(struct rtable *rt)
 647 {
 648         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
 649 }
 650
 651 static inline void rt_drop(struct rtable *rt)
 652 {
 653         ip_rt_put(rt);
 654         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
 655 }
 656
 657 static inline int rt_fast_clean(struct rtable *rth)
 658 {
 659         /* Kill broadcast/multicast entries very aggresively, if they
 660            collide in hash table with more useful entries */
 661         return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
 662                 rt_is_input_route(rth) && rth->dst.rt_next;
 663 }
 664
 665 static inline int rt_valuable(struct rtable *rth)
 666 {
 667         return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
 668                 (rth->peer && rth->peer->pmtu_expires);
 669 }
 670
 671 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
 672 {
 673         unsigned long age;
 674         int ret = 0;
 675
 676         if (atomic_read(&rth->dst.__refcnt))
 677                 goto out;
 678
 679         age = jiffies - rth->dst.lastuse;
 680         if ((age <= tmo1 && !rt_fast_clean(rth)) ||
 681             (age <= tmo2 && rt_valuable(rth)))
 682                 goto out;
 683         ret = 1;
 684 out:    return ret;
 685 }
 686
 687 /* Bits of score are:
 688  * 31: very valuable
 689  * 30: not quite useless
 690  * 29..0: usage counter
 691  */
 692 static inline u32 rt_score(struct rtable *rt)
 693 {
 694         u32 score = jiffies - rt->dst.lastuse;
 695
 696         score = ~score & ~(3<<30);
 697
 698         if (rt_valuable(rt))
 699                 score |= (1<<31);
 700
 701         if (rt_is_output_route(rt) ||
 702             !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
 703                 score |= (1<<30);
 704
 705         return score;
 706 }
 707
 708 static inline bool rt_caching(const struct net *net)
 709 {
 710         return net->ipv4.current_rt_cache_rebuild_count <=
 711                 net->ipv4.sysctl_rt_cache_rebuild_count;
 712 }
 713
 714 static inline bool compare_hash_inputs(const struct rtable *rt1,
 715                                        const struct rtable *rt2)
 716 {
 717         return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
 718                 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
 719                 (rt1->rt_iif ^ rt2->rt_iif)) == 0);
 720 }
 721
 722 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
 723 {
 724         return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
 725                 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
 726                 (rt1->rt_mark ^ rt2->rt_mark) |
 727                 (rt1->rt_tos ^ rt2->rt_tos) |
 728                 (rt1->rt_oif ^ rt2->rt_oif) |
 729                 (rt1->rt_iif ^ rt2->rt_iif)) == 0;
 730 }
 731
 732 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
 733 {
 734         return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
 735 }
 736
 737 static inline int rt_is_expired(struct rtable *rth)
 738 {
 739         return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
 740 }
 741
 742 /*
 743  * Perform a full scan of hash table and free all entries.
 744  * Can be called by a softirq or a process.
 745  * In the later case, we want to be reschedule if necessary
 746  */
 747 static void rt_do_flush(struct net *net, int process_context)
 748 {
 749         unsigned int i;
 750         struct rtable *rth, *next;
 751
 752         for (i = 0; i <= rt_hash_mask; i++) {
 753                 struct rtable __rcu **pprev;
 754                 struct rtable *list;
 755
 756                 if (process_context && need_resched())
 757                         cond_resched();
 758                 rth = rcu_dereference_raw(rt_hash_table[i].chain);
 759                 if (!rth)
 760                         continue;
 761
 762                 spin_lock_bh(rt_hash_lock_addr(i));
 763
 764                 list = NULL;
 765                 pprev = &rt_hash_table[i].chain;
 766                 rth = rcu_dereference_protected(*pprev,
 767                         lockdep_is_held(rt_hash_lock_addr(i)));
 768
 769                 while (rth) {
 770                         next = rcu_dereference_protected(rth->dst.rt_next,
 771                                 lockdep_is_held(rt_hash_lock_addr(i)));
 772
 773                         if (!net ||
 774                             net_eq(dev_net(rth->dst.dev), net)) {
 775                                 rcu_assign_pointer(*pprev, next);
 776                                 rcu_assign_pointer(rth->dst.rt_next, list);
 777                                 list = rth;
 778                         } else {
 779                                 pprev = &rth->dst.rt_next;
 780                         }
 781                         rth = next;
 782                 }
 783
 784                 spin_unlock_bh(rt_hash_lock_addr(i));
 785
 786                 for (; list; list = next) {
 787                         next = rcu_dereference_protected(list->dst.rt_next, 1);
 788                         rt_free(list);
 789                 }
 790         }
 791 }
 792
 793 /*
 794  * While freeing expired entries, we compute average chain length
 795  * and standard deviation, using fixed-point arithmetic.
 796  * This to have an estimation of rt_chain_length_max
 797  *  rt_chain_length_max = max(elasticity, AVG + 4*SD)
 798  * We use 3 bits for frational part, and 29 (or 61) for magnitude.
 799  */
 800
 801 #define FRACT_BITS 3
 802 #define ONE (1UL << FRACT_BITS)
 803
 804 /*
 805  * Given a hash chain and an item in this hash chain,
 806  * find if a previous entry has the same hash_inputs
 807  * (but differs on tos, mark or oif)
 808  * Returns 0 if an alias is found.
 809  * Returns ONE if rth has no alias before itself.
 810  */
 811 static int has_noalias(const struct rtable *head, const struct rtable *rth)
 812 {
 813         const struct rtable *aux = head;
 814
 815         while (aux != rth) {
 816                 if (compare_hash_inputs(aux, rth))
 817                         return 0;
 818                 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
 819         }
 820         return ONE;
 821 }
 822
 823 /*
 824  * Pertubation of rt_genid by a small quantity [1..256]
 825  * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
 826  * many times (2^24) without giving recent rt_genid.
 827  * Jenkins hash is strong enough that litle changes of rt_genid are OK.
 828  */
 829 static void rt_cache_invalidate(struct net *net)
 830 {
 831         unsigned char shuffle;
 832
 833         get_random_bytes(&shuffle, sizeof(shuffle));
 834         atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
 835 }
 836
 837 /*
 838  * delay < 0  : invalidate cache (fast : entries will be deleted later)
 839  * delay >= 0 : invalidate & flush cache (can be long)
 840  */
 841 void rt_cache_flush(struct net *net, int delay)
 842 {
 843         rt_cache_invalidate(net);
 844         if (delay >= 0)
 845                 rt_do_flush(net, !in_softirq());
 846 }
 847
 848 /* Flush previous cache invalidated entries from the cache */
 849 void rt_cache_flush_batch(struct net *net)
 850 {
 851         rt_do_flush(net, !in_softirq());
 852 }
 853
 854 static void rt_emergency_hash_rebuild(struct net *net)
 855 {
 856         if (net_ratelimit())
 857                 printk(KERN_WARNING "Route hash chain too long!\n");
 858         rt_cache_invalidate(net);
 859 }
 860
 861 /*
 862    Short description of GC goals.
 863
 864    We want to build algorithm, which will keep routing cache
 865    at some equilibrium point, when number of aged off entries
 866    is kept approximately equal to newly generated ones.
 867
 868    Current expiration strength is variable "expire".
 869    We try to adjust it dynamically, so that if networking
 870    is idle expires is large enough to keep enough of warm entries,
 871    and when load increases it reduces to limit cache size.
 872  */
 873
 874 static int rt_garbage_collect(struct dst_ops *ops)
 875 {
 876         static unsigned long expire = RT_GC_TIMEOUT;
 877         static unsigned long last_gc;
 878         static int rover;
 879         static int equilibrium;
 880         struct rtable *rth;
 881         struct rtable __rcu **rthp;
 882         unsigned long now = jiffies;
 883         int goal;
 884         int entries = dst_entries_get_fast(&ipv4_dst_ops);
 885
 886         /*
 887          * Garbage collection is pretty expensive,
 888          * do not make it too frequently.
 889          */
 890
 891         RT_CACHE_STAT_INC(gc_total);
 892
 893         if (now - last_gc < ip_rt_gc_min_interval &&
 894             entries < ip_rt_max_size) {
 895                 RT_CACHE_STAT_INC(gc_ignored);
 896                 goto out;
 897         }
 898
 899         entries = dst_entries_get_slow(&ipv4_dst_ops);
 900         /* Calculate number of entries, which we want to expire now. */
 901         goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
 902         if (goal <= 0) {
 903                 if (equilibrium < ipv4_dst_ops.gc_thresh)
 904                         equilibrium = ipv4_dst_ops.gc_thresh;
 905                 goal = entries - equilibrium;
 906                 if (goal > 0) {
 907                         equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
 908                         goal = entries - equilibrium;
 909                 }
 910         } else {
 911                 /* We are in dangerous area. Try to reduce cache really
 912                  * aggressively.
 913                  */
 914                 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
 915                 equilibrium = entries - goal;
 916         }
 917
 918         if (now - last_gc >= ip_rt_gc_min_interval)
 919                 last_gc = now;
 920
 921         if (goal <= 0) {
 922                 equilibrium += goal;
 923                 goto work_done;
 924         }
 925
 926         do {
 927                 int i, k;
 928
 929                 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
 930                         unsigned long tmo = expire;
 931
 932                         k = (k + 1) & rt_hash_mask;
 933                         rthp = &rt_hash_table[k].chain;
 934                         spin_lock_bh(rt_hash_lock_addr(k));
 935                         while ((rth = rcu_dereference_protected(*rthp,
 936                                         lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
 937                                 if (!rt_is_expired(rth) &&
 938                                         !rt_may_expire(rth, tmo, expire)) {
 939                                         tmo >>= 1;
 940                                         rthp = &rth->dst.rt_next;
 941                                         continue;
 942                                 }
 943                                 *rthp = rth->dst.rt_next;
 944                                 rt_free(rth);
 945                                 goal--;
 946                         }
 947                         spin_unlock_bh(rt_hash_lock_addr(k));
 948                         if (goal <= 0)
 949                                 break;
 950                 }
 951                 rover = k;
 952
 953                 if (goal <= 0)
 954                         goto work_done;
 955
 956                 /* Goal is not achieved. We stop process if:
 957
 958                    - if expire reduced to zero. Otherwise, expire is halfed.
 959                    - if table is not full.
 960                    - if we are called from interrupt.
 961                    - jiffies check is just fallback/debug loop breaker.
 962                      We will not spin here for long time in any case.
 963                  */
 964
 965                 RT_CACHE_STAT_INC(gc_goal_miss);
 966
 967                 if (expire == 0)
 968                         break;
 969
 970                 expire >>= 1;
 971 #if RT_CACHE_DEBUG >= 2
 972                 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
 973                                 dst_entries_get_fast(&ipv4_dst_ops), goal, i);
 974 #endif
 975
 976                 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
 977                         goto out;
 978         } while (!in_softirq() && time_before_eq(jiffies, now));
 979
 980         if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
 981                 goto out;
 982         if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
 983                 goto out;
 984         if (net_ratelimit())
 985                 printk(KERN_WARNING "dst cache overflow\n");
 986         RT_CACHE_STAT_INC(gc_dst_overflow);
 987         return 1;
 988
 989 work_done:
 990         expire += ip_rt_gc_min_interval;
 991         if (expire > ip_rt_gc_timeout ||
 992             dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
 993             dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
 994                 expire = ip_rt_gc_timeout;
 995 #if RT_CACHE_DEBUG >= 2
 996         printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
 997                         dst_entries_get_fast(&ipv4_dst_ops), goal, rover);
 998 #endif
 999 out:    return 0;
1000 }
1001
1002 /*
1003  * Returns number of entries in a hash chain that have different hash_inputs
1004  */
1005 static int slow_chain_length(const struct rtable *head)
1006 {
1007         int length = 0;
1008         const struct rtable *rth = head;
1009
1010         while (rth) {
1011                 length += has_noalias(head, rth);
1012                 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1013         }
1014         return length >> FRACT_BITS;
1015 }
1016
1017 static struct rtable *rt_intern_hash(unsigned hash, struct rtable *rt,
1018                                      struct sk_buff *skb, int ifindex)
1019 {
1020         struct rtable   *rth, *cand;
1021         struct rtable __rcu **rthp, **candp;
1022         unsigned long   now;
1023         u32             min_score;
1024         int             chain_length;
1025         int attempts = !in_softirq();
1026
1027 restart:
1028         chain_length = 0;
1029         min_score = ~(u32)0;
1030         cand = NULL;
1031         candp = NULL;
1032         now = jiffies;
1033
1034         if (!rt_caching(dev_net(rt->dst.dev))) {
1035                 /*
1036                  * If we're not caching, just tell the caller we
1037                  * were successful and don't touch the route.  The
1038                  * caller hold the sole reference to the cache entry, and
1039                  * it will be released when the caller is done with it.
1040                  * If we drop it here, the callers have no way to resolve routes
1041                  * when we're not caching.  Instead, just point *rp at rt, so
1042                  * the caller gets a single use out of the route
1043                  * Note that we do rt_free on this new route entry, so that
1044                  * once its refcount hits zero, we are still able to reap it
1045                  * (Thanks Alexey)
1046                  * Note: To avoid expensive rcu stuff for this uncached dst,
1047                  * we set DST_NOCACHE so that dst_release() can free dst without
1048                  * waiting a grace period.
1049                  */
1050
1051                 rt->dst.flags |= DST_NOCACHE;
1052                 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1053                         int err = arp_bind_neighbour(&rt->dst);
1054                         if (err) {
1055                                 if (net_ratelimit())
1056                                         printk(KERN_WARNING
1057                                             "Neighbour table failure & not caching routes.\n");
1058                                 ip_rt_put(rt);
1059                                 return ERR_PTR(err);
1060                         }
1061                 }
1062
1063                 goto skip_hashing;
1064         }
1065
1066         rthp = &rt_hash_table[hash].chain;
1067
1068         spin_lock_bh(rt_hash_lock_addr(hash));
1069         while ((rth = rcu_dereference_protected(*rthp,
1070                         lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1071                 if (rt_is_expired(rth)) {
1072                         *rthp = rth->dst.rt_next;
1073                         rt_free(rth);
1074                         continue;
1075                 }
1076                 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1077                         /* Put it first */
1078                         *rthp = rth->dst.rt_next;
1079                         /*
1080                          * Since lookup is lockfree, the deletion
1081                          * must be visible to another weakly ordered CPU before
1082                          * the insertion at the start of the hash chain.
1083                          */
1084                         rcu_assign_pointer(rth->dst.rt_next,
1085                                            rt_hash_table[hash].chain);
1086                         /*
1087                          * Since lookup is lockfree, the update writes
1088                          * must be ordered for consistency on SMP.
1089                          */
1090                         rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1091
1092                         dst_use(&rth->dst, now);
1093                         spin_unlock_bh(rt_hash_lock_addr(hash));
1094
1095                         rt_drop(rt);
1096                         if (skb)
1097                                 skb_dst_set(skb, &rth->dst);
1098                         return rth;
1099                 }
1100
1101                 if (!atomic_read(&rth->dst.__refcnt)) {
1102                         u32 score = rt_score(rth);
1103
1104                         if (score <= min_score) {
1105                                 cand = rth;
1106                                 candp = rthp;
1107                                 min_score = score;
1108                         }
1109                 }
1110
1111                 chain_length++;
1112
1113                 rthp = &rth->dst.rt_next;
1114         }
1115
1116         if (cand) {
1117                 /* ip_rt_gc_elasticity used to be average length of chain
1118                  * length, when exceeded gc becomes really aggressive.
1119                  *
1120                  * The second limit is less certain. At the moment it allows
1121                  * only 2 entries per bucket. We will see.
1122                  */
1123                 if (chain_length > ip_rt_gc_elasticity) {
1124                         *candp = cand->dst.rt_next;
1125                         rt_free(cand);
1126                 }
1127         } else {
1128                 if (chain_length > rt_chain_length_max &&
1129                     slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1130                         struct net *net = dev_net(rt->dst.dev);
1131                         int num = ++net->ipv4.current_rt_cache_rebuild_count;
1132                         if (!rt_caching(net)) {
1133                                 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1134                                         rt->dst.dev->name, num);
1135                         }
1136                         rt_emergency_hash_rebuild(net);
1137                         spin_unlock_bh(rt_hash_lock_addr(hash));
1138
1139                         hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1140                                         ifindex, rt_genid(net));
1141                         goto restart;
1142                 }
1143         }
1144
1145         /* Try to bind route to arp only if it is output
1146            route or unicast forwarding path.
1147          */
1148         if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1149                 int err = arp_bind_neighbour(&rt->dst);
1150                 if (err) {
1151                         spin_unlock_bh(rt_hash_lock_addr(hash));
1152
1153                         if (err != -ENOBUFS) {
1154                                 rt_drop(rt);
1155                                 return ERR_PTR(err);
1156                         }
1157
1158                         /* Neighbour tables are full and nothing
1159                            can be released. Try to shrink route cache,
1160                            it is most likely it holds some neighbour records.
1161                          */
1162                         if (attempts-- > 0) {
1163                                 int saved_elasticity = ip_rt_gc_elasticity;
1164                                 int saved_int = ip_rt_gc_min_interval;
1165                                 ip_rt_gc_elasticity     = 1;
1166                                 ip_rt_gc_min_interval   = 0;
1167                                 rt_garbage_collect(&ipv4_dst_ops);
1168                                 ip_rt_gc_min_interval   = saved_int;
1169                                 ip_rt_gc_elasticity     = saved_elasticity;
1170                                 goto restart;
1171                         }
1172
1173                         if (net_ratelimit())
1174                                 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1175                         rt_drop(rt);
1176                         return ERR_PTR(-ENOBUFS);
1177                 }
1178         }
1179
1180         rt->dst.rt_next = rt_hash_table[hash].chain;
1181
1182 #if RT_CACHE_DEBUG >= 2
1183         if (rt->dst.rt_next) {
1184                 struct rtable *trt;
1185                 printk(KERN_DEBUG "rt_cache @%02x: %pI4",
1186                        hash, &rt->rt_dst);
1187                 for (trt = rt->dst.rt_next; trt; trt = trt->dst.rt_next)
1188                         printk(" . %pI4", &trt->rt_dst);
1189                 printk("\n");
1190         }
1191 #endif
1192         /*
1193          * Since lookup is lockfree, we must make sure
1194          * previous writes to rt are comitted to memory
1195          * before making rt visible to other CPUS.
1196          */
1197         rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1198
1199         spin_unlock_bh(rt_hash_lock_addr(hash));
1200
1201 skip_hashing:
1202         if (skb)
1203                 skb_dst_set(skb, &rt->dst);
1204         return rt;
1205 }
1206
1207 static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1208
1209 static u32 rt_peer_genid(void)
1210 {
1211         return atomic_read(&__rt_peer_genid);
1212 }
1213
1214 void rt_bind_peer(struct rtable *rt, int create)
1215 {
1216         struct inet_peer *peer;
1217
1218         peer = inet_getpeer_v4(rt->rt_dst, create);
1219
1220         if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1221                 inet_putpeer(peer);
1222         else
1223                 rt->rt_peer_genid = rt_peer_genid();
1224 }
1225
1226 /*
1227  * Peer allocation may fail only in serious out-of-memory conditions.  However
1228  * we still can generate some output.
1229  * Random ID selection looks a bit dangerous because we have no chances to
1230  * select ID being unique in a reasonable period of time.
1231  * But broken packet identifier may be better than no packet at all.
1232  */
1233 static void ip_select_fb_ident(struct iphdr *iph)
1234 {
1235         static DEFINE_SPINLOCK(ip_fb_id_lock);
1236         static u32 ip_fallback_id;
1237         u32 salt;
1238
1239         spin_lock_bh(&ip_fb_id_lock);
1240         salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1241         iph->id = htons(salt & 0xFFFF);
1242         ip_fallback_id = salt;
1243         spin_unlock_bh(&ip_fb_id_lock);
1244 }
1245
1246 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1247 {
1248         struct rtable *rt = (struct rtable *) dst;
1249
1250         if (rt) {
1251                 if (rt->peer == NULL)
1252                         rt_bind_peer(rt, 1);
1253
1254                 /* If peer is attached to destination, it is never detached,
1255                    so that we need not to grab a lock to dereference it.
1256                  */
1257                 if (rt->peer) {
1258                         iph->id = htons(inet_getid(rt->peer, more));
1259                         return;
1260                 }
1261         } else
1262                 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1263                        __builtin_return_address(0));
1264
1265         ip_select_fb_ident(iph);
1266 }
1267 EXPORT_SYMBOL(__ip_select_ident);
1268
1269 static void rt_del(unsigned hash, struct rtable *rt)
1270 {
1271         struct rtable __rcu **rthp;
1272         struct rtable *aux;
1273
1274         rthp = &rt_hash_table[hash].chain;
1275         spin_lock_bh(rt_hash_lock_addr(hash));
1276         ip_rt_put(rt);
1277         while ((aux = rcu_dereference_protected(*rthp,
1278                         lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1279                 if (aux == rt || rt_is_expired(aux)) {
1280                         *rthp = aux->dst.rt_next;
1281                         rt_free(aux);
1282                         continue;
1283                 }
1284                 rthp = &aux->dst.rt_next;
1285         }
1286         spin_unlock_bh(rt_hash_lock_addr(hash));
1287 }
1288
1289 /* called in rcu_read_lock() section */
1290 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1291                     __be32 saddr, struct net_device *dev)
1292 {
1293         struct in_device *in_dev = __in_dev_get_rcu(dev);
1294         struct inet_peer *peer;
1295         struct net *net;
1296
1297         if (!in_dev)
1298                 return;
1299
1300         net = dev_net(dev);
1301         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1302             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1303             ipv4_is_zeronet(new_gw))
1304                 goto reject_redirect;
1305
1306         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1307                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1308                         goto reject_redirect;
1309                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1310                         goto reject_redirect;
1311         } else {
1312                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1313                         goto reject_redirect;
1314         }
1315
1316         peer = inet_getpeer_v4(daddr, 1);
1317         if (peer) {
1318                 peer->redirect_learned.a4 = new_gw;
1319
1320                 inet_putpeer(peer);
1321
1322                 atomic_inc(&__rt_peer_genid);
1323         }
1324         return;
1325
1326 reject_redirect:
1327 #ifdef CONFIG_IP_ROUTE_VERBOSE
1328         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1329                 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1330                         "  Advised path = %pI4 -> %pI4\n",
1331                        &old_gw, dev->name, &new_gw,
1332                        &saddr, &daddr);
1333 #endif
1334         ;
1335 }
1336
1337 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1338 {
1339         struct rtable *rt = (struct rtable *)dst;
1340         struct dst_entry *ret = dst;
1341
1342         if (rt) {
1343                 if (dst->obsolete > 0) {
1344                         ip_rt_put(rt);
1345                         ret = NULL;
1346                 } else if (rt->rt_flags & RTCF_REDIRECTED) {
1347                         unsigned hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1348                                                 rt->rt_oif,
1349                                                 rt_genid(dev_net(dst->dev)));
1350 #if RT_CACHE_DEBUG >= 1
1351                         printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1352                                 &rt->rt_dst, rt->rt_tos);
1353 #endif
1354                         rt_del(hash, rt);
1355                         ret = NULL;
1356                 } else if (rt->peer &&
1357                            rt->peer->pmtu_expires &&
1358                            time_after_eq(jiffies, rt->peer->pmtu_expires)) {
1359                         unsigned long orig = rt->peer->pmtu_expires;
1360
1361                         if (cmpxchg(&rt->peer->pmtu_expires, orig, 0) == orig)
1362                                 dst_metric_set(dst, RTAX_MTU,
1363                                                rt->peer->pmtu_orig);
1364                 }
1365         }
1366         return ret;
1367 }
1368
1369 /*
1370  * Algorithm:
1371  *      1. The first ip_rt_redirect_number redirects are sent
1372  *         with exponential backoff, then we stop sending them at all,
1373  *         assuming that the host ignores our redirects.
1374  *      2. If we did not see packets requiring redirects
1375  *         during ip_rt_redirect_silence, we assume that the host
1376  *         forgot redirected route and start to send redirects again.
1377  *
1378  * This algorithm is much cheaper and more intelligent than dumb load limiting
1379  * in icmp.c.
1380  *
1381  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1382  * and "frag. need" (breaks PMTU discovery) in icmp.c.
1383  */
1384
1385 void ip_rt_send_redirect(struct sk_buff *skb)
1386 {
1387         struct rtable *rt = skb_rtable(skb);
1388         struct in_device *in_dev;
1389         struct inet_peer *peer;
1390         int log_martians;
1391
1392         rcu_read_lock();
1393         in_dev = __in_dev_get_rcu(rt->dst.dev);
1394         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1395                 rcu_read_unlock();
1396                 return;
1397         }
1398         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1399         rcu_read_unlock();
1400
1401         if (!rt->peer)
1402                 rt_bind_peer(rt, 1);
1403         peer = rt->peer;
1404         if (!peer) {
1405                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1406                 return;
1407         }
1408
1409         /* No redirected packets during ip_rt_redirect_silence;
1410          * reset the algorithm.
1411          */
1412         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1413                 peer->rate_tokens = 0;
1414
1415         /* Too many ignored redirects; do not send anything
1416          * set dst.rate_last to the last seen redirected packet.
1417          */
1418         if (peer->rate_tokens >= ip_rt_redirect_number) {
1419                 peer->rate_last = jiffies;
1420                 return;
1421         }
1422
1423         /* Check for load limit; set rate_last to the latest sent
1424          * redirect.
1425          */
1426         if (peer->rate_tokens == 0 ||
1427             time_after(jiffies,
1428                        (peer->rate_last +
1429                         (ip_rt_redirect_load << peer->rate_tokens)))) {
1430                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1431                 peer->rate_last = jiffies;
1432                 ++peer->rate_tokens;
1433 #ifdef CONFIG_IP_ROUTE_VERBOSE
1434                 if (log_martians &&
1435                     peer->rate_tokens == ip_rt_redirect_number &&
1436                     net_ratelimit())
1437                         printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1438                                 &rt->rt_src, rt->rt_iif,
1439                                 &rt->rt_dst, &rt->rt_gateway);
1440 #endif
1441         }
1442 }
1443
1444 static int ip_error(struct sk_buff *skb)
1445 {
1446         struct rtable *rt = skb_rtable(skb);
1447         struct inet_peer *peer;
1448         unsigned long now;
1449         bool send;
1450         int code;
1451
1452         switch (rt->dst.error) {
1453                 case EINVAL:
1454                 default:
1455                         goto out;
1456                 case EHOSTUNREACH:
1457                         code = ICMP_HOST_UNREACH;
1458                         break;
1459                 case ENETUNREACH:
1460                         code = ICMP_NET_UNREACH;
1461                         IP_INC_STATS_BH(dev_net(rt->dst.dev),
1462                                         IPSTATS_MIB_INNOROUTES);
1463                         break;
1464                 case EACCES:
1465                         code = ICMP_PKT_FILTERED;
1466                         break;
1467         }
1468
1469         if (!rt->peer)
1470                 rt_bind_peer(rt, 1);
1471         peer = rt->peer;
1472
1473         send = true;
1474         if (peer) {
1475                 now = jiffies;
1476                 peer->rate_tokens += now - peer->rate_last;
1477                 if (peer->rate_tokens > ip_rt_error_burst)
1478                         peer->rate_tokens = ip_rt_error_burst;
1479                 peer->rate_last = now;
1480                 if (peer->rate_tokens >= ip_rt_error_cost)
1481                         peer->rate_tokens -= ip_rt_error_cost;
1482                 else
1483                         send = false;
1484         }
1485         if (send)
1486                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1487
1488 out:    kfree_skb(skb);
1489         return 0;
1490 }
1491
1492 /*
1493  *      The last two values are not from the RFC but
1494  *      are needed for AMPRnet AX.25 paths.
1495  */
1496
1497 static const unsigned short mtu_plateau[] =
1498 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1499
1500 static inline unsigned short guess_mtu(unsigned short old_mtu)
1501 {
1502         int i;
1503
1504         for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1505                 if (old_mtu > mtu_plateau[i])
1506                         return mtu_plateau[i];
1507         return 68;
1508 }
1509
1510 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1511                                  unsigned short new_mtu,
1512                                  struct net_device *dev)
1513 {
1514         unsigned short old_mtu = ntohs(iph->tot_len);
1515         unsigned short est_mtu = 0;
1516         struct inet_peer *peer;
1517
1518         peer = inet_getpeer_v4(iph->daddr, 1);
1519         if (peer) {
1520                 unsigned short mtu = new_mtu;
1521
1522                 if (new_mtu < 68 || new_mtu >= old_mtu) {
1523                         /* BSD 4.2 derived systems incorrectly adjust
1524                          * tot_len by the IP header length, and report
1525                          * a zero MTU in the ICMP message.
1526                          */
1527                         if (mtu == 0 &&
1528                             old_mtu >= 68 + (iph->ihl << 2))
1529                                 old_mtu -= iph->ihl << 2;
1530                         mtu = guess_mtu(old_mtu);
1531                 }
1532
1533                 if (mtu < ip_rt_min_pmtu)
1534                         mtu = ip_rt_min_pmtu;
1535                 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1536                         est_mtu = mtu;
1537                         peer->pmtu_learned = mtu;
1538                         peer->pmtu_expires = jiffies + ip_rt_mtu_expires;
1539                 }
1540
1541                 inet_putpeer(peer);
1542
1543                 atomic_inc(&__rt_peer_genid);
1544         }
1545         return est_mtu ? : new_mtu;
1546 }
1547
1548 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1549 {
1550         unsigned long expires = peer->pmtu_expires;
1551
1552         if (time_before(expires, jiffies)) {
1553                 u32 orig_dst_mtu = dst_mtu(dst);
1554                 if (peer->pmtu_learned < orig_dst_mtu) {
1555                         if (!peer->pmtu_orig)
1556                                 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1557                         dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1558                 }
1559         } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1560                 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1561 }
1562
1563 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1564 {
1565         struct rtable *rt = (struct rtable *) dst;
1566         struct inet_peer *peer;
1567
1568         dst_confirm(dst);
1569
1570         if (!rt->peer)
1571                 rt_bind_peer(rt, 1);
1572         peer = rt->peer;
1573         if (peer) {
1574                 if (mtu < ip_rt_min_pmtu)
1575                         mtu = ip_rt_min_pmtu;
1576                 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1577                         peer->pmtu_learned = mtu;
1578                         peer->pmtu_expires = jiffies + ip_rt_mtu_expires;
1579
1580                         atomic_inc(&__rt_peer_genid);
1581                         rt->rt_peer_genid = rt_peer_genid();
1582
1583                         check_peer_pmtu(dst, peer);
1584                 }
1585                 inet_putpeer(peer);
1586         }
1587 }
1588
1589 static int check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1590 {
1591         struct rtable *rt = (struct rtable *) dst;
1592         __be32 orig_gw = rt->rt_gateway;
1593
1594         dst_confirm(&rt->dst);
1595
1596         neigh_release(rt->dst.neighbour);
1597         rt->dst.neighbour = NULL;
1598
1599         rt->rt_gateway = peer->redirect_learned.a4;
1600         if (arp_bind_neighbour(&rt->dst) ||
1601             !(rt->dst.neighbour->nud_state & NUD_VALID)) {
1602                 if (rt->dst.neighbour)
1603                         neigh_event_send(rt->dst.neighbour, NULL);
1604                 rt->rt_gateway = orig_gw;
1605                 return -EAGAIN;
1606         } else {
1607                 rt->rt_flags |= RTCF_REDIRECTED;
1608                 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE,
1609                                         rt->dst.neighbour);
1610         }
1611         return 0;
1612 }
1613
1614 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1615 {
1616         struct rtable *rt = (struct rtable *) dst;
1617
1618         if (rt_is_expired(rt))
1619                 return NULL;
1620         if (rt->rt_peer_genid != rt_peer_genid()) {
1621                 struct inet_peer *peer;
1622
1623                 if (!rt->peer)
1624                         rt_bind_peer(rt, 0);
1625
1626                 peer = rt->peer;
1627                 if (peer && peer->pmtu_expires)
1628                         check_peer_pmtu(dst, peer);
1629
1630                 if (peer && peer->redirect_learned.a4 &&
1631                     peer->redirect_learned.a4 != rt->rt_gateway) {
1632                         if (check_peer_redir(dst, peer))
1633                                 return NULL;
1634                 }
1635
1636                 rt->rt_peer_genid = rt_peer_genid();
1637         }
1638         return dst;
1639 }
1640
1641 static void ipv4_dst_destroy(struct dst_entry *dst)
1642 {
1643         struct rtable *rt = (struct rtable *) dst;
1644         struct inet_peer *peer = rt->peer;
1645
1646         if (rt->fi) {
1647                 fib_info_put(rt->fi);
1648                 rt->fi = NULL;
1649         }
1650         if (peer) {
1651                 rt->peer = NULL;
1652                 inet_putpeer(peer);
1653         }
1654 }
1655
1656
1657 static void ipv4_link_failure(struct sk_buff *skb)
1658 {
1659         struct rtable *rt;
1660
1661         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1662
1663         rt = skb_rtable(skb);
1664         if (rt &&
1665             rt->peer &&
1666             rt->peer->pmtu_expires) {
1667                 unsigned long orig = rt->peer->pmtu_expires;
1668
1669                 if (cmpxchg(&rt->peer->pmtu_expires, orig, 0) == orig)
1670                         dst_metric_set(&rt->dst, RTAX_MTU, rt->peer->pmtu_orig);
1671         }
1672 }
1673
1674 static int ip_rt_bug(struct sk_buff *skb)
1675 {
1676         printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1677                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1678                 skb->dev ? skb->dev->name : "?");
1679         kfree_skb(skb);
1680         return 0;
1681 }
1682
1683 /*
1684    We do not cache source address of outgoing interface,
1685    because it is used only by IP RR, TS and SRR options,
1686    so that it out of fast path.
1687
1688    BTW remember: "addr" is allowed to be not aligned
1689    in IP options!
1690  */
1691
1692 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1693 {
1694         __be32 src;
1695         struct fib_result res;
1696
1697         if (rt_is_output_route(rt))
1698                 src = rt->rt_src;
1699         else {
1700                 struct flowi4 fl4 = {
1701                         .daddr = rt->rt_key_dst,
1702                         .saddr = rt->rt_key_src,
1703                         .flowi4_tos = rt->rt_tos,
1704                         .flowi4_oif = rt->rt_oif,
1705                         .flowi4_iif = rt->rt_iif,
1706                         .flowi4_mark = rt->rt_mark,
1707                 };
1708
1709                 rcu_read_lock();
1710                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1711                         src = FIB_RES_PREFSRC(res);
1712                 else
1713                         src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1714                                         RT_SCOPE_UNIVERSE);
1715                 rcu_read_unlock();
1716         }
1717         memcpy(addr, &src, 4);
1718 }
1719
1720 #ifdef CONFIG_IP_ROUTE_CLASSID
1721 static void set_class_tag(struct rtable *rt, u32 tag)
1722 {
1723         if (!(rt->dst.tclassid & 0xFFFF))
1724                 rt->dst.tclassid |= tag & 0xFFFF;
1725         if (!(rt->dst.tclassid & 0xFFFF0000))
1726                 rt->dst.tclassid |= tag & 0xFFFF0000;
1727 }
1728 #endif
1729
1730 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1731 {
1732         unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1733
1734         if (advmss == 0) {
1735                 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1736                                ip_rt_min_advmss);
1737                 if (advmss > 65535 - 40)
1738                         advmss = 65535 - 40;
1739         }
1740         return advmss;
1741 }
1742
1743 static unsigned int ipv4_default_mtu(const struct dst_entry *dst)
1744 {
1745         unsigned int mtu = dst->dev->mtu;
1746
1747         if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1748                 const struct rtable *rt = (const struct rtable *) dst;
1749
1750                 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1751                         mtu = 576;
1752         }
1753
1754         if (mtu > IP_MAX_MTU)
1755                 mtu = IP_MAX_MTU;
1756
1757         return mtu;
1758 }
1759
1760 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *oldflp4,
1761                             struct fib_info *fi)
1762 {
1763         struct inet_peer *peer;
1764         int create = 0;
1765
1766         /* If a peer entry exists for this destination, we must hook
1767          * it up in order to get at cached metrics.
1768          */
1769         if (oldflp4 && (oldflp4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS))
1770                 create = 1;
1771
1772         rt->peer = peer = inet_getpeer_v4(rt->rt_dst, create);
1773         if (peer) {
1774                 rt->rt_peer_genid = rt_peer_genid();
1775                 if (inet_metrics_new(peer))
1776                         memcpy(peer->metrics, fi->fib_metrics,
1777                                sizeof(u32) * RTAX_MAX);
1778                 dst_init_metrics(&rt->dst, peer->metrics, false);
1779
1780                 if (peer->pmtu_expires)
1781                         check_peer_pmtu(&rt->dst, peer);
1782                 if (peer->redirect_learned.a4 &&
1783                     peer->redirect_learned.a4 != rt->rt_gateway) {
1784                         rt->rt_gateway = peer->redirect_learned.a4;
1785                         rt->rt_flags |= RTCF_REDIRECTED;
1786                 }
1787         } else {
1788                 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1789                         rt->fi = fi;
1790                         atomic_inc(&fi->fib_clntref);
1791                 }
1792                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1793         }
1794 }
1795
1796 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *oldflp4,
1797                            const struct fib_result *res,
1798                            struct fib_info *fi, u16 type, u32 itag)
1799 {
1800         struct dst_entry *dst = &rt->dst;
1801
1802         if (fi) {
1803                 if (FIB_RES_GW(*res) &&
1804                     FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1805                         rt->rt_gateway = FIB_RES_GW(*res);
1806                 rt_init_metrics(rt, oldflp4, fi);
1807 #ifdef CONFIG_IP_ROUTE_CLASSID
1808                 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1809 #endif
1810         }
1811
1812         if (dst_mtu(dst) > IP_MAX_MTU)
1813                 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1814         if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40)
1815                 dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40);
1816
1817 #ifdef CONFIG_IP_ROUTE_CLASSID
1818 #ifdef CONFIG_IP_MULTIPLE_TABLES
1819         set_class_tag(rt, fib_rules_tclass(res));
1820 #endif
1821         set_class_tag(rt, itag);
1822 #endif
1823         rt->rt_type = type;
1824 }
1825
1826 static struct rtable *rt_dst_alloc(bool nopolicy, bool noxfrm)
1827 {
1828         struct rtable *rt = dst_alloc(&ipv4_dst_ops, 1);
1829         if (rt) {
1830                 rt->dst.obsolete = -1;
1831
1832                 rt->dst.flags = DST_HOST |
1833                         (nopolicy ? DST_NOPOLICY : 0) |
1834                         (noxfrm ? DST_NOXFRM : 0);
1835         }
1836         return rt;
1837 }
1838
1839 /* called in rcu_read_lock() section */
1840 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1841                                 u8 tos, struct net_device *dev, int our)
1842 {
1843         unsigned int hash;
1844         struct rtable *rth;
1845         __be32 spec_dst;
1846         struct in_device *in_dev = __in_dev_get_rcu(dev);
1847         u32 itag = 0;
1848         int err;
1849
1850         /* Primary sanity checks. */
1851
1852         if (in_dev == NULL)
1853                 return -EINVAL;
1854
1855         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1856             ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1857                 goto e_inval;
1858
1859         if (ipv4_is_zeronet(saddr)) {
1860                 if (!ipv4_is_local_multicast(daddr))
1861                         goto e_inval;
1862                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1863         } else {
1864                 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1865                                           &itag, 0);
1866                 if (err < 0)
1867                         goto e_err;
1868         }
1869         rth = rt_dst_alloc(IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1870         if (!rth)
1871                 goto e_nobufs;
1872
1873         rth->dst.output = ip_rt_bug;
1874
1875         rth->rt_key_dst = daddr;
1876         rth->rt_dst     = daddr;
1877         rth->rt_tos     = tos;
1878         rth->rt_mark    = skb->mark;
1879         rth->rt_key_src = saddr;
1880         rth->rt_src     = saddr;
1881 #ifdef CONFIG_IP_ROUTE_CLASSID
1882         rth->dst.tclassid = itag;
1883 #endif
1884         rth->rt_iif     = dev->ifindex;
1885         rth->dst.dev    = init_net.loopback_dev;
1886         dev_hold(rth->dst.dev);
1887         rth->rt_oif     = 0;
1888         rth->rt_gateway = daddr;
1889         rth->rt_spec_dst= spec_dst;
1890         rth->rt_genid   = rt_genid(dev_net(dev));
1891         rth->rt_flags   = RTCF_MULTICAST;
1892         rth->rt_type    = RTN_MULTICAST;
1893         if (our) {
1894                 rth->dst.input= ip_local_deliver;
1895                 rth->rt_flags |= RTCF_LOCAL;
1896         }
1897
1898 #ifdef CONFIG_IP_MROUTE
1899         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1900                 rth->dst.input = ip_mr_input;
1901 #endif
1902         RT_CACHE_STAT_INC(in_slow_mc);
1903
1904         hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1905         rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
1906         err = 0;
1907         if (IS_ERR(rth))
1908                 err = PTR_ERR(rth);
1909
1910 e_nobufs:
1911         return -ENOBUFS;
1912 e_inval:
1913         return -EINVAL;
1914 e_err:
1915         return err;
1916 }
1917
1918
1919 static void ip_handle_martian_source(struct net_device *dev,
1920                                      struct in_device *in_dev,
1921                                      struct sk_buff *skb,
1922                                      __be32 daddr,
1923                                      __be32 saddr)
1924 {
1925         RT_CACHE_STAT_INC(in_martian_src);
1926 #ifdef CONFIG_IP_ROUTE_VERBOSE
1927         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1928                 /*
1929                  *      RFC1812 recommendation, if source is martian,
1930                  *      the only hint is MAC header.
1931                  */
1932                 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1933                         &daddr, &saddr, dev->name);
1934                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1935                         int i;
1936                         const unsigned char *p = skb_mac_header(skb);
1937                         printk(KERN_WARNING "ll header: ");
1938                         for (i = 0; i < dev->hard_header_len; i++, p++) {
1939                                 printk("%02x", *p);
1940                                 if (i < (dev->hard_header_len - 1))
1941                                         printk(":");
1942                         }
1943                         printk("\n");
1944                 }
1945         }
1946 #endif
1947 }
1948
1949 /* called in rcu_read_lock() section */
1950 static int __mkroute_input(struct sk_buff *skb,
1951                            const struct fib_result *res,
1952                            struct in_device *in_dev,
1953                            __be32 daddr, __be32 saddr, u32 tos,
1954                            struct rtable **result)
1955 {
1956         struct rtable *rth;
1957         int err;
1958         struct in_device *out_dev;
1959         unsigned int flags = 0;
1960         __be32 spec_dst;
1961         u32 itag;
1962
1963         /* get a working reference to the output device */
1964         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1965         if (out_dev == NULL) {
1966                 if (net_ratelimit())
1967                         printk(KERN_CRIT "Bug in ip_route_input" \
1968                                "_slow(). Please, report\n");
1969                 return -EINVAL;
1970         }
1971
1972
1973         err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1974                                   in_dev->dev, &spec_dst, &itag, skb->mark);
1975         if (err < 0) {
1976                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1977                                          saddr);
1978
1979                 goto cleanup;
1980         }
1981
1982         if (err)
1983                 flags |= RTCF_DIRECTSRC;
1984
1985         if (out_dev == in_dev && err &&
1986             (IN_DEV_SHARED_MEDIA(out_dev) ||
1987              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1988                 flags |= RTCF_DOREDIRECT;
1989
1990         if (skb->protocol != htons(ETH_P_IP)) {
1991                 /* Not IP (i.e. ARP). Do not create route, if it is
1992                  * invalid for proxy arp. DNAT routes are always valid.
1993                  *
1994                  * Proxy arp feature have been extended to allow, ARP
1995                  * replies back to the same interface, to support
1996                  * Private VLAN switch technologies. See arp.c.
1997                  */
1998                 if (out_dev == in_dev &&
1999                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2000                         err = -EINVAL;
2001                         goto cleanup;
2002                 }
2003         }
2004
2005         rth = rt_dst_alloc(IN_DEV_CONF_GET(in_dev, NOPOLICY),
2006                            IN_DEV_CONF_GET(out_dev, NOXFRM));
2007         if (!rth) {
2008                 err = -ENOBUFS;
2009                 goto cleanup;
2010         }
2011
2012         rth->rt_key_dst = daddr;
2013         rth->rt_dst     = daddr;
2014         rth->rt_tos     = tos;
2015         rth->rt_mark    = skb->mark;
2016         rth->rt_key_src = saddr;
2017         rth->rt_src     = saddr;
2018         rth->rt_gateway = daddr;
2019         rth->rt_iif     = in_dev->dev->ifindex;
2020         rth->dst.dev    = (out_dev)->dev;
2021         dev_hold(rth->dst.dev);
2022         rth->rt_oif     = 0;
2023         rth->rt_spec_dst= spec_dst;
2024
2025         rth->dst.input = ip_forward;
2026         rth->dst.output = ip_output;
2027         rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2028
2029         rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2030
2031         rth->rt_flags = flags;
2032
2033         *result = rth;
2034         err = 0;
2035  cleanup:
2036         return err;
2037 }
2038
2039 static int ip_mkroute_input(struct sk_buff *skb,
2040                             struct fib_result *res,
2041                             const struct flowi4 *fl4,
2042                             struct in_device *in_dev,
2043                             __be32 daddr, __be32 saddr, u32 tos)
2044 {
2045         struct rtable* rth = NULL;
2046         int err;
2047         unsigned hash;
2048
2049 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2050         if (res->fi && res->fi->fib_nhs > 1)
2051                 fib_select_multipath(res);
2052 #endif
2053
2054         /* create a routing cache entry */
2055         err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2056         if (err)
2057                 return err;
2058
2059         /* put it into the cache */
2060         hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2061                        rt_genid(dev_net(rth->dst.dev)));
2062         rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2063         if (IS_ERR(rth))
2064                 return PTR_ERR(rth);
2065         return 0;
2066 }
2067
2068 /*
2069  *      NOTE. We drop all the packets that has local source
2070  *      addresses, because every properly looped back packet
2071  *      must have correct destination already attached by output routine.
2072  *
2073  *      Such approach solves two big problems:
2074  *      1. Not simplex devices are handled properly.
2075  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2076  *      called with rcu_read_lock()
2077  */
2078
2079 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2080                                u8 tos, struct net_device *dev)
2081 {
2082         struct fib_result res;
2083         struct in_device *in_dev = __in_dev_get_rcu(dev);
2084         struct flowi4   fl4;
2085         unsigned        flags = 0;
2086         u32             itag = 0;
2087         struct rtable * rth;
2088         unsigned        hash;
2089         __be32          spec_dst;
2090         int             err = -EINVAL;
2091         struct net    * net = dev_net(dev);
2092
2093         /* IP on this device is disabled. */
2094
2095         if (!in_dev)
2096                 goto out;
2097
2098         /* Check for the most weird martians, which can be not detected
2099            by fib_lookup.
2100          */
2101
2102         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2103             ipv4_is_loopback(saddr))
2104                 goto martian_source;
2105
2106         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2107                 goto brd_input;
2108
2109         /* Accept zero addresses only to limited broadcast;
2110          * I even do not know to fix it or not. Waiting for complains :-)
2111          */
2112         if (ipv4_is_zeronet(saddr))
2113                 goto martian_source;
2114
2115         if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2116                 goto martian_destination;
2117
2118         /*
2119          *      Now we are ready to route packet.
2120          */
2121         fl4.flowi4_oif = 0;
2122         fl4.flowi4_iif = dev->ifindex;
2123         fl4.flowi4_mark = skb->mark;
2124         fl4.flowi4_tos = tos;
2125         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2126         fl4.daddr = daddr;
2127         fl4.saddr = saddr;
2128         err = fib_lookup(net, &fl4, &res);
2129         if (err != 0) {
2130                 if (!IN_DEV_FORWARD(in_dev))
2131                         goto e_hostunreach;
2132                 goto no_route;
2133         }
2134
2135         RT_CACHE_STAT_INC(in_slow_tot);
2136
2137         if (res.type == RTN_BROADCAST)
2138                 goto brd_input;
2139
2140         if (res.type == RTN_LOCAL) {
2141                 err = fib_validate_source(saddr, daddr, tos,
2142                                           net->loopback_dev->ifindex,
2143                                           dev, &spec_dst, &itag, skb->mark);
2144                 if (err < 0)
2145                         goto martian_source_keep_err;
2146                 if (err)
2147                         flags |= RTCF_DIRECTSRC;
2148                 spec_dst = daddr;
2149                 goto local_input;
2150         }
2151
2152         if (!IN_DEV_FORWARD(in_dev))
2153                 goto e_hostunreach;
2154         if (res.type != RTN_UNICAST)
2155                 goto martian_destination;
2156
2157         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2158 out:    return err;
2159
2160 brd_input:
2161         if (skb->protocol != htons(ETH_P_IP))
2162                 goto e_inval;
2163
2164         if (ipv4_is_zeronet(saddr))
2165                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2166         else {
2167                 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2168                                           &itag, skb->mark);
2169                 if (err < 0)
2170                         goto martian_source_keep_err;
2171                 if (err)
2172                         flags |= RTCF_DIRECTSRC;
2173         }
2174         flags |= RTCF_BROADCAST;
2175         res.type = RTN_BROADCAST;
2176         RT_CACHE_STAT_INC(in_brd);
2177
2178 local_input:
2179         rth = rt_dst_alloc(IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2180         if (!rth)
2181                 goto e_nobufs;
2182
2183         rth->dst.output= ip_rt_bug;
2184         rth->rt_genid = rt_genid(net);
2185
2186         rth->rt_key_dst = daddr;
2187         rth->rt_dst     = daddr;
2188         rth->rt_tos     = tos;
2189         rth->rt_mark    = skb->mark;
2190         rth->rt_key_src = saddr;
2191         rth->rt_src     = saddr;
2192 #ifdef CONFIG_IP_ROUTE_CLASSID
2193         rth->dst.tclassid = itag;
2194 #endif
2195         rth->rt_iif     = dev->ifindex;
2196         rth->dst.dev    = net->loopback_dev;
2197         dev_hold(rth->dst.dev);
2198         rth->rt_gateway = daddr;
2199         rth->rt_spec_dst= spec_dst;
2200         rth->dst.input= ip_local_deliver;
2201         rth->rt_flags   = flags|RTCF_LOCAL;
2202         if (res.type == RTN_UNREACHABLE) {
2203                 rth->dst.input= ip_error;
2204                 rth->dst.error= -err;
2205                 rth->rt_flags   &= ~RTCF_LOCAL;
2206         }
2207         rth->rt_type    = res.type;
2208         hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2209         rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2210         err = 0;
2211         if (IS_ERR(rth))
2212                 err = PTR_ERR(rth);
2213         goto out;
2214
2215 no_route:
2216         RT_CACHE_STAT_INC(in_no_route);
2217         spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2218         res.type = RTN_UNREACHABLE;
2219         if (err == -ESRCH)
2220                 err = -ENETUNREACH;
2221         goto local_input;
2222
2223         /*
2224          *      Do not cache martian addresses: they should be logged (RFC1812)
2225          */
2226 martian_destination:
2227         RT_CACHE_STAT_INC(in_martian_dst);
2228 #ifdef CONFIG_IP_ROUTE_VERBOSE
2229         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2230                 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2231                         &daddr, &saddr, dev->name);
2232 #endif
2233
2234 e_hostunreach:
2235         err = -EHOSTUNREACH;
2236         goto out;
2237
2238 e_inval:
2239         err = -EINVAL;
2240         goto out;
2241
2242 e_nobufs:
2243         err = -ENOBUFS;
2244         goto out;
2245
2246 martian_source:
2247         err = -EINVAL;
2248 martian_source_keep_err:
2249         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2250         goto out;
2251 }
2252
2253 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2254                            u8 tos, struct net_device *dev, bool noref)
2255 {
2256         struct rtable * rth;
2257         unsigned        hash;
2258         int iif = dev->ifindex;
2259         struct net *net;
2260         int res;
2261
2262         net = dev_net(dev);
2263
2264         rcu_read_lock();
2265
2266         if (!rt_caching(net))
2267                 goto skip_cache;
2268
2269         tos &= IPTOS_RT_MASK;
2270         hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2271
2272         for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2273              rth = rcu_dereference(rth->dst.rt_next)) {
2274                 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2275                      ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2276                      (rth->rt_iif ^ iif) |
2277                      rth->rt_oif |
2278                      (rth->rt_tos ^ tos)) == 0 &&
2279                     rth->rt_mark == skb->mark &&
2280                     net_eq(dev_net(rth->dst.dev), net) &&
2281                     !rt_is_expired(rth)) {
2282                         if (noref) {
2283                                 dst_use_noref(&rth->dst, jiffies);
2284                                 skb_dst_set_noref(skb, &rth->dst);
2285                         } else {
2286                                 dst_use(&rth->dst, jiffies);
2287                                 skb_dst_set(skb, &rth->dst);
2288                         }
2289                         RT_CACHE_STAT_INC(in_hit);
2290                         rcu_read_unlock();
2291                         return 0;
2292                 }
2293                 RT_CACHE_STAT_INC(in_hlist_search);
2294         }
2295
2296 skip_cache:
2297         /* Multicast recognition logic is moved from route cache to here.
2298            The problem was that too many Ethernet cards have broken/missing
2299            hardware multicast filters :-( As result the host on multicasting
2300            network acquires a lot of useless route cache entries, sort of
2301            SDR messages from all the world. Now we try to get rid of them.
2302            Really, provided software IP multicast filter is organized
2303            reasonably (at least, hashed), it does not result in a slowdown
2304            comparing with route cache reject entries.
2305            Note, that multicast routers are not affected, because
2306            route cache entry is created eventually.
2307          */
2308         if (ipv4_is_multicast(daddr)) {
2309                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2310
2311                 if (in_dev) {
2312                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2313                                                   ip_hdr(skb)->protocol);
2314                         if (our
2315 #ifdef CONFIG_IP_MROUTE
2316                                 ||
2317                             (!ipv4_is_local_multicast(daddr) &&
2318                              IN_DEV_MFORWARD(in_dev))
2319 #endif
2320                            ) {
2321                                 int res = ip_route_input_mc(skb, daddr, saddr,
2322                                                             tos, dev, our);
2323                                 rcu_read_unlock();
2324                                 return res;
2325                         }
2326                 }
2327                 rcu_read_unlock();
2328                 return -EINVAL;
2329         }
2330         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2331         rcu_read_unlock();
2332         return res;
2333 }
2334 EXPORT_SYMBOL(ip_route_input_common);
2335
2336 /* called with rcu_read_lock() */
2337 static struct rtable *__mkroute_output(const struct fib_result *res,
2338                                        const struct flowi4 *fl4,
2339                                        const struct flowi4 *oldflp4,
2340                                        struct net_device *dev_out,
2341                                        unsigned int flags)
2342 {
2343         struct fib_info *fi = res->fi;
2344         u32 tos = RT_FL_TOS(oldflp4);
2345         struct in_device *in_dev;
2346         u16 type = res->type;
2347         struct rtable *rth;
2348
2349         if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2350                 return ERR_PTR(-EINVAL);
2351
2352         if (ipv4_is_lbcast(fl4->daddr))
2353                 type = RTN_BROADCAST;
2354         else if (ipv4_is_multicast(fl4->daddr))
2355                 type = RTN_MULTICAST;
2356         else if (ipv4_is_zeronet(fl4->daddr))
2357                 return ERR_PTR(-EINVAL);
2358
2359         if (dev_out->flags & IFF_LOOPBACK)
2360                 flags |= RTCF_LOCAL;
2361
2362         in_dev = __in_dev_get_rcu(dev_out);
2363         if (!in_dev)
2364                 return ERR_PTR(-EINVAL);
2365
2366         if (type == RTN_BROADCAST) {
2367                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2368                 fi = NULL;
2369         } else if (type == RTN_MULTICAST) {
2370                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2371                 if (!ip_check_mc_rcu(in_dev, oldflp4->daddr, oldflp4->saddr,
2372                                      oldflp4->flowi4_proto))
2373                         flags &= ~RTCF_LOCAL;
2374                 /* If multicast route do not exist use
2375                  * default one, but do not gateway in this case.
2376                  * Yes, it is hack.
2377                  */
2378                 if (fi && res->prefixlen < 4)
2379                         fi = NULL;
2380         }
2381
2382         rth = rt_dst_alloc(IN_DEV_CONF_GET(in_dev, NOPOLICY),
2383                            IN_DEV_CONF_GET(in_dev, NOXFRM));
2384         if (!rth)
2385                 return ERR_PTR(-ENOBUFS);
2386
2387         rth->rt_key_dst = oldflp4->daddr;
2388         rth->rt_tos     = tos;
2389         rth->rt_key_src = oldflp4->saddr;
2390         rth->rt_oif     = oldflp4->flowi4_oif;
2391         rth->rt_mark    = oldflp4->flowi4_mark;
2392         rth->rt_dst     = fl4->daddr;
2393         rth->rt_src     = fl4->saddr;
2394         rth->rt_iif     = 0;
2395         /* get references to the devices that are to be hold by the routing
2396            cache entry */
2397         rth->dst.dev    = dev_out;
2398         dev_hold(dev_out);
2399         rth->rt_gateway = fl4->daddr;
2400         rth->rt_spec_dst= fl4->saddr;
2401
2402         rth->dst.output=ip_output;
2403         rth->rt_genid = rt_genid(dev_net(dev_out));
2404
2405         RT_CACHE_STAT_INC(out_slow_tot);
2406
2407         if (flags & RTCF_LOCAL) {
2408                 rth->dst.input = ip_local_deliver;
2409                 rth->rt_spec_dst = fl4->daddr;
2410         }
2411         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2412                 rth->rt_spec_dst = fl4->saddr;
2413                 if (flags & RTCF_LOCAL &&
2414                     !(dev_out->flags & IFF_LOOPBACK)) {
2415                         rth->dst.output = ip_mc_output;
2416                         RT_CACHE_STAT_INC(out_slow_mc);
2417                 }
2418 #ifdef CONFIG_IP_MROUTE
2419                 if (type == RTN_MULTICAST) {
2420                         if (IN_DEV_MFORWARD(in_dev) &&
2421                             !ipv4_is_local_multicast(oldflp4->daddr)) {
2422                                 rth->dst.input = ip_mr_input;
2423                                 rth->dst.output = ip_mc_output;
2424                         }
2425                 }
2426 #endif
2427         }
2428
2429         rt_set_nexthop(rth, oldflp4, res, fi, type, 0);
2430
2431         rth->rt_flags = flags;
2432         return rth;
2433 }
2434
2435 /*
2436  * Major route resolver routine.
2437  * called with rcu_read_lock();
2438  */
2439
2440 static struct rtable *ip_route_output_slow(struct net *net,
2441                                            const struct flowi4 *oldflp4)
2442 {
2443         u32 tos = RT_FL_TOS(oldflp4);
2444         struct flowi4 fl4;
2445         struct fib_result res;
2446         unsigned int flags = 0;
2447         struct net_device *dev_out = NULL;
2448         struct rtable *rth;
2449
2450         res.fi          = NULL;
2451 #ifdef CONFIG_IP_MULTIPLE_TABLES
2452         res.r           = NULL;
2453 #endif
2454
2455         fl4.flowi4_oif = oldflp4->flowi4_oif;
2456         fl4.flowi4_iif = net->loopback_dev->ifindex;
2457         fl4.flowi4_mark = oldflp4->flowi4_mark;
2458         fl4.daddr = oldflp4->daddr;
2459         fl4.saddr = oldflp4->saddr;
2460         fl4.flowi4_tos = tos & IPTOS_RT_MASK;
2461         fl4.flowi4_scope = ((tos & RTO_ONLINK) ?
2462                         RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2463
2464         rcu_read_lock();
2465         if (oldflp4->saddr) {
2466                 rth = ERR_PTR(-EINVAL);
2467                 if (ipv4_is_multicast(oldflp4->saddr) ||
2468                     ipv4_is_lbcast(oldflp4->saddr) ||
2469                     ipv4_is_zeronet(oldflp4->saddr))
2470                         goto out;
2471
2472                 /* I removed check for oif == dev_out->oif here.
2473                    It was wrong for two reasons:
2474                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2475                       is assigned to multiple interfaces.
2476                    2. Moreover, we are allowed to send packets with saddr
2477                       of another iface. --ANK
2478                  */
2479
2480                 if (oldflp4->flowi4_oif == 0 &&
2481                     (ipv4_is_multicast(oldflp4->daddr) ||
2482                      ipv4_is_lbcast(oldflp4->daddr))) {
2483                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2484                         dev_out = __ip_dev_find(net, oldflp4->saddr, false);
2485                         if (dev_out == NULL)
2486                                 goto out;
2487
2488                         /* Special hack: user can direct multicasts
2489                            and limited broadcast via necessary interface
2490                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2491                            This hack is not just for fun, it allows
2492                            vic,vat and friends to work.
2493                            They bind socket to loopback, set ttl to zero
2494                            and expect that it will work.
2495                            From the viewpoint of routing cache they are broken,
2496                            because we are not allowed to build multicast path
2497                            with loopback source addr (look, routing cache
2498                            cannot know, that ttl is zero, so that packet
2499                            will not leave this host and route is valid).
2500                            Luckily, this hack is good workaround.
2501                          */
2502
2503                         fl4.flowi4_oif = dev_out->ifindex;
2504                         goto make_route;
2505                 }
2506
2507                 if (!(oldflp4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2508                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2509                         if (!__ip_dev_find(net, oldflp4->saddr, false))
2510                                 goto out;
2511                 }
2512         }
2513
2514
2515         if (oldflp4->flowi4_oif) {
2516                 dev_out = dev_get_by_index_rcu(net, oldflp4->flowi4_oif);
2517                 rth = ERR_PTR(-ENODEV);
2518                 if (dev_out == NULL)
2519                         goto out;
2520
2521                 /* RACE: Check return value of inet_select_addr instead. */
2522                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2523                         rth = ERR_PTR(-ENETUNREACH);
2524                         goto out;
2525                 }
2526                 if (ipv4_is_local_multicast(oldflp4->daddr) ||
2527                     ipv4_is_lbcast(oldflp4->daddr)) {
2528                         if (!fl4.saddr)
2529                                 fl4.saddr = inet_select_addr(dev_out, 0,
2530                                                              RT_SCOPE_LINK);
2531                         goto make_route;
2532                 }
2533                 if (!fl4.saddr) {
2534                         if (ipv4_is_multicast(oldflp4->daddr))
2535                                 fl4.saddr = inet_select_addr(dev_out, 0,
2536                                                              fl4.flowi4_scope);
2537                         else if (!oldflp4->daddr)
2538                                 fl4.saddr = inet_select_addr(dev_out, 0,
2539                                                              RT_SCOPE_HOST);
2540                 }
2541         }
2542
2543         if (!fl4.daddr) {
2544                 fl4.daddr = fl4.saddr;
2545                 if (!fl4.daddr)
2546                         fl4.daddr = fl4.saddr = htonl(INADDR_LOOPBACK);
2547                 dev_out = net->loopback_dev;
2548                 fl4.flowi4_oif = net->loopback_dev->ifindex;
2549                 res.type = RTN_LOCAL;
2550                 flags |= RTCF_LOCAL;
2551                 goto make_route;
2552         }
2553
2554         if (fib_lookup(net, &fl4, &res)) {
2555                 res.fi = NULL;
2556                 if (oldflp4->flowi4_oif) {
2557                         /* Apparently, routing tables are wrong. Assume,
2558                            that the destination is on link.
2559
2560                            WHY? DW.
2561                            Because we are allowed to send to iface
2562                            even if it has NO routes and NO assigned
2563                            addresses. When oif is specified, routing
2564                            tables are looked up with only one purpose:
2565                            to catch if destination is gatewayed, rather than
2566                            direct. Moreover, if MSG_DONTROUTE is set,
2567                            we send packet, ignoring both routing tables
2568                            and ifaddr state. --ANK
2569
2570
2571                            We could make it even if oif is unknown,
2572                            likely IPv6, but we do not.
2573                          */
2574
2575                         if (fl4.saddr == 0)
2576                                 fl4.saddr = inet_select_addr(dev_out, 0,
2577                                                              RT_SCOPE_LINK);
2578                         res.type = RTN_UNICAST;
2579                         goto make_route;
2580                 }
2581                 rth = ERR_PTR(-ENETUNREACH);
2582                 goto out;
2583         }
2584
2585         if (res.type == RTN_LOCAL) {
2586                 if (!fl4.saddr) {
2587                         if (res.fi->fib_prefsrc)
2588                                 fl4.saddr = res.fi->fib_prefsrc;
2589                         else
2590                                 fl4.saddr = fl4.daddr;
2591                 }
2592                 dev_out = net->loopback_dev;
2593                 fl4.flowi4_oif = dev_out->ifindex;
2594                 res.fi = NULL;
2595                 flags |= RTCF_LOCAL;
2596                 goto make_route;
2597         }
2598
2599 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2600         if (res.fi->fib_nhs > 1 && fl4.flowi4_oif == 0)
2601                 fib_select_multipath(&res);
2602         else
2603 #endif
2604         if (!res.prefixlen && res.type == RTN_UNICAST && !fl4.flowi4_oif)
2605                 fib_select_default(&res);
2606
2607         if (!fl4.saddr)
2608                 fl4.saddr = FIB_RES_PREFSRC(res);
2609
2610         dev_out = FIB_RES_DEV(res);
2611         fl4.flowi4_oif = dev_out->ifindex;
2612
2613
2614 make_route:
2615         rth = __mkroute_output(&res, &fl4, oldflp4, dev_out, flags);
2616         if (!IS_ERR(rth)) {
2617                 unsigned int hash;
2618
2619                 hash = rt_hash(oldflp4->daddr, oldflp4->saddr, oldflp4->flowi4_oif,
2620                                rt_genid(dev_net(dev_out)));
2621                 rth = rt_intern_hash(hash, rth, NULL, oldflp4->flowi4_oif);
2622         }
2623
2624 out:
2625         rcu_read_unlock();
2626         return rth;
2627 }
2628
2629 struct rtable *__ip_route_output_key(struct net *net, const struct flowi4 *flp4)
2630 {
2631         struct rtable *rth;
2632         unsigned int hash;
2633
2634         if (!rt_caching(net))
2635                 goto slow_output;
2636
2637         hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2638
2639         rcu_read_lock_bh();
2640         for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2641                 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2642                 if (rth->rt_key_dst == flp4->daddr &&
2643                     rth->rt_key_src == flp4->saddr &&
2644                     rt_is_output_route(rth) &&
2645                     rth->rt_oif == flp4->flowi4_oif &&
2646                     rth->rt_mark == flp4->flowi4_mark &&
2647                     !((rth->rt_tos ^ flp4->flowi4_tos) &
2648                             (IPTOS_RT_MASK | RTO_ONLINK)) &&
2649                     net_eq(dev_net(rth->dst.dev), net) &&
2650                     !rt_is_expired(rth)) {
2651                         dst_use(&rth->dst, jiffies);
2652                         RT_CACHE_STAT_INC(out_hit);
2653                         rcu_read_unlock_bh();
2654                         return rth;
2655                 }
2656                 RT_CACHE_STAT_INC(out_hlist_search);
2657         }
2658         rcu_read_unlock_bh();
2659
2660 slow_output:
2661         return ip_route_output_slow(net, flp4);
2662 }
2663 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2664
2665 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2666 {
2667         return NULL;
2668 }
2669
2670 static unsigned int ipv4_blackhole_default_mtu(const struct dst_entry *dst)
2671 {
2672         return 0;
2673 }
2674
2675 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2676 {
2677 }
2678
2679 static struct dst_ops ipv4_dst_blackhole_ops = {
2680         .family                 =       AF_INET,
2681         .protocol               =       cpu_to_be16(ETH_P_IP),
2682         .destroy                =       ipv4_dst_destroy,
2683         .check                  =       ipv4_blackhole_dst_check,
2684         .default_mtu            =       ipv4_blackhole_default_mtu,
2685         .default_advmss         =       ipv4_default_advmss,
2686         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2687 };
2688
2689 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2690 {
2691         struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, 1);
2692         struct rtable *ort = (struct rtable *) dst_orig;
2693
2694         if (rt) {
2695                 struct dst_entry *new = &rt->dst;
2696
2697                 new->__use = 1;
2698                 new->input = dst_discard;
2699                 new->output = dst_discard;
2700                 dst_copy_metrics(new, &ort->dst);
2701
2702                 new->dev = ort->dst.dev;
2703                 if (new->dev)
2704                         dev_hold(new->dev);
2705
2706                 rt->rt_key_dst = ort->rt_key_dst;
2707                 rt->rt_key_src = ort->rt_key_src;
2708                 rt->rt_tos = ort->rt_tos;
2709                 rt->rt_iif = ort->rt_iif;
2710                 rt->rt_oif = ort->rt_oif;
2711                 rt->rt_mark = ort->rt_mark;
2712
2713                 rt->rt_genid = rt_genid(net);
2714                 rt->rt_flags = ort->rt_flags;
2715                 rt->rt_type = ort->rt_type;
2716                 rt->rt_dst = ort->rt_dst;
2717                 rt->rt_src = ort->rt_src;
2718                 rt->rt_iif = ort->rt_iif;
2719                 rt->rt_gateway = ort->rt_gateway;
2720                 rt->rt_spec_dst = ort->rt_spec_dst;
2721                 rt->peer = ort->peer;
2722                 if (rt->peer)
2723                         atomic_inc(&rt->peer->refcnt);
2724                 rt->fi = ort->fi;
2725                 if (rt->fi)
2726                         atomic_inc(&rt->fi->fib_clntref);
2727
2728                 dst_free(new);
2729         }
2730
2731         dst_release(dst_orig);
2732
2733         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2734 }
2735
2736 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2737                                     struct sock *sk)
2738 {
2739         struct rtable *rt = __ip_route_output_key(net, flp4);
2740
2741         if (IS_ERR(rt))
2742                 return rt;
2743
2744         if (flp4->flowi4_proto) {
2745                 if (!flp4->saddr)
2746                         flp4->saddr = rt->rt_src;
2747                 if (!flp4->daddr)
2748                         flp4->daddr = rt->rt_dst;
2749                 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2750                                                    flowi4_to_flowi(flp4),
2751                                                    sk, 0);
2752         }
2753
2754         return rt;
2755 }
2756 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2757
2758 static int rt_fill_info(struct net *net,
2759                         struct sk_buff *skb, u32 pid, u32 seq, int event,
2760                         int nowait, unsigned int flags)
2761 {
2762         struct rtable *rt = skb_rtable(skb);
2763         struct rtmsg *r;
2764         struct nlmsghdr *nlh;
2765         long expires;
2766         u32 id = 0, ts = 0, tsage = 0, error;
2767
2768         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2769         if (nlh == NULL)
2770                 return -EMSGSIZE;
2771
2772         r = nlmsg_data(nlh);
2773         r->rtm_family    = AF_INET;
2774         r->rtm_dst_len  = 32;
2775         r->rtm_src_len  = 0;
2776         r->rtm_tos      = rt->rt_tos;
2777         r->rtm_table    = RT_TABLE_MAIN;
2778         NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2779         r->rtm_type     = rt->rt_type;
2780         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2781         r->rtm_protocol = RTPROT_UNSPEC;
2782         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2783         if (rt->rt_flags & RTCF_NOTIFY)
2784                 r->rtm_flags |= RTM_F_NOTIFY;
2785
2786         NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2787
2788         if (rt->rt_key_src) {
2789                 r->rtm_src_len = 32;
2790                 NLA_PUT_BE32(skb, RTA_SRC, rt->rt_key_src);
2791         }
2792         if (rt->dst.dev)
2793                 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2794 #ifdef CONFIG_IP_ROUTE_CLASSID
2795         if (rt->dst.tclassid)
2796                 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2797 #endif
2798         if (rt_is_input_route(rt))
2799                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2800         else if (rt->rt_src != rt->rt_key_src)
2801                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2802
2803         if (rt->rt_dst != rt->rt_gateway)
2804                 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2805
2806         if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2807                 goto nla_put_failure;
2808
2809         if (rt->rt_mark)
2810                 NLA_PUT_BE32(skb, RTA_MARK, rt->rt_mark);
2811
2812         error = rt->dst.error;
2813         expires = (rt->peer && rt->peer->pmtu_expires) ?
2814                 rt->peer->pmtu_expires - jiffies : 0;
2815         if (rt->peer) {
2816                 inet_peer_refcheck(rt->peer);
2817                 id = atomic_read(&rt->peer->ip_id_count) & 0xffff;
2818                 if (rt->peer->tcp_ts_stamp) {
2819                         ts = rt->peer->tcp_ts;
2820                         tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2821                 }
2822         }
2823
2824         if (rt_is_input_route(rt)) {
2825 #ifdef CONFIG_IP_MROUTE
2826                 __be32 dst = rt->rt_dst;
2827
2828                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2829                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2830                         int err = ipmr_get_route(net, skb, r, nowait);
2831                         if (err <= 0) {
2832                                 if (!nowait) {
2833                                         if (err == 0)
2834                                                 return 0;
2835                                         goto nla_put_failure;
2836                                 } else {
2837                                         if (err == -EMSGSIZE)
2838                                                 goto nla_put_failure;
2839                                         error = err;
2840                                 }
2841                         }
2842                 } else
2843 #endif
2844                         NLA_PUT_U32(skb, RTA_IIF, rt->rt_iif);
2845         }
2846
2847         if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2848                                expires, error) < 0)
2849                 goto nla_put_failure;
2850
2851         return nlmsg_end(skb, nlh);
2852
2853 nla_put_failure:
2854         nlmsg_cancel(skb, nlh);
2855         return -EMSGSIZE;
2856 }
2857
2858 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2859 {
2860         struct net *net = sock_net(in_skb->sk);
2861         struct rtmsg *rtm;
2862         struct nlattr *tb[RTA_MAX+1];
2863         struct rtable *rt = NULL;
2864         __be32 dst = 0;
2865         __be32 src = 0;
2866         u32 iif;
2867         int err;
2868         int mark;
2869         struct sk_buff *skb;
2870
2871         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2872         if (err < 0)
2873                 goto errout;
2874
2875         rtm = nlmsg_data(nlh);
2876
2877         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2878         if (skb == NULL) {
2879                 err = -ENOBUFS;
2880                 goto errout;
2881         }
2882
2883         /* Reserve room for dummy headers, this skb can pass
2884            through good chunk of routing engine.
2885          */
2886         skb_reset_mac_header(skb);
2887         skb_reset_network_header(skb);
2888
2889         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2890         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2891         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2892
2893         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2894         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2895         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2896         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2897
2898         if (iif) {
2899                 struct net_device *dev;
2900
2901                 dev = __dev_get_by_index(net, iif);
2902                 if (dev == NULL) {
2903                         err = -ENODEV;
2904                         goto errout_free;
2905                 }
2906
2907                 skb->protocol   = htons(ETH_P_IP);
2908                 skb->dev        = dev;
2909                 skb->mark       = mark;
2910                 local_bh_disable();
2911                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2912                 local_bh_enable();
2913
2914                 rt = skb_rtable(skb);
2915                 if (err == 0 && rt->dst.error)
2916                         err = -rt->dst.error;
2917         } else {
2918                 struct flowi4 fl4 = {
2919                         .daddr = dst,
2920                         .saddr = src,
2921                         .flowi4_tos = rtm->rtm_tos,
2922                         .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2923                         .flowi4_mark = mark,
2924                 };
2925                 rt = ip_route_output_key(net, &fl4);
2926
2927                 err = 0;
2928                 if (IS_ERR(rt))
2929                         err = PTR_ERR(rt);
2930         }
2931
2932         if (err)
2933                 goto errout_free;
2934
2935         skb_dst_set(skb, &rt->dst);
2936         if (rtm->rtm_flags & RTM_F_NOTIFY)
2937                 rt->rt_flags |= RTCF_NOTIFY;
2938
2939         err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2940                            RTM_NEWROUTE, 0, 0);
2941         if (err <= 0)
2942                 goto errout_free;
2943
2944         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2945 errout:
2946         return err;
2947
2948 errout_free:
2949         kfree_skb(skb);
2950         goto errout;
2951 }
2952
2953 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
2954 {
2955         struct rtable *rt;
2956         int h, s_h;
2957         int idx, s_idx;
2958         struct net *net;
2959
2960         net = sock_net(skb->sk);
2961
2962         s_h = cb->args[0];
2963         if (s_h < 0)
2964                 s_h = 0;
2965         s_idx = idx = cb->args[1];
2966         for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
2967                 if (!rt_hash_table[h].chain)
2968                         continue;
2969                 rcu_read_lock_bh();
2970                 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
2971                      rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
2972                         if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
2973                                 continue;
2974                         if (rt_is_expired(rt))
2975                                 continue;
2976                         skb_dst_set_noref(skb, &rt->dst);
2977                         if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
2978                                          cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2979                                          1, NLM_F_MULTI) <= 0) {
2980                                 skb_dst_drop(skb);
2981                                 rcu_read_unlock_bh();
2982                                 goto done;
2983                         }
2984                         skb_dst_drop(skb);
2985                 }
2986                 rcu_read_unlock_bh();
2987         }
2988
2989 done:
2990         cb->args[0] = h;
2991         cb->args[1] = idx;
2992         return skb->len;
2993 }
2994
2995 void ip_rt_multicast_event(struct in_device *in_dev)
2996 {
2997         rt_cache_flush(dev_net(in_dev->dev), 0);
2998 }
2999
3000 #ifdef CONFIG_SYSCTL
3001 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3002                                         void __user *buffer,
3003                                         size_t *lenp, loff_t *ppos)
3004 {
3005         if (write) {
3006                 int flush_delay;
3007                 ctl_table ctl;
3008                 struct net *net;
3009
3010                 memcpy(&ctl, __ctl, sizeof(ctl));
3011                 ctl.data = &flush_delay;
3012                 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3013
3014                 net = (struct net *)__ctl->extra1;
3015                 rt_cache_flush(net, flush_delay);
3016                 return 0;
3017         }
3018
3019         return -EINVAL;
3020 }
3021
3022 static ctl_table ipv4_route_table[] = {
3023         {
3024                 .procname       = "gc_thresh",
3025                 .data           = &ipv4_dst_ops.gc_thresh,
3026                 .maxlen         = sizeof(int),
3027                 .mode           = 0644,
3028                 .proc_handler   = proc_dointvec,
3029         },
3030         {
3031                 .procname       = "max_size",
3032                 .data           = &ip_rt_max_size,
3033                 .maxlen         = sizeof(int),
3034                 .mode           = 0644,
3035                 .proc_handler   = proc_dointvec,
3036         },
3037         {
3038                 /*  Deprecated. Use gc_min_interval_ms */
3039
3040                 .procname       = "gc_min_interval",
3041                 .data           = &ip_rt_gc_min_interval,
3042                 .maxlen         = sizeof(int),
3043                 .mode           = 0644,
3044                 .proc_handler   = proc_dointvec_jiffies,
3045         },
3046         {
3047                 .procname       = "gc_min_interval_ms",
3048                 .data           = &ip_rt_gc_min_interval,
3049                 .maxlen         = sizeof(int),
3050                 .mode           = 0644,
3051                 .proc_handler   = proc_dointvec_ms_jiffies,
3052         },
3053         {
3054                 .procname       = "gc_timeout",
3055                 .data           = &ip_rt_gc_timeout,
3056                 .maxlen         = sizeof(int),
3057                 .mode           = 0644,
3058                 .proc_handler   = proc_dointvec_jiffies,
3059         },
3060         {
3061                 .procname       = "gc_interval",
3062                 .data           = &ip_rt_gc_interval,
3063                 .maxlen         = sizeof(int),
3064                 .mode           = 0644,
3065                 .proc_handler   = proc_dointvec_jiffies,
3066         },
3067         {
3068                 .procname       = "redirect_load",
3069                 .data           = &ip_rt_redirect_load,
3070                 .maxlen         = sizeof(int),
3071                 .mode           = 0644,
3072                 .proc_handler   = proc_dointvec,
3073         },
3074         {
3075                 .procname       = "redirect_number",
3076                 .data           = &ip_rt_redirect_number,
3077                 .maxlen         = sizeof(int),
3078                 .mode           = 0644,
3079                 .proc_handler   = proc_dointvec,
3080         },
3081         {
3082                 .procname       = "redirect_silence",
3083                 .data           = &ip_rt_redirect_silence,
3084                 .maxlen         = sizeof(int),
3085                 .mode           = 0644,
3086                 .proc_handler   = proc_dointvec,
3087         },
3088         {
3089                 .procname       = "error_cost",
3090                 .data           = &ip_rt_error_cost,
3091                 .maxlen         = sizeof(int),
3092                 .mode           = 0644,
3093                 .proc_handler   = proc_dointvec,
3094         },
3095         {
3096                 .procname       = "error_burst",
3097                 .data           = &ip_rt_error_burst,
3098                 .maxlen         = sizeof(int),
3099                 .mode           = 0644,
3100                 .proc_handler   = proc_dointvec,
3101         },
3102         {
3103                 .procname       = "gc_elasticity",
3104                 .data           = &ip_rt_gc_elasticity,
3105                 .maxlen         = sizeof(int),
3106                 .mode           = 0644,
3107                 .proc_handler   = proc_dointvec,
3108         },
3109         {
3110                 .procname       = "mtu_expires",
3111                 .data           = &ip_rt_mtu_expires,
3112                 .maxlen         = sizeof(int),
3113                 .mode           = 0644,
3114                 .proc_handler   = proc_dointvec_jiffies,
3115         },
3116         {
3117                 .procname       = "min_pmtu",
3118                 .data           = &ip_rt_min_pmtu,
3119                 .maxlen         = sizeof(int),
3120                 .mode           = 0644,
3121                 .proc_handler   = proc_dointvec,
3122         },
3123         {
3124                 .procname       = "min_adv_mss",
3125                 .data           = &ip_rt_min_advmss,
3126                 .maxlen         = sizeof(int),
3127                 .mode           = 0644,
3128                 .proc_handler   = proc_dointvec,
3129         },
3130         { }
3131 };
3132
3133 static struct ctl_table empty[1];
3134
3135 static struct ctl_table ipv4_skeleton[] =
3136 {
3137         { .procname = "route",
3138           .mode = 0555, .child = ipv4_route_table},
3139         { .procname = "neigh",
3140           .mode = 0555, .child = empty},
3141         { }
3142 };
3143
3144 static __net_initdata struct ctl_path ipv4_path[] = {
3145         { .procname = "net", },
3146         { .procname = "ipv4", },
3147         { },
3148 };
3149
3150 static struct ctl_table ipv4_route_flush_table[] = {
3151         {
3152                 .procname       = "flush",
3153                 .maxlen         = sizeof(int),
3154                 .mode           = 0200,
3155                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3156         },
3157         { },
3158 };
3159
3160 static __net_initdata struct ctl_path ipv4_route_path[] = {
3161         { .procname = "net", },
3162         { .procname = "ipv4", },
3163         { .procname = "route", },
3164         { },
3165 };
3166
3167 static __net_init int sysctl_route_net_init(struct net *net)
3168 {
3169         struct ctl_table *tbl;
3170
3171         tbl = ipv4_route_flush_table;
3172         if (!net_eq(net, &init_net)) {
3173                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3174                 if (tbl == NULL)
3175                         goto err_dup;
3176         }
3177         tbl[0].extra1 = net;
3178
3179         net->ipv4.route_hdr =
3180                 register_net_sysctl_table(net, ipv4_route_path, tbl);
3181         if (net->ipv4.route_hdr == NULL)
3182                 goto err_reg;
3183         return 0;
3184
3185 err_reg:
3186         if (tbl != ipv4_route_flush_table)
3187                 kfree(tbl);
3188 err_dup:
3189         return -ENOMEM;
3190 }
3191
3192 static __net_exit void sysctl_route_net_exit(struct net *net)
3193 {
3194         struct ctl_table *tbl;
3195
3196         tbl = net->ipv4.route_hdr->ctl_table_arg;
3197         unregister_net_sysctl_table(net->ipv4.route_hdr);
3198         BUG_ON(tbl == ipv4_route_flush_table);
3199         kfree(tbl);
3200 }
3201
3202 static __net_initdata struct pernet_operations sysctl_route_ops = {
3203         .init = sysctl_route_net_init,
3204         .exit = sysctl_route_net_exit,
3205 };
3206 #endif
3207
3208 static __net_init int rt_genid_init(struct net *net)
3209 {
3210         get_random_bytes(&net->ipv4.rt_genid,
3211                          sizeof(net->ipv4.rt_genid));
3212         return 0;
3213 }
3214
3215 static __net_initdata struct pernet_operations rt_genid_ops = {
3216         .init = rt_genid_init,
3217 };
3218
3219
3220 #ifdef CONFIG_IP_ROUTE_CLASSID
3221 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3222 #endif /* CONFIG_IP_ROUTE_CLASSID */
3223
3224 static __initdata unsigned long rhash_entries;
3225 static int __init set_rhash_entries(char *str)
3226 {
3227         if (!str)
3228                 return 0;
3229         rhash_entries = simple_strtoul(str, &str, 0);
3230         return 1;
3231 }
3232 __setup("rhash_entries=", set_rhash_entries);
3233
3234 int __init ip_rt_init(void)
3235 {
3236         int rc = 0;
3237
3238 #ifdef CONFIG_IP_ROUTE_CLASSID
3239         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3240         if (!ip_rt_acct)
3241                 panic("IP: failed to allocate ip_rt_acct\n");
3242 #endif
3243
3244         ipv4_dst_ops.kmem_cachep =
3245                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3246                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3247
3248         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3249
3250         if (dst_entries_init(&ipv4_dst_ops) < 0)
3251                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3252
3253         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3254                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3255
3256         rt_hash_table = (struct rt_hash_bucket *)
3257                 alloc_large_system_hash("IP route cache",
3258                                         sizeof(struct rt_hash_bucket),
3259                                         rhash_entries,
3260                                         (totalram_pages >= 128 * 1024) ?
3261                                         15 : 17,
3262                                         0,
3263                                         &rt_hash_log,
3264                                         &rt_hash_mask,
3265                                         rhash_entries ? 0 : 512 * 1024);
3266         memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3267         rt_hash_lock_init();
3268
3269         ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3270         ip_rt_max_size = (rt_hash_mask + 1) * 16;
3271
3272         devinet_init();
3273         ip_fib_init();
3274
3275         if (ip_rt_proc_init())
3276                 printk(KERN_ERR "Unable to create route proc files\n");
3277 #ifdef CONFIG_XFRM
3278         xfrm_init();
3279         xfrm4_init(ip_rt_max_size);
3280 #endif
3281         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3282
3283 #ifdef CONFIG_SYSCTL
3284         register_pernet_subsys(&sysctl_route_ops);
3285 #endif
3286         register_pernet_subsys(&rt_genid_ops);
3287         return rc;
3288 }
3289
3290 #ifdef CONFIG_SYSCTL
3291 /*
3292  * We really need to sanitize the damn ipv4 init order, then all
3293  * this nonsense will go away.
3294  */
3295 void __init ip_static_sysctl_init(void)
3296 {
3297         register_sysctl_paths(ipv4_path, ipv4_skeleton);
3298 }
3299 #endif