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 void              ipv4_dst_destroy(struct dst_entry *dst);
 143 static void              ipv4_dst_ifdown(struct dst_entry *dst,
 144                                          struct net_device *dev, int how);
 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
 151 static struct dst_ops ipv4_dst_ops = {
 152         .family =               AF_INET,
 153         .protocol =             cpu_to_be16(ETH_P_IP),
 154         .gc =                   rt_garbage_collect,
 155         .check =                ipv4_dst_check,
 156         .destroy =              ipv4_dst_destroy,
 157         .ifdown =               ipv4_dst_ifdown,
 158         .negative_advice =      ipv4_negative_advice,
 159         .link_failure =         ipv4_link_failure,
 160         .update_pmtu =          ip_rt_update_pmtu,
 161         .local_out =            __ip_local_out,
 162         .entries =              ATOMIC_INIT(0),
 163 };
 164
 165 #define ECN_OR_COST(class)      TC_PRIO_##class
 166
 167 const __u8 ip_tos2prio[16] = {
 168         TC_PRIO_BESTEFFORT,
 169         ECN_OR_COST(FILLER),
 170         TC_PRIO_BESTEFFORT,
 171         ECN_OR_COST(BESTEFFORT),
 172         TC_PRIO_BULK,
 173         ECN_OR_COST(BULK),
 174         TC_PRIO_BULK,
 175         ECN_OR_COST(BULK),
 176         TC_PRIO_INTERACTIVE,
 177         ECN_OR_COST(INTERACTIVE),
 178         TC_PRIO_INTERACTIVE,
 179         ECN_OR_COST(INTERACTIVE),
 180         TC_PRIO_INTERACTIVE_BULK,
 181         ECN_OR_COST(INTERACTIVE_BULK),
 182         TC_PRIO_INTERACTIVE_BULK,
 183         ECN_OR_COST(INTERACTIVE_BULK)
 184 };
 185
 186
 187 /*
 188  * Route cache.
 189  */
 190
 191 /* The locking scheme is rather straight forward:
 192  *
 193  * 1) Read-Copy Update protects the buckets of the central route hash.
 194  * 2) Only writers remove entries, and they hold the lock
 195  *    as they look at rtable reference counts.
 196  * 3) Only readers acquire references to rtable entries,
 197  *    they do so with atomic increments and with the
 198  *    lock held.
 199  */
 200
 201 struct rt_hash_bucket {
 202         struct rtable   *chain;
 203 };
 204
 205 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
 206         defined(CONFIG_PROVE_LOCKING)
 207 /*
 208  * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
 209  * The size of this table is a power of two and depends on the number of CPUS.
 210  * (on lockdep we have a quite big spinlock_t, so keep the size down there)
 211  */
 212 #ifdef CONFIG_LOCKDEP
 213 # define RT_HASH_LOCK_SZ        256
 214 #else
 215 # if NR_CPUS >= 32
 216 #  define RT_HASH_LOCK_SZ       4096
 217 # elif NR_CPUS >= 16
 218 #  define RT_HASH_LOCK_SZ       2048
 219 # elif NR_CPUS >= 8
 220 #  define RT_HASH_LOCK_SZ       1024
 221 # elif NR_CPUS >= 4
 222 #  define RT_HASH_LOCK_SZ       512
 223 # else
 224 #  define RT_HASH_LOCK_SZ       256
 225 # endif
 226 #endif
 227
 228 static spinlock_t       *rt_hash_locks;
 229 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
 230
 231 static __init void rt_hash_lock_init(void)
 232 {
 233         int i;
 234
 235         rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
 236                         GFP_KERNEL);
 237         if (!rt_hash_locks)
 238                 panic("IP: failed to allocate rt_hash_locks\n");
 239
 240         for (i = 0; i < RT_HASH_LOCK_SZ; i++)
 241                 spin_lock_init(&rt_hash_locks[i]);
 242 }
 243 #else
 244 # define rt_hash_lock_addr(slot) NULL
 245
 246 static inline void rt_hash_lock_init(void)
 247 {
 248 }
 249 #endif
 250
 251 static struct rt_hash_bucket    *rt_hash_table __read_mostly;
 252 static unsigned                 rt_hash_mask __read_mostly;
 253 static unsigned int             rt_hash_log  __read_mostly;
 254
 255 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
 256 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
 257
 258 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
 259                                    int genid)
 260 {
 261         return jhash_3words((__force u32)daddr, (__force u32)saddr,
 262                             idx, genid)
 263                 & rt_hash_mask;
 264 }
 265
 266 static inline int rt_genid(struct net *net)
 267 {
 268         return atomic_read(&net->ipv4.rt_genid);
 269 }
 270
 271 #ifdef CONFIG_PROC_FS
 272 struct rt_cache_iter_state {
 273         struct seq_net_private p;
 274         int bucket;
 275         int genid;
 276 };
 277
 278 static struct rtable *rt_cache_get_first(struct seq_file *seq)
 279 {
 280         struct rt_cache_iter_state *st = seq->private;
 281         struct rtable *r = NULL;
 282
 283         for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
 284                 if (!rt_hash_table[st->bucket].chain)
 285                         continue;
 286                 rcu_read_lock_bh();
 287                 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
 288                 while (r) {
 289                         if (dev_net(r->dst.dev) == seq_file_net(seq) &&
 290                             r->rt_genid == st->genid)
 291                                 return r;
 292                         r = rcu_dereference_bh(r->dst.rt_next);
 293                 }
 294                 rcu_read_unlock_bh();
 295         }
 296         return r;
 297 }
 298
 299 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
 300                                           struct rtable *r)
 301 {
 302         struct rt_cache_iter_state *st = seq->private;
 303
 304         r = r->dst.rt_next;
 305         while (!r) {
 306                 rcu_read_unlock_bh();
 307                 do {
 308                         if (--st->bucket < 0)
 309                                 return NULL;
 310                 } while (!rt_hash_table[st->bucket].chain);
 311                 rcu_read_lock_bh();
 312                 r = rt_hash_table[st->bucket].chain;
 313         }
 314         return rcu_dereference_bh(r);
 315 }
 316
 317 static struct rtable *rt_cache_get_next(struct seq_file *seq,
 318                                         struct rtable *r)
 319 {
 320         struct rt_cache_iter_state *st = seq->private;
 321         while ((r = __rt_cache_get_next(seq, r)) != NULL) {
 322                 if (dev_net(r->dst.dev) != seq_file_net(seq))
 323                         continue;
 324                 if (r->rt_genid == st->genid)
 325                         break;
 326         }
 327         return r;
 328 }
 329
 330 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
 331 {
 332         struct rtable *r = rt_cache_get_first(seq);
 333
 334         if (r)
 335                 while (pos && (r = rt_cache_get_next(seq, r)))
 336                         --pos;
 337         return pos ? NULL : r;
 338 }
 339
 340 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
 341 {
 342         struct rt_cache_iter_state *st = seq->private;
 343         if (*pos)
 344                 return rt_cache_get_idx(seq, *pos - 1);
 345         st->genid = rt_genid(seq_file_net(seq));
 346         return SEQ_START_TOKEN;
 347 }
 348
 349 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 350 {
 351         struct rtable *r;
 352
 353         if (v == SEQ_START_TOKEN)
 354                 r = rt_cache_get_first(seq);
 355         else
 356                 r = rt_cache_get_next(seq, v);
 357         ++*pos;
 358         return r;
 359 }
 360
 361 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
 362 {
 363         if (v && v != SEQ_START_TOKEN)
 364                 rcu_read_unlock_bh();
 365 }
 366
 367 static int rt_cache_seq_show(struct seq_file *seq, void *v)
 368 {
 369         if (v == SEQ_START_TOKEN)
 370                 seq_printf(seq, "%-127s\n",
 371                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
 372                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
 373                            "HHUptod\tSpecDst");
 374         else {
 375                 struct rtable *r = v;
 376                 int len;
 377
 378                 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
 379                               "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
 380                         r->dst.dev ? r->dst.dev->name : "*",
 381                         (__force u32)r->rt_dst,
 382                         (__force u32)r->rt_gateway,
 383                         r->rt_flags, atomic_read(&r->dst.__refcnt),
 384                         r->dst.__use, 0, (__force u32)r->rt_src,
 385                         (dst_metric(&r->dst, RTAX_ADVMSS) ?
 386                              (int)dst_metric(&r->dst, RTAX_ADVMSS) + 40 : 0),
 387                         dst_metric(&r->dst, RTAX_WINDOW),
 388                         (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
 389                               dst_metric(&r->dst, RTAX_RTTVAR)),
 390                         r->fl.fl4_tos,
 391                         r->dst.hh ? atomic_read(&r->dst.hh->hh_refcnt) : -1,
 392                         r->dst.hh ? (r->dst.hh->hh_output ==
 393                                        dev_queue_xmit) : 0,
 394                         r->rt_spec_dst, &len);
 395
 396                 seq_printf(seq, "%*s\n", 127 - len, "");
 397         }
 398         return 0;
 399 }
 400
 401 static const struct seq_operations rt_cache_seq_ops = {
 402         .start  = rt_cache_seq_start,
 403         .next   = rt_cache_seq_next,
 404         .stop   = rt_cache_seq_stop,
 405         .show   = rt_cache_seq_show,
 406 };
 407
 408 static int rt_cache_seq_open(struct inode *inode, struct file *file)
 409 {
 410         return seq_open_net(inode, file, &rt_cache_seq_ops,
 411                         sizeof(struct rt_cache_iter_state));
 412 }
 413
 414 static const struct file_operations rt_cache_seq_fops = {
 415         .owner   = THIS_MODULE,
 416         .open    = rt_cache_seq_open,
 417         .read    = seq_read,
 418         .llseek  = seq_lseek,
 419         .release = seq_release_net,
 420 };
 421
 422
 423 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
 424 {
 425         int cpu;
 426
 427         if (*pos == 0)
 428                 return SEQ_START_TOKEN;
 429
 430         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
 431                 if (!cpu_possible(cpu))
 432                         continue;
 433                 *pos = cpu+1;
 434                 return &per_cpu(rt_cache_stat, cpu);
 435         }
 436         return NULL;
 437 }
 438
 439 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 440 {
 441         int cpu;
 442
 443         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
 444                 if (!cpu_possible(cpu))
 445                         continue;
 446                 *pos = cpu+1;
 447                 return &per_cpu(rt_cache_stat, cpu);
 448         }
 449         return NULL;
 450
 451 }
 452
 453 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
 454 {
 455
 456 }
 457
 458 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
 459 {
 460         struct rt_cache_stat *st = v;
 461
 462         if (v == SEQ_START_TOKEN) {
 463                 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");
 464                 return 0;
 465         }
 466
 467         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
 468                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
 469                    atomic_read(&ipv4_dst_ops.entries),
 470                    st->in_hit,
 471                    st->in_slow_tot,
 472                    st->in_slow_mc,
 473                    st->in_no_route,
 474                    st->in_brd,
 475                    st->in_martian_dst,
 476                    st->in_martian_src,
 477
 478                    st->out_hit,
 479                    st->out_slow_tot,
 480                    st->out_slow_mc,
 481
 482                    st->gc_total,
 483                    st->gc_ignored,
 484                    st->gc_goal_miss,
 485                    st->gc_dst_overflow,
 486                    st->in_hlist_search,
 487                    st->out_hlist_search
 488                 );
 489         return 0;
 490 }
 491
 492 static const struct seq_operations rt_cpu_seq_ops = {
 493         .start  = rt_cpu_seq_start,
 494         .next   = rt_cpu_seq_next,
 495         .stop   = rt_cpu_seq_stop,
 496         .show   = rt_cpu_seq_show,
 497 };
 498
 499
 500 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
 501 {
 502         return seq_open(file, &rt_cpu_seq_ops);
 503 }
 504
 505 static const struct file_operations rt_cpu_seq_fops = {
 506         .owner   = THIS_MODULE,
 507         .open    = rt_cpu_seq_open,
 508         .read    = seq_read,
 509         .llseek  = seq_lseek,
 510         .release = seq_release,
 511 };
 512
 513 #ifdef CONFIG_NET_CLS_ROUTE
 514 static int rt_acct_proc_show(struct seq_file *m, void *v)
 515 {
 516         struct ip_rt_acct *dst, *src;
 517         unsigned int i, j;
 518
 519         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
 520         if (!dst)
 521                 return -ENOMEM;
 522
 523         for_each_possible_cpu(i) {
 524                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
 525                 for (j = 0; j < 256; j++) {
 526                         dst[j].o_bytes   += src[j].o_bytes;
 527                         dst[j].o_packets += src[j].o_packets;
 528                         dst[j].i_bytes   += src[j].i_bytes;
 529                         dst[j].i_packets += src[j].i_packets;
 530                 }
 531         }
 532
 533         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
 534         kfree(dst);
 535         return 0;
 536 }
 537
 538 static int rt_acct_proc_open(struct inode *inode, struct file *file)
 539 {
 540         return single_open(file, rt_acct_proc_show, NULL);
 541 }
 542
 543 static const struct file_operations rt_acct_proc_fops = {
 544         .owner          = THIS_MODULE,
 545         .open           = rt_acct_proc_open,
 546         .read           = seq_read,
 547         .llseek         = seq_lseek,
 548         .release        = single_release,
 549 };
 550 #endif
 551
 552 static int __net_init ip_rt_do_proc_init(struct net *net)
 553 {
 554         struct proc_dir_entry *pde;
 555
 556         pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
 557                         &rt_cache_seq_fops);
 558         if (!pde)
 559                 goto err1;
 560
 561         pde = proc_create("rt_cache", S_IRUGO,
 562                           net->proc_net_stat, &rt_cpu_seq_fops);
 563         if (!pde)
 564                 goto err2;
 565
 566 #ifdef CONFIG_NET_CLS_ROUTE
 567         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
 568         if (!pde)
 569                 goto err3;
 570 #endif
 571         return 0;
 572
 573 #ifdef CONFIG_NET_CLS_ROUTE
 574 err3:
 575         remove_proc_entry("rt_cache", net->proc_net_stat);
 576 #endif
 577 err2:
 578         remove_proc_entry("rt_cache", net->proc_net);
 579 err1:
 580         return -ENOMEM;
 581 }
 582
 583 static void __net_exit ip_rt_do_proc_exit(struct net *net)
 584 {
 585         remove_proc_entry("rt_cache", net->proc_net_stat);
 586         remove_proc_entry("rt_cache", net->proc_net);
 587 #ifdef CONFIG_NET_CLS_ROUTE
 588         remove_proc_entry("rt_acct", net->proc_net);
 589 #endif
 590 }
 591
 592 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
 593         .init = ip_rt_do_proc_init,
 594         .exit = ip_rt_do_proc_exit,
 595 };
 596
 597 static int __init ip_rt_proc_init(void)
 598 {
 599         return register_pernet_subsys(&ip_rt_proc_ops);
 600 }
 601
 602 #else
 603 static inline int ip_rt_proc_init(void)
 604 {
 605         return 0;
 606 }
 607 #endif /* CONFIG_PROC_FS */
 608
 609 static inline void rt_free(struct rtable *rt)
 610 {
 611         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
 612 }
 613
 614 static inline void rt_drop(struct rtable *rt)
 615 {
 616         ip_rt_put(rt);
 617         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
 618 }
 619
 620 static inline int rt_fast_clean(struct rtable *rth)
 621 {
 622         /* Kill broadcast/multicast entries very aggresively, if they
 623            collide in hash table with more useful entries */
 624         return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
 625                 rth->fl.iif && rth->dst.rt_next;
 626 }
 627
 628 static inline int rt_valuable(struct rtable *rth)
 629 {
 630         return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
 631                 rth->dst.expires;
 632 }
 633
 634 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
 635 {
 636         unsigned long age;
 637         int ret = 0;
 638
 639         if (atomic_read(&rth->dst.__refcnt))
 640                 goto out;
 641
 642         ret = 1;
 643         if (rth->dst.expires &&
 644             time_after_eq(jiffies, rth->dst.expires))
 645                 goto out;
 646
 647         age = jiffies - rth->dst.lastuse;
 648         ret = 0;
 649         if ((age <= tmo1 && !rt_fast_clean(rth)) ||
 650             (age <= tmo2 && rt_valuable(rth)))
 651                 goto out;
 652         ret = 1;
 653 out:    return ret;
 654 }
 655
 656 /* Bits of score are:
 657  * 31: very valuable
 658  * 30: not quite useless
 659  * 29..0: usage counter
 660  */
 661 static inline u32 rt_score(struct rtable *rt)
 662 {
 663         u32 score = jiffies - rt->dst.lastuse;
 664
 665         score = ~score & ~(3<<30);
 666
 667         if (rt_valuable(rt))
 668                 score |= (1<<31);
 669
 670         if (!rt->fl.iif ||
 671             !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
 672                 score |= (1<<30);
 673
 674         return score;
 675 }
 676
 677 static inline bool rt_caching(const struct net *net)
 678 {
 679         return net->ipv4.current_rt_cache_rebuild_count <=
 680                 net->ipv4.sysctl_rt_cache_rebuild_count;
 681 }
 682
 683 static inline bool compare_hash_inputs(const struct flowi *fl1,
 684                                         const struct flowi *fl2)
 685 {
 686         return ((((__force u32)fl1->nl_u.ip4_u.daddr ^ (__force u32)fl2->nl_u.ip4_u.daddr) |
 687                 ((__force u32)fl1->nl_u.ip4_u.saddr ^ (__force u32)fl2->nl_u.ip4_u.saddr) |
 688                 (fl1->iif ^ fl2->iif)) == 0);
 689 }
 690
 691 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
 692 {
 693         return (((__force u32)fl1->nl_u.ip4_u.daddr ^ (__force u32)fl2->nl_u.ip4_u.daddr) |
 694                 ((__force u32)fl1->nl_u.ip4_u.saddr ^ (__force u32)fl2->nl_u.ip4_u.saddr) |
 695                 (fl1->mark ^ fl2->mark) |
 696                 (*(u16 *)&fl1->nl_u.ip4_u.tos ^ *(u16 *)&fl2->nl_u.ip4_u.tos) |
 697                 (fl1->oif ^ fl2->oif) |
 698                 (fl1->iif ^ fl2->iif)) == 0;
 699 }
 700
 701 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
 702 {
 703         return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
 704 }
 705
 706 static inline int rt_is_expired(struct rtable *rth)
 707 {
 708         return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
 709 }
 710
 711 /*
 712  * Perform a full scan of hash table and free all entries.
 713  * Can be called by a softirq or a process.
 714  * In the later case, we want to be reschedule if necessary
 715  */
 716 static void rt_do_flush(int process_context)
 717 {
 718         unsigned int i;
 719         struct rtable *rth, *next;
 720         struct rtable * tail;
 721
 722         for (i = 0; i <= rt_hash_mask; i++) {
 723                 if (process_context && need_resched())
 724                         cond_resched();
 725                 rth = rt_hash_table[i].chain;
 726                 if (!rth)
 727                         continue;
 728
 729                 spin_lock_bh(rt_hash_lock_addr(i));
 730 #ifdef CONFIG_NET_NS
 731                 {
 732                 struct rtable ** prev, * p;
 733
 734                 rth = rt_hash_table[i].chain;
 735
 736                 /* defer releasing the head of the list after spin_unlock */
 737                 for (tail = rth; tail; tail = tail->dst.rt_next)
 738                         if (!rt_is_expired(tail))
 739                                 break;
 740                 if (rth != tail)
 741                         rt_hash_table[i].chain = tail;
 742
 743                 /* call rt_free on entries after the tail requiring flush */
 744                 prev = &rt_hash_table[i].chain;
 745                 for (p = *prev; p; p = next) {
 746                         next = p->dst.rt_next;
 747                         if (!rt_is_expired(p)) {
 748                                 prev = &p->dst.rt_next;
 749                         } else {
 750                                 *prev = next;
 751                                 rt_free(p);
 752                         }
 753                 }
 754                 }
 755 #else
 756                 rth = rt_hash_table[i].chain;
 757                 rt_hash_table[i].chain = NULL;
 758                 tail = NULL;
 759 #endif
 760                 spin_unlock_bh(rt_hash_lock_addr(i));
 761
 762                 for (; rth != tail; rth = next) {
 763                         next = rth->dst.rt_next;
 764                         rt_free(rth);
 765                 }
 766         }
 767 }
 768
 769 /*
 770  * While freeing expired entries, we compute average chain length
 771  * and standard deviation, using fixed-point arithmetic.
 772  * This to have an estimation of rt_chain_length_max
 773  *  rt_chain_length_max = max(elasticity, AVG + 4*SD)
 774  * We use 3 bits for frational part, and 29 (or 61) for magnitude.
 775  */
 776
 777 #define FRACT_BITS 3
 778 #define ONE (1UL << FRACT_BITS)
 779
 780 /*
 781  * Given a hash chain and an item in this hash chain,
 782  * find if a previous entry has the same hash_inputs
 783  * (but differs on tos, mark or oif)
 784  * Returns 0 if an alias is found.
 785  * Returns ONE if rth has no alias before itself.
 786  */
 787 static int has_noalias(const struct rtable *head, const struct rtable *rth)
 788 {
 789         const struct rtable *aux = head;
 790
 791         while (aux != rth) {
 792                 if (compare_hash_inputs(&aux->fl, &rth->fl))
 793                         return 0;
 794                 aux = aux->dst.rt_next;
 795         }
 796         return ONE;
 797 }
 798
 799 static void rt_check_expire(void)
 800 {
 801         static unsigned int rover;
 802         unsigned int i = rover, goal;
 803         struct rtable *rth, **rthp;
 804         unsigned long samples = 0;
 805         unsigned long sum = 0, sum2 = 0;
 806         unsigned long delta;
 807         u64 mult;
 808
 809         delta = jiffies - expires_ljiffies;
 810         expires_ljiffies = jiffies;
 811         mult = ((u64)delta) << rt_hash_log;
 812         if (ip_rt_gc_timeout > 1)
 813                 do_div(mult, ip_rt_gc_timeout);
 814         goal = (unsigned int)mult;
 815         if (goal > rt_hash_mask)
 816                 goal = rt_hash_mask + 1;
 817         for (; goal > 0; goal--) {
 818                 unsigned long tmo = ip_rt_gc_timeout;
 819                 unsigned long length;
 820
 821                 i = (i + 1) & rt_hash_mask;
 822                 rthp = &rt_hash_table[i].chain;
 823
 824                 if (need_resched())
 825                         cond_resched();
 826
 827                 samples++;
 828
 829                 if (*rthp == NULL)
 830                         continue;
 831                 length = 0;
 832                 spin_lock_bh(rt_hash_lock_addr(i));
 833                 while ((rth = *rthp) != NULL) {
 834                         prefetch(rth->dst.rt_next);
 835                         if (rt_is_expired(rth)) {
 836                                 *rthp = rth->dst.rt_next;
 837                                 rt_free(rth);
 838                                 continue;
 839                         }
 840                         if (rth->dst.expires) {
 841                                 /* Entry is expired even if it is in use */
 842                                 if (time_before_eq(jiffies, rth->dst.expires)) {
 843 nofree:
 844                                         tmo >>= 1;
 845                                         rthp = &rth->dst.rt_next;
 846                                         /*
 847                                          * We only count entries on
 848                                          * a chain with equal hash inputs once
 849                                          * so that entries for different QOS
 850                                          * levels, and other non-hash input
 851                                          * attributes don't unfairly skew
 852                                          * the length computation
 853                                          */
 854                                         length += has_noalias(rt_hash_table[i].chain, rth);
 855                                         continue;
 856                                 }
 857                         } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout))
 858                                 goto nofree;
 859
 860                         /* Cleanup aged off entries. */
 861                         *rthp = rth->dst.rt_next;
 862                         rt_free(rth);
 863                 }
 864                 spin_unlock_bh(rt_hash_lock_addr(i));
 865                 sum += length;
 866                 sum2 += length*length;
 867         }
 868         if (samples) {
 869                 unsigned long avg = sum / samples;
 870                 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
 871                 rt_chain_length_max = max_t(unsigned long,
 872                                         ip_rt_gc_elasticity,
 873                                         (avg + 4*sd) >> FRACT_BITS);
 874         }
 875         rover = i;
 876 }
 877
 878 /*
 879  * rt_worker_func() is run in process context.
 880  * we call rt_check_expire() to scan part of the hash table
 881  */
 882 static void rt_worker_func(struct work_struct *work)
 883 {
 884         rt_check_expire();
 885         schedule_delayed_work(&expires_work, ip_rt_gc_interval);
 886 }
 887
 888 /*
 889  * Pertubation of rt_genid by a small quantity [1..256]
 890  * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
 891  * many times (2^24) without giving recent rt_genid.
 892  * Jenkins hash is strong enough that litle changes of rt_genid are OK.
 893  */
 894 static void rt_cache_invalidate(struct net *net)
 895 {
 896         unsigned char shuffle;
 897
 898         get_random_bytes(&shuffle, sizeof(shuffle));
 899         atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
 900 }
 901
 902 /*
 903  * delay < 0  : invalidate cache (fast : entries will be deleted later)
 904  * delay >= 0 : invalidate & flush cache (can be long)
 905  */
 906 void rt_cache_flush(struct net *net, int delay)
 907 {
 908         rt_cache_invalidate(net);
 909         if (delay >= 0)
 910                 rt_do_flush(!in_softirq());
 911 }
 912
 913 /* Flush previous cache invalidated entries from the cache */
 914 void rt_cache_flush_batch(void)
 915 {
 916         rt_do_flush(!in_softirq());
 917 }
 918
 919 static void rt_emergency_hash_rebuild(struct net *net)
 920 {
 921         if (net_ratelimit())
 922                 printk(KERN_WARNING "Route hash chain too long!\n");
 923         rt_cache_invalidate(net);
 924 }
 925
 926 /*
 927    Short description of GC goals.
 928
 929    We want to build algorithm, which will keep routing cache
 930    at some equilibrium point, when number of aged off entries
 931    is kept approximately equal to newly generated ones.
 932
 933    Current expiration strength is variable "expire".
 934    We try to adjust it dynamically, so that if networking
 935    is idle expires is large enough to keep enough of warm entries,
 936    and when load increases it reduces to limit cache size.
 937  */
 938
 939 static int rt_garbage_collect(struct dst_ops *ops)
 940 {
 941         static unsigned long expire = RT_GC_TIMEOUT;
 942         static unsigned long last_gc;
 943         static int rover;
 944         static int equilibrium;
 945         struct rtable *rth, **rthp;
 946         unsigned long now = jiffies;
 947         int goal;
 948
 949         /*
 950          * Garbage collection is pretty expensive,
 951          * do not make it too frequently.
 952          */
 953
 954         RT_CACHE_STAT_INC(gc_total);
 955
 956         if (now - last_gc < ip_rt_gc_min_interval &&
 957             atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) {
 958                 RT_CACHE_STAT_INC(gc_ignored);
 959                 goto out;
 960         }
 961
 962         /* Calculate number of entries, which we want to expire now. */
 963         goal = atomic_read(&ipv4_dst_ops.entries) -
 964                 (ip_rt_gc_elasticity << rt_hash_log);
 965         if (goal <= 0) {
 966                 if (equilibrium < ipv4_dst_ops.gc_thresh)
 967                         equilibrium = ipv4_dst_ops.gc_thresh;
 968                 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
 969                 if (goal > 0) {
 970                         equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
 971                         goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
 972                 }
 973         } else {
 974                 /* We are in dangerous area. Try to reduce cache really
 975                  * aggressively.
 976                  */
 977                 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
 978                 equilibrium = atomic_read(&ipv4_dst_ops.entries) - goal;
 979         }
 980
 981         if (now - last_gc >= ip_rt_gc_min_interval)
 982                 last_gc = now;
 983
 984         if (goal <= 0) {
 985                 equilibrium += goal;
 986                 goto work_done;
 987         }
 988
 989         do {
 990                 int i, k;
 991
 992                 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
 993                         unsigned long tmo = expire;
 994
 995                         k = (k + 1) & rt_hash_mask;
 996                         rthp = &rt_hash_table[k].chain;
 997                         spin_lock_bh(rt_hash_lock_addr(k));
 998                         while ((rth = *rthp) != NULL) {
 999                                 if (!rt_is_expired(rth) &&
1000                                         !rt_may_expire(rth, tmo, expire)) {
1001                                         tmo >>= 1;
1002                                         rthp = &rth->dst.rt_next;
1003                                         continue;
1004                                 }
1005                                 *rthp = rth->dst.rt_next;
1006                                 rt_free(rth);
1007                                 goal--;
1008                         }
1009                         spin_unlock_bh(rt_hash_lock_addr(k));
1010                         if (goal <= 0)
1011                                 break;
1012                 }
1013                 rover = k;
1014
1015                 if (goal <= 0)
1016                         goto work_done;
1017
1018                 /* Goal is not achieved. We stop process if:
1019
1020                    - if expire reduced to zero. Otherwise, expire is halfed.
1021                    - if table is not full.
1022                    - if we are called from interrupt.
1023                    - jiffies check is just fallback/debug loop breaker.
1024                      We will not spin here for long time in any case.
1025                  */
1026
1027                 RT_CACHE_STAT_INC(gc_goal_miss);
1028
1029                 if (expire == 0)
1030                         break;
1031
1032                 expire >>= 1;
1033 #if RT_CACHE_DEBUG >= 2
1034                 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
1035                                 atomic_read(&ipv4_dst_ops.entries), goal, i);
1036 #endif
1037
1038                 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
1039                         goto out;
1040         } while (!in_softirq() && time_before_eq(jiffies, now));
1041
1042         if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
1043                 goto out;
1044         if (net_ratelimit())
1045                 printk(KERN_WARNING "dst cache overflow\n");
1046         RT_CACHE_STAT_INC(gc_dst_overflow);
1047         return 1;
1048
1049 work_done:
1050         expire += ip_rt_gc_min_interval;
1051         if (expire > ip_rt_gc_timeout ||
1052             atomic_read(&ipv4_dst_ops.entries) < ipv4_dst_ops.gc_thresh)
1053                 expire = ip_rt_gc_timeout;
1054 #if RT_CACHE_DEBUG >= 2
1055         printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
1056                         atomic_read(&ipv4_dst_ops.entries), goal, rover);
1057 #endif
1058 out:    return 0;
1059 }
1060
1061 /*
1062  * Returns number of entries in a hash chain that have different hash_inputs
1063  */
1064 static int slow_chain_length(const struct rtable *head)
1065 {
1066         int length = 0;
1067         const struct rtable *rth = head;
1068
1069         while (rth) {
1070                 length += has_noalias(head, rth);
1071                 rth = rth->dst.rt_next;
1072         }
1073         return length >> FRACT_BITS;
1074 }
1075
1076 static int rt_intern_hash(unsigned hash, struct rtable *rt,
1077                           struct rtable **rp, struct sk_buff *skb, int ifindex)
1078 {
1079         struct rtable   *rth, **rthp;
1080         unsigned long   now;
1081         struct rtable *cand, **candp;
1082         u32             min_score;
1083         int             chain_length;
1084         int attempts = !in_softirq();
1085
1086 restart:
1087         chain_length = 0;
1088         min_score = ~(u32)0;
1089         cand = NULL;
1090         candp = NULL;
1091         now = jiffies;
1092
1093         if (!rt_caching(dev_net(rt->dst.dev))) {
1094                 /*
1095                  * If we're not caching, just tell the caller we
1096                  * were successful and don't touch the route.  The
1097                  * caller hold the sole reference to the cache entry, and
1098                  * it will be released when the caller is done with it.
1099                  * If we drop it here, the callers have no way to resolve routes
1100                  * when we're not caching.  Instead, just point *rp at rt, so
1101                  * the caller gets a single use out of the route
1102                  * Note that we do rt_free on this new route entry, so that
1103                  * once its refcount hits zero, we are still able to reap it
1104                  * (Thanks Alexey)
1105                  * Note also the rt_free uses call_rcu.  We don't actually
1106                  * need rcu protection here, this is just our path to get
1107                  * on the route gc list.
1108                  */
1109
1110                 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1111                         int err = arp_bind_neighbour(&rt->dst);
1112                         if (err) {
1113                                 if (net_ratelimit())
1114                                         printk(KERN_WARNING
1115                                             "Neighbour table failure & not caching routes.\n");
1116                                 rt_drop(rt);
1117                                 return err;
1118                         }
1119                 }
1120
1121                 rt_free(rt);
1122                 goto skip_hashing;
1123         }
1124
1125         rthp = &rt_hash_table[hash].chain;
1126
1127         spin_lock_bh(rt_hash_lock_addr(hash));
1128         while ((rth = *rthp) != NULL) {
1129                 if (rt_is_expired(rth)) {
1130                         *rthp = rth->dst.rt_next;
1131                         rt_free(rth);
1132                         continue;
1133                 }
1134                 if (compare_keys(&rth->fl, &rt->fl) && compare_netns(rth, rt)) {
1135                         /* Put it first */
1136                         *rthp = rth->dst.rt_next;
1137                         /*
1138                          * Since lookup is lockfree, the deletion
1139                          * must be visible to another weakly ordered CPU before
1140                          * the insertion at the start of the hash chain.
1141                          */
1142                         rcu_assign_pointer(rth->dst.rt_next,
1143                                            rt_hash_table[hash].chain);
1144                         /*
1145                          * Since lookup is lockfree, the update writes
1146                          * must be ordered for consistency on SMP.
1147                          */
1148                         rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1149
1150                         dst_use(&rth->dst, now);
1151                         spin_unlock_bh(rt_hash_lock_addr(hash));
1152
1153                         rt_drop(rt);
1154                         if (rp)
1155                                 *rp = rth;
1156                         else
1157                                 skb_dst_set(skb, &rth->dst);
1158                         return 0;
1159                 }
1160
1161                 if (!atomic_read(&rth->dst.__refcnt)) {
1162                         u32 score = rt_score(rth);
1163
1164                         if (score <= min_score) {
1165                                 cand = rth;
1166                                 candp = rthp;
1167                                 min_score = score;
1168                         }
1169                 }
1170
1171                 chain_length++;
1172
1173                 rthp = &rth->dst.rt_next;
1174         }
1175
1176         if (cand) {
1177                 /* ip_rt_gc_elasticity used to be average length of chain
1178                  * length, when exceeded gc becomes really aggressive.
1179                  *
1180                  * The second limit is less certain. At the moment it allows
1181                  * only 2 entries per bucket. We will see.
1182                  */
1183                 if (chain_length > ip_rt_gc_elasticity) {
1184                         *candp = cand->dst.rt_next;
1185                         rt_free(cand);
1186                 }
1187         } else {
1188                 if (chain_length > rt_chain_length_max &&
1189                     slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1190                         struct net *net = dev_net(rt->dst.dev);
1191                         int num = ++net->ipv4.current_rt_cache_rebuild_count;
1192                         if (!rt_caching(net)) {
1193                                 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1194                                         rt->dst.dev->name, num);
1195                         }
1196                         rt_emergency_hash_rebuild(net);
1197                         spin_unlock_bh(rt_hash_lock_addr(hash));
1198
1199                         hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1200                                         ifindex, rt_genid(net));
1201                         goto restart;
1202                 }
1203         }
1204
1205         /* Try to bind route to arp only if it is output
1206            route or unicast forwarding path.
1207          */
1208         if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1209                 int err = arp_bind_neighbour(&rt->dst);
1210                 if (err) {
1211                         spin_unlock_bh(rt_hash_lock_addr(hash));
1212
1213                         if (err != -ENOBUFS) {
1214                                 rt_drop(rt);
1215                                 return err;
1216                         }
1217
1218                         /* Neighbour tables are full and nothing
1219                            can be released. Try to shrink route cache,
1220                            it is most likely it holds some neighbour records.
1221                          */
1222                         if (attempts-- > 0) {
1223                                 int saved_elasticity = ip_rt_gc_elasticity;
1224                                 int saved_int = ip_rt_gc_min_interval;
1225                                 ip_rt_gc_elasticity     = 1;
1226                                 ip_rt_gc_min_interval   = 0;
1227                                 rt_garbage_collect(&ipv4_dst_ops);
1228                                 ip_rt_gc_min_interval   = saved_int;
1229                                 ip_rt_gc_elasticity     = saved_elasticity;
1230                                 goto restart;
1231                         }
1232
1233                         if (net_ratelimit())
1234                                 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1235                         rt_drop(rt);
1236                         return -ENOBUFS;
1237                 }
1238         }
1239
1240         rt->dst.rt_next = rt_hash_table[hash].chain;
1241
1242 #if RT_CACHE_DEBUG >= 2
1243         if (rt->dst.rt_next) {
1244                 struct rtable *trt;
1245                 printk(KERN_DEBUG "rt_cache @%02x: %pI4",
1246                        hash, &rt->rt_dst);
1247                 for (trt = rt->dst.rt_next; trt; trt = trt->dst.rt_next)
1248                         printk(" . %pI4", &trt->rt_dst);
1249                 printk("\n");
1250         }
1251 #endif
1252         /*
1253          * Since lookup is lockfree, we must make sure
1254          * previous writes to rt are comitted to memory
1255          * before making rt visible to other CPUS.
1256          */
1257         rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1258
1259         spin_unlock_bh(rt_hash_lock_addr(hash));
1260
1261 skip_hashing:
1262         if (rp)
1263                 *rp = rt;
1264         else
1265                 skb_dst_set(skb, &rt->dst);
1266         return 0;
1267 }
1268
1269 void rt_bind_peer(struct rtable *rt, int create)
1270 {
1271         static DEFINE_SPINLOCK(rt_peer_lock);
1272         struct inet_peer *peer;
1273
1274         peer = inet_getpeer(rt->rt_dst, create);
1275
1276         spin_lock_bh(&rt_peer_lock);
1277         if (rt->peer == NULL) {
1278                 rt->peer = peer;
1279                 peer = NULL;
1280         }
1281         spin_unlock_bh(&rt_peer_lock);
1282         if (peer)
1283                 inet_putpeer(peer);
1284 }
1285
1286 /*
1287  * Peer allocation may fail only in serious out-of-memory conditions.  However
1288  * we still can generate some output.
1289  * Random ID selection looks a bit dangerous because we have no chances to
1290  * select ID being unique in a reasonable period of time.
1291  * But broken packet identifier may be better than no packet at all.
1292  */
1293 static void ip_select_fb_ident(struct iphdr *iph)
1294 {
1295         static DEFINE_SPINLOCK(ip_fb_id_lock);
1296         static u32 ip_fallback_id;
1297         u32 salt;
1298
1299         spin_lock_bh(&ip_fb_id_lock);
1300         salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1301         iph->id = htons(salt & 0xFFFF);
1302         ip_fallback_id = salt;
1303         spin_unlock_bh(&ip_fb_id_lock);
1304 }
1305
1306 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1307 {
1308         struct rtable *rt = (struct rtable *) dst;
1309
1310         if (rt) {
1311                 if (rt->peer == NULL)
1312                         rt_bind_peer(rt, 1);
1313
1314                 /* If peer is attached to destination, it is never detached,
1315                    so that we need not to grab a lock to dereference it.
1316                  */
1317                 if (rt->peer) {
1318                         iph->id = htons(inet_getid(rt->peer, more));
1319                         return;
1320                 }
1321         } else
1322                 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1323                        __builtin_return_address(0));
1324
1325         ip_select_fb_ident(iph);
1326 }
1327 EXPORT_SYMBOL(__ip_select_ident);
1328
1329 static void rt_del(unsigned hash, struct rtable *rt)
1330 {
1331         struct rtable **rthp, *aux;
1332
1333         rthp = &rt_hash_table[hash].chain;
1334         spin_lock_bh(rt_hash_lock_addr(hash));
1335         ip_rt_put(rt);
1336         while ((aux = *rthp) != NULL) {
1337                 if (aux == rt || rt_is_expired(aux)) {
1338                         *rthp = aux->dst.rt_next;
1339                         rt_free(aux);
1340                         continue;
1341                 }
1342                 rthp = &aux->dst.rt_next;
1343         }
1344         spin_unlock_bh(rt_hash_lock_addr(hash));
1345 }
1346
1347 /* called in rcu_read_lock() section */
1348 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1349                     __be32 saddr, struct net_device *dev)
1350 {
1351         int i, k;
1352         struct in_device *in_dev = __in_dev_get_rcu(dev);
1353         struct rtable *rth, **rthp;
1354         __be32  skeys[2] = { saddr, 0 };
1355         int  ikeys[2] = { dev->ifindex, 0 };
1356         struct netevent_redirect netevent;
1357         struct net *net;
1358
1359         if (!in_dev)
1360                 return;
1361
1362         net = dev_net(dev);
1363         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1364             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1365             ipv4_is_zeronet(new_gw))
1366                 goto reject_redirect;
1367
1368         if (!rt_caching(net))
1369                 goto reject_redirect;
1370
1371         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1372                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1373                         goto reject_redirect;
1374                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1375                         goto reject_redirect;
1376         } else {
1377                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1378                         goto reject_redirect;
1379         }
1380
1381         for (i = 0; i < 2; i++) {
1382                 for (k = 0; k < 2; k++) {
1383                         unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1384                                                 rt_genid(net));
1385
1386                         rthp=&rt_hash_table[hash].chain;
1387
1388                         while ((rth = rcu_dereference(*rthp)) != NULL) {
1389                                 struct rtable *rt;
1390
1391                                 if (rth->fl.fl4_dst != daddr ||
1392                                     rth->fl.fl4_src != skeys[i] ||
1393                                     rth->fl.oif != ikeys[k] ||
1394                                     rth->fl.iif != 0 ||
1395                                     rt_is_expired(rth) ||
1396                                     !net_eq(dev_net(rth->dst.dev), net)) {
1397                                         rthp = &rth->dst.rt_next;
1398                                         continue;
1399                                 }
1400
1401                                 if (rth->rt_dst != daddr ||
1402                                     rth->rt_src != saddr ||
1403                                     rth->dst.error ||
1404                                     rth->rt_gateway != old_gw ||
1405                                     rth->dst.dev != dev)
1406                                         break;
1407
1408                                 dst_hold(&rth->dst);
1409
1410                                 rt = dst_alloc(&ipv4_dst_ops);
1411                                 if (rt == NULL) {
1412                                         ip_rt_put(rth);
1413                                         return;
1414                                 }
1415
1416                                 /* Copy all the information. */
1417                                 *rt = *rth;
1418                                 rt->dst.__use           = 1;
1419                                 atomic_set(&rt->dst.__refcnt, 1);
1420                                 rt->dst.child           = NULL;
1421                                 if (rt->dst.dev)
1422                                         dev_hold(rt->dst.dev);
1423                                 if (rt->idev)
1424                                         in_dev_hold(rt->idev);
1425                                 rt->dst.obsolete        = -1;
1426                                 rt->dst.lastuse = jiffies;
1427                                 rt->dst.path            = &rt->dst;
1428                                 rt->dst.neighbour       = NULL;
1429                                 rt->dst.hh              = NULL;
1430 #ifdef CONFIG_XFRM
1431                                 rt->dst.xfrm            = NULL;
1432 #endif
1433                                 rt->rt_genid            = rt_genid(net);
1434                                 rt->rt_flags            |= RTCF_REDIRECTED;
1435
1436                                 /* Gateway is different ... */
1437                                 rt->rt_gateway          = new_gw;
1438
1439                                 /* Redirect received -> path was valid */
1440                                 dst_confirm(&rth->dst);
1441
1442                                 if (rt->peer)
1443                                         atomic_inc(&rt->peer->refcnt);
1444
1445                                 if (arp_bind_neighbour(&rt->dst) ||
1446                                     !(rt->dst.neighbour->nud_state &
1447                                             NUD_VALID)) {
1448                                         if (rt->dst.neighbour)
1449                                                 neigh_event_send(rt->dst.neighbour, NULL);
1450                                         ip_rt_put(rth);
1451                                         rt_drop(rt);
1452                                         goto do_next;
1453                                 }
1454
1455                                 netevent.old = &rth->dst;
1456                                 netevent.new = &rt->dst;
1457                                 call_netevent_notifiers(NETEVENT_REDIRECT,
1458                                                         &netevent);
1459
1460                                 rt_del(hash, rth);
1461                                 if (!rt_intern_hash(hash, rt, &rt, NULL, rt->fl.oif))
1462                                         ip_rt_put(rt);
1463                                 goto do_next;
1464                         }
1465                 do_next:
1466                         ;
1467                 }
1468         }
1469         return;
1470
1471 reject_redirect:
1472 #ifdef CONFIG_IP_ROUTE_VERBOSE
1473         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1474                 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1475                         "  Advised path = %pI4 -> %pI4\n",
1476                        &old_gw, dev->name, &new_gw,
1477                        &saddr, &daddr);
1478 #endif
1479         ;
1480 }
1481
1482 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1483 {
1484         struct rtable *rt = (struct rtable *)dst;
1485         struct dst_entry *ret = dst;
1486
1487         if (rt) {
1488                 if (dst->obsolete > 0) {
1489                         ip_rt_put(rt);
1490                         ret = NULL;
1491                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1492                            (rt->dst.expires &&
1493                             time_after_eq(jiffies, rt->dst.expires))) {
1494                         unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1495                                                 rt->fl.oif,
1496                                                 rt_genid(dev_net(dst->dev)));
1497 #if RT_CACHE_DEBUG >= 1
1498                         printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1499                                 &rt->rt_dst, rt->fl.fl4_tos);
1500 #endif
1501                         rt_del(hash, rt);
1502                         ret = NULL;
1503                 }
1504         }
1505         return ret;
1506 }
1507
1508 /*
1509  * Algorithm:
1510  *      1. The first ip_rt_redirect_number redirects are sent
1511  *         with exponential backoff, then we stop sending them at all,
1512  *         assuming that the host ignores our redirects.
1513  *      2. If we did not see packets requiring redirects
1514  *         during ip_rt_redirect_silence, we assume that the host
1515  *         forgot redirected route and start to send redirects again.
1516  *
1517  * This algorithm is much cheaper and more intelligent than dumb load limiting
1518  * in icmp.c.
1519  *
1520  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1521  * and "frag. need" (breaks PMTU discovery) in icmp.c.
1522  */
1523
1524 void ip_rt_send_redirect(struct sk_buff *skb)
1525 {
1526         struct rtable *rt = skb_rtable(skb);
1527         struct in_device *in_dev;
1528         int log_martians;
1529
1530         rcu_read_lock();
1531         in_dev = __in_dev_get_rcu(rt->dst.dev);
1532         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1533                 rcu_read_unlock();
1534                 return;
1535         }
1536         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1537         rcu_read_unlock();
1538
1539         /* No redirected packets during ip_rt_redirect_silence;
1540          * reset the algorithm.
1541          */
1542         if (time_after(jiffies, rt->dst.rate_last + ip_rt_redirect_silence))
1543                 rt->dst.rate_tokens = 0;
1544
1545         /* Too many ignored redirects; do not send anything
1546          * set dst.rate_last to the last seen redirected packet.
1547          */
1548         if (rt->dst.rate_tokens >= ip_rt_redirect_number) {
1549                 rt->dst.rate_last = jiffies;
1550                 return;
1551         }
1552
1553         /* Check for load limit; set rate_last to the latest sent
1554          * redirect.
1555          */
1556         if (rt->dst.rate_tokens == 0 ||
1557             time_after(jiffies,
1558                        (rt->dst.rate_last +
1559                         (ip_rt_redirect_load << rt->dst.rate_tokens)))) {
1560                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1561                 rt->dst.rate_last = jiffies;
1562                 ++rt->dst.rate_tokens;
1563 #ifdef CONFIG_IP_ROUTE_VERBOSE
1564                 if (log_martians &&
1565                     rt->dst.rate_tokens == ip_rt_redirect_number &&
1566                     net_ratelimit())
1567                         printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1568                                 &rt->rt_src, rt->rt_iif,
1569                                 &rt->rt_dst, &rt->rt_gateway);
1570 #endif
1571         }
1572 }
1573
1574 static int ip_error(struct sk_buff *skb)
1575 {
1576         struct rtable *rt = skb_rtable(skb);
1577         unsigned long now;
1578         int code;
1579
1580         switch (rt->dst.error) {
1581                 case EINVAL:
1582                 default:
1583                         goto out;
1584                 case EHOSTUNREACH:
1585                         code = ICMP_HOST_UNREACH;
1586                         break;
1587                 case ENETUNREACH:
1588                         code = ICMP_NET_UNREACH;
1589                         IP_INC_STATS_BH(dev_net(rt->dst.dev),
1590                                         IPSTATS_MIB_INNOROUTES);
1591                         break;
1592                 case EACCES:
1593                         code = ICMP_PKT_FILTERED;
1594                         break;
1595         }
1596
1597         now = jiffies;
1598         rt->dst.rate_tokens += now - rt->dst.rate_last;
1599         if (rt->dst.rate_tokens > ip_rt_error_burst)
1600                 rt->dst.rate_tokens = ip_rt_error_burst;
1601         rt->dst.rate_last = now;
1602         if (rt->dst.rate_tokens >= ip_rt_error_cost) {
1603                 rt->dst.rate_tokens -= ip_rt_error_cost;
1604                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1605         }
1606
1607 out:    kfree_skb(skb);
1608         return 0;
1609 }
1610
1611 /*
1612  *      The last two values are not from the RFC but
1613  *      are needed for AMPRnet AX.25 paths.
1614  */
1615
1616 static const unsigned short mtu_plateau[] =
1617 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1618
1619 static inline unsigned short guess_mtu(unsigned short old_mtu)
1620 {
1621         int i;
1622
1623         for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1624                 if (old_mtu > mtu_plateau[i])
1625                         return mtu_plateau[i];
1626         return 68;
1627 }
1628
1629 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1630                                  unsigned short new_mtu,
1631                                  struct net_device *dev)
1632 {
1633         int i, k;
1634         unsigned short old_mtu = ntohs(iph->tot_len);
1635         struct rtable *rth;
1636         int  ikeys[2] = { dev->ifindex, 0 };
1637         __be32  skeys[2] = { iph->saddr, 0, };
1638         __be32  daddr = iph->daddr;
1639         unsigned short est_mtu = 0;
1640
1641         for (k = 0; k < 2; k++) {
1642                 for (i = 0; i < 2; i++) {
1643                         unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1644                                                 rt_genid(net));
1645
1646                         rcu_read_lock();
1647                         for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1648                              rth = rcu_dereference(rth->dst.rt_next)) {
1649                                 unsigned short mtu = new_mtu;
1650
1651                                 if (rth->fl.fl4_dst != daddr ||
1652                                     rth->fl.fl4_src != skeys[i] ||
1653                                     rth->rt_dst != daddr ||
1654                                     rth->rt_src != iph->saddr ||
1655                                     rth->fl.oif != ikeys[k] ||
1656                                     rth->fl.iif != 0 ||
1657                                     dst_metric_locked(&rth->dst, RTAX_MTU) ||
1658                                     !net_eq(dev_net(rth->dst.dev), net) ||
1659                                     rt_is_expired(rth))
1660                                         continue;
1661
1662                                 if (new_mtu < 68 || new_mtu >= old_mtu) {
1663
1664                                         /* BSD 4.2 compatibility hack :-( */
1665                                         if (mtu == 0 &&
1666                                             old_mtu >= dst_mtu(&rth->dst) &&
1667                                             old_mtu >= 68 + (iph->ihl << 2))
1668                                                 old_mtu -= iph->ihl << 2;
1669
1670                                         mtu = guess_mtu(old_mtu);
1671                                 }
1672                                 if (mtu <= dst_mtu(&rth->dst)) {
1673                                         if (mtu < dst_mtu(&rth->dst)) {
1674                                                 dst_confirm(&rth->dst);
1675                                                 if (mtu < ip_rt_min_pmtu) {
1676                                                         mtu = ip_rt_min_pmtu;
1677                                                         rth->dst.metrics[RTAX_LOCK-1] |=
1678                                                                 (1 << RTAX_MTU);
1679                                                 }
1680                                                 rth->dst.metrics[RTAX_MTU-1] = mtu;
1681                                                 dst_set_expires(&rth->dst,
1682                                                         ip_rt_mtu_expires);
1683                                         }
1684                                         est_mtu = mtu;
1685                                 }
1686                         }
1687                         rcu_read_unlock();
1688                 }
1689         }
1690         return est_mtu ? : new_mtu;
1691 }
1692
1693 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1694 {
1695         if (dst_mtu(dst) > mtu && mtu >= 68 &&
1696             !(dst_metric_locked(dst, RTAX_MTU))) {
1697                 if (mtu < ip_rt_min_pmtu) {
1698                         mtu = ip_rt_min_pmtu;
1699                         dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1700                 }
1701                 dst->metrics[RTAX_MTU-1] = mtu;
1702                 dst_set_expires(dst, ip_rt_mtu_expires);
1703                 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1704         }
1705 }
1706
1707 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1708 {
1709         if (rt_is_expired((struct rtable *)dst))
1710                 return NULL;
1711         return dst;
1712 }
1713
1714 static void ipv4_dst_destroy(struct dst_entry *dst)
1715 {
1716         struct rtable *rt = (struct rtable *) dst;
1717         struct inet_peer *peer = rt->peer;
1718         struct in_device *idev = rt->idev;
1719
1720         if (peer) {
1721                 rt->peer = NULL;
1722                 inet_putpeer(peer);
1723         }
1724
1725         if (idev) {
1726                 rt->idev = NULL;
1727                 in_dev_put(idev);
1728         }
1729 }
1730
1731 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1732                             int how)
1733 {
1734         struct rtable *rt = (struct rtable *) dst;
1735         struct in_device *idev = rt->idev;
1736         if (dev != dev_net(dev)->loopback_dev && idev && idev->dev == dev) {
1737                 struct in_device *loopback_idev =
1738                         in_dev_get(dev_net(dev)->loopback_dev);
1739                 if (loopback_idev) {
1740                         rt->idev = loopback_idev;
1741                         in_dev_put(idev);
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->fl.iif == 0)
1781                 src = rt->rt_src;
1782         else if (fib_lookup(dev_net(rt->dst.dev), &rt->fl, &res) == 0) {
1783                 src = FIB_RES_PREFSRC(res);
1784                 fib_res_put(&res);
1785         } else
1786                 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1787                                         RT_SCOPE_UNIVERSE);
1788         memcpy(addr, &src, 4);
1789 }
1790
1791 #ifdef CONFIG_NET_CLS_ROUTE
1792 static void set_class_tag(struct rtable *rt, u32 tag)
1793 {
1794         if (!(rt->dst.tclassid & 0xFFFF))
1795                 rt->dst.tclassid |= tag & 0xFFFF;
1796         if (!(rt->dst.tclassid & 0xFFFF0000))
1797                 rt->dst.tclassid |= tag & 0xFFFF0000;
1798 }
1799 #endif
1800
1801 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1802 {
1803         struct fib_info *fi = res->fi;
1804
1805         if (fi) {
1806                 if (FIB_RES_GW(*res) &&
1807                     FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1808                         rt->rt_gateway = FIB_RES_GW(*res);
1809                 memcpy(rt->dst.metrics, fi->fib_metrics,
1810                        sizeof(rt->dst.metrics));
1811                 if (fi->fib_mtu == 0) {
1812                         rt->dst.metrics[RTAX_MTU-1] = rt->dst.dev->mtu;
1813                         if (dst_metric_locked(&rt->dst, RTAX_MTU) &&
1814                             rt->rt_gateway != rt->rt_dst &&
1815                             rt->dst.dev->mtu > 576)
1816                                 rt->dst.metrics[RTAX_MTU-1] = 576;
1817                 }
1818 #ifdef CONFIG_NET_CLS_ROUTE
1819                 rt->dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1820 #endif
1821         } else
1822                 rt->dst.metrics[RTAX_MTU-1]= rt->dst.dev->mtu;
1823
1824         if (dst_metric(&rt->dst, RTAX_HOPLIMIT) == 0)
1825                 rt->dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1826         if (dst_mtu(&rt->dst) > IP_MAX_MTU)
1827                 rt->dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1828         if (dst_metric(&rt->dst, RTAX_ADVMSS) == 0)
1829                 rt->dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->dst.dev->mtu - 40,
1830                                        ip_rt_min_advmss);
1831         if (dst_metric(&rt->dst, RTAX_ADVMSS) > 65535 - 40)
1832                 rt->dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1833
1834 #ifdef CONFIG_NET_CLS_ROUTE
1835 #ifdef CONFIG_IP_MULTIPLE_TABLES
1836         set_class_tag(rt, fib_rules_tclass(res));
1837 #endif
1838         set_class_tag(rt, itag);
1839 #endif
1840         rt->rt_type = res->type;
1841 }
1842
1843 /* called in rcu_read_lock() section */
1844 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1845                                 u8 tos, struct net_device *dev, int our)
1846 {
1847         unsigned int hash;
1848         struct rtable *rth;
1849         __be32 spec_dst;
1850         struct in_device *in_dev = __in_dev_get_rcu(dev);
1851         u32 itag = 0;
1852         int err;
1853
1854         /* Primary sanity checks. */
1855
1856         if (in_dev == NULL)
1857                 return -EINVAL;
1858
1859         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1860             ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1861                 goto e_inval;
1862
1863         if (ipv4_is_zeronet(saddr)) {
1864                 if (!ipv4_is_local_multicast(daddr))
1865                         goto e_inval;
1866                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1867         } else {
1868                 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1869                                           &itag, 0);
1870                 if (err < 0)
1871                         goto e_err;
1872         }
1873         rth = dst_alloc(&ipv4_dst_ops);
1874         if (!rth)
1875                 goto e_nobufs;
1876
1877         rth->dst.output = ip_rt_bug;
1878         rth->dst.obsolete = -1;
1879
1880         atomic_set(&rth->dst.__refcnt, 1);
1881         rth->dst.flags= DST_HOST;
1882         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1883                 rth->dst.flags |= DST_NOPOLICY;
1884         rth->fl.fl4_dst = daddr;
1885         rth->rt_dst     = daddr;
1886         rth->fl.fl4_tos = tos;
1887         rth->fl.mark    = skb->mark;
1888         rth->fl.fl4_src = saddr;
1889         rth->rt_src     = saddr;
1890 #ifdef CONFIG_NET_CLS_ROUTE
1891         rth->dst.tclassid = itag;
1892 #endif
1893         rth->rt_iif     =
1894         rth->fl.iif     = dev->ifindex;
1895         rth->dst.dev    = init_net.loopback_dev;
1896         dev_hold(rth->dst.dev);
1897         rth->idev       = in_dev_get(rth->dst.dev);
1898         rth->fl.oif     = 0;
1899         rth->rt_gateway = daddr;
1900         rth->rt_spec_dst= spec_dst;
1901         rth->rt_genid   = rt_genid(dev_net(dev));
1902         rth->rt_flags   = RTCF_MULTICAST;
1903         rth->rt_type    = RTN_MULTICAST;
1904         if (our) {
1905                 rth->dst.input= ip_local_deliver;
1906                 rth->rt_flags |= RTCF_LOCAL;
1907         }
1908
1909 #ifdef CONFIG_IP_MROUTE
1910         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1911                 rth->dst.input = ip_mr_input;
1912 #endif
1913         RT_CACHE_STAT_INC(in_slow_mc);
1914
1915         hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1916         return rt_intern_hash(hash, rth, NULL, skb, dev->ifindex);
1917
1918 e_nobufs:
1919         return -ENOBUFS;
1920 e_inval:
1921         return -EINVAL;
1922 e_err:
1923         return err;
1924 }
1925
1926
1927 static void ip_handle_martian_source(struct net_device *dev,
1928                                      struct in_device *in_dev,
1929                                      struct sk_buff *skb,
1930                                      __be32 daddr,
1931                                      __be32 saddr)
1932 {
1933         RT_CACHE_STAT_INC(in_martian_src);
1934 #ifdef CONFIG_IP_ROUTE_VERBOSE
1935         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1936                 /*
1937                  *      RFC1812 recommendation, if source is martian,
1938                  *      the only hint is MAC header.
1939                  */
1940                 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1941                         &daddr, &saddr, dev->name);
1942                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1943                         int i;
1944                         const unsigned char *p = skb_mac_header(skb);
1945                         printk(KERN_WARNING "ll header: ");
1946                         for (i = 0; i < dev->hard_header_len; i++, p++) {
1947                                 printk("%02x", *p);
1948                                 if (i < (dev->hard_header_len - 1))
1949                                         printk(":");
1950                         }
1951                         printk("\n");
1952                 }
1953         }
1954 #endif
1955 }
1956
1957 /* called in rcu_read_lock() section */
1958 static int __mkroute_input(struct sk_buff *skb,
1959                            struct fib_result *res,
1960                            struct in_device *in_dev,
1961                            __be32 daddr, __be32 saddr, u32 tos,
1962                            struct rtable **result)
1963 {
1964         struct rtable *rth;
1965         int err;
1966         struct in_device *out_dev;
1967         unsigned int flags = 0;
1968         __be32 spec_dst;
1969         u32 itag;
1970
1971         /* get a working reference to the output device */
1972         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1973         if (out_dev == NULL) {
1974                 if (net_ratelimit())
1975                         printk(KERN_CRIT "Bug in ip_route_input" \
1976                                "_slow(). Please, report\n");
1977                 return -EINVAL;
1978         }
1979
1980
1981         err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1982                                   in_dev->dev, &spec_dst, &itag, skb->mark);
1983         if (err < 0) {
1984                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1985                                          saddr);
1986
1987                 goto cleanup;
1988         }
1989
1990         if (err)
1991                 flags |= RTCF_DIRECTSRC;
1992
1993         if (out_dev == in_dev && err &&
1994             (IN_DEV_SHARED_MEDIA(out_dev) ||
1995              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1996                 flags |= RTCF_DOREDIRECT;
1997
1998         if (skb->protocol != htons(ETH_P_IP)) {
1999                 /* Not IP (i.e. ARP). Do not create route, if it is
2000                  * invalid for proxy arp. DNAT routes are always valid.
2001                  *
2002                  * Proxy arp feature have been extended to allow, ARP
2003                  * replies back to the same interface, to support
2004                  * Private VLAN switch technologies. See arp.c.
2005                  */
2006                 if (out_dev == in_dev &&
2007                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2008                         err = -EINVAL;
2009                         goto cleanup;
2010                 }
2011         }
2012
2013
2014         rth = dst_alloc(&ipv4_dst_ops);
2015         if (!rth) {
2016                 err = -ENOBUFS;
2017                 goto cleanup;
2018         }
2019
2020         atomic_set(&rth->dst.__refcnt, 1);
2021         rth->dst.flags= DST_HOST;
2022         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2023                 rth->dst.flags |= DST_NOPOLICY;
2024         if (IN_DEV_CONF_GET(out_dev, NOXFRM))
2025                 rth->dst.flags |= DST_NOXFRM;
2026         rth->fl.fl4_dst = daddr;
2027         rth->rt_dst     = daddr;
2028         rth->fl.fl4_tos = tos;
2029         rth->fl.mark    = skb->mark;
2030         rth->fl.fl4_src = saddr;
2031         rth->rt_src     = saddr;
2032         rth->rt_gateway = daddr;
2033         rth->rt_iif     =
2034                 rth->fl.iif     = in_dev->dev->ifindex;
2035         rth->dst.dev    = (out_dev)->dev;
2036         dev_hold(rth->dst.dev);
2037         rth->idev       = in_dev_get(rth->dst.dev);
2038         rth->fl.oif     = 0;
2039         rth->rt_spec_dst= spec_dst;
2040
2041         rth->dst.obsolete = -1;
2042         rth->dst.input = ip_forward;
2043         rth->dst.output = ip_output;
2044         rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2045
2046         rt_set_nexthop(rth, res, itag);
2047
2048         rth->rt_flags = flags;
2049
2050         *result = rth;
2051         err = 0;
2052  cleanup:
2053         return err;
2054 }
2055
2056 static int ip_mkroute_input(struct sk_buff *skb,
2057                             struct fib_result *res,
2058                             const struct flowi *fl,
2059                             struct in_device *in_dev,
2060                             __be32 daddr, __be32 saddr, u32 tos)
2061 {
2062         struct rtable* rth = NULL;
2063         int err;
2064         unsigned hash;
2065
2066 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2067         if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
2068                 fib_select_multipath(fl, res);
2069 #endif
2070
2071         /* create a routing cache entry */
2072         err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2073         if (err)
2074                 return err;
2075
2076         /* put it into the cache */
2077         hash = rt_hash(daddr, saddr, fl->iif,
2078                        rt_genid(dev_net(rth->dst.dev)));
2079         return rt_intern_hash(hash, rth, NULL, skb, fl->iif);
2080 }
2081
2082 /*
2083  *      NOTE. We drop all the packets that has local source
2084  *      addresses, because every properly looped back packet
2085  *      must have correct destination already attached by output routine.
2086  *
2087  *      Such approach solves two big problems:
2088  *      1. Not simplex devices are handled properly.
2089  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2090  */
2091
2092 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2093                                u8 tos, struct net_device *dev)
2094 {
2095         struct fib_result res;
2096         struct in_device *in_dev = __in_dev_get_rcu(dev);
2097         struct flowi fl = { .nl_u = { .ip4_u =
2098                                       { .daddr = daddr,
2099                                         .saddr = saddr,
2100                                         .tos = tos,
2101                                         .scope = RT_SCOPE_UNIVERSE,
2102                                       } },
2103                             .mark = skb->mark,
2104                             .iif = dev->ifindex };
2105         unsigned        flags = 0;
2106         u32             itag = 0;
2107         struct rtable * rth;
2108         unsigned        hash;
2109         __be32          spec_dst;
2110         int             err = -EINVAL;
2111         int             free_res = 0;
2112         struct net    * net = dev_net(dev);
2113
2114         /* IP on this device is disabled. */
2115
2116         if (!in_dev)
2117                 goto out;
2118
2119         /* Check for the most weird martians, which can be not detected
2120            by fib_lookup.
2121          */
2122
2123         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2124             ipv4_is_loopback(saddr))
2125                 goto martian_source;
2126
2127         if (daddr == htonl(0xFFFFFFFF) || (saddr == 0 && daddr == 0))
2128                 goto brd_input;
2129
2130         /* Accept zero addresses only to limited broadcast;
2131          * I even do not know to fix it or not. Waiting for complains :-)
2132          */
2133         if (ipv4_is_zeronet(saddr))
2134                 goto martian_source;
2135
2136         if (ipv4_is_lbcast(daddr) || ipv4_is_zeronet(daddr) ||
2137             ipv4_is_loopback(daddr))
2138                 goto martian_destination;
2139
2140         /*
2141          *      Now we are ready to route packet.
2142          */
2143         if ((err = fib_lookup(net, &fl, &res)) != 0) {
2144                 if (!IN_DEV_FORWARD(in_dev))
2145                         goto e_hostunreach;
2146                 goto no_route;
2147         }
2148         free_res = 1;
2149
2150         RT_CACHE_STAT_INC(in_slow_tot);
2151
2152         if (res.type == RTN_BROADCAST)
2153                 goto brd_input;
2154
2155         if (res.type == RTN_LOCAL) {
2156                 err = fib_validate_source(saddr, daddr, tos,
2157                                              net->loopback_dev->ifindex,
2158                                              dev, &spec_dst, &itag, skb->mark);
2159                 if (err < 0)
2160                         goto martian_source_keep_err;
2161                 if (err)
2162                         flags |= RTCF_DIRECTSRC;
2163                 spec_dst = daddr;
2164                 goto local_input;
2165         }
2166
2167         if (!IN_DEV_FORWARD(in_dev))
2168                 goto e_hostunreach;
2169         if (res.type != RTN_UNICAST)
2170                 goto martian_destination;
2171
2172         err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
2173 done:
2174         if (free_res)
2175                 fib_res_put(&res);
2176 out:    return err;
2177
2178 brd_input:
2179         if (skb->protocol != htons(ETH_P_IP))
2180                 goto e_inval;
2181
2182         if (ipv4_is_zeronet(saddr))
2183                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2184         else {
2185                 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2186                                           &itag, skb->mark);
2187                 if (err < 0)
2188                         goto martian_source_keep_err;
2189                 if (err)
2190                         flags |= RTCF_DIRECTSRC;
2191         }
2192         flags |= RTCF_BROADCAST;
2193         res.type = RTN_BROADCAST;
2194         RT_CACHE_STAT_INC(in_brd);
2195
2196 local_input:
2197         rth = dst_alloc(&ipv4_dst_ops);
2198         if (!rth)
2199                 goto e_nobufs;
2200
2201         rth->dst.output= ip_rt_bug;
2202         rth->dst.obsolete = -1;
2203         rth->rt_genid = rt_genid(net);
2204
2205         atomic_set(&rth->dst.__refcnt, 1);
2206         rth->dst.flags= DST_HOST;
2207         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2208                 rth->dst.flags |= DST_NOPOLICY;
2209         rth->fl.fl4_dst = daddr;
2210         rth->rt_dst     = daddr;
2211         rth->fl.fl4_tos = tos;
2212         rth->fl.mark    = skb->mark;
2213         rth->fl.fl4_src = saddr;
2214         rth->rt_src     = saddr;
2215 #ifdef CONFIG_NET_CLS_ROUTE
2216         rth->dst.tclassid = itag;
2217 #endif
2218         rth->rt_iif     =
2219         rth->fl.iif     = dev->ifindex;
2220         rth->dst.dev    = net->loopback_dev;
2221         dev_hold(rth->dst.dev);
2222         rth->idev       = in_dev_get(rth->dst.dev);
2223         rth->rt_gateway = daddr;
2224         rth->rt_spec_dst= spec_dst;
2225         rth->dst.input= ip_local_deliver;
2226         rth->rt_flags   = flags|RTCF_LOCAL;
2227         if (res.type == RTN_UNREACHABLE) {
2228                 rth->dst.input= ip_error;
2229                 rth->dst.error= -err;
2230                 rth->rt_flags   &= ~RTCF_LOCAL;
2231         }
2232         rth->rt_type    = res.type;
2233         hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));
2234         err = rt_intern_hash(hash, rth, NULL, skb, fl.iif);
2235         goto done;
2236
2237 no_route:
2238         RT_CACHE_STAT_INC(in_no_route);
2239         spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2240         res.type = RTN_UNREACHABLE;
2241         if (err == -ESRCH)
2242                 err = -ENETUNREACH;
2243         goto local_input;
2244
2245         /*
2246          *      Do not cache martian addresses: they should be logged (RFC1812)
2247          */
2248 martian_destination:
2249         RT_CACHE_STAT_INC(in_martian_dst);
2250 #ifdef CONFIG_IP_ROUTE_VERBOSE
2251         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2252                 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2253                         &daddr, &saddr, dev->name);
2254 #endif
2255
2256 e_hostunreach:
2257         err = -EHOSTUNREACH;
2258         goto done;
2259
2260 e_inval:
2261         err = -EINVAL;
2262         goto done;
2263
2264 e_nobufs:
2265         err = -ENOBUFS;
2266         goto done;
2267
2268 martian_source:
2269         err = -EINVAL;
2270 martian_source_keep_err:
2271         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2272         goto done;
2273 }
2274
2275 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2276                            u8 tos, struct net_device *dev, bool noref)
2277 {
2278         struct rtable * rth;
2279         unsigned        hash;
2280         int iif = dev->ifindex;
2281         struct net *net;
2282         int res;
2283
2284         net = dev_net(dev);
2285
2286         rcu_read_lock();
2287
2288         if (!rt_caching(net))
2289                 goto skip_cache;
2290
2291         tos &= IPTOS_RT_MASK;
2292         hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2293
2294         for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2295              rth = rcu_dereference(rth->dst.rt_next)) {
2296                 if ((((__force u32)rth->fl.fl4_dst ^ (__force u32)daddr) |
2297                      ((__force u32)rth->fl.fl4_src ^ (__force u32)saddr) |
2298                      (rth->fl.iif ^ iif) |
2299                      rth->fl.oif |
2300                      (rth->fl.fl4_tos ^ tos)) == 0 &&
2301                     rth->fl.mark == skb->mark &&
2302                     net_eq(dev_net(rth->dst.dev), net) &&
2303                     !rt_is_expired(rth)) {
2304                         if (noref) {
2305                                 dst_use_noref(&rth->dst, jiffies);
2306                                 skb_dst_set_noref(skb, &rth->dst);
2307                         } else {
2308                                 dst_use(&rth->dst, jiffies);
2309                                 skb_dst_set(skb, &rth->dst);
2310                         }
2311                         RT_CACHE_STAT_INC(in_hit);
2312                         rcu_read_unlock();
2313                         return 0;
2314                 }
2315                 RT_CACHE_STAT_INC(in_hlist_search);
2316         }
2317
2318 skip_cache:
2319         /* Multicast recognition logic is moved from route cache to here.
2320            The problem was that too many Ethernet cards have broken/missing
2321            hardware multicast filters :-( As result the host on multicasting
2322            network acquires a lot of useless route cache entries, sort of
2323            SDR messages from all the world. Now we try to get rid of them.
2324            Really, provided software IP multicast filter is organized
2325            reasonably (at least, hashed), it does not result in a slowdown
2326            comparing with route cache reject entries.
2327            Note, that multicast routers are not affected, because
2328            route cache entry is created eventually.
2329          */
2330         if (ipv4_is_multicast(daddr)) {
2331                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2332
2333                 if (in_dev) {
2334                         int our = ip_check_mc(in_dev, daddr, saddr,
2335                                               ip_hdr(skb)->protocol);
2336                         if (our
2337 #ifdef CONFIG_IP_MROUTE
2338                                 ||
2339                             (!ipv4_is_local_multicast(daddr) &&
2340                              IN_DEV_MFORWARD(in_dev))
2341 #endif
2342                            ) {
2343                                 int res = ip_route_input_mc(skb, daddr, saddr,
2344                                                             tos, dev, our);
2345                                 rcu_read_unlock();
2346                                 return res;
2347                         }
2348                 }
2349                 rcu_read_unlock();
2350                 return -EINVAL;
2351         }
2352         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2353         rcu_read_unlock();
2354         return res;
2355 }
2356 EXPORT_SYMBOL(ip_route_input_common);
2357
2358 static int __mkroute_output(struct rtable **result,
2359                             struct fib_result *res,
2360                             const struct flowi *fl,
2361                             const struct flowi *oldflp,
2362                             struct net_device *dev_out,
2363                             unsigned flags)
2364 {
2365         struct rtable *rth;
2366         struct in_device *in_dev;
2367         u32 tos = RT_FL_TOS(oldflp);
2368         int err = 0;
2369
2370         if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags&IFF_LOOPBACK))
2371                 return -EINVAL;
2372
2373         if (fl->fl4_dst == htonl(0xFFFFFFFF))
2374                 res->type = RTN_BROADCAST;
2375         else if (ipv4_is_multicast(fl->fl4_dst))
2376                 res->type = RTN_MULTICAST;
2377         else if (ipv4_is_lbcast(fl->fl4_dst) || ipv4_is_zeronet(fl->fl4_dst))
2378                 return -EINVAL;
2379
2380         if (dev_out->flags & IFF_LOOPBACK)
2381                 flags |= RTCF_LOCAL;
2382
2383         /* get work reference to inet device */
2384         in_dev = in_dev_get(dev_out);
2385         if (!in_dev)
2386                 return -EINVAL;
2387
2388         if (res->type == RTN_BROADCAST) {
2389                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2390                 if (res->fi) {
2391                         fib_info_put(res->fi);
2392                         res->fi = NULL;
2393                 }
2394         } else if (res->type == RTN_MULTICAST) {
2395                 flags |= RTCF_MULTICAST|RTCF_LOCAL;
2396                 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2397                                  oldflp->proto))
2398                         flags &= ~RTCF_LOCAL;
2399                 /* If multicast route do not exist use
2400                    default one, but do not gateway in this case.
2401                    Yes, it is hack.
2402                  */
2403                 if (res->fi && res->prefixlen < 4) {
2404                         fib_info_put(res->fi);
2405                         res->fi = NULL;
2406                 }
2407         }
2408
2409
2410         rth = dst_alloc(&ipv4_dst_ops);
2411         if (!rth) {
2412                 err = -ENOBUFS;
2413                 goto cleanup;
2414         }
2415
2416         atomic_set(&rth->dst.__refcnt, 1);
2417         rth->dst.flags= DST_HOST;
2418         if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2419                 rth->dst.flags |= DST_NOXFRM;
2420         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2421                 rth->dst.flags |= DST_NOPOLICY;
2422
2423         rth->fl.fl4_dst = oldflp->fl4_dst;
2424         rth->fl.fl4_tos = tos;
2425         rth->fl.fl4_src = oldflp->fl4_src;
2426         rth->fl.oif     = oldflp->oif;
2427         rth->fl.mark    = oldflp->mark;
2428         rth->rt_dst     = fl->fl4_dst;
2429         rth->rt_src     = fl->fl4_src;
2430         rth->rt_iif     = oldflp->oif ? : dev_out->ifindex;
2431         /* get references to the devices that are to be hold by the routing
2432            cache entry */
2433         rth->dst.dev    = dev_out;
2434         dev_hold(dev_out);
2435         rth->idev       = in_dev_get(dev_out);
2436         rth->rt_gateway = fl->fl4_dst;
2437         rth->rt_spec_dst= fl->fl4_src;
2438
2439         rth->dst.output=ip_output;
2440         rth->dst.obsolete = -1;
2441         rth->rt_genid = rt_genid(dev_net(dev_out));
2442
2443         RT_CACHE_STAT_INC(out_slow_tot);
2444
2445         if (flags & RTCF_LOCAL) {
2446                 rth->dst.input = ip_local_deliver;
2447                 rth->rt_spec_dst = fl->fl4_dst;
2448         }
2449         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2450                 rth->rt_spec_dst = fl->fl4_src;
2451                 if (flags & RTCF_LOCAL &&
2452                     !(dev_out->flags & IFF_LOOPBACK)) {
2453                         rth->dst.output = ip_mc_output;
2454                         RT_CACHE_STAT_INC(out_slow_mc);
2455                 }
2456 #ifdef CONFIG_IP_MROUTE
2457                 if (res->type == RTN_MULTICAST) {
2458                         if (IN_DEV_MFORWARD(in_dev) &&
2459                             !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2460                                 rth->dst.input = ip_mr_input;
2461                                 rth->dst.output = ip_mc_output;
2462                         }
2463                 }
2464 #endif
2465         }
2466
2467         rt_set_nexthop(rth, res, 0);
2468
2469         rth->rt_flags = flags;
2470
2471         *result = rth;
2472  cleanup:
2473         /* release work reference to inet device */
2474         in_dev_put(in_dev);
2475
2476         return err;
2477 }
2478
2479 static int ip_mkroute_output(struct rtable **rp,
2480                              struct fib_result *res,
2481                              const struct flowi *fl,
2482                              const struct flowi *oldflp,
2483                              struct net_device *dev_out,
2484                              unsigned flags)
2485 {
2486         struct rtable *rth = NULL;
2487         int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2488         unsigned hash;
2489         if (err == 0) {
2490                 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif,
2491                                rt_genid(dev_net(dev_out)));
2492                 err = rt_intern_hash(hash, rth, rp, NULL, oldflp->oif);
2493         }
2494
2495         return err;
2496 }
2497
2498 /*
2499  * Major route resolver routine.
2500  */
2501
2502 static int ip_route_output_slow(struct net *net, struct rtable **rp,
2503                                 const struct flowi *oldflp)
2504 {
2505         u32 tos = RT_FL_TOS(oldflp);
2506         struct flowi fl = { .nl_u = { .ip4_u =
2507                                       { .daddr = oldflp->fl4_dst,
2508                                         .saddr = oldflp->fl4_src,
2509                                         .tos = tos & IPTOS_RT_MASK,
2510                                         .scope = ((tos & RTO_ONLINK) ?
2511                                                   RT_SCOPE_LINK :
2512                                                   RT_SCOPE_UNIVERSE),
2513                                       } },
2514                             .mark = oldflp->mark,
2515                             .iif = net->loopback_dev->ifindex,
2516                             .oif = oldflp->oif };
2517         struct fib_result res;
2518         unsigned flags = 0;
2519         struct net_device *dev_out = NULL;
2520         int free_res = 0;
2521         int err;
2522
2523
2524         res.fi          = NULL;
2525 #ifdef CONFIG_IP_MULTIPLE_TABLES
2526         res.r           = NULL;
2527 #endif
2528
2529         if (oldflp->fl4_src) {
2530                 err = -EINVAL;
2531                 if (ipv4_is_multicast(oldflp->fl4_src) ||
2532                     ipv4_is_lbcast(oldflp->fl4_src) ||
2533                     ipv4_is_zeronet(oldflp->fl4_src))
2534                         goto out;
2535
2536                 /* I removed check for oif == dev_out->oif here.
2537                    It was wrong for two reasons:
2538                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2539                       is assigned to multiple interfaces.
2540                    2. Moreover, we are allowed to send packets with saddr
2541                       of another iface. --ANK
2542                  */
2543
2544                 if (oldflp->oif == 0 &&
2545                     (ipv4_is_multicast(oldflp->fl4_dst) ||
2546                      oldflp->fl4_dst == htonl(0xFFFFFFFF))) {
2547                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2548                         dev_out = ip_dev_find(net, oldflp->fl4_src);
2549                         if (dev_out == NULL)
2550                                 goto out;
2551
2552                         /* Special hack: user can direct multicasts
2553                            and limited broadcast via necessary interface
2554                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2555                            This hack is not just for fun, it allows
2556                            vic,vat and friends to work.
2557                            They bind socket to loopback, set ttl to zero
2558                            and expect that it will work.
2559                            From the viewpoint of routing cache they are broken,
2560                            because we are not allowed to build multicast path
2561                            with loopback source addr (look, routing cache
2562                            cannot know, that ttl is zero, so that packet
2563                            will not leave this host and route is valid).
2564                            Luckily, this hack is good workaround.
2565                          */
2566
2567                         fl.oif = dev_out->ifindex;
2568                         goto make_route;
2569                 }
2570
2571                 if (!(oldflp->flags & FLOWI_FLAG_ANYSRC)) {
2572                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2573                         dev_out = ip_dev_find(net, oldflp->fl4_src);
2574                         if (dev_out == NULL)
2575                                 goto out;
2576                         dev_put(dev_out);
2577                         dev_out = NULL;
2578                 }
2579         }
2580
2581
2582         if (oldflp->oif) {
2583                 dev_out = dev_get_by_index(net, oldflp->oif);
2584                 err = -ENODEV;
2585                 if (dev_out == NULL)
2586                         goto out;
2587
2588                 /* RACE: Check return value of inet_select_addr instead. */
2589                 if (__in_dev_get_rtnl(dev_out) == NULL) {
2590                         dev_put(dev_out);
2591                         goto out;       /* Wrong error code */
2592                 }
2593
2594                 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2595                     oldflp->fl4_dst == htonl(0xFFFFFFFF)) {
2596                         if (!fl.fl4_src)
2597                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2598                                                               RT_SCOPE_LINK);
2599                         goto make_route;
2600                 }
2601                 if (!fl.fl4_src) {
2602                         if (ipv4_is_multicast(oldflp->fl4_dst))
2603                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2604                                                               fl.fl4_scope);
2605                         else if (!oldflp->fl4_dst)
2606                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2607                                                               RT_SCOPE_HOST);
2608                 }
2609         }
2610
2611         if (!fl.fl4_dst) {
2612                 fl.fl4_dst = fl.fl4_src;
2613                 if (!fl.fl4_dst)
2614                         fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2615                 if (dev_out)
2616                         dev_put(dev_out);
2617                 dev_out = net->loopback_dev;
2618                 dev_hold(dev_out);
2619                 fl.oif = net->loopback_dev->ifindex;
2620                 res.type = RTN_LOCAL;
2621                 flags |= RTCF_LOCAL;
2622                 goto make_route;
2623         }
2624
2625         if (fib_lookup(net, &fl, &res)) {
2626                 res.fi = NULL;
2627                 if (oldflp->oif) {
2628                         /* Apparently, routing tables are wrong. Assume,
2629                            that the destination is on link.
2630
2631                            WHY? DW.
2632                            Because we are allowed to send to iface
2633                            even if it has NO routes and NO assigned
2634                            addresses. When oif is specified, routing
2635                            tables are looked up with only one purpose:
2636                            to catch if destination is gatewayed, rather than
2637                            direct. Moreover, if MSG_DONTROUTE is set,
2638                            we send packet, ignoring both routing tables
2639                            and ifaddr state. --ANK
2640
2641
2642                            We could make it even if oif is unknown,
2643                            likely IPv6, but we do not.
2644                          */
2645
2646                         if (fl.fl4_src == 0)
2647                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2648                                                               RT_SCOPE_LINK);
2649                         res.type = RTN_UNICAST;
2650                         goto make_route;
2651                 }
2652                 if (dev_out)
2653                         dev_put(dev_out);
2654                 err = -ENETUNREACH;
2655                 goto out;
2656         }
2657         free_res = 1;
2658
2659         if (res.type == RTN_LOCAL) {
2660                 if (!fl.fl4_src)
2661                         fl.fl4_src = fl.fl4_dst;
2662                 if (dev_out)
2663                         dev_put(dev_out);
2664                 dev_out = net->loopback_dev;
2665                 dev_hold(dev_out);
2666                 fl.oif = dev_out->ifindex;
2667                 if (res.fi)
2668                         fib_info_put(res.fi);
2669                 res.fi = NULL;
2670                 flags |= RTCF_LOCAL;
2671                 goto make_route;
2672         }
2673
2674 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2675         if (res.fi->fib_nhs > 1 && fl.oif == 0)
2676                 fib_select_multipath(&fl, &res);
2677         else
2678 #endif
2679         if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2680                 fib_select_default(net, &fl, &res);
2681
2682         if (!fl.fl4_src)
2683                 fl.fl4_src = FIB_RES_PREFSRC(res);
2684
2685         if (dev_out)
2686                 dev_put(dev_out);
2687         dev_out = FIB_RES_DEV(res);
2688         dev_hold(dev_out);
2689         fl.oif = dev_out->ifindex;
2690
2691
2692 make_route:
2693         err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2694
2695
2696         if (free_res)
2697                 fib_res_put(&res);
2698         if (dev_out)
2699                 dev_put(dev_out);
2700 out:    return err;
2701 }
2702
2703 int __ip_route_output_key(struct net *net, struct rtable **rp,
2704                           const struct flowi *flp)
2705 {
2706         unsigned hash;
2707         struct rtable *rth;
2708
2709         if (!rt_caching(net))
2710                 goto slow_output;
2711
2712         hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, rt_genid(net));
2713
2714         rcu_read_lock_bh();
2715         for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2716                 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2717                 if (rth->fl.fl4_dst == flp->fl4_dst &&
2718                     rth->fl.fl4_src == flp->fl4_src &&
2719                     rth->fl.iif == 0 &&
2720                     rth->fl.oif == flp->oif &&
2721                     rth->fl.mark == flp->mark &&
2722                     !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2723                             (IPTOS_RT_MASK | RTO_ONLINK)) &&
2724                     net_eq(dev_net(rth->dst.dev), net) &&
2725                     !rt_is_expired(rth)) {
2726                         dst_use(&rth->dst, jiffies);
2727                         RT_CACHE_STAT_INC(out_hit);
2728                         rcu_read_unlock_bh();
2729                         *rp = rth;
2730                         return 0;
2731                 }
2732                 RT_CACHE_STAT_INC(out_hlist_search);
2733         }
2734         rcu_read_unlock_bh();
2735
2736 slow_output:
2737         return ip_route_output_slow(net, rp, flp);
2738 }
2739 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2740
2741 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2742 {
2743         return NULL;
2744 }
2745
2746 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2747 {
2748 }
2749
2750 static struct dst_ops ipv4_dst_blackhole_ops = {
2751         .family                 =       AF_INET,
2752         .protocol               =       cpu_to_be16(ETH_P_IP),
2753         .destroy                =       ipv4_dst_destroy,
2754         .check                  =       ipv4_blackhole_dst_check,
2755         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2756         .entries                =       ATOMIC_INIT(0),
2757 };
2758
2759
2760 static int ipv4_dst_blackhole(struct net *net, struct rtable **rp, struct flowi *flp)
2761 {
2762         struct rtable *ort = *rp;
2763         struct rtable *rt = (struct rtable *)
2764                 dst_alloc(&ipv4_dst_blackhole_ops);
2765
2766         if (rt) {
2767                 struct dst_entry *new = &rt->dst;
2768
2769                 atomic_set(&new->__refcnt, 1);
2770                 new->__use = 1;
2771                 new->input = dst_discard;
2772                 new->output = dst_discard;
2773                 memcpy(new->metrics, ort->dst.metrics, RTAX_MAX*sizeof(u32));
2774
2775                 new->dev = ort->dst.dev;
2776                 if (new->dev)
2777                         dev_hold(new->dev);
2778
2779                 rt->fl = ort->fl;
2780
2781                 rt->idev = ort->idev;
2782                 if (rt->idev)
2783                         in_dev_hold(rt->idev);
2784                 rt->rt_genid = rt_genid(net);
2785                 rt->rt_flags = ort->rt_flags;
2786                 rt->rt_type = ort->rt_type;
2787                 rt->rt_dst = ort->rt_dst;
2788                 rt->rt_src = ort->rt_src;
2789                 rt->rt_iif = ort->rt_iif;
2790                 rt->rt_gateway = ort->rt_gateway;
2791                 rt->rt_spec_dst = ort->rt_spec_dst;
2792                 rt->peer = ort->peer;
2793                 if (rt->peer)
2794                         atomic_inc(&rt->peer->refcnt);
2795
2796                 dst_free(new);
2797         }
2798
2799         dst_release(&(*rp)->dst);
2800         *rp = rt;
2801         return (rt ? 0 : -ENOMEM);
2802 }
2803
2804 int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2805                          struct sock *sk, int flags)
2806 {
2807         int err;
2808
2809         if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2810                 return err;
2811
2812         if (flp->proto) {
2813                 if (!flp->fl4_src)
2814                         flp->fl4_src = (*rp)->rt_src;
2815                 if (!flp->fl4_dst)
2816                         flp->fl4_dst = (*rp)->rt_dst;
2817                 err = __xfrm_lookup(net, (struct dst_entry **)rp, flp, sk,
2818                                     flags ? XFRM_LOOKUP_WAIT : 0);
2819                 if (err == -EREMOTE)
2820                         err = ipv4_dst_blackhole(net, rp, flp);
2821
2822                 return err;
2823         }
2824
2825         return 0;
2826 }
2827 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2828
2829 int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2830 {
2831         return ip_route_output_flow(net, rp, flp, NULL, 0);
2832 }
2833 EXPORT_SYMBOL(ip_route_output_key);
2834
2835 static int rt_fill_info(struct net *net,
2836                         struct sk_buff *skb, u32 pid, u32 seq, int event,
2837                         int nowait, unsigned int flags)
2838 {
2839         struct rtable *rt = skb_rtable(skb);
2840         struct rtmsg *r;
2841         struct nlmsghdr *nlh;
2842         long expires;
2843         u32 id = 0, ts = 0, tsage = 0, error;
2844
2845         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2846         if (nlh == NULL)
2847                 return -EMSGSIZE;
2848
2849         r = nlmsg_data(nlh);
2850         r->rtm_family    = AF_INET;
2851         r->rtm_dst_len  = 32;
2852         r->rtm_src_len  = 0;
2853         r->rtm_tos      = rt->fl.fl4_tos;
2854         r->rtm_table    = RT_TABLE_MAIN;
2855         NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2856         r->rtm_type     = rt->rt_type;
2857         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2858         r->rtm_protocol = RTPROT_UNSPEC;
2859         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2860         if (rt->rt_flags & RTCF_NOTIFY)
2861                 r->rtm_flags |= RTM_F_NOTIFY;
2862
2863         NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2864
2865         if (rt->fl.fl4_src) {
2866                 r->rtm_src_len = 32;
2867                 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2868         }
2869         if (rt->dst.dev)
2870                 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2871 #ifdef CONFIG_NET_CLS_ROUTE
2872         if (rt->dst.tclassid)
2873                 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2874 #endif
2875         if (rt->fl.iif)
2876                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2877         else if (rt->rt_src != rt->fl.fl4_src)
2878                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2879
2880         if (rt->rt_dst != rt->rt_gateway)
2881                 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2882
2883         if (rtnetlink_put_metrics(skb, rt->dst.metrics) < 0)
2884                 goto nla_put_failure;
2885
2886         if (rt->fl.mark)
2887                 NLA_PUT_BE32(skb, RTA_MARK, rt->fl.mark);
2888
2889         error = rt->dst.error;
2890         expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
2891         if (rt->peer) {
2892                 inet_peer_refcheck(rt->peer);
2893                 id = atomic_read(&rt->peer->ip_id_count) & 0xffff;
2894                 if (rt->peer->tcp_ts_stamp) {
2895                         ts = rt->peer->tcp_ts;
2896                         tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2897                 }
2898         }
2899
2900         if (rt->fl.iif) {
2901 #ifdef CONFIG_IP_MROUTE
2902                 __be32 dst = rt->rt_dst;
2903
2904                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2905                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2906                         int err = ipmr_get_route(net, skb, r, nowait);
2907                         if (err <= 0) {
2908                                 if (!nowait) {
2909                                         if (err == 0)
2910                                                 return 0;
2911                                         goto nla_put_failure;
2912                                 } else {
2913                                         if (err == -EMSGSIZE)
2914                                                 goto nla_put_failure;
2915                                         error = err;
2916                                 }
2917                         }
2918                 } else
2919 #endif
2920                         NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2921         }
2922
2923         if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2924                                expires, error) < 0)
2925                 goto nla_put_failure;
2926
2927         return nlmsg_end(skb, nlh);
2928
2929 nla_put_failure:
2930         nlmsg_cancel(skb, nlh);
2931         return -EMSGSIZE;
2932 }
2933
2934 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2935 {
2936         struct net *net = sock_net(in_skb->sk);
2937         struct rtmsg *rtm;
2938         struct nlattr *tb[RTA_MAX+1];
2939         struct rtable *rt = NULL;
2940         __be32 dst = 0;
2941         __be32 src = 0;
2942         u32 iif;
2943         int err;
2944         int mark;
2945         struct sk_buff *skb;
2946
2947         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2948         if (err < 0)
2949                 goto errout;
2950
2951         rtm = nlmsg_data(nlh);
2952
2953         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2954         if (skb == NULL) {
2955                 err = -ENOBUFS;
2956                 goto errout;
2957         }
2958
2959         /* Reserve room for dummy headers, this skb can pass
2960            through good chunk of routing engine.
2961          */
2962         skb_reset_mac_header(skb);
2963         skb_reset_network_header(skb);
2964
2965         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2966         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2967         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2968
2969         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2970         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2971         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2972         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2973
2974         if (iif) {
2975                 struct net_device *dev;
2976
2977                 dev = __dev_get_by_index(net, iif);
2978                 if (dev == NULL) {
2979                         err = -ENODEV;
2980                         goto errout_free;
2981                 }
2982
2983                 skb->protocol   = htons(ETH_P_IP);
2984                 skb->dev        = dev;
2985                 skb->mark       = mark;
2986                 local_bh_disable();
2987                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2988                 local_bh_enable();
2989
2990                 rt = skb_rtable(skb);
2991                 if (err == 0 && rt->dst.error)
2992                         err = -rt->dst.error;
2993         } else {
2994                 struct flowi fl = {
2995                         .nl_u = {
2996                                 .ip4_u = {
2997                                         .daddr = dst,
2998                                         .saddr = src,
2999                                         .tos = rtm->rtm_tos,
3000                                 },
3001                         },
3002                         .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
3003                         .mark = mark,
3004                 };
3005                 err = ip_route_output_key(net, &rt, &fl);
3006         }
3007
3008         if (err)
3009                 goto errout_free;
3010
3011         skb_dst_set(skb, &rt->dst);
3012         if (rtm->rtm_flags & RTM_F_NOTIFY)
3013                 rt->rt_flags |= RTCF_NOTIFY;
3014
3015         err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3016                            RTM_NEWROUTE, 0, 0);
3017         if (err <= 0)
3018                 goto errout_free;
3019
3020         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3021 errout:
3022         return err;
3023
3024 errout_free:
3025         kfree_skb(skb);
3026         goto errout;
3027 }
3028
3029 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
3030 {
3031         struct rtable *rt;
3032         int h, s_h;
3033         int idx, s_idx;
3034         struct net *net;
3035
3036         net = sock_net(skb->sk);
3037
3038         s_h = cb->args[0];
3039         if (s_h < 0)
3040                 s_h = 0;
3041         s_idx = idx = cb->args[1];
3042         for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3043                 if (!rt_hash_table[h].chain)
3044                         continue;
3045                 rcu_read_lock_bh();
3046                 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3047                      rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3048                         if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3049                                 continue;
3050                         if (rt_is_expired(rt))
3051                                 continue;
3052                         skb_dst_set_noref(skb, &rt->dst);
3053                         if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3054                                          cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3055                                          1, NLM_F_MULTI) <= 0) {
3056                                 skb_dst_drop(skb);
3057                                 rcu_read_unlock_bh();
3058                                 goto done;
3059                         }
3060                         skb_dst_drop(skb);
3061                 }
3062                 rcu_read_unlock_bh();
3063         }
3064
3065 done:
3066         cb->args[0] = h;
3067         cb->args[1] = idx;
3068         return skb->len;
3069 }
3070
3071 void ip_rt_multicast_event(struct in_device *in_dev)
3072 {
3073         rt_cache_flush(dev_net(in_dev->dev), 0);
3074 }
3075
3076 #ifdef CONFIG_SYSCTL
3077 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3078                                         void __user *buffer,
3079                                         size_t *lenp, loff_t *ppos)
3080 {
3081         if (write) {
3082                 int flush_delay;
3083                 ctl_table ctl;
3084                 struct net *net;
3085
3086                 memcpy(&ctl, __ctl, sizeof(ctl));
3087                 ctl.data = &flush_delay;
3088                 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3089
3090                 net = (struct net *)__ctl->extra1;
3091                 rt_cache_flush(net, flush_delay);
3092                 return 0;
3093         }
3094
3095         return -EINVAL;
3096 }
3097
3098 static ctl_table ipv4_route_table[] = {
3099         {
3100                 .procname       = "gc_thresh",
3101                 .data           = &ipv4_dst_ops.gc_thresh,
3102                 .maxlen         = sizeof(int),
3103                 .mode           = 0644,
3104                 .proc_handler   = proc_dointvec,
3105         },
3106         {
3107                 .procname       = "max_size",
3108                 .data           = &ip_rt_max_size,
3109                 .maxlen         = sizeof(int),
3110                 .mode           = 0644,
3111                 .proc_handler   = proc_dointvec,
3112         },
3113         {
3114                 /*  Deprecated. Use gc_min_interval_ms */
3115
3116                 .procname       = "gc_min_interval",
3117                 .data           = &ip_rt_gc_min_interval,
3118                 .maxlen         = sizeof(int),
3119                 .mode           = 0644,
3120                 .proc_handler   = proc_dointvec_jiffies,
3121         },
3122         {
3123                 .procname       = "gc_min_interval_ms",
3124                 .data           = &ip_rt_gc_min_interval,
3125                 .maxlen         = sizeof(int),
3126                 .mode           = 0644,
3127                 .proc_handler   = proc_dointvec_ms_jiffies,
3128         },
3129         {
3130                 .procname       = "gc_timeout",
3131                 .data           = &ip_rt_gc_timeout,
3132                 .maxlen         = sizeof(int),
3133                 .mode           = 0644,
3134                 .proc_handler   = proc_dointvec_jiffies,
3135         },
3136         {
3137                 .procname       = "gc_interval",
3138                 .data           = &ip_rt_gc_interval,
3139                 .maxlen         = sizeof(int),
3140                 .mode           = 0644,
3141                 .proc_handler   = proc_dointvec_jiffies,
3142         },
3143         {
3144                 .procname       = "redirect_load",
3145                 .data           = &ip_rt_redirect_load,
3146                 .maxlen         = sizeof(int),
3147                 .mode           = 0644,
3148                 .proc_handler   = proc_dointvec,
3149         },
3150         {
3151                 .procname       = "redirect_number",
3152                 .data           = &ip_rt_redirect_number,
3153                 .maxlen         = sizeof(int),
3154                 .mode           = 0644,
3155                 .proc_handler   = proc_dointvec,
3156         },
3157         {
3158                 .procname       = "redirect_silence",
3159                 .data           = &ip_rt_redirect_silence,
3160                 .maxlen         = sizeof(int),
3161                 .mode           = 0644,
3162                 .proc_handler   = proc_dointvec,
3163         },
3164         {
3165                 .procname       = "error_cost",
3166                 .data           = &ip_rt_error_cost,
3167                 .maxlen         = sizeof(int),
3168                 .mode           = 0644,
3169                 .proc_handler   = proc_dointvec,
3170         },
3171         {
3172                 .procname       = "error_burst",
3173                 .data           = &ip_rt_error_burst,
3174                 .maxlen         = sizeof(int),
3175                 .mode           = 0644,
3176                 .proc_handler   = proc_dointvec,
3177         },
3178         {
3179                 .procname       = "gc_elasticity",
3180                 .data           = &ip_rt_gc_elasticity,
3181                 .maxlen         = sizeof(int),
3182                 .mode           = 0644,
3183                 .proc_handler   = proc_dointvec,
3184         },
3185         {
3186                 .procname       = "mtu_expires",
3187                 .data           = &ip_rt_mtu_expires,
3188                 .maxlen         = sizeof(int),
3189                 .mode           = 0644,
3190                 .proc_handler   = proc_dointvec_jiffies,
3191         },
3192         {
3193                 .procname       = "min_pmtu",
3194                 .data           = &ip_rt_min_pmtu,
3195                 .maxlen         = sizeof(int),
3196                 .mode           = 0644,
3197                 .proc_handler   = proc_dointvec,
3198         },
3199         {
3200                 .procname       = "min_adv_mss",
3201                 .data           = &ip_rt_min_advmss,
3202                 .maxlen         = sizeof(int),
3203                 .mode           = 0644,
3204                 .proc_handler   = proc_dointvec,
3205         },
3206         { }
3207 };
3208
3209 static struct ctl_table empty[1];
3210
3211 static struct ctl_table ipv4_skeleton[] =
3212 {
3213         { .procname = "route",
3214           .mode = 0555, .child = ipv4_route_table},
3215         { .procname = "neigh",
3216           .mode = 0555, .child = empty},
3217         { }
3218 };
3219
3220 static __net_initdata struct ctl_path ipv4_path[] = {
3221         { .procname = "net", },
3222         { .procname = "ipv4", },
3223         { },
3224 };
3225
3226 static struct ctl_table ipv4_route_flush_table[] = {
3227         {
3228                 .procname       = "flush",
3229                 .maxlen         = sizeof(int),
3230                 .mode           = 0200,
3231                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3232         },
3233         { },
3234 };
3235
3236 static __net_initdata struct ctl_path ipv4_route_path[] = {
3237         { .procname = "net", },
3238         { .procname = "ipv4", },
3239         { .procname = "route", },
3240         { },
3241 };
3242
3243 static __net_init int sysctl_route_net_init(struct net *net)
3244 {
3245         struct ctl_table *tbl;
3246
3247         tbl = ipv4_route_flush_table;
3248         if (!net_eq(net, &init_net)) {
3249                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3250                 if (tbl == NULL)
3251                         goto err_dup;
3252         }
3253         tbl[0].extra1 = net;
3254
3255         net->ipv4.route_hdr =
3256                 register_net_sysctl_table(net, ipv4_route_path, tbl);
3257         if (net->ipv4.route_hdr == NULL)
3258                 goto err_reg;
3259         return 0;
3260
3261 err_reg:
3262         if (tbl != ipv4_route_flush_table)
3263                 kfree(tbl);
3264 err_dup:
3265         return -ENOMEM;
3266 }
3267
3268 static __net_exit void sysctl_route_net_exit(struct net *net)
3269 {
3270         struct ctl_table *tbl;
3271
3272         tbl = net->ipv4.route_hdr->ctl_table_arg;
3273         unregister_net_sysctl_table(net->ipv4.route_hdr);
3274         BUG_ON(tbl == ipv4_route_flush_table);
3275         kfree(tbl);
3276 }
3277
3278 static __net_initdata struct pernet_operations sysctl_route_ops = {
3279         .init = sysctl_route_net_init,
3280         .exit = sysctl_route_net_exit,
3281 };
3282 #endif
3283
3284 static __net_init int rt_genid_init(struct net *net)
3285 {
3286         get_random_bytes(&net->ipv4.rt_genid,
3287                          sizeof(net->ipv4.rt_genid));
3288         return 0;
3289 }
3290
3291 static __net_initdata struct pernet_operations rt_genid_ops = {
3292         .init = rt_genid_init,
3293 };
3294
3295
3296 #ifdef CONFIG_NET_CLS_ROUTE
3297 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3298 #endif /* CONFIG_NET_CLS_ROUTE */
3299
3300 static __initdata unsigned long rhash_entries;
3301 static int __init set_rhash_entries(char *str)
3302 {
3303         if (!str)
3304                 return 0;
3305         rhash_entries = simple_strtoul(str, &str, 0);
3306         return 1;
3307 }
3308 __setup("rhash_entries=", set_rhash_entries);
3309
3310 int __init ip_rt_init(void)
3311 {
3312         int rc = 0;
3313
3314 #ifdef CONFIG_NET_CLS_ROUTE
3315         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3316         if (!ip_rt_acct)
3317                 panic("IP: failed to allocate ip_rt_acct\n");
3318 #endif
3319
3320         ipv4_dst_ops.kmem_cachep =
3321                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3322                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3323
3324         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3325
3326         rt_hash_table = (struct rt_hash_bucket *)
3327                 alloc_large_system_hash("IP route cache",
3328                                         sizeof(struct rt_hash_bucket),
3329                                         rhash_entries,
3330                                         (totalram_pages >= 128 * 1024) ?
3331                                         15 : 17,
3332                                         0,
3333                                         &rt_hash_log,
3334                                         &rt_hash_mask,
3335                                         rhash_entries ? 0 : 512 * 1024);
3336         memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3337         rt_hash_lock_init();
3338
3339         ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3340         ip_rt_max_size = (rt_hash_mask + 1) * 16;
3341
3342         devinet_init();
3343         ip_fib_init();
3344
3345         /* All the timers, started at system startup tend
3346            to synchronize. Perturb it a bit.
3347          */
3348         INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3349         expires_ljiffies = jiffies;
3350         schedule_delayed_work(&expires_work,
3351                 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3352
3353         if (ip_rt_proc_init())
3354                 printk(KERN_ERR "Unable to create route proc files\n");
3355 #ifdef CONFIG_XFRM
3356         xfrm_init();
3357         xfrm4_init(ip_rt_max_size);
3358 #endif
3359         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3360
3361 #ifdef CONFIG_SYSCTL
3362         register_pernet_subsys(&sysctl_route_ops);
3363 #endif
3364         register_pernet_subsys(&rt_genid_ops);
3365         return rc;
3366 }
3367
3368 #ifdef CONFIG_SYSCTL
3369 /*
3370  * We really need to sanitize the damn ipv4 init order, then all
3371  * this nonsense will go away.
3372  */
3373 void __init ip_static_sysctl_init(void)
3374 {
3375         register_sysctl_paths(ipv4_path, ipv4_skeleton);
3376 }
3377 #endif