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(oldflp) \
 113     ((u32)(oldflp->fl4_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 static struct delayed_work expires_work;
 135 static unsigned long expires_ljiffies;
 136
 137 /*
 138  *      Interface to generic destination cache.
 139  */
 140
 141 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
 142 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
 143 static unsigned int      ipv4_default_mtu(const struct dst_entry *dst);
 144 static void              ipv4_dst_destroy(struct dst_entry *dst);
 145 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
 146 static void              ipv4_link_failure(struct sk_buff *skb);
 147 static void              ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
 148 static int rt_garbage_collect(struct dst_ops *ops);
 149
 150 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 151                             int how)
 152 {
 153 }
 154
 155 static struct dst_ops ipv4_dst_ops = {
 156         .family =               AF_INET,
 157         .protocol =             cpu_to_be16(ETH_P_IP),
 158         .gc =                   rt_garbage_collect,
 159         .check =                ipv4_dst_check,
 160         .default_advmss =       ipv4_default_advmss,
 161         .default_mtu =          ipv4_default_mtu,
 162         .destroy =              ipv4_dst_destroy,
 163         .ifdown =               ipv4_dst_ifdown,
 164         .negative_advice =      ipv4_negative_advice,
 165         .link_failure =         ipv4_link_failure,
 166         .update_pmtu =          ip_rt_update_pmtu,
 167         .local_out =            __ip_local_out,
 168 };
 169
 170 #define ECN_OR_COST(class)      TC_PRIO_##class
 171
 172 const __u8 ip_tos2prio[16] = {
 173         TC_PRIO_BESTEFFORT,
 174         ECN_OR_COST(FILLER),
 175         TC_PRIO_BESTEFFORT,
 176         ECN_OR_COST(BESTEFFORT),
 177         TC_PRIO_BULK,
 178         ECN_OR_COST(BULK),
 179         TC_PRIO_BULK,
 180         ECN_OR_COST(BULK),
 181         TC_PRIO_INTERACTIVE,
 182         ECN_OR_COST(INTERACTIVE),
 183         TC_PRIO_INTERACTIVE,
 184         ECN_OR_COST(INTERACTIVE),
 185         TC_PRIO_INTERACTIVE_BULK,
 186         ECN_OR_COST(INTERACTIVE_BULK),
 187         TC_PRIO_INTERACTIVE_BULK,
 188         ECN_OR_COST(INTERACTIVE_BULK)
 189 };
 190
 191
 192 /*
 193  * Route cache.
 194  */
 195
 196 /* The locking scheme is rather straight forward:
 197  *
 198  * 1) Read-Copy Update protects the buckets of the central route hash.
 199  * 2) Only writers remove entries, and they hold the lock
 200  *    as they look at rtable reference counts.
 201  * 3) Only readers acquire references to rtable entries,
 202  *    they do so with atomic increments and with the
 203  *    lock held.
 204  */
 205
 206 struct rt_hash_bucket {
 207         struct rtable __rcu     *chain;
 208 };
 209
 210 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
 211         defined(CONFIG_PROVE_LOCKING)
 212 /*
 213  * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
 214  * The size of this table is a power of two and depends on the number of CPUS.
 215  * (on lockdep we have a quite big spinlock_t, so keep the size down there)
 216  */
 217 #ifdef CONFIG_LOCKDEP
 218 # define RT_HASH_LOCK_SZ        256
 219 #else
 220 # if NR_CPUS >= 32
 221 #  define RT_HASH_LOCK_SZ       4096
 222 # elif NR_CPUS >= 16
 223 #  define RT_HASH_LOCK_SZ       2048
 224 # elif NR_CPUS >= 8
 225 #  define RT_HASH_LOCK_SZ       1024
 226 # elif NR_CPUS >= 4
 227 #  define RT_HASH_LOCK_SZ       512
 228 # else
 229 #  define RT_HASH_LOCK_SZ       256
 230 # endif
 231 #endif
 232
 233 static spinlock_t       *rt_hash_locks;
 234 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
 235
 236 static __init void rt_hash_lock_init(void)
 237 {
 238         int i;
 239
 240         rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
 241                         GFP_KERNEL);
 242         if (!rt_hash_locks)
 243                 panic("IP: failed to allocate rt_hash_locks\n");
 244
 245         for (i = 0; i < RT_HASH_LOCK_SZ; i++)
 246                 spin_lock_init(&rt_hash_locks[i]);
 247 }
 248 #else
 249 # define rt_hash_lock_addr(slot) NULL
 250
 251 static inline void rt_hash_lock_init(void)
 252 {
 253 }
 254 #endif
 255
 256 static struct rt_hash_bucket    *rt_hash_table __read_mostly;
 257 static unsigned                 rt_hash_mask __read_mostly;
 258 static unsigned int             rt_hash_log  __read_mostly;
 259
 260 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 261 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
 262
 263 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
 264                                    int genid)
 265 {
 266         return jhash_3words((__force u32)daddr, (__force u32)saddr,
 267                             idx, genid)
 268                 & rt_hash_mask;
 269 }
 270
 271 static inline int rt_genid(struct net *net)
 272 {
 273         return atomic_read(&net->ipv4.rt_genid);
 274 }
 275
 276 #ifdef CONFIG_PROC_FS
 277 struct rt_cache_iter_state {
 278         struct seq_net_private p;
 279         int bucket;
 280         int genid;
 281 };
 282
 283 static struct rtable *rt_cache_get_first(struct seq_file *seq)
 284 {
 285         struct rt_cache_iter_state *st = seq->private;
 286         struct rtable *r = NULL;
 287
 288         for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
 289                 if (!rcu_dereference_raw(rt_hash_table[st->bucket].chain))
 290                         continue;
 291                 rcu_read_lock_bh();
 292                 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
 293                 while (r) {
 294                         if (dev_net(r->dst.dev) == seq_file_net(seq) &&
 295                             r->rt_genid == st->genid)
 296                                 return r;
 297                         r = rcu_dereference_bh(r->dst.rt_next);
 298                 }
 299                 rcu_read_unlock_bh();
 300         }
 301         return r;
 302 }
 303
 304 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
 305                                           struct rtable *r)
 306 {
 307         struct rt_cache_iter_state *st = seq->private;
 308
 309         r = rcu_dereference_bh(r->dst.rt_next);
 310         while (!r) {
 311                 rcu_read_unlock_bh();
 312                 do {
 313                         if (--st->bucket < 0)
 314                                 return NULL;
 315                 } while (!rcu_dereference_raw(rt_hash_table[st->bucket].chain));
 316                 rcu_read_lock_bh();
 317                 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
 318         }
 319         return r;
 320 }
 321
 322 static struct rtable *rt_cache_get_next(struct seq_file *seq,
 323                                         struct rtable *r)
 324 {
 325         struct rt_cache_iter_state *st = seq->private;
 326         while ((r = __rt_cache_get_next(seq, r)) != NULL) {
 327                 if (dev_net(r->dst.dev) != seq_file_net(seq))
 328                         continue;
 329                 if (r->rt_genid == st->genid)
 330                         break;
 331         }
 332         return r;
 333 }
 334
 335 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
 336 {
 337         struct rtable *r = rt_cache_get_first(seq);
 338
 339         if (r)
 340                 while (pos && (r = rt_cache_get_next(seq, r)))
 341                         --pos;
 342         return pos ? NULL : r;
 343 }
 344
 345 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 346 {
 347         struct rt_cache_iter_state *st = seq->private;
 348         if (*pos)
 349                 return rt_cache_get_idx(seq, *pos - 1);
 350         st->genid = rt_genid(seq_file_net(seq));
 351         return SEQ_START_TOKEN;
 352 }
 353
 354 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 355 {
 356         struct rtable *r;
 357
 358         if (v == SEQ_START_TOKEN)
 359                 r = rt_cache_get_first(seq);
 360         else
 361                 r = rt_cache_get_next(seq, v);
 362         ++*pos;
 363         return r;
 364 }
 365
 366 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 367 {
 368         if (v && v != SEQ_START_TOKEN)
 369                 rcu_read_unlock_bh();
 370 }
 371
 372 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 373 {
 374         if (v == SEQ_START_TOKEN)
 375                 seq_printf(seq, "%-127s\n",
 376                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 377                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 378                            "HHUptod\tSpecDst");
 379         else {
 380                 struct rtable *r = v;
 381                 int len;
 382
 383                 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
 384                               "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
 385                         r->dst.dev ? r->dst.dev->name : "*",
 386                         (__force u32)r->rt_dst,
 387                         (__force u32)r->rt_gateway,
 388                         r->rt_flags, atomic_read(&r->dst.__refcnt),
 389                         r->dst.__use, 0, (__force u32)r->rt_src,
 390                         dst_metric_advmss(&r->dst) + 40,
 391                         dst_metric(&r->dst, RTAX_WINDOW),
 392                         (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
 393                               dst_metric(&r->dst, RTAX_RTTVAR)),
 394                         r->fl.fl4_tos,
 395                         r->dst.hh ? atomic_read(&r->dst.hh->hh_refcnt) : -1,
 396                         r->dst.hh ? (r->dst.hh->hh_output ==
 397                                        dev_queue_xmit) : 0,
 398                         r->rt_spec_dst, &len);
 399
 400                 seq_printf(seq, "%*s\n", 127 - len, "");
 401         }
 402         return 0;
 403 }
 404
 405 static const struct seq_operations rt_cache_seq_ops = {
 406         .start  = rt_cache_seq_start,
 407         .next   = rt_cache_seq_next,
 408         .stop   = rt_cache_seq_stop,
 409         .show   = rt_cache_seq_show,
 410 };
 411
 412 static int rt_cache_seq_open(struct inode *inode, struct file *file)
 413 {
 414         return seq_open_net(inode, file, &rt_cache_seq_ops,
 415                         sizeof(struct rt_cache_iter_state));
 416 }
 417
 418 static const struct file_operations rt_cache_seq_fops = {
 419         .owner   = THIS_MODULE,
 420         .open    = rt_cache_seq_open,
 421         .read    = seq_read,
 422         .llseek  = seq_lseek,
 423         .release = seq_release_net,
 424 };
 425
 426
 427 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 428 {
 429         int cpu;
 430
 431         if (*pos == 0)
 432                 return SEQ_START_TOKEN;
 433
 434         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 435                 if (!cpu_possible(cpu))
 436                         continue;
 437                 *pos = cpu+1;
 438                 return &per_cpu(rt_cache_stat, cpu);
 439         }
 440         return NULL;
 441 }
 442
 443 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 444 {
 445         int cpu;
 446
 447         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 448                 if (!cpu_possible(cpu))
 449                         continue;
 450                 *pos = cpu+1;
 451                 return &per_cpu(rt_cache_stat, cpu);
 452         }
 453         return NULL;
 454
 455 }
 456
 457 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 458 {
 459
 460 }
 461
 462 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 463 {
 464         struct rt_cache_stat *st = v;
 465
 466         if (v == SEQ_START_TOKEN) {
 467                 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");
 468                 return 0;
 469         }
 470
 471         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 472                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 473                    dst_entries_get_slow(&ipv4_dst_ops),
 474                    st->in_hit,
 475                    st->in_slow_tot,
 476                    st->in_slow_mc,
 477                    st->in_no_route,
 478                    st->in_brd,
 479                    st->in_martian_dst,
 480                    st->in_martian_src,
 481
 482                    st->out_hit,
 483                    st->out_slow_tot,
 484                    st->out_slow_mc,
 485
 486                    st->gc_total,
 487                    st->gc_ignored,
 488                    st->gc_goal_miss,
 489                    st->gc_dst_overflow,
 490                    st->in_hlist_search,
 491                    st->out_hlist_search
 492                 );
 493         return 0;
 494 }
 495
 496 static const struct seq_operations rt_cpu_seq_ops = {
 497         .start  = rt_cpu_seq_start,
 498         .next   = rt_cpu_seq_next,
 499         .stop   = rt_cpu_seq_stop,
 500         .show   = rt_cpu_seq_show,
 501 };
 502
 503
 504 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
 505 {
 506         return seq_open(file, &rt_cpu_seq_ops);
 507 }
 508
 509 static const struct file_operations rt_cpu_seq_fops = {
 510         .owner   = THIS_MODULE,
 511         .open    = rt_cpu_seq_open,
 512         .read    = seq_read,
 513         .llseek  = seq_lseek,
 514         .release = seq_release,
 515 };
 516
 517 #ifdef CONFIG_NET_CLS_ROUTE
 518 static int rt_acct_proc_show(struct seq_file *m, void *v)
 519 {
 520         struct ip_rt_acct *dst, *src;
 521         unsigned int i, j;
 522
 523         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 524         if (!dst)
 525                 return -ENOMEM;
 526
 527         for_each_possible_cpu(i) {
 528                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 529                 for (j = 0; j < 256; j++) {
 530                         dst[j].o_bytes   += src[j].o_bytes;
 531                         dst[j].o_packets += src[j].o_packets;
 532                         dst[j].i_bytes   += src[j].i_bytes;
 533                         dst[j].i_packets += src[j].i_packets;
 534                 }
 535         }
 536
 537         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 538         kfree(dst);
 539         return 0;
 540 }
 541
 542 static int rt_acct_proc_open(struct inode *inode, struct file *file)
 543 {
 544         return single_open(file, rt_acct_proc_show, NULL);
 545 }
 546
 547 static const struct file_operations rt_acct_proc_fops = {
 548         .owner          = THIS_MODULE,
 549         .open           = rt_acct_proc_open,
 550         .read           = seq_read,
 551         .llseek         = seq_lseek,
 552         .release        = single_release,
 553 };
 554 #endif
 555
 556 static int __net_init ip_rt_do_proc_init(struct net *net)
 557 {
 558         struct proc_dir_entry *pde;
 559
 560         pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
 561                         &rt_cache_seq_fops);
 562         if (!pde)
 563                 goto err1;
 564
 565         pde = proc_create("rt_cache", S_IRUGO,
 566                           net->proc_net_stat, &rt_cpu_seq_fops);
 567         if (!pde)
 568                 goto err2;
 569
 570 #ifdef CONFIG_NET_CLS_ROUTE
 571         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
 572         if (!pde)
 573                 goto err3;
 574 #endif
 575         return 0;
 576
 577 #ifdef CONFIG_NET_CLS_ROUTE
 578 err3:
 579         remove_proc_entry("rt_cache", net->proc_net_stat);
 580 #endif
 581 err2:
 582         remove_proc_entry("rt_cache", net->proc_net);
 583 err1:
 584         return -ENOMEM;
 585 }
 586
 587 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 588 {
 589         remove_proc_entry("rt_cache", net->proc_net_stat);
 590         remove_proc_entry("rt_cache", net->proc_net);
 591 #ifdef CONFIG_NET_CLS_ROUTE
 592         remove_proc_entry("rt_acct", net->proc_net);
 593 #endif
 594 }
 595
 596 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 597         .init = ip_rt_do_proc_init,
 598         .exit = ip_rt_do_proc_exit,
 599 };
 600
 601 static int __init ip_rt_proc_init(void)
 602 {
 603         return register_pernet_subsys(&ip_rt_proc_ops);
 604 }
 605
 606 #else
 607 static inline int ip_rt_proc_init(void)
 608 {
 609         return 0;
 610 }
 611 #endif /* CONFIG_PROC_FS */
 612
 613 static inline void rt_free(struct rtable *rt)
 614 {
 615         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
 616 }
 617
 618 static inline void rt_drop(struct rtable *rt)
 619 {
 620         ip_rt_put(rt);
 621         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
 622 }
 623
 624 static inline int rt_fast_clean(struct rtable *rth)
 625 {
 626         /* Kill broadcast/multicast entries very aggresively, if they
 627            collide in hash table with more useful entries */
 628         return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
 629                 rt_is_input_route(rth) && rth->dst.rt_next;
 630 }
 631
 632 static inline int rt_valuable(struct rtable *rth)
 633 {
 634         return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
 635                 rth->dst.expires;
 636 }
 637
 638 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
 639 {
 640         unsigned long age;
 641         int ret = 0;
 642
 643         if (atomic_read(&rth->dst.__refcnt))
 644                 goto out;
 645
 646         ret = 1;
 647         if (rth->dst.expires &&
 648             time_after_eq(jiffies, rth->dst.expires))
 649                 goto out;
 650
 651         age = jiffies - rth->dst.lastuse;
 652         ret = 0;
 653         if ((age <= tmo1 && !rt_fast_clean(rth)) ||
 654             (age <= tmo2 && rt_valuable(rth)))
 655                 goto out;
 656         ret = 1;
 657 out:    return ret;
 658 }
 659
 660 /* Bits of score are:
 661  * 31: very valuable
 662  * 30: not quite useless
 663  * 29..0: usage counter
 664  */
 665 static inline u32 rt_score(struct rtable *rt)
 666 {
 667         u32 score = jiffies - rt->dst.lastuse;
 668
 669         score = ~score & ~(3<<30);
 670
 671         if (rt_valuable(rt))
 672                 score |= (1<<31);
 673
 674         if (rt_is_output_route(rt) ||
 675             !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
 676                 score |= (1<<30);
 677
 678         return score;
 679 }
 680
 681 static inline bool rt_caching(const struct net *net)
 682 {
 683         return net->ipv4.current_rt_cache_rebuild_count <=
 684                 net->ipv4.sysctl_rt_cache_rebuild_count;
 685 }
 686
 687 static inline bool compare_hash_inputs(const struct flowi *fl1,
 688                                         const struct flowi *fl2)
 689 {
 690         return ((((__force u32)fl1->fl4_dst ^ (__force u32)fl2->fl4_dst) |
 691                 ((__force u32)fl1->fl4_src ^ (__force u32)fl2->fl4_src) |
 692                 (fl1->iif ^ fl2->iif)) == 0);
 693 }
 694
 695 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
 696 {
 697         return (((__force u32)fl1->fl4_dst ^ (__force u32)fl2->fl4_dst) |
 698                 ((__force u32)fl1->fl4_src ^ (__force u32)fl2->fl4_src) |
 699                 (fl1->mark ^ fl2->mark) |
 700                 (*(u16 *)&fl1->fl4_tos ^ *(u16 *)&fl2->fl4_tos) |
 701                 (fl1->oif ^ fl2->oif) |
 702                 (fl1->iif ^ fl2->iif)) == 0;
 703 }
 704
 705 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
 706 {
 707         return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
 708 }
 709
 710 static inline int rt_is_expired(struct rtable *rth)
 711 {
 712         return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
 713 }
 714
 715 /*
 716  * Perform a full scan of hash table and free all entries.
 717  * Can be called by a softirq or a process.
 718  * In the later case, we want to be reschedule if necessary
 719  */
 720 static void rt_do_flush(int process_context)
 721 {
 722         unsigned int i;
 723         struct rtable *rth, *next;
 724         struct rtable * tail;
 725
 726         for (i = 0; i <= rt_hash_mask; i++) {
 727                 if (process_context && need_resched())
 728                         cond_resched();
 729                 rth = rcu_dereference_raw(rt_hash_table[i].chain);
 730                 if (!rth)
 731                         continue;
 732
 733                 spin_lock_bh(rt_hash_lock_addr(i));
 734 #ifdef CONFIG_NET_NS
 735                 {
 736                 struct rtable __rcu **prev;
 737                 struct rtable *p;
 738
 739                 rth = rcu_dereference_protected(rt_hash_table[i].chain,
 740                         lockdep_is_held(rt_hash_lock_addr(i)));
 741
 742                 /* defer releasing the head of the list after spin_unlock */
 743                 for (tail = rth; tail;
 744                      tail = rcu_dereference_protected(tail->dst.rt_next,
 745                                 lockdep_is_held(rt_hash_lock_addr(i))))
 746                         if (!rt_is_expired(tail))
 747                                 break;
 748                 if (rth != tail)
 749                         rt_hash_table[i].chain = tail;
 750
 751                 /* call rt_free on entries after the tail requiring flush */
 752                 prev = &rt_hash_table[i].chain;
 753                 for (p = rcu_dereference_protected(*prev,
 754                                 lockdep_is_held(rt_hash_lock_addr(i)));
 755                      p != NULL;
 756                      p = next) {
 757                         next = rcu_dereference_protected(p->dst.rt_next,
 758                                 lockdep_is_held(rt_hash_lock_addr(i)));
 759                         if (!rt_is_expired(p)) {
 760                                 prev = &p->dst.rt_next;
 761                         } else {
 762                                 *prev = next;
 763                                 rt_free(p);
 764                         }
 765                 }
 766                 }
 767 #else
 768                 rth = rcu_dereference_protected(rt_hash_table[i].chain,
 769                         lockdep_is_held(rt_hash_lock_addr(i)));
 770                 rcu_assign_pointer(rt_hash_table[i].chain, NULL);
 771                 tail = NULL;
 772 #endif
 773                 spin_unlock_bh(rt_hash_lock_addr(i));
 774
 775                 for (; rth != tail; rth = next) {
 776                         next = rcu_dereference_protected(rth->dst.rt_next, 1);
 777                         rt_free(rth);
 778                 }
 779         }
 780 }
 781
 782 /*
 783  * While freeing expired entries, we compute average chain length
 784  * and standard deviation, using fixed-point arithmetic.
 785  * This to have an estimation of rt_chain_length_max
 786  *  rt_chain_length_max = max(elasticity, AVG + 4*SD)
 787  * We use 3 bits for frational part, and 29 (or 61) for magnitude.
 788  */
 789
 790 #define FRACT_BITS 3
 791 #define ONE (1UL << FRACT_BITS)
 792
 793 /*
 794  * Given a hash chain and an item in this hash chain,
 795  * find if a previous entry has the same hash_inputs
 796  * (but differs on tos, mark or oif)
 797  * Returns 0 if an alias is found.
 798  * Returns ONE if rth has no alias before itself.
 799  */
 800 static int has_noalias(const struct rtable *head, const struct rtable *rth)
 801 {
 802         const struct rtable *aux = head;
 803
 804         while (aux != rth) {
 805                 if (compare_hash_inputs(&aux->fl, &rth->fl))
 806                         return 0;
 807                 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
 808         }
 809         return ONE;
 810 }
 811
 812 static void rt_check_expire(void)
 813 {
 814         static unsigned int rover;
 815         unsigned int i = rover, goal;
 816         struct rtable *rth;
 817         struct rtable __rcu **rthp;
 818         unsigned long samples = 0;
 819         unsigned long sum = 0, sum2 = 0;
 820         unsigned long delta;
 821         u64 mult;
 822
 823         delta = jiffies - expires_ljiffies;
 824         expires_ljiffies = jiffies;
 825         mult = ((u64)delta) << rt_hash_log;
 826         if (ip_rt_gc_timeout > 1)
 827                 do_div(mult, ip_rt_gc_timeout);
 828         goal = (unsigned int)mult;
 829         if (goal > rt_hash_mask)
 830                 goal = rt_hash_mask + 1;
 831         for (; goal > 0; goal--) {
 832                 unsigned long tmo = ip_rt_gc_timeout;
 833                 unsigned long length;
 834
 835                 i = (i + 1) & rt_hash_mask;
 836                 rthp = &rt_hash_table[i].chain;
 837
 838                 if (need_resched())
 839                         cond_resched();
 840
 841                 samples++;
 842
 843                 if (rcu_dereference_raw(*rthp) == NULL)
 844                         continue;
 845                 length = 0;
 846                 spin_lock_bh(rt_hash_lock_addr(i));
 847                 while ((rth = rcu_dereference_protected(*rthp,
 848                                         lockdep_is_held(rt_hash_lock_addr(i)))) != NULL) {
 849                         prefetch(rth->dst.rt_next);
 850                         if (rt_is_expired(rth)) {
 851                                 *rthp = rth->dst.rt_next;
 852                                 rt_free(rth);
 853                                 continue;
 854                         }
 855                         if (rth->dst.expires) {
 856                                 /* Entry is expired even if it is in use */
 857                                 if (time_before_eq(jiffies, rth->dst.expires)) {
 858 nofree:
 859                                         tmo >>= 1;
 860                                         rthp = &rth->dst.rt_next;
 861                                         /*
 862                                          * We only count entries on
 863                                          * a chain with equal hash inputs once
 864                                          * so that entries for different QOS
 865                                          * levels, and other non-hash input
 866                                          * attributes don't unfairly skew
 867                                          * the length computation
 868                                          */
 869                                         length += has_noalias(rt_hash_table[i].chain, rth);
 870                                         continue;
 871                                 }
 872                         } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout))
 873                                 goto nofree;
 874
 875                         /* Cleanup aged off entries. */
 876                         *rthp = rth->dst.rt_next;
 877                         rt_free(rth);
 878                 }
 879                 spin_unlock_bh(rt_hash_lock_addr(i));
 880                 sum += length;
 881                 sum2 += length*length;
 882         }
 883         if (samples) {
 884                 unsigned long avg = sum / samples;
 885                 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
 886                 rt_chain_length_max = max_t(unsigned long,
 887                                         ip_rt_gc_elasticity,
 888                                         (avg + 4*sd) >> FRACT_BITS);
 889         }
 890         rover = i;
 891 }
 892
 893 /*
 894  * rt_worker_func() is run in process context.
 895  * we call rt_check_expire() to scan part of the hash table
 896  */
 897 static void rt_worker_func(struct work_struct *work)
 898 {
 899         rt_check_expire();
 900         schedule_delayed_work(&expires_work, ip_rt_gc_interval);
 901 }
 902
 903 /*
 904  * Pertubation of rt_genid by a small quantity [1..256]
 905  * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
 906  * many times (2^24) without giving recent rt_genid.
 907  * Jenkins hash is strong enough that litle changes of rt_genid are OK.
 908  */
 909 static void rt_cache_invalidate(struct net *net)
 910 {
 911         unsigned char shuffle;
 912
 913         get_random_bytes(&shuffle, sizeof(shuffle));
 914         atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
 915 }
 916
 917 /*
 918  * delay < 0  : invalidate cache (fast : entries will be deleted later)
 919  * delay >= 0 : invalidate & flush cache (can be long)
 920  */
 921 void rt_cache_flush(struct net *net, int delay)
 922 {
 923         rt_cache_invalidate(net);
 924         if (delay >= 0)
 925                 rt_do_flush(!in_softirq());
 926 }
 927
 928 /* Flush previous cache invalidated entries from the cache */
 929 void rt_cache_flush_batch(void)
 930 {
 931         rt_do_flush(!in_softirq());
 932 }
 933
 934 static void rt_emergency_hash_rebuild(struct net *net)
 935 {
 936         if (net_ratelimit())
 937                 printk(KERN_WARNING "Route hash chain too long!\n");
 938         rt_cache_invalidate(net);
 939 }
 940
 941 /*
 942    Short description of GC goals.
 943
 944    We want to build algorithm, which will keep routing cache
 945    at some equilibrium point, when number of aged off entries
 946    is kept approximately equal to newly generated ones.
 947
 948    Current expiration strength is variable "expire".
 949    We try to adjust it dynamically, so that if networking
 950    is idle expires is large enough to keep enough of warm entries,
 951    and when load increases it reduces to limit cache size.
 952  */
 953
 954 static int rt_garbage_collect(struct dst_ops *ops)
 955 {
 956         static unsigned long expire = RT_GC_TIMEOUT;
 957         static unsigned long last_gc;
 958         static int rover;
 959         static int equilibrium;
 960         struct rtable *rth;
 961         struct rtable __rcu **rthp;
 962         unsigned long now = jiffies;
 963         int goal;
 964         int entries = dst_entries_get_fast(&ipv4_dst_ops);
 965
 966         /*
 967          * Garbage collection is pretty expensive,
 968          * do not make it too frequently.
 969          */
 970
 971         RT_CACHE_STAT_INC(gc_total);
 972
 973         if (now - last_gc < ip_rt_gc_min_interval &&
 974             entries < ip_rt_max_size) {
 975                 RT_CACHE_STAT_INC(gc_ignored);
 976                 goto out;
 977         }
 978
 979         entries = dst_entries_get_slow(&ipv4_dst_ops);
 980         /* Calculate number of entries, which we want to expire now. */
 981         goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
 982         if (goal <= 0) {
 983                 if (equilibrium < ipv4_dst_ops.gc_thresh)
 984                         equilibrium = ipv4_dst_ops.gc_thresh;
 985                 goal = entries - equilibrium;
 986                 if (goal > 0) {
 987                         equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
 988                         goal = entries - equilibrium;
 989                 }
 990         } else {
 991                 /* We are in dangerous area. Try to reduce cache really
 992                  * aggressively.
 993                  */
 994                 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
 995                 equilibrium = entries - goal;
 996         }
 997
 998         if (now - last_gc >= ip_rt_gc_min_interval)
 999                 last_gc = now;
1000
1001         if (goal <= 0) {
1002                 equilibrium += goal;
1003                 goto work_done;
1004         }
1005
1006         do {
1007                 int i, k;
1008
1009                 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
1010                         unsigned long tmo = expire;
1011
1012                         k = (k + 1) & rt_hash_mask;
1013                         rthp = &rt_hash_table[k].chain;
1014                         spin_lock_bh(rt_hash_lock_addr(k));
1015                         while ((rth = rcu_dereference_protected(*rthp,
1016                                         lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
1017                                 if (!rt_is_expired(rth) &&
1018                                         !rt_may_expire(rth, tmo, expire)) {
1019                                         tmo >>= 1;
1020                                         rthp = &rth->dst.rt_next;
1021                                         continue;
1022                                 }
1023                                 *rthp = rth->dst.rt_next;
1024                                 rt_free(rth);
1025                                 goal--;
1026                         }
1027                         spin_unlock_bh(rt_hash_lock_addr(k));
1028                         if (goal <= 0)
1029                                 break;
1030                 }
1031                 rover = k;
1032
1033                 if (goal <= 0)
1034                         goto work_done;
1035
1036                 /* Goal is not achieved. We stop process if:
1037
1038                    - if expire reduced to zero. Otherwise, expire is halfed.
1039                    - if table is not full.
1040                    - if we are called from interrupt.
1041                    - jiffies check is just fallback/debug loop breaker.
1042                      We will not spin here for long time in any case.
1043                  */
1044
1045                 RT_CACHE_STAT_INC(gc_goal_miss);
1046
1047                 if (expire == 0)
1048                         break;
1049
1050                 expire >>= 1;
1051 #if RT_CACHE_DEBUG >= 2
1052                 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
1053                                 dst_entries_get_fast(&ipv4_dst_ops), goal, i);
1054 #endif
1055
1056                 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1057                         goto out;
1058         } while (!in_softirq() && time_before_eq(jiffies, now));
1059
1060         if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1061                 goto out;
1062         if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
1063                 goto out;
1064         if (net_ratelimit())
1065                 printk(KERN_WARNING "dst cache overflow\n");
1066         RT_CACHE_STAT_INC(gc_dst_overflow);
1067         return 1;
1068
1069 work_done:
1070         expire += ip_rt_gc_min_interval;
1071         if (expire > ip_rt_gc_timeout ||
1072             dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
1073             dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1074                 expire = ip_rt_gc_timeout;
1075 #if RT_CACHE_DEBUG >= 2
1076         printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
1077                         dst_entries_get_fast(&ipv4_dst_ops), goal, rover);
1078 #endif
1079 out:    return 0;
1080 }
1081
1082 /*
1083  * Returns number of entries in a hash chain that have different hash_inputs
1084  */
1085 static int slow_chain_length(const struct rtable *head)
1086 {
1087         int length = 0;
1088         const struct rtable *rth = head;
1089
1090         while (rth) {
1091                 length += has_noalias(head, rth);
1092                 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1093         }
1094         return length >> FRACT_BITS;
1095 }
1096
1097 static int rt_intern_hash(unsigned hash, struct rtable *rt,
1098                           struct rtable **rp, struct sk_buff *skb, int ifindex)
1099 {
1100         struct rtable   *rth, *cand;
1101         struct rtable __rcu **rthp, **candp;
1102         unsigned long   now;
1103         u32             min_score;
1104         int             chain_length;
1105         int attempts = !in_softirq();
1106
1107 restart:
1108         chain_length = 0;
1109         min_score = ~(u32)0;
1110         cand = NULL;
1111         candp = NULL;
1112         now = jiffies;
1113
1114         if (!rt_caching(dev_net(rt->dst.dev))) {
1115                 /*
1116                  * If we're not caching, just tell the caller we
1117                  * were successful and don't touch the route.  The
1118                  * caller hold the sole reference to the cache entry, and
1119                  * it will be released when the caller is done with it.
1120                  * If we drop it here, the callers have no way to resolve routes
1121                  * when we're not caching.  Instead, just point *rp at rt, so
1122                  * the caller gets a single use out of the route
1123                  * Note that we do rt_free on this new route entry, so that
1124                  * once its refcount hits zero, we are still able to reap it
1125                  * (Thanks Alexey)
1126                  * Note: To avoid expensive rcu stuff for this uncached dst,
1127                  * we set DST_NOCACHE so that dst_release() can free dst without
1128                  * waiting a grace period.
1129                  */
1130
1131                 rt->dst.flags |= DST_NOCACHE;
1132                 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1133                         int err = arp_bind_neighbour(&rt->dst);
1134                         if (err) {
1135                                 if (net_ratelimit())
1136                                         printk(KERN_WARNING
1137                                             "Neighbour table failure & not caching routes.\n");
1138                                 ip_rt_put(rt);
1139                                 return err;
1140                         }
1141                 }
1142
1143                 goto skip_hashing;
1144         }
1145
1146         rthp = &rt_hash_table[hash].chain;
1147
1148         spin_lock_bh(rt_hash_lock_addr(hash));
1149         while ((rth = rcu_dereference_protected(*rthp,
1150                         lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1151                 if (rt_is_expired(rth)) {
1152                         *rthp = rth->dst.rt_next;
1153                         rt_free(rth);
1154                         continue;
1155                 }
1156                 if (compare_keys(&rth->fl, &rt->fl) && compare_netns(rth, rt)) {
1157                         /* Put it first */
1158                         *rthp = rth->dst.rt_next;
1159                         /*
1160                          * Since lookup is lockfree, the deletion
1161                          * must be visible to another weakly ordered CPU before
1162                          * the insertion at the start of the hash chain.
1163                          */
1164                         rcu_assign_pointer(rth->dst.rt_next,
1165                                            rt_hash_table[hash].chain);
1166                         /*
1167                          * Since lookup is lockfree, the update writes
1168                          * must be ordered for consistency on SMP.
1169                          */
1170                         rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1171
1172                         dst_use(&rth->dst, now);
1173                         spin_unlock_bh(rt_hash_lock_addr(hash));
1174
1175                         rt_drop(rt);
1176                         if (rp)
1177                                 *rp = rth;
1178                         else
1179                                 skb_dst_set(skb, &rth->dst);
1180                         return 0;
1181                 }
1182
1183                 if (!atomic_read(&rth->dst.__refcnt)) {
1184                         u32 score = rt_score(rth);
1185
1186                         if (score <= min_score) {
1187                                 cand = rth;
1188                                 candp = rthp;
1189                                 min_score = score;
1190                         }
1191                 }
1192
1193                 chain_length++;
1194
1195                 rthp = &rth->dst.rt_next;
1196         }
1197
1198         if (cand) {
1199                 /* ip_rt_gc_elasticity used to be average length of chain
1200                  * length, when exceeded gc becomes really aggressive.
1201                  *
1202                  * The second limit is less certain. At the moment it allows
1203                  * only 2 entries per bucket. We will see.
1204                  */
1205                 if (chain_length > ip_rt_gc_elasticity) {
1206                         *candp = cand->dst.rt_next;
1207                         rt_free(cand);
1208                 }
1209         } else {
1210                 if (chain_length > rt_chain_length_max &&
1211                     slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1212                         struct net *net = dev_net(rt->dst.dev);
1213                         int num = ++net->ipv4.current_rt_cache_rebuild_count;
1214                         if (!rt_caching(net)) {
1215                                 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1216                                         rt->dst.dev->name, num);
1217                         }
1218                         rt_emergency_hash_rebuild(net);
1219                         spin_unlock_bh(rt_hash_lock_addr(hash));
1220
1221                         hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1222                                         ifindex, rt_genid(net));
1223                         goto restart;
1224                 }
1225         }
1226
1227         /* Try to bind route to arp only if it is output
1228            route or unicast forwarding path.
1229          */
1230         if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1231                 int err = arp_bind_neighbour(&rt->dst);
1232                 if (err) {
1233                         spin_unlock_bh(rt_hash_lock_addr(hash));
1234
1235                         if (err != -ENOBUFS) {
1236                                 rt_drop(rt);
1237                                 return err;
1238                         }
1239
1240                         /* Neighbour tables are full and nothing
1241                            can be released. Try to shrink route cache,
1242                            it is most likely it holds some neighbour records.
1243                          */
1244                         if (attempts-- > 0) {
1245                                 int saved_elasticity = ip_rt_gc_elasticity;
1246                                 int saved_int = ip_rt_gc_min_interval;
1247                                 ip_rt_gc_elasticity     = 1;
1248                                 ip_rt_gc_min_interval   = 0;
1249                                 rt_garbage_collect(&ipv4_dst_ops);
1250                                 ip_rt_gc_min_interval   = saved_int;
1251                                 ip_rt_gc_elasticity     = saved_elasticity;
1252                                 goto restart;
1253                         }
1254
1255                         if (net_ratelimit())
1256                                 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1257                         rt_drop(rt);
1258                         return -ENOBUFS;
1259                 }
1260         }
1261
1262         rt->dst.rt_next = rt_hash_table[hash].chain;
1263
1264 #if RT_CACHE_DEBUG >= 2
1265         if (rt->dst.rt_next) {
1266                 struct rtable *trt;
1267                 printk(KERN_DEBUG "rt_cache @%02x: %pI4",
1268                        hash, &rt->rt_dst);
1269                 for (trt = rt->dst.rt_next; trt; trt = trt->dst.rt_next)
1270                         printk(" . %pI4", &trt->rt_dst);
1271                 printk("\n");
1272         }
1273 #endif
1274         /*
1275          * Since lookup is lockfree, we must make sure
1276          * previous writes to rt are comitted to memory
1277          * before making rt visible to other CPUS.
1278          */
1279         rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1280
1281         spin_unlock_bh(rt_hash_lock_addr(hash));
1282
1283 skip_hashing:
1284         if (rp)
1285                 *rp = rt;
1286         else
1287                 skb_dst_set(skb, &rt->dst);
1288         return 0;
1289 }
1290
1291 void rt_bind_peer(struct rtable *rt, int create)
1292 {
1293         struct inet_peer *peer;
1294
1295         peer = inet_getpeer_v4(rt->rt_dst, create);
1296
1297         if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1298                 inet_putpeer(peer);
1299 }
1300
1301 /*
1302  * Peer allocation may fail only in serious out-of-memory conditions.  However
1303  * we still can generate some output.
1304  * Random ID selection looks a bit dangerous because we have no chances to
1305  * select ID being unique in a reasonable period of time.
1306  * But broken packet identifier may be better than no packet at all.
1307  */
1308 static void ip_select_fb_ident(struct iphdr *iph)
1309 {
1310         static DEFINE_SPINLOCK(ip_fb_id_lock);
1311         static u32 ip_fallback_id;
1312         u32 salt;
1313
1314         spin_lock_bh(&ip_fb_id_lock);
1315         salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1316         iph->id = htons(salt & 0xFFFF);
1317         ip_fallback_id = salt;
1318         spin_unlock_bh(&ip_fb_id_lock);
1319 }
1320
1321 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1322 {
1323         struct rtable *rt = (struct rtable *) dst;
1324
1325         if (rt) {
1326                 if (rt->peer == NULL)
1327                         rt_bind_peer(rt, 1);
1328
1329                 /* If peer is attached to destination, it is never detached,
1330                    so that we need not to grab a lock to dereference it.
1331                  */
1332                 if (rt->peer) {
1333                         iph->id = htons(inet_getid(rt->peer, more));
1334                         return;
1335                 }
1336         } else
1337                 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1338                        __builtin_return_address(0));
1339
1340         ip_select_fb_ident(iph);
1341 }
1342 EXPORT_SYMBOL(__ip_select_ident);
1343
1344 static void rt_del(unsigned hash, struct rtable *rt)
1345 {
1346         struct rtable __rcu **rthp;
1347         struct rtable *aux;
1348
1349         rthp = &rt_hash_table[hash].chain;
1350         spin_lock_bh(rt_hash_lock_addr(hash));
1351         ip_rt_put(rt);
1352         while ((aux = rcu_dereference_protected(*rthp,
1353                         lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1354                 if (aux == rt || rt_is_expired(aux)) {
1355                         *rthp = aux->dst.rt_next;
1356                         rt_free(aux);
1357                         continue;
1358                 }
1359                 rthp = &aux->dst.rt_next;
1360         }
1361         spin_unlock_bh(rt_hash_lock_addr(hash));
1362 }
1363
1364 /* called in rcu_read_lock() section */
1365 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1366                     __be32 saddr, struct net_device *dev)
1367 {
1368         int i, k;
1369         struct in_device *in_dev = __in_dev_get_rcu(dev);
1370         struct rtable *rth;
1371         struct rtable __rcu **rthp;
1372         __be32  skeys[2] = { saddr, 0 };
1373         int  ikeys[2] = { dev->ifindex, 0 };
1374         struct netevent_redirect netevent;
1375         struct net *net;
1376
1377         if (!in_dev)
1378                 return;
1379
1380         net = dev_net(dev);
1381         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1382             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1383             ipv4_is_zeronet(new_gw))
1384                 goto reject_redirect;
1385
1386         if (!rt_caching(net))
1387                 goto reject_redirect;
1388
1389         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1390                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1391                         goto reject_redirect;
1392                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1393                         goto reject_redirect;
1394         } else {
1395                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1396                         goto reject_redirect;
1397         }
1398
1399         for (i = 0; i < 2; i++) {
1400                 for (k = 0; k < 2; k++) {
1401                         unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1402                                                 rt_genid(net));
1403
1404                         rthp = &rt_hash_table[hash].chain;
1405
1406                         while ((rth = rcu_dereference(*rthp)) != NULL) {
1407                                 struct rtable *rt;
1408
1409                                 if (rth->fl.fl4_dst != daddr ||
1410                                     rth->fl.fl4_src != skeys[i] ||
1411                                     rth->fl.oif != ikeys[k] ||
1412                                     rt_is_input_route(rth) ||
1413                                     rt_is_expired(rth) ||
1414                                     !net_eq(dev_net(rth->dst.dev), net)) {
1415                                         rthp = &rth->dst.rt_next;
1416                                         continue;
1417                                 }
1418
1419                                 if (rth->rt_dst != daddr ||
1420                                     rth->rt_src != saddr ||
1421                                     rth->dst.error ||
1422                                     rth->rt_gateway != old_gw ||
1423                                     rth->dst.dev != dev)
1424                                         break;
1425
1426                                 dst_hold(&rth->dst);
1427
1428                                 rt = dst_alloc(&ipv4_dst_ops);
1429                                 if (rt == NULL) {
1430                                         ip_rt_put(rth);
1431                                         return;
1432                                 }
1433
1434                                 /* Copy all the information. */
1435                                 *rt = *rth;
1436                                 rt->dst.__use           = 1;
1437                                 atomic_set(&rt->dst.__refcnt, 1);
1438                                 rt->dst.child           = NULL;
1439                                 if (rt->dst.dev)
1440                                         dev_hold(rt->dst.dev);
1441                                 rt->dst.obsolete        = -1;
1442                                 rt->dst.lastuse = jiffies;
1443                                 rt->dst.path            = &rt->dst;
1444                                 rt->dst.neighbour       = NULL;
1445                                 rt->dst.hh              = NULL;
1446 #ifdef CONFIG_XFRM
1447                                 rt->dst.xfrm            = NULL;
1448 #endif
1449                                 rt->rt_genid            = rt_genid(net);
1450                                 rt->rt_flags            |= RTCF_REDIRECTED;
1451
1452                                 /* Gateway is different ... */
1453                                 rt->rt_gateway          = new_gw;
1454
1455                                 /* Redirect received -> path was valid */
1456                                 dst_confirm(&rth->dst);
1457
1458                                 if (rt->peer)
1459                                         atomic_inc(&rt->peer->refcnt);
1460
1461                                 if (arp_bind_neighbour(&rt->dst) ||
1462                                     !(rt->dst.neighbour->nud_state &
1463                                             NUD_VALID)) {
1464                                         if (rt->dst.neighbour)
1465                                                 neigh_event_send(rt->dst.neighbour, NULL);
1466                                         ip_rt_put(rth);
1467                                         rt_drop(rt);
1468                                         goto do_next;
1469                                 }
1470
1471                                 netevent.old = &rth->dst;
1472                                 netevent.new = &rt->dst;
1473                                 call_netevent_notifiers(NETEVENT_REDIRECT,
1474                                                         &netevent);
1475
1476                                 rt_del(hash, rth);
1477                                 if (!rt_intern_hash(hash, rt, &rt, NULL, rt->fl.oif))
1478                                         ip_rt_put(rt);
1479                                 goto do_next;
1480                         }
1481                 do_next:
1482                         ;
1483                 }
1484         }
1485         return;
1486
1487 reject_redirect:
1488 #ifdef CONFIG_IP_ROUTE_VERBOSE
1489         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1490                 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1491                         "  Advised path = %pI4 -> %pI4\n",
1492                        &old_gw, dev->name, &new_gw,
1493                        &saddr, &daddr);
1494 #endif
1495         ;
1496 }
1497
1498 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1499 {
1500         struct rtable *rt = (struct rtable *)dst;
1501         struct dst_entry *ret = dst;
1502
1503         if (rt) {
1504                 if (dst->obsolete > 0) {
1505                         ip_rt_put(rt);
1506                         ret = NULL;
1507                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1508                            (rt->dst.expires &&
1509                             time_after_eq(jiffies, rt->dst.expires))) {
1510                         unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1511                                                 rt->fl.oif,
1512                                                 rt_genid(dev_net(dst->dev)));
1513 #if RT_CACHE_DEBUG >= 1
1514                         printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1515                                 &rt->rt_dst, rt->fl.fl4_tos);
1516 #endif
1517                         rt_del(hash, rt);
1518                         ret = NULL;
1519                 }
1520         }
1521         return ret;
1522 }
1523
1524 /*
1525  * Algorithm:
1526  *      1. The first ip_rt_redirect_number redirects are sent
1527  *         with exponential backoff, then we stop sending them at all,
1528  *         assuming that the host ignores our redirects.
1529  *      2. If we did not see packets requiring redirects
1530  *         during ip_rt_redirect_silence, we assume that the host
1531  *         forgot redirected route and start to send redirects again.
1532  *
1533  * This algorithm is much cheaper and more intelligent than dumb load limiting
1534  * in icmp.c.
1535  *
1536  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1537  * and "frag. need" (breaks PMTU discovery) in icmp.c.
1538  */
1539
1540 void ip_rt_send_redirect(struct sk_buff *skb)
1541 {
1542         struct rtable *rt = skb_rtable(skb);
1543         struct in_device *in_dev;
1544         int log_martians;
1545
1546         rcu_read_lock();
1547         in_dev = __in_dev_get_rcu(rt->dst.dev);
1548         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1549                 rcu_read_unlock();
1550                 return;
1551         }
1552         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1553         rcu_read_unlock();
1554
1555         /* No redirected packets during ip_rt_redirect_silence;
1556          * reset the algorithm.
1557          */
1558         if (time_after(jiffies, rt->dst.rate_last + ip_rt_redirect_silence))
1559                 rt->dst.rate_tokens = 0;
1560
1561         /* Too many ignored redirects; do not send anything
1562          * set dst.rate_last to the last seen redirected packet.
1563          */
1564         if (rt->dst.rate_tokens >= ip_rt_redirect_number) {
1565                 rt->dst.rate_last = jiffies;
1566                 return;
1567         }
1568
1569         /* Check for load limit; set rate_last to the latest sent
1570          * redirect.
1571          */
1572         if (rt->dst.rate_tokens == 0 ||
1573             time_after(jiffies,
1574                        (rt->dst.rate_last +
1575                         (ip_rt_redirect_load << rt->dst.rate_tokens)))) {
1576                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1577                 rt->dst.rate_last = jiffies;
1578                 ++rt->dst.rate_tokens;
1579 #ifdef CONFIG_IP_ROUTE_VERBOSE
1580                 if (log_martians &&
1581                     rt->dst.rate_tokens == ip_rt_redirect_number &&
1582                     net_ratelimit())
1583                         printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1584                                 &rt->rt_src, rt->rt_iif,
1585                                 &rt->rt_dst, &rt->rt_gateway);
1586 #endif
1587         }
1588 }
1589
1590 static int ip_error(struct sk_buff *skb)
1591 {
1592         struct rtable *rt = skb_rtable(skb);
1593         unsigned long now;
1594         int code;
1595
1596         switch (rt->dst.error) {
1597                 case EINVAL:
1598                 default:
1599                         goto out;
1600                 case EHOSTUNREACH:
1601                         code = ICMP_HOST_UNREACH;
1602                         break;
1603                 case ENETUNREACH:
1604                         code = ICMP_NET_UNREACH;
1605                         IP_INC_STATS_BH(dev_net(rt->dst.dev),
1606                                         IPSTATS_MIB_INNOROUTES);
1607                         break;
1608                 case EACCES:
1609                         code = ICMP_PKT_FILTERED;
1610                         break;
1611         }
1612
1613         now = jiffies;
1614         rt->dst.rate_tokens += now - rt->dst.rate_last;
1615         if (rt->dst.rate_tokens > ip_rt_error_burst)
1616                 rt->dst.rate_tokens = ip_rt_error_burst;
1617         rt->dst.rate_last = now;
1618         if (rt->dst.rate_tokens >= ip_rt_error_cost) {
1619                 rt->dst.rate_tokens -= ip_rt_error_cost;
1620                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1621         }
1622
1623 out:    kfree_skb(skb);
1624         return 0;
1625 }
1626
1627 /*
1628  *      The last two values are not from the RFC but
1629  *      are needed for AMPRnet AX.25 paths.
1630  */
1631
1632 static const unsigned short mtu_plateau[] =
1633 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1634
1635 static inline unsigned short guess_mtu(unsigned short old_mtu)
1636 {
1637         int i;
1638
1639         for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1640                 if (old_mtu > mtu_plateau[i])
1641                         return mtu_plateau[i];
1642         return 68;
1643 }
1644
1645 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1646                                  unsigned short new_mtu,
1647                                  struct net_device *dev)
1648 {
1649         int i, k;
1650         unsigned short old_mtu = ntohs(iph->tot_len);
1651         struct rtable *rth;
1652         int  ikeys[2] = { dev->ifindex, 0 };
1653         __be32  skeys[2] = { iph->saddr, 0, };
1654         __be32  daddr = iph->daddr;
1655         unsigned short est_mtu = 0;
1656
1657         for (k = 0; k < 2; k++) {
1658                 for (i = 0; i < 2; i++) {
1659                         unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1660                                                 rt_genid(net));
1661
1662                         rcu_read_lock();
1663                         for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1664                              rth = rcu_dereference(rth->dst.rt_next)) {
1665                                 unsigned short mtu = new_mtu;
1666
1667                                 if (rth->fl.fl4_dst != daddr ||
1668                                     rth->fl.fl4_src != skeys[i] ||
1669                                     rth->rt_dst != daddr ||
1670                                     rth->rt_src != iph->saddr ||
1671                                     rth->fl.oif != ikeys[k] ||
1672                                     rt_is_input_route(rth) ||
1673                                     dst_metric_locked(&rth->dst, RTAX_MTU) ||
1674                                     !net_eq(dev_net(rth->dst.dev), net) ||
1675                                     rt_is_expired(rth))
1676                                         continue;
1677
1678                                 if (new_mtu < 68 || new_mtu >= old_mtu) {
1679
1680                                         /* BSD 4.2 compatibility hack :-( */
1681                                         if (mtu == 0 &&
1682                                             old_mtu >= dst_mtu(&rth->dst) &&
1683                                             old_mtu >= 68 + (iph->ihl << 2))
1684                                                 old_mtu -= iph->ihl << 2;
1685
1686                                         mtu = guess_mtu(old_mtu);
1687                                 }
1688                                 if (mtu <= dst_mtu(&rth->dst)) {
1689                                         if (mtu < dst_mtu(&rth->dst)) {
1690                                                 dst_confirm(&rth->dst);
1691                                                 if (mtu < ip_rt_min_pmtu) {
1692                                                         u32 lock = dst_metric(&rth->dst,
1693                                                                               RTAX_LOCK);
1694                                                         mtu = ip_rt_min_pmtu;
1695                                                         lock |= (1 << RTAX_MTU);
1696                                                         dst_metric_set(&rth->dst, RTAX_LOCK,
1697                                                                        lock);
1698                                                 }
1699                                                 dst_metric_set(&rth->dst, RTAX_MTU, mtu);
1700                                                 dst_set_expires(&rth->dst,
1701                                                         ip_rt_mtu_expires);
1702                                         }
1703                                         est_mtu = mtu;
1704                                 }
1705                         }
1706                         rcu_read_unlock();
1707                 }
1708         }
1709         return est_mtu ? : new_mtu;
1710 }
1711
1712 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1713 {
1714         if (dst_mtu(dst) > mtu && mtu >= 68 &&
1715             !(dst_metric_locked(dst, RTAX_MTU))) {
1716                 if (mtu < ip_rt_min_pmtu) {
1717                         u32 lock = dst_metric(dst, RTAX_LOCK);
1718                         mtu = ip_rt_min_pmtu;
1719                         dst_metric_set(dst, RTAX_LOCK, lock | (1 << RTAX_MTU));
1720                 }
1721                 dst_metric_set(dst, RTAX_MTU, mtu);
1722                 dst_set_expires(dst, ip_rt_mtu_expires);
1723                 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1724         }
1725 }
1726
1727 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1728 {
1729         if (rt_is_expired((struct rtable *)dst))
1730                 return NULL;
1731         return dst;
1732 }
1733
1734 static void ipv4_dst_destroy(struct dst_entry *dst)
1735 {
1736         struct rtable *rt = (struct rtable *) dst;
1737         struct inet_peer *peer = rt->peer;
1738
1739         if (peer) {
1740                 rt->peer = NULL;
1741                 inet_putpeer(peer);
1742         }
1743 }
1744
1745
1746 static void ipv4_link_failure(struct sk_buff *skb)
1747 {
1748         struct rtable *rt;
1749
1750         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1751
1752         rt = skb_rtable(skb);
1753         if (rt)
1754                 dst_set_expires(&rt->dst, 0);
1755 }
1756
1757 static int ip_rt_bug(struct sk_buff *skb)
1758 {
1759         printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1760                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1761                 skb->dev ? skb->dev->name : "?");
1762         kfree_skb(skb);
1763         return 0;
1764 }
1765
1766 /*
1767    We do not cache source address of outgoing interface,
1768    because it is used only by IP RR, TS and SRR options,
1769    so that it out of fast path.
1770
1771    BTW remember: "addr" is allowed to be not aligned
1772    in IP options!
1773  */
1774
1775 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1776 {
1777         __be32 src;
1778         struct fib_result res;
1779
1780         if (rt_is_output_route(rt))
1781                 src = rt->rt_src;
1782         else {
1783                 rcu_read_lock();
1784                 if (fib_lookup(dev_net(rt->dst.dev), &rt->fl, &res) == 0)
1785                         src = FIB_RES_PREFSRC(res);
1786                 else
1787                         src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1788                                         RT_SCOPE_UNIVERSE);
1789                 rcu_read_unlock();
1790         }
1791         memcpy(addr, &src, 4);
1792 }
1793
1794 #ifdef CONFIG_NET_CLS_ROUTE
1795 static void set_class_tag(struct rtable *rt, u32 tag)
1796 {
1797         if (!(rt->dst.tclassid & 0xFFFF))
1798                 rt->dst.tclassid |= tag & 0xFFFF;
1799         if (!(rt->dst.tclassid & 0xFFFF0000))
1800                 rt->dst.tclassid |= tag & 0xFFFF0000;
1801 }
1802 #endif
1803
1804 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1805 {
1806         unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1807
1808         if (advmss == 0) {
1809                 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1810                                ip_rt_min_advmss);
1811                 if (advmss > 65535 - 40)
1812                         advmss = 65535 - 40;
1813         }
1814         return advmss;
1815 }
1816
1817 static unsigned int ipv4_default_mtu(const struct dst_entry *dst)
1818 {
1819         unsigned int mtu = dst->dev->mtu;
1820
1821         if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1822                 const struct rtable *rt = (const struct rtable *) dst;
1823
1824                 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1825                         mtu = 576;
1826         }
1827
1828         if (mtu > IP_MAX_MTU)
1829                 mtu = IP_MAX_MTU;
1830
1831         return mtu;
1832 }
1833
1834 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1835 {
1836         struct dst_entry *dst = &rt->dst;
1837         struct fib_info *fi = res->fi;
1838
1839         if (fi) {
1840                 if (FIB_RES_GW(*res) &&
1841                     FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1842                         rt->rt_gateway = FIB_RES_GW(*res);
1843                 dst_import_metrics(dst, fi->fib_metrics);
1844 #ifdef CONFIG_NET_CLS_ROUTE
1845                 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1846 #endif
1847         }
1848
1849         if (dst_mtu(dst) > IP_MAX_MTU)
1850                 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1851         if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40)
1852                 dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40);
1853
1854 #ifdef CONFIG_NET_CLS_ROUTE
1855 #ifdef CONFIG_IP_MULTIPLE_TABLES
1856         set_class_tag(rt, fib_rules_tclass(res));
1857 #endif
1858         set_class_tag(rt, itag);
1859 #endif
1860         rt->rt_type = res->type;
1861 }
1862
1863 /* called in rcu_read_lock() section */
1864 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1865                                 u8 tos, struct net_device *dev, int our)
1866 {
1867         unsigned int hash;
1868         struct rtable *rth;
1869         __be32 spec_dst;
1870         struct in_device *in_dev = __in_dev_get_rcu(dev);
1871         u32 itag = 0;
1872         int err;
1873
1874         /* Primary sanity checks. */
1875
1876         if (in_dev == NULL)
1877                 return -EINVAL;
1878
1879         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1880             ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1881                 goto e_inval;
1882
1883         if (ipv4_is_zeronet(saddr)) {
1884                 if (!ipv4_is_local_multicast(daddr))
1885                         goto e_inval;
1886                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1887         } else {
1888                 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1889                                           &itag, 0);
1890                 if (err < 0)
1891                         goto e_err;
1892         }
1893         rth = dst_alloc(&ipv4_dst_ops);
1894         if (!rth)
1895                 goto e_nobufs;
1896
1897         rth->dst.output = ip_rt_bug;
1898         rth->dst.obsolete = -1;
1899
1900         atomic_set(&rth->dst.__refcnt, 1);
1901         rth->dst.flags= DST_HOST;
1902         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1903                 rth->dst.flags |= DST_NOPOLICY;
1904         rth->fl.fl4_dst = daddr;
1905         rth->rt_dst     = daddr;
1906         rth->fl.fl4_tos = tos;
1907         rth->fl.mark    = skb->mark;
1908         rth->fl.fl4_src = saddr;
1909         rth->rt_src     = saddr;
1910 #ifdef CONFIG_NET_CLS_ROUTE
1911         rth->dst.tclassid = itag;
1912 #endif
1913         rth->rt_iif     =
1914         rth->fl.iif     = dev->ifindex;
1915         rth->dst.dev    = init_net.loopback_dev;
1916         dev_hold(rth->dst.dev);
1917         rth->fl.oif     = 0;
1918         rth->rt_gateway = daddr;
1919         rth->rt_spec_dst= spec_dst;
1920         rth->rt_genid   = rt_genid(dev_net(dev));
1921         rth->rt_flags   = RTCF_MULTICAST;
1922         rth->rt_type    = RTN_MULTICAST;
1923         if (our) {
1924                 rth->dst.input= ip_local_deliver;
1925                 rth->rt_flags |= RTCF_LOCAL;
1926         }
1927
1928 #ifdef CONFIG_IP_MROUTE
1929         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1930                 rth->dst.input = ip_mr_input;
1931 #endif
1932         RT_CACHE_STAT_INC(in_slow_mc);
1933
1934         hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1935         return rt_intern_hash(hash, rth, NULL, skb, dev->ifindex);
1936
1937 e_nobufs:
1938         return -ENOBUFS;
1939 e_inval:
1940         return -EINVAL;
1941 e_err:
1942         return err;
1943 }
1944
1945
1946 static void ip_handle_martian_source(struct net_device *dev,
1947                                      struct in_device *in_dev,
1948                                      struct sk_buff *skb,
1949                                      __be32 daddr,
1950                                      __be32 saddr)
1951 {
1952         RT_CACHE_STAT_INC(in_martian_src);
1953 #ifdef CONFIG_IP_ROUTE_VERBOSE
1954         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1955                 /*
1956                  *      RFC1812 recommendation, if source is martian,
1957                  *      the only hint is MAC header.
1958                  */
1959                 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1960                         &daddr, &saddr, dev->name);
1961                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1962                         int i;
1963                         const unsigned char *p = skb_mac_header(skb);
1964                         printk(KERN_WARNING "ll header: ");
1965                         for (i = 0; i < dev->hard_header_len; i++, p++) {
1966                                 printk("%02x", *p);
1967                                 if (i < (dev->hard_header_len - 1))
1968                                         printk(":");
1969                         }
1970                         printk("\n");
1971                 }
1972         }
1973 #endif
1974 }
1975
1976 /* called in rcu_read_lock() section */
1977 static int __mkroute_input(struct sk_buff *skb,
1978                            struct fib_result *res,
1979                            struct in_device *in_dev,
1980                            __be32 daddr, __be32 saddr, u32 tos,
1981                            struct rtable **result)
1982 {
1983         struct rtable *rth;
1984         int err;
1985         struct in_device *out_dev;
1986         unsigned int flags = 0;
1987         __be32 spec_dst;
1988         u32 itag;
1989
1990         /* get a working reference to the output device */
1991         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1992         if (out_dev == NULL) {
1993                 if (net_ratelimit())
1994                         printk(KERN_CRIT "Bug in ip_route_input" \
1995                                "_slow(). Please, report\n");
1996                 return -EINVAL;
1997         }
1998
1999
2000         err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
2001                                   in_dev->dev, &spec_dst, &itag, skb->mark);
2002         if (err < 0) {
2003                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2004                                          saddr);
2005
2006                 goto cleanup;
2007         }
2008
2009         if (err)
2010                 flags |= RTCF_DIRECTSRC;
2011
2012         if (out_dev == in_dev && err &&
2013             (IN_DEV_SHARED_MEDIA(out_dev) ||
2014              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2015                 flags |= RTCF_DOREDIRECT;
2016
2017         if (skb->protocol != htons(ETH_P_IP)) {
2018                 /* Not IP (i.e. ARP). Do not create route, if it is
2019                  * invalid for proxy arp. DNAT routes are always valid.
2020                  *
2021                  * Proxy arp feature have been extended to allow, ARP
2022                  * replies back to the same interface, to support
2023                  * Private VLAN switch technologies. See arp.c.
2024                  */
2025                 if (out_dev == in_dev &&
2026                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2027                         err = -EINVAL;
2028                         goto cleanup;
2029                 }
2030         }
2031
2032
2033         rth = dst_alloc(&ipv4_dst_ops);
2034         if (!rth) {
2035                 err = -ENOBUFS;
2036                 goto cleanup;
2037         }
2038
2039         atomic_set(&rth->dst.__refcnt, 1);
2040         rth->dst.flags= DST_HOST;
2041         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2042                 rth->dst.flags |= DST_NOPOLICY;
2043         if (IN_DEV_CONF_GET(out_dev, NOXFRM))
2044                 rth->dst.flags |= DST_NOXFRM;
2045         rth->fl.fl4_dst = daddr;
2046         rth->rt_dst     = daddr;
2047         rth->fl.fl4_tos = tos;
2048         rth->fl.mark    = skb->mark;
2049         rth->fl.fl4_src = saddr;
2050         rth->rt_src     = saddr;
2051         rth->rt_gateway = daddr;
2052         rth->rt_iif     =
2053                 rth->fl.iif     = in_dev->dev->ifindex;
2054         rth->dst.dev    = (out_dev)->dev;
2055         dev_hold(rth->dst.dev);
2056         rth->fl.oif     = 0;
2057         rth->rt_spec_dst= spec_dst;
2058
2059         rth->dst.obsolete = -1;
2060         rth->dst.input = ip_forward;
2061         rth->dst.output = ip_output;
2062         rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2063
2064         rt_set_nexthop(rth, res, itag);
2065
2066         rth->rt_flags = flags;
2067
2068         *result = rth;
2069         err = 0;
2070  cleanup:
2071         return err;
2072 }
2073
2074 static int ip_mkroute_input(struct sk_buff *skb,
2075                             struct fib_result *res,
2076                             const struct flowi *fl,
2077                             struct in_device *in_dev,
2078                             __be32 daddr, __be32 saddr, u32 tos)
2079 {
2080         struct rtable* rth = NULL;
2081         int err;
2082         unsigned hash;
2083
2084 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2085         if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
2086                 fib_select_multipath(fl, res);
2087 #endif
2088
2089         /* create a routing cache entry */
2090         err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2091         if (err)
2092                 return err;
2093
2094         /* put it into the cache */
2095         hash = rt_hash(daddr, saddr, fl->iif,
2096                        rt_genid(dev_net(rth->dst.dev)));
2097         return rt_intern_hash(hash, rth, NULL, skb, fl->iif);
2098 }
2099
2100 /*
2101  *      NOTE. We drop all the packets that has local source
2102  *      addresses, because every properly looped back packet
2103  *      must have correct destination already attached by output routine.
2104  *
2105  *      Such approach solves two big problems:
2106  *      1. Not simplex devices are handled properly.
2107  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2108  *      called with rcu_read_lock()
2109  */
2110
2111 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2112                                u8 tos, struct net_device *dev)
2113 {
2114         struct fib_result res;
2115         struct in_device *in_dev = __in_dev_get_rcu(dev);
2116         struct flowi fl = { .fl4_dst    = daddr,
2117                             .fl4_src    = saddr,
2118                             .fl4_tos    = tos,
2119                             .fl4_scope  = RT_SCOPE_UNIVERSE,
2120                             .mark = skb->mark,
2121                             .iif = dev->ifindex };
2122         unsigned        flags = 0;
2123         u32             itag = 0;
2124         struct rtable * rth;
2125         unsigned        hash;
2126         __be32          spec_dst;
2127         int             err = -EINVAL;
2128         struct net    * net = dev_net(dev);
2129
2130         /* IP on this device is disabled. */
2131
2132         if (!in_dev)
2133                 goto out;
2134
2135         /* Check for the most weird martians, which can be not detected
2136            by fib_lookup.
2137          */
2138
2139         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2140             ipv4_is_loopback(saddr))
2141                 goto martian_source;
2142
2143         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2144                 goto brd_input;
2145
2146         /* Accept zero addresses only to limited broadcast;
2147          * I even do not know to fix it or not. Waiting for complains :-)
2148          */
2149         if (ipv4_is_zeronet(saddr))
2150                 goto martian_source;
2151
2152         if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2153                 goto martian_destination;
2154
2155         /*
2156          *      Now we are ready to route packet.
2157          */
2158         err = fib_lookup(net, &fl, &res);
2159         if (err != 0) {
2160                 if (!IN_DEV_FORWARD(in_dev))
2161                         goto e_hostunreach;
2162                 goto no_route;
2163         }
2164
2165         RT_CACHE_STAT_INC(in_slow_tot);
2166
2167         if (res.type == RTN_BROADCAST)
2168                 goto brd_input;
2169
2170         if (res.type == RTN_LOCAL) {
2171                 err = fib_validate_source(saddr, daddr, tos,
2172                                           net->loopback_dev->ifindex,
2173                                           dev, &spec_dst, &itag, skb->mark);
2174                 if (err < 0)
2175                         goto martian_source_keep_err;
2176                 if (err)
2177                         flags |= RTCF_DIRECTSRC;
2178                 spec_dst = daddr;
2179                 goto local_input;
2180         }
2181
2182         if (!IN_DEV_FORWARD(in_dev))
2183                 goto e_hostunreach;
2184         if (res.type != RTN_UNICAST)
2185                 goto martian_destination;
2186
2187         err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
2188 out:    return err;
2189
2190 brd_input:
2191         if (skb->protocol != htons(ETH_P_IP))
2192                 goto e_inval;
2193
2194         if (ipv4_is_zeronet(saddr))
2195                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2196         else {
2197                 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2198                                           &itag, skb->mark);
2199                 if (err < 0)
2200                         goto martian_source_keep_err;
2201                 if (err)
2202                         flags |= RTCF_DIRECTSRC;
2203         }
2204         flags |= RTCF_BROADCAST;
2205         res.type = RTN_BROADCAST;
2206         RT_CACHE_STAT_INC(in_brd);
2207
2208 local_input:
2209         rth = dst_alloc(&ipv4_dst_ops);
2210         if (!rth)
2211                 goto e_nobufs;
2212
2213         rth->dst.output= ip_rt_bug;
2214         rth->dst.obsolete = -1;
2215         rth->rt_genid = rt_genid(net);
2216
2217         atomic_set(&rth->dst.__refcnt, 1);
2218         rth->dst.flags= DST_HOST;
2219         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2220                 rth->dst.flags |= DST_NOPOLICY;
2221         rth->fl.fl4_dst = daddr;
2222         rth->rt_dst     = daddr;
2223         rth->fl.fl4_tos = tos;
2224         rth->fl.mark    = skb->mark;
2225         rth->fl.fl4_src = saddr;
2226         rth->rt_src     = saddr;
2227 #ifdef CONFIG_NET_CLS_ROUTE
2228         rth->dst.tclassid = itag;
2229 #endif
2230         rth->rt_iif     =
2231         rth->fl.iif     = dev->ifindex;
2232         rth->dst.dev    = net->loopback_dev;
2233         dev_hold(rth->dst.dev);
2234         rth->rt_gateway = daddr;
2235         rth->rt_spec_dst= spec_dst;
2236         rth->dst.input= ip_local_deliver;
2237         rth->rt_flags   = flags|RTCF_LOCAL;
2238         if (res.type == RTN_UNREACHABLE) {
2239                 rth->dst.input= ip_error;
2240                 rth->dst.error= -err;
2241                 rth->rt_flags   &= ~RTCF_LOCAL;
2242         }
2243         rth->rt_type    = res.type;
2244         hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));
2245         err = rt_intern_hash(hash, rth, NULL, skb, fl.iif);
2246         goto out;
2247
2248 no_route:
2249         RT_CACHE_STAT_INC(in_no_route);
2250         spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2251         res.type = RTN_UNREACHABLE;
2252         if (err == -ESRCH)
2253                 err = -ENETUNREACH;
2254         goto local_input;
2255
2256         /*
2257          *      Do not cache martian addresses: they should be logged (RFC1812)
2258          */
2259 martian_destination:
2260         RT_CACHE_STAT_INC(in_martian_dst);
2261 #ifdef CONFIG_IP_ROUTE_VERBOSE
2262         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2263                 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2264                         &daddr, &saddr, dev->name);
2265 #endif
2266
2267 e_hostunreach:
2268         err = -EHOSTUNREACH;
2269         goto out;
2270
2271 e_inval:
2272         err = -EINVAL;
2273         goto out;
2274
2275 e_nobufs:
2276         err = -ENOBUFS;
2277         goto out;
2278
2279 martian_source:
2280         err = -EINVAL;
2281 martian_source_keep_err:
2282         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2283         goto out;
2284 }
2285
2286 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2287                            u8 tos, struct net_device *dev, bool noref)
2288 {
2289         struct rtable * rth;
2290         unsigned        hash;
2291         int iif = dev->ifindex;
2292         struct net *net;
2293         int res;
2294
2295         net = dev_net(dev);
2296
2297         rcu_read_lock();
2298
2299         if (!rt_caching(net))
2300                 goto skip_cache;
2301
2302         tos &= IPTOS_RT_MASK;
2303         hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2304
2305         for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2306              rth = rcu_dereference(rth->dst.rt_next)) {
2307                 if ((((__force u32)rth->fl.fl4_dst ^ (__force u32)daddr) |
2308                      ((__force u32)rth->fl.fl4_src ^ (__force u32)saddr) |
2309                      (rth->fl.iif ^ iif) |
2310                      rth->fl.oif |
2311                      (rth->fl.fl4_tos ^ tos)) == 0 &&
2312                     rth->fl.mark == skb->mark &&
2313                     net_eq(dev_net(rth->dst.dev), net) &&
2314                     !rt_is_expired(rth)) {
2315                         if (noref) {
2316                                 dst_use_noref(&rth->dst, jiffies);
2317                                 skb_dst_set_noref(skb, &rth->dst);
2318                         } else {
2319                                 dst_use(&rth->dst, jiffies);
2320                                 skb_dst_set(skb, &rth->dst);
2321                         }
2322                         RT_CACHE_STAT_INC(in_hit);
2323                         rcu_read_unlock();
2324                         return 0;
2325                 }
2326                 RT_CACHE_STAT_INC(in_hlist_search);
2327         }
2328
2329 skip_cache:
2330         /* Multicast recognition logic is moved from route cache to here.
2331            The problem was that too many Ethernet cards have broken/missing
2332            hardware multicast filters :-( As result the host on multicasting
2333            network acquires a lot of useless route cache entries, sort of
2334            SDR messages from all the world. Now we try to get rid of them.
2335            Really, provided software IP multicast filter is organized
2336            reasonably (at least, hashed), it does not result in a slowdown
2337            comparing with route cache reject entries.
2338            Note, that multicast routers are not affected, because
2339            route cache entry is created eventually.
2340          */
2341         if (ipv4_is_multicast(daddr)) {
2342                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2343
2344                 if (in_dev) {
2345                         int our = ip_check_mc(in_dev, daddr, saddr,
2346                                               ip_hdr(skb)->protocol);
2347                         if (our
2348 #ifdef CONFIG_IP_MROUTE
2349                                 ||
2350                             (!ipv4_is_local_multicast(daddr) &&
2351                              IN_DEV_MFORWARD(in_dev))
2352 #endif
2353                            ) {
2354                                 int res = ip_route_input_mc(skb, daddr, saddr,
2355                                                             tos, dev, our);
2356                                 rcu_read_unlock();
2357                                 return res;
2358                         }
2359                 }
2360                 rcu_read_unlock();
2361                 return -EINVAL;
2362         }
2363         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2364         rcu_read_unlock();
2365         return res;
2366 }
2367 EXPORT_SYMBOL(ip_route_input_common);
2368
2369 /* called with rcu_read_lock() */
2370 static int __mkroute_output(struct rtable **result,
2371                             struct fib_result *res,
2372                             const struct flowi *fl,
2373                             const struct flowi *oldflp,
2374                             struct net_device *dev_out,
2375                             unsigned flags)
2376 {
2377         struct rtable *rth;
2378         struct in_device *in_dev;
2379         u32 tos = RT_FL_TOS(oldflp);
2380
2381         if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags & IFF_LOOPBACK))
2382                 return -EINVAL;
2383
2384         if (ipv4_is_lbcast(fl->fl4_dst))
2385                 res->type = RTN_BROADCAST;
2386         else if (ipv4_is_multicast(fl->fl4_dst))
2387                 res->type = RTN_MULTICAST;
2388         else if (ipv4_is_zeronet(fl->fl4_dst))
2389                 return -EINVAL;
2390
2391         if (dev_out->flags & IFF_LOOPBACK)
2392                 flags |= RTCF_LOCAL;
2393
2394         in_dev = __in_dev_get_rcu(dev_out);
2395         if (!in_dev)
2396                 return -EINVAL;
2397
2398         if (res->type == RTN_BROADCAST) {
2399                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2400                 res->fi = NULL;
2401         } else if (res->type == RTN_MULTICAST) {
2402                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2403                 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2404                                  oldflp->proto))
2405                         flags &= ~RTCF_LOCAL;
2406                 /* If multicast route do not exist use
2407                  * default one, but do not gateway in this case.
2408                  * Yes, it is hack.
2409                  */
2410                 if (res->fi && res->prefixlen < 4)
2411                         res->fi = NULL;
2412         }
2413
2414
2415         rth = dst_alloc(&ipv4_dst_ops);
2416         if (!rth)
2417                 return -ENOBUFS;
2418
2419         atomic_set(&rth->dst.__refcnt, 1);
2420         rth->dst.flags= DST_HOST;
2421         if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2422                 rth->dst.flags |= DST_NOXFRM;
2423         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2424                 rth->dst.flags |= DST_NOPOLICY;
2425
2426         rth->fl.fl4_dst = oldflp->fl4_dst;
2427         rth->fl.fl4_tos = tos;
2428         rth->fl.fl4_src = oldflp->fl4_src;
2429         rth->fl.oif     = oldflp->oif;
2430         rth->fl.mark    = oldflp->mark;
2431         rth->rt_dst     = fl->fl4_dst;
2432         rth->rt_src     = fl->fl4_src;
2433         rth->rt_iif     = oldflp->oif ? : dev_out->ifindex;
2434         /* get references to the devices that are to be hold by the routing
2435            cache entry */
2436         rth->dst.dev    = dev_out;
2437         dev_hold(dev_out);
2438         rth->rt_gateway = fl->fl4_dst;
2439         rth->rt_spec_dst= fl->fl4_src;
2440
2441         rth->dst.output=ip_output;
2442         rth->dst.obsolete = -1;
2443         rth->rt_genid = rt_genid(dev_net(dev_out));
2444
2445         RT_CACHE_STAT_INC(out_slow_tot);
2446
2447         if (flags & RTCF_LOCAL) {
2448                 rth->dst.input = ip_local_deliver;
2449                 rth->rt_spec_dst = fl->fl4_dst;
2450         }
2451         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2452                 rth->rt_spec_dst = fl->fl4_src;
2453                 if (flags & RTCF_LOCAL &&
2454                     !(dev_out->flags & IFF_LOOPBACK)) {
2455                         rth->dst.output = ip_mc_output;
2456                         RT_CACHE_STAT_INC(out_slow_mc);
2457                 }
2458 #ifdef CONFIG_IP_MROUTE
2459                 if (res->type == RTN_MULTICAST) {
2460                         if (IN_DEV_MFORWARD(in_dev) &&
2461                             !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2462                                 rth->dst.input = ip_mr_input;
2463                                 rth->dst.output = ip_mc_output;
2464                         }
2465                 }
2466 #endif
2467         }
2468
2469         rt_set_nexthop(rth, res, 0);
2470
2471         rth->rt_flags = flags;
2472         *result = rth;
2473         return 0;
2474 }
2475
2476 /* called with rcu_read_lock() */
2477 static int ip_mkroute_output(struct rtable **rp,
2478                              struct fib_result *res,
2479                              const struct flowi *fl,
2480                              const struct flowi *oldflp,
2481                              struct net_device *dev_out,
2482                              unsigned flags)
2483 {
2484         struct rtable *rth = NULL;
2485         int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2486         unsigned hash;
2487         if (err == 0) {
2488                 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif,
2489                                rt_genid(dev_net(dev_out)));
2490                 err = rt_intern_hash(hash, rth, rp, NULL, oldflp->oif);
2491         }
2492
2493         return err;
2494 }
2495
2496 /*
2497  * Major route resolver routine.
2498  * called with rcu_read_lock();
2499  */
2500
2501 static int ip_route_output_slow(struct net *net, struct rtable **rp,
2502                                 const struct flowi *oldflp)
2503 {
2504         u32 tos = RT_FL_TOS(oldflp);
2505         struct flowi fl = { .fl4_dst = oldflp->fl4_dst,
2506                             .fl4_src = oldflp->fl4_src,
2507                             .fl4_tos = tos & IPTOS_RT_MASK,
2508                             .fl4_scope = ((tos & RTO_ONLINK) ?
2509                                           RT_SCOPE_LINK : RT_SCOPE_UNIVERSE),
2510                             .mark = oldflp->mark,
2511                             .iif = net->loopback_dev->ifindex,
2512                             .oif = oldflp->oif };
2513         struct fib_result res;
2514         unsigned int flags = 0;
2515         struct net_device *dev_out = NULL;
2516         int err;
2517
2518
2519         res.fi          = NULL;
2520 #ifdef CONFIG_IP_MULTIPLE_TABLES
2521         res.r           = NULL;
2522 #endif
2523
2524         if (oldflp->fl4_src) {
2525                 err = -EINVAL;
2526                 if (ipv4_is_multicast(oldflp->fl4_src) ||
2527                     ipv4_is_lbcast(oldflp->fl4_src) ||
2528                     ipv4_is_zeronet(oldflp->fl4_src))
2529                         goto out;
2530
2531                 /* I removed check for oif == dev_out->oif here.
2532                    It was wrong for two reasons:
2533                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2534                       is assigned to multiple interfaces.
2535                    2. Moreover, we are allowed to send packets with saddr
2536                       of another iface. --ANK
2537                  */
2538
2539                 if (oldflp->oif == 0 &&
2540                     (ipv4_is_multicast(oldflp->fl4_dst) ||
2541                      ipv4_is_lbcast(oldflp->fl4_dst))) {
2542                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2543                         dev_out = __ip_dev_find(net, oldflp->fl4_src, false);
2544                         if (dev_out == NULL)
2545                                 goto out;
2546
2547                         /* Special hack: user can direct multicasts
2548                            and limited broadcast via necessary interface
2549                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2550                            This hack is not just for fun, it allows
2551                            vic,vat and friends to work.
2552                            They bind socket to loopback, set ttl to zero
2553                            and expect that it will work.
2554                            From the viewpoint of routing cache they are broken,
2555                            because we are not allowed to build multicast path
2556                            with loopback source addr (look, routing cache
2557                            cannot know, that ttl is zero, so that packet
2558                            will not leave this host and route is valid).
2559                            Luckily, this hack is good workaround.
2560                          */
2561
2562                         fl.oif = dev_out->ifindex;
2563                         goto make_route;
2564                 }
2565
2566                 if (!(oldflp->flags & FLOWI_FLAG_ANYSRC)) {
2567                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2568                         if (!__ip_dev_find(net, oldflp->fl4_src, false))
2569                                 goto out;
2570                 }
2571         }
2572
2573
2574         if (oldflp->oif) {
2575                 dev_out = dev_get_by_index_rcu(net, oldflp->oif);
2576                 err = -ENODEV;
2577                 if (dev_out == NULL)
2578                         goto out;
2579
2580                 /* RACE: Check return value of inet_select_addr instead. */
2581                 if (rcu_dereference(dev_out->ip_ptr) == NULL)
2582                         goto out;       /* Wrong error code */
2583
2584                 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2585                     ipv4_is_lbcast(oldflp->fl4_dst)) {
2586                         if (!fl.fl4_src)
2587                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2588                                                               RT_SCOPE_LINK);
2589                         goto make_route;
2590                 }
2591                 if (!fl.fl4_src) {
2592                         if (ipv4_is_multicast(oldflp->fl4_dst))
2593                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2594                                                               fl.fl4_scope);
2595                         else if (!oldflp->fl4_dst)
2596                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2597                                                               RT_SCOPE_HOST);
2598                 }
2599         }
2600
2601         if (!fl.fl4_dst) {
2602                 fl.fl4_dst = fl.fl4_src;
2603                 if (!fl.fl4_dst)
2604                         fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2605                 dev_out = net->loopback_dev;
2606                 fl.oif = net->loopback_dev->ifindex;
2607                 res.type = RTN_LOCAL;
2608                 flags |= RTCF_LOCAL;
2609                 goto make_route;
2610         }
2611
2612         if (fib_lookup(net, &fl, &res)) {
2613                 res.fi = NULL;
2614                 if (oldflp->oif) {
2615                         /* Apparently, routing tables are wrong. Assume,
2616                            that the destination is on link.
2617
2618                            WHY? DW.
2619                            Because we are allowed to send to iface
2620                            even if it has NO routes and NO assigned
2621                            addresses. When oif is specified, routing
2622                            tables are looked up with only one purpose:
2623                            to catch if destination is gatewayed, rather than
2624                            direct. Moreover, if MSG_DONTROUTE is set,
2625                            we send packet, ignoring both routing tables
2626                            and ifaddr state. --ANK
2627
2628
2629                            We could make it even if oif is unknown,
2630                            likely IPv6, but we do not.
2631                          */
2632
2633                         if (fl.fl4_src == 0)
2634                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2635                                                               RT_SCOPE_LINK);
2636                         res.type = RTN_UNICAST;
2637                         goto make_route;
2638                 }
2639                 err = -ENETUNREACH;
2640                 goto out;
2641         }
2642
2643         if (res.type == RTN_LOCAL) {
2644                 if (!fl.fl4_src)
2645                         fl.fl4_src = fl.fl4_dst;
2646                 dev_out = net->loopback_dev;
2647                 fl.oif = dev_out->ifindex;
2648                 res.fi = NULL;
2649                 flags |= RTCF_LOCAL;
2650                 goto make_route;
2651         }
2652
2653 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2654         if (res.fi->fib_nhs > 1 && fl.oif == 0)
2655                 fib_select_multipath(&fl, &res);
2656         else
2657 #endif
2658         if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2659                 fib_select_default(net, &fl, &res);
2660
2661         if (!fl.fl4_src)
2662                 fl.fl4_src = FIB_RES_PREFSRC(res);
2663
2664         dev_out = FIB_RES_DEV(res);
2665         fl.oif = dev_out->ifindex;
2666
2667
2668 make_route:
2669         err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2670
2671 out:    return err;
2672 }
2673
2674 int __ip_route_output_key(struct net *net, struct rtable **rp,
2675                           const struct flowi *flp)
2676 {
2677         unsigned int hash;
2678         int res;
2679         struct rtable *rth;
2680
2681         if (!rt_caching(net))
2682                 goto slow_output;
2683
2684         hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, rt_genid(net));
2685
2686         rcu_read_lock_bh();
2687         for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2688                 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2689                 if (rth->fl.fl4_dst == flp->fl4_dst &&
2690                     rth->fl.fl4_src == flp->fl4_src &&
2691                     rt_is_output_route(rth) &&
2692                     rth->fl.oif == flp->oif &&
2693                     rth->fl.mark == flp->mark &&
2694                     !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2695                             (IPTOS_RT_MASK | RTO_ONLINK)) &&
2696                     net_eq(dev_net(rth->dst.dev), net) &&
2697                     !rt_is_expired(rth)) {
2698                         dst_use(&rth->dst, jiffies);
2699                         RT_CACHE_STAT_INC(out_hit);
2700                         rcu_read_unlock_bh();
2701                         *rp = rth;
2702                         return 0;
2703                 }
2704                 RT_CACHE_STAT_INC(out_hlist_search);
2705         }
2706         rcu_read_unlock_bh();
2707
2708 slow_output:
2709         rcu_read_lock();
2710         res = ip_route_output_slow(net, rp, flp);
2711         rcu_read_unlock();
2712         return res;
2713 }
2714 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2715
2716 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2717 {
2718         return NULL;
2719 }
2720
2721 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2722 {
2723 }
2724
2725 static struct dst_ops ipv4_dst_blackhole_ops = {
2726         .family                 =       AF_INET,
2727         .protocol               =       cpu_to_be16(ETH_P_IP),
2728         .destroy                =       ipv4_dst_destroy,
2729         .check                  =       ipv4_blackhole_dst_check,
2730         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2731 };
2732
2733
2734 static int ipv4_dst_blackhole(struct net *net, struct rtable **rp, struct flowi *flp)
2735 {
2736         struct rtable *ort = *rp;
2737         struct rtable *rt = (struct rtable *)
2738                 dst_alloc(&ipv4_dst_blackhole_ops);
2739
2740         if (rt) {
2741                 struct dst_entry *new = &rt->dst;
2742
2743                 atomic_set(&new->__refcnt, 1);
2744                 new->__use = 1;
2745                 new->input = dst_discard;
2746                 new->output = dst_discard;
2747                 dst_copy_metrics(new, &ort->dst);
2748
2749                 new->dev = ort->dst.dev;
2750                 if (new->dev)
2751                         dev_hold(new->dev);
2752
2753                 rt->fl = ort->fl;
2754
2755                 rt->rt_genid = rt_genid(net);
2756                 rt->rt_flags = ort->rt_flags;
2757                 rt->rt_type = ort->rt_type;
2758                 rt->rt_dst = ort->rt_dst;
2759                 rt->rt_src = ort->rt_src;
2760                 rt->rt_iif = ort->rt_iif;
2761                 rt->rt_gateway = ort->rt_gateway;
2762                 rt->rt_spec_dst = ort->rt_spec_dst;
2763                 rt->peer = ort->peer;
2764                 if (rt->peer)
2765                         atomic_inc(&rt->peer->refcnt);
2766
2767                 dst_free(new);
2768         }
2769
2770         dst_release(&(*rp)->dst);
2771         *rp = rt;
2772         return rt ? 0 : -ENOMEM;
2773 }
2774
2775 int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2776                          struct sock *sk, int flags)
2777 {
2778         int err;
2779
2780         if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2781                 return err;
2782
2783         if (flp->proto) {
2784                 if (!flp->fl4_src)
2785                         flp->fl4_src = (*rp)->rt_src;
2786                 if (!flp->fl4_dst)
2787                         flp->fl4_dst = (*rp)->rt_dst;
2788                 err = __xfrm_lookup(net, (struct dst_entry **)rp, flp, sk,
2789                                     flags ? XFRM_LOOKUP_WAIT : 0);
2790                 if (err == -EREMOTE)
2791                         err = ipv4_dst_blackhole(net, rp, flp);
2792
2793                 return err;
2794         }
2795
2796         return 0;
2797 }
2798 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2799
2800 int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2801 {
2802         return ip_route_output_flow(net, rp, flp, NULL, 0);
2803 }
2804 EXPORT_SYMBOL(ip_route_output_key);
2805
2806 static int rt_fill_info(struct net *net,
2807                         struct sk_buff *skb, u32 pid, u32 seq, int event,
2808                         int nowait, unsigned int flags)
2809 {
2810         struct rtable *rt = skb_rtable(skb);
2811         struct rtmsg *r;
2812         struct nlmsghdr *nlh;
2813         long expires;
2814         u32 id = 0, ts = 0, tsage = 0, error;
2815
2816         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2817         if (nlh == NULL)
2818                 return -EMSGSIZE;
2819
2820         r = nlmsg_data(nlh);
2821         r->rtm_family    = AF_INET;
2822         r->rtm_dst_len  = 32;
2823         r->rtm_src_len  = 0;
2824         r->rtm_tos      = rt->fl.fl4_tos;
2825         r->rtm_table    = RT_TABLE_MAIN;
2826         NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2827         r->rtm_type     = rt->rt_type;
2828         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2829         r->rtm_protocol = RTPROT_UNSPEC;
2830         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2831         if (rt->rt_flags & RTCF_NOTIFY)
2832                 r->rtm_flags |= RTM_F_NOTIFY;
2833
2834         NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2835
2836         if (rt->fl.fl4_src) {
2837                 r->rtm_src_len = 32;
2838                 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2839         }
2840         if (rt->dst.dev)
2841                 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2842 #ifdef CONFIG_NET_CLS_ROUTE
2843         if (rt->dst.tclassid)
2844                 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2845 #endif
2846         if (rt_is_input_route(rt))
2847                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2848         else if (rt->rt_src != rt->fl.fl4_src)
2849                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2850
2851         if (rt->rt_dst != rt->rt_gateway)
2852                 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2853
2854         if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2855                 goto nla_put_failure;
2856
2857         if (rt->fl.mark)
2858                 NLA_PUT_BE32(skb, RTA_MARK, rt->fl.mark);
2859
2860         error = rt->dst.error;
2861         expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
2862         if (rt->peer) {
2863                 inet_peer_refcheck(rt->peer);
2864                 id = atomic_read(&rt->peer->ip_id_count) & 0xffff;
2865                 if (rt->peer->tcp_ts_stamp) {
2866                         ts = rt->peer->tcp_ts;
2867                         tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2868                 }
2869         }
2870
2871         if (rt_is_input_route(rt)) {
2872 #ifdef CONFIG_IP_MROUTE
2873                 __be32 dst = rt->rt_dst;
2874
2875                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2876                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2877                         int err = ipmr_get_route(net, skb, r, nowait);
2878                         if (err <= 0) {
2879                                 if (!nowait) {
2880                                         if (err == 0)
2881                                                 return 0;
2882                                         goto nla_put_failure;
2883                                 } else {
2884                                         if (err == -EMSGSIZE)
2885                                                 goto nla_put_failure;
2886                                         error = err;
2887                                 }
2888                         }
2889                 } else
2890 #endif
2891                         NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2892         }
2893
2894         if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2895                                expires, error) < 0)
2896                 goto nla_put_failure;
2897
2898         return nlmsg_end(skb, nlh);
2899
2900 nla_put_failure:
2901         nlmsg_cancel(skb, nlh);
2902         return -EMSGSIZE;
2903 }
2904
2905 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2906 {
2907         struct net *net = sock_net(in_skb->sk);
2908         struct rtmsg *rtm;
2909         struct nlattr *tb[RTA_MAX+1];
2910         struct rtable *rt = NULL;
2911         __be32 dst = 0;
2912         __be32 src = 0;
2913         u32 iif;
2914         int err;
2915         int mark;
2916         struct sk_buff *skb;
2917
2918         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2919         if (err < 0)
2920                 goto errout;
2921
2922         rtm = nlmsg_data(nlh);
2923
2924         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2925         if (skb == NULL) {
2926                 err = -ENOBUFS;
2927                 goto errout;
2928         }
2929
2930         /* Reserve room for dummy headers, this skb can pass
2931            through good chunk of routing engine.
2932          */
2933         skb_reset_mac_header(skb);
2934         skb_reset_network_header(skb);
2935
2936         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2937         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2938         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2939
2940         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2941         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2942         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2943         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2944
2945         if (iif) {
2946                 struct net_device *dev;
2947
2948                 dev = __dev_get_by_index(net, iif);
2949                 if (dev == NULL) {
2950                         err = -ENODEV;
2951                         goto errout_free;
2952                 }
2953
2954                 skb->protocol   = htons(ETH_P_IP);
2955                 skb->dev        = dev;
2956                 skb->mark       = mark;
2957                 local_bh_disable();
2958                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2959                 local_bh_enable();
2960
2961                 rt = skb_rtable(skb);
2962                 if (err == 0 && rt->dst.error)
2963                         err = -rt->dst.error;
2964         } else {
2965                 struct flowi fl = {
2966                         .fl4_dst = dst,
2967                         .fl4_src = src,
2968                         .fl4_tos = rtm->rtm_tos,
2969                         .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2970                         .mark = mark,
2971                 };
2972                 err = ip_route_output_key(net, &rt, &fl);
2973         }
2974
2975         if (err)
2976                 goto errout_free;
2977
2978         skb_dst_set(skb, &rt->dst);
2979         if (rtm->rtm_flags & RTM_F_NOTIFY)
2980                 rt->rt_flags |= RTCF_NOTIFY;
2981
2982         err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2983                            RTM_NEWROUTE, 0, 0);
2984         if (err <= 0)
2985                 goto errout_free;
2986
2987         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2988 errout:
2989         return err;
2990
2991 errout_free:
2992         kfree_skb(skb);
2993         goto errout;
2994 }
2995
2996 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
2997 {
2998         struct rtable *rt;
2999         int h, s_h;
3000         int idx, s_idx;
3001         struct net *net;
3002
3003         net = sock_net(skb->sk);
3004
3005         s_h = cb->args[0];
3006         if (s_h < 0)
3007                 s_h = 0;
3008         s_idx = idx = cb->args[1];
3009         for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3010                 if (!rt_hash_table[h].chain)
3011                         continue;
3012                 rcu_read_lock_bh();
3013                 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3014                      rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3015                         if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3016                                 continue;
3017                         if (rt_is_expired(rt))
3018                                 continue;
3019                         skb_dst_set_noref(skb, &rt->dst);
3020                         if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3021                                          cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3022                                          1, NLM_F_MULTI) <= 0) {
3023                                 skb_dst_drop(skb);
3024                                 rcu_read_unlock_bh();
3025                                 goto done;
3026                         }
3027                         skb_dst_drop(skb);
3028                 }
3029                 rcu_read_unlock_bh();
3030         }
3031
3032 done:
3033         cb->args[0] = h;
3034         cb->args[1] = idx;
3035         return skb->len;
3036 }
3037
3038 void ip_rt_multicast_event(struct in_device *in_dev)
3039 {
3040         rt_cache_flush(dev_net(in_dev->dev), 0);
3041 }
3042
3043 #ifdef CONFIG_SYSCTL
3044 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3045                                         void __user *buffer,
3046                                         size_t *lenp, loff_t *ppos)
3047 {
3048         if (write) {
3049                 int flush_delay;
3050                 ctl_table ctl;
3051                 struct net *net;
3052
3053                 memcpy(&ctl, __ctl, sizeof(ctl));
3054                 ctl.data = &flush_delay;
3055                 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3056
3057                 net = (struct net *)__ctl->extra1;
3058                 rt_cache_flush(net, flush_delay);
3059                 return 0;
3060         }
3061
3062         return -EINVAL;
3063 }
3064
3065 static ctl_table ipv4_route_table[] = {
3066         {
3067                 .procname       = "gc_thresh",
3068                 .data           = &ipv4_dst_ops.gc_thresh,
3069                 .maxlen         = sizeof(int),
3070                 .mode           = 0644,
3071                 .proc_handler   = proc_dointvec,
3072         },
3073         {
3074                 .procname       = "max_size",
3075                 .data           = &ip_rt_max_size,
3076                 .maxlen         = sizeof(int),
3077                 .mode           = 0644,
3078                 .proc_handler   = proc_dointvec,
3079         },
3080         {
3081                 /*  Deprecated. Use gc_min_interval_ms */
3082
3083                 .procname       = "gc_min_interval",
3084                 .data           = &ip_rt_gc_min_interval,
3085                 .maxlen         = sizeof(int),
3086                 .mode           = 0644,
3087                 .proc_handler   = proc_dointvec_jiffies,
3088         },
3089         {
3090                 .procname       = "gc_min_interval_ms",
3091                 .data           = &ip_rt_gc_min_interval,
3092                 .maxlen         = sizeof(int),
3093                 .mode           = 0644,
3094                 .proc_handler   = proc_dointvec_ms_jiffies,
3095         },
3096         {
3097                 .procname       = "gc_timeout",
3098                 .data           = &ip_rt_gc_timeout,
3099                 .maxlen         = sizeof(int),
3100                 .mode           = 0644,
3101                 .proc_handler   = proc_dointvec_jiffies,
3102         },
3103         {
3104                 .procname       = "gc_interval",
3105                 .data           = &ip_rt_gc_interval,
3106                 .maxlen         = sizeof(int),
3107                 .mode           = 0644,
3108                 .proc_handler   = proc_dointvec_jiffies,
3109         },
3110         {
3111                 .procname       = "redirect_load",
3112                 .data           = &ip_rt_redirect_load,
3113                 .maxlen         = sizeof(int),
3114                 .mode           = 0644,
3115                 .proc_handler   = proc_dointvec,
3116         },
3117         {
3118                 .procname       = "redirect_number",
3119                 .data           = &ip_rt_redirect_number,
3120                 .maxlen         = sizeof(int),
3121                 .mode           = 0644,
3122                 .proc_handler   = proc_dointvec,
3123         },
3124         {
3125                 .procname       = "redirect_silence",
3126                 .data           = &ip_rt_redirect_silence,
3127                 .maxlen         = sizeof(int),
3128                 .mode           = 0644,
3129                 .proc_handler   = proc_dointvec,
3130         },
3131         {
3132                 .procname       = "error_cost",
3133                 .data           = &ip_rt_error_cost,
3134                 .maxlen         = sizeof(int),
3135                 .mode           = 0644,
3136                 .proc_handler   = proc_dointvec,
3137         },
3138         {
3139                 .procname       = "error_burst",
3140                 .data           = &ip_rt_error_burst,
3141                 .maxlen         = sizeof(int),
3142                 .mode           = 0644,
3143                 .proc_handler   = proc_dointvec,
3144         },
3145         {
3146                 .procname       = "gc_elasticity",
3147                 .data           = &ip_rt_gc_elasticity,
3148                 .maxlen         = sizeof(int),
3149                 .mode           = 0644,
3150                 .proc_handler   = proc_dointvec,
3151         },
3152         {
3153                 .procname       = "mtu_expires",
3154                 .data           = &ip_rt_mtu_expires,
3155                 .maxlen         = sizeof(int),
3156                 .mode           = 0644,
3157                 .proc_handler   = proc_dointvec_jiffies,
3158         },
3159         {
3160                 .procname       = "min_pmtu",
3161                 .data           = &ip_rt_min_pmtu,
3162                 .maxlen         = sizeof(int),
3163                 .mode           = 0644,
3164                 .proc_handler   = proc_dointvec,
3165         },
3166         {
3167                 .procname       = "min_adv_mss",
3168                 .data           = &ip_rt_min_advmss,
3169                 .maxlen         = sizeof(int),
3170                 .mode           = 0644,
3171                 .proc_handler   = proc_dointvec,
3172         },
3173         { }
3174 };
3175
3176 static struct ctl_table empty[1];
3177
3178 static struct ctl_table ipv4_skeleton[] =
3179 {
3180         { .procname = "route",
3181           .mode = 0555, .child = ipv4_route_table},
3182         { .procname = "neigh",
3183           .mode = 0555, .child = empty},
3184         { }
3185 };
3186
3187 static __net_initdata struct ctl_path ipv4_path[] = {
3188         { .procname = "net", },
3189         { .procname = "ipv4", },
3190         { },
3191 };
3192
3193 static struct ctl_table ipv4_route_flush_table[] = {
3194         {
3195                 .procname       = "flush",
3196                 .maxlen         = sizeof(int),
3197                 .mode           = 0200,
3198                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3199         },
3200         { },
3201 };
3202
3203 static __net_initdata struct ctl_path ipv4_route_path[] = {
3204         { .procname = "net", },
3205         { .procname = "ipv4", },
3206         { .procname = "route", },
3207         { },
3208 };
3209
3210 static __net_init int sysctl_route_net_init(struct net *net)
3211 {
3212         struct ctl_table *tbl;
3213
3214         tbl = ipv4_route_flush_table;
3215         if (!net_eq(net, &init_net)) {
3216                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3217                 if (tbl == NULL)
3218                         goto err_dup;
3219         }
3220         tbl[0].extra1 = net;
3221
3222         net->ipv4.route_hdr =
3223                 register_net_sysctl_table(net, ipv4_route_path, tbl);
3224         if (net->ipv4.route_hdr == NULL)
3225                 goto err_reg;
3226         return 0;
3227
3228 err_reg:
3229         if (tbl != ipv4_route_flush_table)
3230                 kfree(tbl);
3231 err_dup:
3232         return -ENOMEM;
3233 }
3234
3235 static __net_exit void sysctl_route_net_exit(struct net *net)
3236 {
3237         struct ctl_table *tbl;
3238
3239         tbl = net->ipv4.route_hdr->ctl_table_arg;
3240         unregister_net_sysctl_table(net->ipv4.route_hdr);
3241         BUG_ON(tbl == ipv4_route_flush_table);
3242         kfree(tbl);
3243 }
3244
3245 static __net_initdata struct pernet_operations sysctl_route_ops = {
3246         .init = sysctl_route_net_init,
3247         .exit = sysctl_route_net_exit,
3248 };
3249 #endif
3250
3251 static __net_init int rt_genid_init(struct net *net)
3252 {
3253         get_random_bytes(&net->ipv4.rt_genid,
3254                          sizeof(net->ipv4.rt_genid));
3255         return 0;
3256 }
3257
3258 static __net_initdata struct pernet_operations rt_genid_ops = {
3259         .init = rt_genid_init,
3260 };
3261
3262
3263 #ifdef CONFIG_NET_CLS_ROUTE
3264 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3265 #endif /* CONFIG_NET_CLS_ROUTE */
3266
3267 static __initdata unsigned long rhash_entries;
3268 static int __init set_rhash_entries(char *str)
3269 {
3270         if (!str)
3271                 return 0;
3272         rhash_entries = simple_strtoul(str, &str, 0);
3273         return 1;
3274 }
3275 __setup("rhash_entries=", set_rhash_entries);
3276
3277 int __init ip_rt_init(void)
3278 {
3279         int rc = 0;
3280
3281 #ifdef CONFIG_NET_CLS_ROUTE
3282         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3283         if (!ip_rt_acct)
3284                 panic("IP: failed to allocate ip_rt_acct\n");
3285 #endif
3286
3287         ipv4_dst_ops.kmem_cachep =
3288                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3289                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3290
3291         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3292
3293         if (dst_entries_init(&ipv4_dst_ops) < 0)
3294                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3295
3296         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3297                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3298
3299         rt_hash_table = (struct rt_hash_bucket *)
3300                 alloc_large_system_hash("IP route cache",
3301                                         sizeof(struct rt_hash_bucket),
3302                                         rhash_entries,
3303                                         (totalram_pages >= 128 * 1024) ?
3304                                         15 : 17,
3305                                         0,
3306                                         &rt_hash_log,
3307                                         &rt_hash_mask,
3308                                         rhash_entries ? 0 : 512 * 1024);
3309         memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3310         rt_hash_lock_init();
3311
3312         ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3313         ip_rt_max_size = (rt_hash_mask + 1) * 16;
3314
3315         devinet_init();
3316         ip_fib_init();
3317
3318         /* All the timers, started at system startup tend
3319            to synchronize. Perturb it a bit.
3320          */
3321         INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3322         expires_ljiffies = jiffies;
3323         schedule_delayed_work(&expires_work,
3324                 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3325
3326         if (ip_rt_proc_init())
3327                 printk(KERN_ERR "Unable to create route proc files\n");
3328 #ifdef CONFIG_XFRM
3329         xfrm_init();
3330         xfrm4_init(ip_rt_max_size);
3331 #endif
3332         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3333
3334 #ifdef CONFIG_SYSCTL
3335         register_pernet_subsys(&sysctl_route_ops);
3336 #endif
3337         register_pernet_subsys(&rt_genid_ops);
3338         return rc;
3339 }
3340
3341 #ifdef CONFIG_SYSCTL
3342 /*
3343  * We really need to sanitize the damn ipv4 init order, then all
3344  * this nonsense will go away.
3345  */
3346 void __init ip_static_sysctl_init(void)
3347 {
3348         register_sysctl_paths(ipv4_path, ipv4_skeleton);
3349 }
3350 #endif