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->u.dst.dev) == seq_file_net(seq) &&
 290                             r->rt_genid == st->genid)
 291                                 return r;
 292                         r = rcu_dereference_bh(r->u.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->u.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->u.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->u.dst.dev ? r->u.dst.dev->name : "*",
 381                         (__force u32)r->rt_dst,
 382                         (__force u32)r->rt_gateway,
 383                         r->rt_flags, atomic_read(&r->u.dst.__refcnt),
 384                         r->u.dst.__use, 0, (__force u32)r->rt_src,
 385                         (dst_metric(&r->u.dst, RTAX_ADVMSS) ?
 386                              (int)dst_metric(&r->u.dst, RTAX_ADVMSS) + 40 : 0),
 387                         dst_metric(&r->u.dst, RTAX_WINDOW),
 388                         (int)((dst_metric(&r->u.dst, RTAX_RTT) >> 3) +
 389                               dst_metric(&r->u.dst, RTAX_RTTVAR)),
 390                         r->fl.fl4_tos,
 391                         r->u.dst.hh ? atomic_read(&r->u.dst.hh->hh_refcnt) : -1,
 392                         r->u.dst.hh ? (r->u.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->u.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->u.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->u.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->u.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->u.dst.__refcnt))
 640                 goto out;
 641
 642         ret = 1;
 643         if (rth->u.dst.expires &&
 644             time_after_eq(jiffies, rth->u.dst.expires))
 645                 goto out;
 646
 647         age = jiffies - rth->u.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->u.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->u.dst.dev), dev_net(rt2->u.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->u.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->u.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->u.dst.rt_next;
 747                         if (!rt_is_expired(p)) {
 748                                 prev = &p->u.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->u.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->u.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->u.dst.rt_next);
 835                         if (rt_is_expired(rth)) {
 836                                 *rthp = rth->u.dst.rt_next;
 837                                 rt_free(rth);
 838                                 continue;
 839                         }
 840                         if (rth->u.dst.expires) {
 841                                 /* Entry is expired even if it is in use */
 842                                 if (time_before_eq(jiffies, rth->u.dst.expires)) {
 843 nofree:
 844                                         tmo >>= 1;
 845                                         rthp = &rth->u.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->u.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->u.dst.rt_next;
1003                                         continue;
1004                                 }
1005                                 *rthp = rth->u.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->u.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->u.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->u.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->u.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->u.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->u.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->u.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->u.dst);
1158                         return 0;
1159                 }
1160
1161                 if (!atomic_read(&rth->u.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->u.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->u.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->u.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->u.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->u.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 "Neighbour table overflow.\n");
1235                         rt_drop(rt);
1236                         return -ENOBUFS;
1237                 }
1238         }
1239
1240         rt->u.dst.rt_next = rt_hash_table[hash].chain;
1241
1242 #if RT_CACHE_DEBUG >= 2
1243         if (rt->u.dst.rt_next) {
1244                 struct rtable *trt;
1245                 printk(KERN_DEBUG "rt_cache @%02x: %pI4",
1246                        hash, &rt->rt_dst);
1247                 for (trt = rt->u.dst.rt_next; trt; trt = trt->u.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->u.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
1328 static void rt_del(unsigned hash, struct rtable *rt)
1329 {
1330         struct rtable **rthp, *aux;
1331
1332         rthp = &rt_hash_table[hash].chain;
1333         spin_lock_bh(rt_hash_lock_addr(hash));
1334         ip_rt_put(rt);
1335         while ((aux = *rthp) != NULL) {
1336                 if (aux == rt || rt_is_expired(aux)) {
1337                         *rthp = aux->u.dst.rt_next;
1338                         rt_free(aux);
1339                         continue;
1340                 }
1341                 rthp = &aux->u.dst.rt_next;
1342         }
1343         spin_unlock_bh(rt_hash_lock_addr(hash));
1344 }
1345
1346 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1347                     __be32 saddr, struct net_device *dev)
1348 {
1349         int i, k;
1350         struct in_device *in_dev = in_dev_get(dev);
1351         struct rtable *rth, **rthp;
1352         __be32  skeys[2] = { saddr, 0 };
1353         int  ikeys[2] = { dev->ifindex, 0 };
1354         struct netevent_redirect netevent;
1355         struct net *net;
1356
1357         if (!in_dev)
1358                 return;
1359
1360         net = dev_net(dev);
1361         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1362             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1363             ipv4_is_zeronet(new_gw))
1364                 goto reject_redirect;
1365
1366         if (!rt_caching(net))
1367                 goto reject_redirect;
1368
1369         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1370                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1371                         goto reject_redirect;
1372                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1373                         goto reject_redirect;
1374         } else {
1375                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1376                         goto reject_redirect;
1377         }
1378
1379         for (i = 0; i < 2; i++) {
1380                 for (k = 0; k < 2; k++) {
1381                         unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1382                                                 rt_genid(net));
1383
1384                         rthp=&rt_hash_table[hash].chain;
1385
1386                         rcu_read_lock();
1387                         while ((rth = rcu_dereference(*rthp)) != NULL) {
1388                                 struct rtable *rt;
1389
1390                                 if (rth->fl.fl4_dst != daddr ||
1391                                     rth->fl.fl4_src != skeys[i] ||
1392                                     rth->fl.oif != ikeys[k] ||
1393                                     rth->fl.iif != 0 ||
1394                                     rt_is_expired(rth) ||
1395                                     !net_eq(dev_net(rth->u.dst.dev), net)) {
1396                                         rthp = &rth->u.dst.rt_next;
1397                                         continue;
1398                                 }
1399
1400                                 if (rth->rt_dst != daddr ||
1401                                     rth->rt_src != saddr ||
1402                                     rth->u.dst.error ||
1403                                     rth->rt_gateway != old_gw ||
1404                                     rth->u.dst.dev != dev)
1405                                         break;
1406
1407                                 dst_hold(&rth->u.dst);
1408                                 rcu_read_unlock();
1409
1410                                 rt = dst_alloc(&ipv4_dst_ops);
1411                                 if (rt == NULL) {
1412                                         ip_rt_put(rth);
1413                                         in_dev_put(in_dev);
1414                                         return;
1415                                 }
1416
1417                                 /* Copy all the information. */
1418                                 *rt = *rth;
1419                                 rt->u.dst.__use         = 1;
1420                                 atomic_set(&rt->u.dst.__refcnt, 1);
1421                                 rt->u.dst.child         = NULL;
1422                                 if (rt->u.dst.dev)
1423                                         dev_hold(rt->u.dst.dev);
1424                                 if (rt->idev)
1425                                         in_dev_hold(rt->idev);
1426                                 rt->u.dst.obsolete      = -1;
1427                                 rt->u.dst.lastuse       = jiffies;
1428                                 rt->u.dst.path          = &rt->u.dst;
1429                                 rt->u.dst.neighbour     = NULL;
1430                                 rt->u.dst.hh            = NULL;
1431 #ifdef CONFIG_XFRM
1432                                 rt->u.dst.xfrm          = NULL;
1433 #endif
1434                                 rt->rt_genid            = rt_genid(net);
1435                                 rt->rt_flags            |= RTCF_REDIRECTED;
1436
1437                                 /* Gateway is different ... */
1438                                 rt->rt_gateway          = new_gw;
1439
1440                                 /* Redirect received -> path was valid */
1441                                 dst_confirm(&rth->u.dst);
1442
1443                                 if (rt->peer)
1444                                         atomic_inc(&rt->peer->refcnt);
1445
1446                                 if (arp_bind_neighbour(&rt->u.dst) ||
1447                                     !(rt->u.dst.neighbour->nud_state &
1448                                             NUD_VALID)) {
1449                                         if (rt->u.dst.neighbour)
1450                                                 neigh_event_send(rt->u.dst.neighbour, NULL);
1451                                         ip_rt_put(rth);
1452                                         rt_drop(rt);
1453                                         goto do_next;
1454                                 }
1455
1456                                 netevent.old = &rth->u.dst;
1457                                 netevent.new = &rt->u.dst;
1458                                 call_netevent_notifiers(NETEVENT_REDIRECT,
1459                                                         &netevent);
1460
1461                                 rt_del(hash, rth);
1462                                 if (!rt_intern_hash(hash, rt, &rt, NULL, rt->fl.oif))
1463                                         ip_rt_put(rt);
1464                                 goto do_next;
1465                         }
1466                         rcu_read_unlock();
1467                 do_next:
1468                         ;
1469                 }
1470         }
1471         in_dev_put(in_dev);
1472         return;
1473
1474 reject_redirect:
1475 #ifdef CONFIG_IP_ROUTE_VERBOSE
1476         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1477                 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1478                         "  Advised path = %pI4 -> %pI4\n",
1479                        &old_gw, dev->name, &new_gw,
1480                        &saddr, &daddr);
1481 #endif
1482         in_dev_put(in_dev);
1483 }
1484
1485 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1486 {
1487         struct rtable *rt = (struct rtable *)dst;
1488         struct dst_entry *ret = dst;
1489
1490         if (rt) {
1491                 if (dst->obsolete > 0) {
1492                         ip_rt_put(rt);
1493                         ret = NULL;
1494                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1495                            (rt->u.dst.expires &&
1496                             time_after_eq(jiffies, rt->u.dst.expires))) {
1497                         unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1498                                                 rt->fl.oif,
1499                                                 rt_genid(dev_net(dst->dev)));
1500 #if RT_CACHE_DEBUG >= 1
1501                         printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1502                                 &rt->rt_dst, rt->fl.fl4_tos);
1503 #endif
1504                         rt_del(hash, rt);
1505                         ret = NULL;
1506                 }
1507         }
1508         return ret;
1509 }
1510
1511 /*
1512  * Algorithm:
1513  *      1. The first ip_rt_redirect_number redirects are sent
1514  *         with exponential backoff, then we stop sending them at all,
1515  *         assuming that the host ignores our redirects.
1516  *      2. If we did not see packets requiring redirects
1517  *         during ip_rt_redirect_silence, we assume that the host
1518  *         forgot redirected route and start to send redirects again.
1519  *
1520  * This algorithm is much cheaper and more intelligent than dumb load limiting
1521  * in icmp.c.
1522  *
1523  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1524  * and "frag. need" (breaks PMTU discovery) in icmp.c.
1525  */
1526
1527 void ip_rt_send_redirect(struct sk_buff *skb)
1528 {
1529         struct rtable *rt = skb_rtable(skb);
1530         struct in_device *in_dev;
1531         int log_martians;
1532
1533         rcu_read_lock();
1534         in_dev = __in_dev_get_rcu(rt->u.dst.dev);
1535         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1536                 rcu_read_unlock();
1537                 return;
1538         }
1539         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1540         rcu_read_unlock();
1541
1542         /* No redirected packets during ip_rt_redirect_silence;
1543          * reset the algorithm.
1544          */
1545         if (time_after(jiffies, rt->u.dst.rate_last + ip_rt_redirect_silence))
1546                 rt->u.dst.rate_tokens = 0;
1547
1548         /* Too many ignored redirects; do not send anything
1549          * set u.dst.rate_last to the last seen redirected packet.
1550          */
1551         if (rt->u.dst.rate_tokens >= ip_rt_redirect_number) {
1552                 rt->u.dst.rate_last = jiffies;
1553                 return;
1554         }
1555
1556         /* Check for load limit; set rate_last to the latest sent
1557          * redirect.
1558          */
1559         if (rt->u.dst.rate_tokens == 0 ||
1560             time_after(jiffies,
1561                        (rt->u.dst.rate_last +
1562                         (ip_rt_redirect_load << rt->u.dst.rate_tokens)))) {
1563                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1564                 rt->u.dst.rate_last = jiffies;
1565                 ++rt->u.dst.rate_tokens;
1566 #ifdef CONFIG_IP_ROUTE_VERBOSE
1567                 if (log_martians &&
1568                     rt->u.dst.rate_tokens == ip_rt_redirect_number &&
1569                     net_ratelimit())
1570                         printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1571                                 &rt->rt_src, rt->rt_iif,
1572                                 &rt->rt_dst, &rt->rt_gateway);
1573 #endif
1574         }
1575 }
1576
1577 static int ip_error(struct sk_buff *skb)
1578 {
1579         struct rtable *rt = skb_rtable(skb);
1580         unsigned long now;
1581         int code;
1582
1583         switch (rt->u.dst.error) {
1584                 case EINVAL:
1585                 default:
1586                         goto out;
1587                 case EHOSTUNREACH:
1588                         code = ICMP_HOST_UNREACH;
1589                         break;
1590                 case ENETUNREACH:
1591                         code = ICMP_NET_UNREACH;
1592                         IP_INC_STATS_BH(dev_net(rt->u.dst.dev),
1593                                         IPSTATS_MIB_INNOROUTES);
1594                         break;
1595                 case EACCES:
1596                         code = ICMP_PKT_FILTERED;
1597                         break;
1598         }
1599
1600         now = jiffies;
1601         rt->u.dst.rate_tokens += now - rt->u.dst.rate_last;
1602         if (rt->u.dst.rate_tokens > ip_rt_error_burst)
1603                 rt->u.dst.rate_tokens = ip_rt_error_burst;
1604         rt->u.dst.rate_last = now;
1605         if (rt->u.dst.rate_tokens >= ip_rt_error_cost) {
1606                 rt->u.dst.rate_tokens -= ip_rt_error_cost;
1607                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1608         }
1609
1610 out:    kfree_skb(skb);
1611         return 0;
1612 }
1613
1614 /*
1615  *      The last two values are not from the RFC but
1616  *      are needed for AMPRnet AX.25 paths.
1617  */
1618
1619 static const unsigned short mtu_plateau[] =
1620 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1621
1622 static inline unsigned short guess_mtu(unsigned short old_mtu)
1623 {
1624         int i;
1625
1626         for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1627                 if (old_mtu > mtu_plateau[i])
1628                         return mtu_plateau[i];
1629         return 68;
1630 }
1631
1632 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1633                                  unsigned short new_mtu,
1634                                  struct net_device *dev)
1635 {
1636         int i, k;
1637         unsigned short old_mtu = ntohs(iph->tot_len);
1638         struct rtable *rth;
1639         int  ikeys[2] = { dev->ifindex, 0 };
1640         __be32  skeys[2] = { iph->saddr, 0, };
1641         __be32  daddr = iph->daddr;
1642         unsigned short est_mtu = 0;
1643
1644         for (k = 0; k < 2; k++) {
1645                 for (i = 0; i < 2; i++) {
1646                         unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1647                                                 rt_genid(net));
1648
1649                         rcu_read_lock();
1650                         for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1651                              rth = rcu_dereference(rth->u.dst.rt_next)) {
1652                                 unsigned short mtu = new_mtu;
1653
1654                                 if (rth->fl.fl4_dst != daddr ||
1655                                     rth->fl.fl4_src != skeys[i] ||
1656                                     rth->rt_dst != daddr ||
1657                                     rth->rt_src != iph->saddr ||
1658                                     rth->fl.oif != ikeys[k] ||
1659                                     rth->fl.iif != 0 ||
1660                                     dst_metric_locked(&rth->u.dst, RTAX_MTU) ||
1661                                     !net_eq(dev_net(rth->u.dst.dev), net) ||
1662                                     rt_is_expired(rth))
1663                                         continue;
1664
1665                                 if (new_mtu < 68 || new_mtu >= old_mtu) {
1666
1667                                         /* BSD 4.2 compatibility hack :-( */
1668                                         if (mtu == 0 &&
1669                                             old_mtu >= dst_mtu(&rth->u.dst) &&
1670                                             old_mtu >= 68 + (iph->ihl << 2))
1671                                                 old_mtu -= iph->ihl << 2;
1672
1673                                         mtu = guess_mtu(old_mtu);
1674                                 }
1675                                 if (mtu <= dst_mtu(&rth->u.dst)) {
1676                                         if (mtu < dst_mtu(&rth->u.dst)) {
1677                                                 dst_confirm(&rth->u.dst);
1678                                                 if (mtu < ip_rt_min_pmtu) {
1679                                                         mtu = ip_rt_min_pmtu;
1680                                                         rth->u.dst.metrics[RTAX_LOCK-1] |=
1681                                                                 (1 << RTAX_MTU);
1682                                                 }
1683                                                 rth->u.dst.metrics[RTAX_MTU-1] = mtu;
1684                                                 dst_set_expires(&rth->u.dst,
1685                                                         ip_rt_mtu_expires);
1686                                         }
1687                                         est_mtu = mtu;
1688                                 }
1689                         }
1690                         rcu_read_unlock();
1691                 }
1692         }
1693         return est_mtu ? : new_mtu;
1694 }
1695
1696 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1697 {
1698         if (dst_mtu(dst) > mtu && mtu >= 68 &&
1699             !(dst_metric_locked(dst, RTAX_MTU))) {
1700                 if (mtu < ip_rt_min_pmtu) {
1701                         mtu = ip_rt_min_pmtu;
1702                         dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1703                 }
1704                 dst->metrics[RTAX_MTU-1] = mtu;
1705                 dst_set_expires(dst, ip_rt_mtu_expires);
1706                 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1707         }
1708 }
1709
1710 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1711 {
1712         if (rt_is_expired((struct rtable *)dst))
1713                 return NULL;
1714         return dst;
1715 }
1716
1717 static void ipv4_dst_destroy(struct dst_entry *dst)
1718 {
1719         struct rtable *rt = (struct rtable *) dst;
1720         struct inet_peer *peer = rt->peer;
1721         struct in_device *idev = rt->idev;
1722
1723         if (peer) {
1724                 rt->peer = NULL;
1725                 inet_putpeer(peer);
1726         }
1727
1728         if (idev) {
1729                 rt->idev = NULL;
1730                 in_dev_put(idev);
1731         }
1732 }
1733
1734 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1735                             int how)
1736 {
1737         struct rtable *rt = (struct rtable *) dst;
1738         struct in_device *idev = rt->idev;
1739         if (dev != dev_net(dev)->loopback_dev && idev && idev->dev == dev) {
1740                 struct in_device *loopback_idev =
1741                         in_dev_get(dev_net(dev)->loopback_dev);
1742                 if (loopback_idev) {
1743                         rt->idev = loopback_idev;
1744                         in_dev_put(idev);
1745                 }
1746         }
1747 }
1748
1749 static void ipv4_link_failure(struct sk_buff *skb)
1750 {
1751         struct rtable *rt;
1752
1753         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1754
1755         rt = skb_rtable(skb);
1756         if (rt)
1757                 dst_set_expires(&rt->u.dst, 0);
1758 }
1759
1760 static int ip_rt_bug(struct sk_buff *skb)
1761 {
1762         printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1763                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1764                 skb->dev ? skb->dev->name : "?");
1765         kfree_skb(skb);
1766         return 0;
1767 }
1768
1769 /*
1770    We do not cache source address of outgoing interface,
1771    because it is used only by IP RR, TS and SRR options,
1772    so that it out of fast path.
1773
1774    BTW remember: "addr" is allowed to be not aligned
1775    in IP options!
1776  */
1777
1778 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1779 {
1780         __be32 src;
1781         struct fib_result res;
1782
1783         if (rt->fl.iif == 0)
1784                 src = rt->rt_src;
1785         else if (fib_lookup(dev_net(rt->u.dst.dev), &rt->fl, &res) == 0) {
1786                 src = FIB_RES_PREFSRC(res);
1787                 fib_res_put(&res);
1788         } else
1789                 src = inet_select_addr(rt->u.dst.dev, rt->rt_gateway,
1790                                         RT_SCOPE_UNIVERSE);
1791         memcpy(addr, &src, 4);
1792 }
1793
1794 #ifdef CONFIG_NET_CLS_ROUTE
1795 static void set_class_tag(struct rtable *rt, u32 tag)
1796 {
1797         if (!(rt->u.dst.tclassid & 0xFFFF))
1798                 rt->u.dst.tclassid |= tag & 0xFFFF;
1799         if (!(rt->u.dst.tclassid & 0xFFFF0000))
1800                 rt->u.dst.tclassid |= tag & 0xFFFF0000;
1801 }
1802 #endif
1803
1804 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1805 {
1806         struct fib_info *fi = res->fi;
1807
1808         if (fi) {
1809                 if (FIB_RES_GW(*res) &&
1810                     FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1811                         rt->rt_gateway = FIB_RES_GW(*res);
1812                 memcpy(rt->u.dst.metrics, fi->fib_metrics,
1813                        sizeof(rt->u.dst.metrics));
1814                 if (fi->fib_mtu == 0) {
1815                         rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
1816                         if (dst_metric_locked(&rt->u.dst, RTAX_MTU) &&
1817                             rt->rt_gateway != rt->rt_dst &&
1818                             rt->u.dst.dev->mtu > 576)
1819                                 rt->u.dst.metrics[RTAX_MTU-1] = 576;
1820                 }
1821 #ifdef CONFIG_NET_CLS_ROUTE
1822                 rt->u.dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1823 #endif
1824         } else
1825                 rt->u.dst.metrics[RTAX_MTU-1]= rt->u.dst.dev->mtu;
1826
1827         if (dst_metric(&rt->u.dst, RTAX_HOPLIMIT) == 0)
1828                 rt->u.dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1829         if (dst_mtu(&rt->u.dst) > IP_MAX_MTU)
1830                 rt->u.dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1831         if (dst_metric(&rt->u.dst, RTAX_ADVMSS) == 0)
1832                 rt->u.dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->u.dst.dev->mtu - 40,
1833                                        ip_rt_min_advmss);
1834         if (dst_metric(&rt->u.dst, RTAX_ADVMSS) > 65535 - 40)
1835                 rt->u.dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1836
1837 #ifdef CONFIG_NET_CLS_ROUTE
1838 #ifdef CONFIG_IP_MULTIPLE_TABLES
1839         set_class_tag(rt, fib_rules_tclass(res));
1840 #endif
1841         set_class_tag(rt, itag);
1842 #endif
1843         rt->rt_type = res->type;
1844 }
1845
1846 /* called in rcu_read_lock() section */
1847 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1848                                 u8 tos, struct net_device *dev, int our)
1849 {
1850         unsigned int hash;
1851         struct rtable *rth;
1852         __be32 spec_dst;
1853         struct in_device *in_dev = __in_dev_get_rcu(dev);
1854         u32 itag = 0;
1855         int err;
1856
1857         /* Primary sanity checks. */
1858
1859         if (in_dev == NULL)
1860                 return -EINVAL;
1861
1862         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1863             ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1864                 goto e_inval;
1865
1866         if (ipv4_is_zeronet(saddr)) {
1867                 if (!ipv4_is_local_multicast(daddr))
1868                         goto e_inval;
1869                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1870         } else {
1871                 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1872                                           &itag, 0);
1873                 if (err < 0)
1874                         goto e_err;
1875         }
1876         rth = dst_alloc(&ipv4_dst_ops);
1877         if (!rth)
1878                 goto e_nobufs;
1879
1880         rth->u.dst.output = ip_rt_bug;
1881         rth->u.dst.obsolete = -1;
1882
1883         atomic_set(&rth->u.dst.__refcnt, 1);
1884         rth->u.dst.flags= DST_HOST;
1885         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1886                 rth->u.dst.flags |= DST_NOPOLICY;
1887         rth->fl.fl4_dst = daddr;
1888         rth->rt_dst     = daddr;
1889         rth->fl.fl4_tos = tos;
1890         rth->fl.mark    = skb->mark;
1891         rth->fl.fl4_src = saddr;
1892         rth->rt_src     = saddr;
1893 #ifdef CONFIG_NET_CLS_ROUTE
1894         rth->u.dst.tclassid = itag;
1895 #endif
1896         rth->rt_iif     =
1897         rth->fl.iif     = dev->ifindex;
1898         rth->u.dst.dev  = init_net.loopback_dev;
1899         dev_hold(rth->u.dst.dev);
1900         rth->idev       = in_dev_get(rth->u.dst.dev);
1901         rth->fl.oif     = 0;
1902         rth->rt_gateway = daddr;
1903         rth->rt_spec_dst= spec_dst;
1904         rth->rt_genid   = rt_genid(dev_net(dev));
1905         rth->rt_flags   = RTCF_MULTICAST;
1906         rth->rt_type    = RTN_MULTICAST;
1907         if (our) {
1908                 rth->u.dst.input= ip_local_deliver;
1909                 rth->rt_flags |= RTCF_LOCAL;
1910         }
1911
1912 #ifdef CONFIG_IP_MROUTE
1913         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1914                 rth->u.dst.input = ip_mr_input;
1915 #endif
1916         RT_CACHE_STAT_INC(in_slow_mc);
1917
1918         hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1919         return rt_intern_hash(hash, rth, NULL, skb, dev->ifindex);
1920
1921 e_nobufs:
1922         return -ENOBUFS;
1923 e_inval:
1924         return -EINVAL;
1925 e_err:
1926         return err;
1927 }
1928
1929
1930 static void ip_handle_martian_source(struct net_device *dev,
1931                                      struct in_device *in_dev,
1932                                      struct sk_buff *skb,
1933                                      __be32 daddr,
1934                                      __be32 saddr)
1935 {
1936         RT_CACHE_STAT_INC(in_martian_src);
1937 #ifdef CONFIG_IP_ROUTE_VERBOSE
1938         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1939                 /*
1940                  *      RFC1812 recommendation, if source is martian,
1941                  *      the only hint is MAC header.
1942                  */
1943                 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1944                         &daddr, &saddr, dev->name);
1945                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1946                         int i;
1947                         const unsigned char *p = skb_mac_header(skb);
1948                         printk(KERN_WARNING "ll header: ");
1949                         for (i = 0; i < dev->hard_header_len; i++, p++) {
1950                                 printk("%02x", *p);
1951                                 if (i < (dev->hard_header_len - 1))
1952                                         printk(":");
1953                         }
1954                         printk("\n");
1955                 }
1956         }
1957 #endif
1958 }
1959
1960 static int __mkroute_input(struct sk_buff *skb,
1961                            struct fib_result *res,
1962                            struct in_device *in_dev,
1963                            __be32 daddr, __be32 saddr, u32 tos,
1964                            struct rtable **result)
1965 {
1966
1967         struct rtable *rth;
1968         int err;
1969         struct in_device *out_dev;
1970         unsigned flags = 0;
1971         __be32 spec_dst;
1972         u32 itag;
1973
1974         /* get a working reference to the output device */
1975         out_dev = in_dev_get(FIB_RES_DEV(*res));
1976         if (out_dev == NULL) {
1977                 if (net_ratelimit())
1978                         printk(KERN_CRIT "Bug in ip_route_input" \
1979                                "_slow(). Please, report\n");
1980                 return -EINVAL;
1981         }
1982
1983
1984         err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1985                                   in_dev->dev, &spec_dst, &itag, skb->mark);
1986         if (err < 0) {
1987                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1988                                          saddr);
1989
1990                 goto cleanup;
1991         }
1992
1993         if (err)
1994                 flags |= RTCF_DIRECTSRC;
1995
1996         if (out_dev == in_dev && err &&
1997             (IN_DEV_SHARED_MEDIA(out_dev) ||
1998              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1999                 flags |= RTCF_DOREDIRECT;
2000
2001         if (skb->protocol != htons(ETH_P_IP)) {
2002                 /* Not IP (i.e. ARP). Do not create route, if it is
2003                  * invalid for proxy arp. DNAT routes are always valid.
2004                  *
2005                  * Proxy arp feature have been extended to allow, ARP
2006                  * replies back to the same interface, to support
2007                  * Private VLAN switch technologies. See arp.c.
2008                  */
2009                 if (out_dev == in_dev &&
2010                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2011                         err = -EINVAL;
2012                         goto cleanup;
2013                 }
2014         }
2015
2016
2017         rth = dst_alloc(&ipv4_dst_ops);
2018         if (!rth) {
2019                 err = -ENOBUFS;
2020                 goto cleanup;
2021         }
2022
2023         atomic_set(&rth->u.dst.__refcnt, 1);
2024         rth->u.dst.flags= DST_HOST;
2025         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2026                 rth->u.dst.flags |= DST_NOPOLICY;
2027         if (IN_DEV_CONF_GET(out_dev, NOXFRM))
2028                 rth->u.dst.flags |= DST_NOXFRM;
2029         rth->fl.fl4_dst = daddr;
2030         rth->rt_dst     = daddr;
2031         rth->fl.fl4_tos = tos;
2032         rth->fl.mark    = skb->mark;
2033         rth->fl.fl4_src = saddr;
2034         rth->rt_src     = saddr;
2035         rth->rt_gateway = daddr;
2036         rth->rt_iif     =
2037                 rth->fl.iif     = in_dev->dev->ifindex;
2038         rth->u.dst.dev  = (out_dev)->dev;
2039         dev_hold(rth->u.dst.dev);
2040         rth->idev       = in_dev_get(rth->u.dst.dev);
2041         rth->fl.oif     = 0;
2042         rth->rt_spec_dst= spec_dst;
2043
2044         rth->u.dst.obsolete = -1;
2045         rth->u.dst.input = ip_forward;
2046         rth->u.dst.output = ip_output;
2047         rth->rt_genid = rt_genid(dev_net(rth->u.dst.dev));
2048
2049         rt_set_nexthop(rth, res, itag);
2050
2051         rth->rt_flags = flags;
2052
2053         *result = rth;
2054         err = 0;
2055  cleanup:
2056         /* release the working reference to the output device */
2057         in_dev_put(out_dev);
2058         return err;
2059 }
2060
2061 static int ip_mkroute_input(struct sk_buff *skb,
2062                             struct fib_result *res,
2063                             const struct flowi *fl,
2064                             struct in_device *in_dev,
2065                             __be32 daddr, __be32 saddr, u32 tos)
2066 {
2067         struct rtable* rth = NULL;
2068         int err;
2069         unsigned hash;
2070
2071 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2072         if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
2073                 fib_select_multipath(fl, res);
2074 #endif
2075
2076         /* create a routing cache entry */
2077         err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2078         if (err)
2079                 return err;
2080
2081         /* put it into the cache */
2082         hash = rt_hash(daddr, saddr, fl->iif,
2083                        rt_genid(dev_net(rth->u.dst.dev)));
2084         return rt_intern_hash(hash, rth, NULL, skb, fl->iif);
2085 }
2086
2087 /*
2088  *      NOTE. We drop all the packets that has local source
2089  *      addresses, because every properly looped back packet
2090  *      must have correct destination already attached by output routine.
2091  *
2092  *      Such approach solves two big problems:
2093  *      1. Not simplex devices are handled properly.
2094  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2095  */
2096
2097 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2098                                u8 tos, struct net_device *dev)
2099 {
2100         struct fib_result res;
2101         struct in_device *in_dev = __in_dev_get_rcu(dev);
2102         struct flowi fl = { .nl_u = { .ip4_u =
2103                                       { .daddr = daddr,
2104                                         .saddr = saddr,
2105                                         .tos = tos,
2106                                         .scope = RT_SCOPE_UNIVERSE,
2107                                       } },
2108                             .mark = skb->mark,
2109                             .iif = dev->ifindex };
2110         unsigned        flags = 0;
2111         u32             itag = 0;
2112         struct rtable * rth;
2113         unsigned        hash;
2114         __be32          spec_dst;
2115         int             err = -EINVAL;
2116         int             free_res = 0;
2117         struct net    * net = dev_net(dev);
2118
2119         /* IP on this device is disabled. */
2120
2121         if (!in_dev)
2122                 goto out;
2123
2124         /* Check for the most weird martians, which can be not detected
2125            by fib_lookup.
2126          */
2127
2128         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2129             ipv4_is_loopback(saddr))
2130                 goto martian_source;
2131
2132         if (daddr == htonl(0xFFFFFFFF) || (saddr == 0 && daddr == 0))
2133                 goto brd_input;
2134
2135         /* Accept zero addresses only to limited broadcast;
2136          * I even do not know to fix it or not. Waiting for complains :-)
2137          */
2138         if (ipv4_is_zeronet(saddr))
2139                 goto martian_source;
2140
2141         if (ipv4_is_lbcast(daddr) || ipv4_is_zeronet(daddr) ||
2142             ipv4_is_loopback(daddr))
2143                 goto martian_destination;
2144
2145         /*
2146          *      Now we are ready to route packet.
2147          */
2148         if ((err = fib_lookup(net, &fl, &res)) != 0) {
2149                 if (!IN_DEV_FORWARD(in_dev))
2150                         goto e_hostunreach;
2151                 goto no_route;
2152         }
2153         free_res = 1;
2154
2155         RT_CACHE_STAT_INC(in_slow_tot);
2156
2157         if (res.type == RTN_BROADCAST)
2158                 goto brd_input;
2159
2160         if (res.type == RTN_LOCAL) {
2161                 err = fib_validate_source(saddr, daddr, tos,
2162                                              net->loopback_dev->ifindex,
2163                                              dev, &spec_dst, &itag, skb->mark);
2164                 if (err < 0)
2165                         goto martian_source_keep_err;
2166                 if (err)
2167                         flags |= RTCF_DIRECTSRC;
2168                 spec_dst = daddr;
2169                 goto local_input;
2170         }
2171
2172         if (!IN_DEV_FORWARD(in_dev))
2173                 goto e_hostunreach;
2174         if (res.type != RTN_UNICAST)
2175                 goto martian_destination;
2176
2177         err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
2178 done:
2179         if (free_res)
2180                 fib_res_put(&res);
2181 out:    return err;
2182
2183 brd_input:
2184         if (skb->protocol != htons(ETH_P_IP))
2185                 goto e_inval;
2186
2187         if (ipv4_is_zeronet(saddr))
2188                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2189         else {
2190                 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2191                                           &itag, skb->mark);
2192                 if (err < 0)
2193                         goto martian_source_keep_err;
2194                 if (err)
2195                         flags |= RTCF_DIRECTSRC;
2196         }
2197         flags |= RTCF_BROADCAST;
2198         res.type = RTN_BROADCAST;
2199         RT_CACHE_STAT_INC(in_brd);
2200
2201 local_input:
2202         rth = dst_alloc(&ipv4_dst_ops);
2203         if (!rth)
2204                 goto e_nobufs;
2205
2206         rth->u.dst.output= ip_rt_bug;
2207         rth->u.dst.obsolete = -1;
2208         rth->rt_genid = rt_genid(net);
2209
2210         atomic_set(&rth->u.dst.__refcnt, 1);
2211         rth->u.dst.flags= DST_HOST;
2212         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2213                 rth->u.dst.flags |= DST_NOPOLICY;
2214         rth->fl.fl4_dst = daddr;
2215         rth->rt_dst     = daddr;
2216         rth->fl.fl4_tos = tos;
2217         rth->fl.mark    = skb->mark;
2218         rth->fl.fl4_src = saddr;
2219         rth->rt_src     = saddr;
2220 #ifdef CONFIG_NET_CLS_ROUTE
2221         rth->u.dst.tclassid = itag;
2222 #endif
2223         rth->rt_iif     =
2224         rth->fl.iif     = dev->ifindex;
2225         rth->u.dst.dev  = net->loopback_dev;
2226         dev_hold(rth->u.dst.dev);
2227         rth->idev       = in_dev_get(rth->u.dst.dev);
2228         rth->rt_gateway = daddr;
2229         rth->rt_spec_dst= spec_dst;
2230         rth->u.dst.input= ip_local_deliver;
2231         rth->rt_flags   = flags|RTCF_LOCAL;
2232         if (res.type == RTN_UNREACHABLE) {
2233                 rth->u.dst.input= ip_error;
2234                 rth->u.dst.error= -err;
2235                 rth->rt_flags   &= ~RTCF_LOCAL;
2236         }
2237         rth->rt_type    = res.type;
2238         hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));
2239         err = rt_intern_hash(hash, rth, NULL, skb, fl.iif);
2240         goto done;
2241
2242 no_route:
2243         RT_CACHE_STAT_INC(in_no_route);
2244         spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2245         res.type = RTN_UNREACHABLE;
2246         if (err == -ESRCH)
2247                 err = -ENETUNREACH;
2248         goto local_input;
2249
2250         /*
2251          *      Do not cache martian addresses: they should be logged (RFC1812)
2252          */
2253 martian_destination:
2254         RT_CACHE_STAT_INC(in_martian_dst);
2255 #ifdef CONFIG_IP_ROUTE_VERBOSE
2256         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2257                 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2258                         &daddr, &saddr, dev->name);
2259 #endif
2260
2261 e_hostunreach:
2262         err = -EHOSTUNREACH;
2263         goto done;
2264
2265 e_inval:
2266         err = -EINVAL;
2267         goto done;
2268
2269 e_nobufs:
2270         err = -ENOBUFS;
2271         goto done;
2272
2273 martian_source:
2274         err = -EINVAL;
2275 martian_source_keep_err:
2276         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2277         goto done;
2278 }
2279
2280 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2281                            u8 tos, struct net_device *dev, bool noref)
2282 {
2283         struct rtable * rth;
2284         unsigned        hash;
2285         int iif = dev->ifindex;
2286         struct net *net;
2287         int res;
2288
2289         net = dev_net(dev);
2290
2291         rcu_read_lock();
2292
2293         if (!rt_caching(net))
2294                 goto skip_cache;
2295
2296         tos &= IPTOS_RT_MASK;
2297         hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2298
2299         for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2300              rth = rcu_dereference(rth->u.dst.rt_next)) {
2301                 if ((((__force u32)rth->fl.fl4_dst ^ (__force u32)daddr) |
2302                      ((__force u32)rth->fl.fl4_src ^ (__force u32)saddr) |
2303                      (rth->fl.iif ^ iif) |
2304                      rth->fl.oif |
2305                      (rth->fl.fl4_tos ^ tos)) == 0 &&
2306                     rth->fl.mark == skb->mark &&
2307                     net_eq(dev_net(rth->u.dst.dev), net) &&
2308                     !rt_is_expired(rth)) {
2309                         if (noref) {
2310                                 dst_use_noref(&rth->u.dst, jiffies);
2311                                 skb_dst_set_noref(skb, &rth->u.dst);
2312                         } else {
2313                                 dst_use(&rth->u.dst, jiffies);
2314                                 skb_dst_set(skb, &rth->u.dst);
2315                         }
2316                         RT_CACHE_STAT_INC(in_hit);
2317                         rcu_read_unlock();
2318                         return 0;
2319                 }
2320                 RT_CACHE_STAT_INC(in_hlist_search);
2321         }
2322
2323 skip_cache:
2324         /* Multicast recognition logic is moved from route cache to here.
2325            The problem was that too many Ethernet cards have broken/missing
2326            hardware multicast filters :-( As result the host on multicasting
2327            network acquires a lot of useless route cache entries, sort of
2328            SDR messages from all the world. Now we try to get rid of them.
2329            Really, provided software IP multicast filter is organized
2330            reasonably (at least, hashed), it does not result in a slowdown
2331            comparing with route cache reject entries.
2332            Note, that multicast routers are not affected, because
2333            route cache entry is created eventually.
2334          */
2335         if (ipv4_is_multicast(daddr)) {
2336                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2337
2338                 if (in_dev) {
2339                         int our = ip_check_mc(in_dev, daddr, saddr,
2340                                               ip_hdr(skb)->protocol);
2341                         if (our
2342 #ifdef CONFIG_IP_MROUTE
2343                                 ||
2344                             (!ipv4_is_local_multicast(daddr) &&
2345                              IN_DEV_MFORWARD(in_dev))
2346 #endif
2347                            ) {
2348                                 int res = ip_route_input_mc(skb, daddr, saddr,
2349                                                             tos, dev, our);
2350                                 rcu_read_unlock();
2351                                 return res;
2352                         }
2353                 }
2354                 rcu_read_unlock();
2355                 return -EINVAL;
2356         }
2357         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2358         rcu_read_unlock();
2359         return res;
2360 }
2361 EXPORT_SYMBOL(ip_route_input_common);
2362
2363 static int __mkroute_output(struct rtable **result,
2364                             struct fib_result *res,
2365                             const struct flowi *fl,
2366                             const struct flowi *oldflp,
2367                             struct net_device *dev_out,
2368                             unsigned flags)
2369 {
2370         struct rtable *rth;
2371         struct in_device *in_dev;
2372         u32 tos = RT_FL_TOS(oldflp);
2373         int err = 0;
2374
2375         if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags&IFF_LOOPBACK))
2376                 return -EINVAL;
2377
2378         if (fl->fl4_dst == htonl(0xFFFFFFFF))
2379                 res->type = RTN_BROADCAST;
2380         else if (ipv4_is_multicast(fl->fl4_dst))
2381                 res->type = RTN_MULTICAST;
2382         else if (ipv4_is_lbcast(fl->fl4_dst) || ipv4_is_zeronet(fl->fl4_dst))
2383                 return -EINVAL;
2384
2385         if (dev_out->flags & IFF_LOOPBACK)
2386                 flags |= RTCF_LOCAL;
2387
2388         /* get work reference to inet device */
2389         in_dev = in_dev_get(dev_out);
2390         if (!in_dev)
2391                 return -EINVAL;
2392
2393         if (res->type == RTN_BROADCAST) {
2394                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2395                 if (res->fi) {
2396                         fib_info_put(res->fi);
2397                         res->fi = NULL;
2398                 }
2399         } else if (res->type == RTN_MULTICAST) {
2400                 flags |= RTCF_MULTICAST|RTCF_LOCAL;
2401                 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2402                                  oldflp->proto))
2403                         flags &= ~RTCF_LOCAL;
2404                 /* If multicast route do not exist use
2405                    default one, but do not gateway in this case.
2406                    Yes, it is hack.
2407                  */
2408                 if (res->fi && res->prefixlen < 4) {
2409                         fib_info_put(res->fi);
2410                         res->fi = NULL;
2411                 }
2412         }
2413
2414
2415         rth = dst_alloc(&ipv4_dst_ops);
2416         if (!rth) {
2417                 err = -ENOBUFS;
2418                 goto cleanup;
2419         }
2420
2421         atomic_set(&rth->u.dst.__refcnt, 1);
2422         rth->u.dst.flags= DST_HOST;
2423         if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2424                 rth->u.dst.flags |= DST_NOXFRM;
2425         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2426                 rth->u.dst.flags |= DST_NOPOLICY;
2427
2428         rth->fl.fl4_dst = oldflp->fl4_dst;
2429         rth->fl.fl4_tos = tos;
2430         rth->fl.fl4_src = oldflp->fl4_src;
2431         rth->fl.oif     = oldflp->oif;
2432         rth->fl.mark    = oldflp->mark;
2433         rth->rt_dst     = fl->fl4_dst;
2434         rth->rt_src     = fl->fl4_src;
2435         rth->rt_iif     = oldflp->oif ? : dev_out->ifindex;
2436         /* get references to the devices that are to be hold by the routing
2437            cache entry */
2438         rth->u.dst.dev  = dev_out;
2439         dev_hold(dev_out);
2440         rth->idev       = in_dev_get(dev_out);
2441         rth->rt_gateway = fl->fl4_dst;
2442         rth->rt_spec_dst= fl->fl4_src;
2443
2444         rth->u.dst.output=ip_output;
2445         rth->u.dst.obsolete = -1;
2446         rth->rt_genid = rt_genid(dev_net(dev_out));
2447
2448         RT_CACHE_STAT_INC(out_slow_tot);
2449
2450         if (flags & RTCF_LOCAL) {
2451                 rth->u.dst.input = ip_local_deliver;
2452                 rth->rt_spec_dst = fl->fl4_dst;
2453         }
2454         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2455                 rth->rt_spec_dst = fl->fl4_src;
2456                 if (flags & RTCF_LOCAL &&
2457                     !(dev_out->flags & IFF_LOOPBACK)) {
2458                         rth->u.dst.output = ip_mc_output;
2459                         RT_CACHE_STAT_INC(out_slow_mc);
2460                 }
2461 #ifdef CONFIG_IP_MROUTE
2462                 if (res->type == RTN_MULTICAST) {
2463                         if (IN_DEV_MFORWARD(in_dev) &&
2464                             !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2465                                 rth->u.dst.input = ip_mr_input;
2466                                 rth->u.dst.output = ip_mc_output;
2467                         }
2468                 }
2469 #endif
2470         }
2471
2472         rt_set_nexthop(rth, res, 0);
2473
2474         rth->rt_flags = flags;
2475
2476         *result = rth;
2477  cleanup:
2478         /* release work reference to inet device */
2479         in_dev_put(in_dev);
2480
2481         return err;
2482 }
2483
2484 static int ip_mkroute_output(struct rtable **rp,
2485                              struct fib_result *res,
2486                              const struct flowi *fl,
2487                              const struct flowi *oldflp,
2488                              struct net_device *dev_out,
2489                              unsigned flags)
2490 {
2491         struct rtable *rth = NULL;
2492         int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2493         unsigned hash;
2494         if (err == 0) {
2495                 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif,
2496                                rt_genid(dev_net(dev_out)));
2497                 err = rt_intern_hash(hash, rth, rp, NULL, oldflp->oif);
2498         }
2499
2500         return err;
2501 }
2502
2503 /*
2504  * Major route resolver routine.
2505  */
2506
2507 static int ip_route_output_slow(struct net *net, struct rtable **rp,
2508                                 const struct flowi *oldflp)
2509 {
2510         u32 tos = RT_FL_TOS(oldflp);
2511         struct flowi fl = { .nl_u = { .ip4_u =
2512                                       { .daddr = oldflp->fl4_dst,
2513                                         .saddr = oldflp->fl4_src,
2514                                         .tos = tos & IPTOS_RT_MASK,
2515                                         .scope = ((tos & RTO_ONLINK) ?
2516                                                   RT_SCOPE_LINK :
2517                                                   RT_SCOPE_UNIVERSE),
2518                                       } },
2519                             .mark = oldflp->mark,
2520                             .iif = net->loopback_dev->ifindex,
2521                             .oif = oldflp->oif };
2522         struct fib_result res;
2523         unsigned flags = 0;
2524         struct net_device *dev_out = NULL;
2525         int free_res = 0;
2526         int err;
2527
2528
2529         res.fi          = NULL;
2530 #ifdef CONFIG_IP_MULTIPLE_TABLES
2531         res.r           = NULL;
2532 #endif
2533
2534         if (oldflp->fl4_src) {
2535                 err = -EINVAL;
2536                 if (ipv4_is_multicast(oldflp->fl4_src) ||
2537                     ipv4_is_lbcast(oldflp->fl4_src) ||
2538                     ipv4_is_zeronet(oldflp->fl4_src))
2539                         goto out;
2540
2541                 /* I removed check for oif == dev_out->oif here.
2542                    It was wrong for two reasons:
2543                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2544                       is assigned to multiple interfaces.
2545                    2. Moreover, we are allowed to send packets with saddr
2546                       of another iface. --ANK
2547                  */
2548
2549                 if (oldflp->oif == 0 &&
2550                     (ipv4_is_multicast(oldflp->fl4_dst) ||
2551                      oldflp->fl4_dst == htonl(0xFFFFFFFF))) {
2552                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2553                         dev_out = ip_dev_find(net, oldflp->fl4_src);
2554                         if (dev_out == NULL)
2555                                 goto out;
2556
2557                         /* Special hack: user can direct multicasts
2558                            and limited broadcast via necessary interface
2559                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2560                            This hack is not just for fun, it allows
2561                            vic,vat and friends to work.
2562                            They bind socket to loopback, set ttl to zero
2563                            and expect that it will work.
2564                            From the viewpoint of routing cache they are broken,
2565                            because we are not allowed to build multicast path
2566                            with loopback source addr (look, routing cache
2567                            cannot know, that ttl is zero, so that packet
2568                            will not leave this host and route is valid).
2569                            Luckily, this hack is good workaround.
2570                          */
2571
2572                         fl.oif = dev_out->ifindex;
2573                         goto make_route;
2574                 }
2575
2576                 if (!(oldflp->flags & FLOWI_FLAG_ANYSRC)) {
2577                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2578                         dev_out = ip_dev_find(net, oldflp->fl4_src);
2579                         if (dev_out == NULL)
2580                                 goto out;
2581                         dev_put(dev_out);
2582                         dev_out = NULL;
2583                 }
2584         }
2585
2586
2587         if (oldflp->oif) {
2588                 dev_out = dev_get_by_index(net, oldflp->oif);
2589                 err = -ENODEV;
2590                 if (dev_out == NULL)
2591                         goto out;
2592
2593                 /* RACE: Check return value of inet_select_addr instead. */
2594                 if (__in_dev_get_rtnl(dev_out) == NULL) {
2595                         dev_put(dev_out);
2596                         goto out;       /* Wrong error code */
2597                 }
2598
2599                 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2600                     oldflp->fl4_dst == htonl(0xFFFFFFFF)) {
2601                         if (!fl.fl4_src)
2602                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2603                                                               RT_SCOPE_LINK);
2604                         goto make_route;
2605                 }
2606                 if (!fl.fl4_src) {
2607                         if (ipv4_is_multicast(oldflp->fl4_dst))
2608                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2609                                                               fl.fl4_scope);
2610                         else if (!oldflp->fl4_dst)
2611                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2612                                                               RT_SCOPE_HOST);
2613                 }
2614         }
2615
2616         if (!fl.fl4_dst) {
2617                 fl.fl4_dst = fl.fl4_src;
2618                 if (!fl.fl4_dst)
2619                         fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2620                 if (dev_out)
2621                         dev_put(dev_out);
2622                 dev_out = net->loopback_dev;
2623                 dev_hold(dev_out);
2624                 fl.oif = net->loopback_dev->ifindex;
2625                 res.type = RTN_LOCAL;
2626                 flags |= RTCF_LOCAL;
2627                 goto make_route;
2628         }
2629
2630         if (fib_lookup(net, &fl, &res)) {
2631                 res.fi = NULL;
2632                 if (oldflp->oif) {
2633                         /* Apparently, routing tables are wrong. Assume,
2634                            that the destination is on link.
2635
2636                            WHY? DW.
2637                            Because we are allowed to send to iface
2638                            even if it has NO routes and NO assigned
2639                            addresses. When oif is specified, routing
2640                            tables are looked up with only one purpose:
2641                            to catch if destination is gatewayed, rather than
2642                            direct. Moreover, if MSG_DONTROUTE is set,
2643                            we send packet, ignoring both routing tables
2644                            and ifaddr state. --ANK
2645
2646
2647                            We could make it even if oif is unknown,
2648                            likely IPv6, but we do not.
2649                          */
2650
2651                         if (fl.fl4_src == 0)
2652                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2653                                                               RT_SCOPE_LINK);
2654                         res.type = RTN_UNICAST;
2655                         goto make_route;
2656                 }
2657                 if (dev_out)
2658                         dev_put(dev_out);
2659                 err = -ENETUNREACH;
2660                 goto out;
2661         }
2662         free_res = 1;
2663
2664         if (res.type == RTN_LOCAL) {
2665                 if (!fl.fl4_src)
2666                         fl.fl4_src = fl.fl4_dst;
2667                 if (dev_out)
2668                         dev_put(dev_out);
2669                 dev_out = net->loopback_dev;
2670                 dev_hold(dev_out);
2671                 fl.oif = dev_out->ifindex;
2672                 if (res.fi)
2673                         fib_info_put(res.fi);
2674                 res.fi = NULL;
2675                 flags |= RTCF_LOCAL;
2676                 goto make_route;
2677         }
2678
2679 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2680         if (res.fi->fib_nhs > 1 && fl.oif == 0)
2681                 fib_select_multipath(&fl, &res);
2682         else
2683 #endif
2684         if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2685                 fib_select_default(net, &fl, &res);
2686
2687         if (!fl.fl4_src)
2688                 fl.fl4_src = FIB_RES_PREFSRC(res);
2689
2690         if (dev_out)
2691                 dev_put(dev_out);
2692         dev_out = FIB_RES_DEV(res);
2693         dev_hold(dev_out);
2694         fl.oif = dev_out->ifindex;
2695
2696
2697 make_route:
2698         err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2699
2700
2701         if (free_res)
2702                 fib_res_put(&res);
2703         if (dev_out)
2704                 dev_put(dev_out);
2705 out:    return err;
2706 }
2707
2708 int __ip_route_output_key(struct net *net, struct rtable **rp,
2709                           const struct flowi *flp)
2710 {
2711         unsigned hash;
2712         struct rtable *rth;
2713
2714         if (!rt_caching(net))
2715                 goto slow_output;
2716
2717         hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, rt_genid(net));
2718
2719         rcu_read_lock_bh();
2720         for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2721                 rth = rcu_dereference_bh(rth->u.dst.rt_next)) {
2722                 if (rth->fl.fl4_dst == flp->fl4_dst &&
2723                     rth->fl.fl4_src == flp->fl4_src &&
2724                     rth->fl.iif == 0 &&
2725                     rth->fl.oif == flp->oif &&
2726                     rth->fl.mark == flp->mark &&
2727                     !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2728                             (IPTOS_RT_MASK | RTO_ONLINK)) &&
2729                     net_eq(dev_net(rth->u.dst.dev), net) &&
2730                     !rt_is_expired(rth)) {
2731                         dst_use(&rth->u.dst, jiffies);
2732                         RT_CACHE_STAT_INC(out_hit);
2733                         rcu_read_unlock_bh();
2734                         *rp = rth;
2735                         return 0;
2736                 }
2737                 RT_CACHE_STAT_INC(out_hlist_search);
2738         }
2739         rcu_read_unlock_bh();
2740
2741 slow_output:
2742         return ip_route_output_slow(net, rp, flp);
2743 }
2744
2745 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2746
2747 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2748 {
2749 }
2750
2751 static struct dst_ops ipv4_dst_blackhole_ops = {
2752         .family                 =       AF_INET,
2753         .protocol               =       cpu_to_be16(ETH_P_IP),
2754         .destroy                =       ipv4_dst_destroy,
2755         .check                  =       ipv4_dst_check,
2756         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2757         .entries                =       ATOMIC_INIT(0),
2758 };
2759
2760
2761 static int ipv4_dst_blackhole(struct net *net, struct rtable **rp, struct flowi *flp)
2762 {
2763         struct rtable *ort = *rp;
2764         struct rtable *rt = (struct rtable *)
2765                 dst_alloc(&ipv4_dst_blackhole_ops);
2766
2767         if (rt) {
2768                 struct dst_entry *new = &rt->u.dst;
2769
2770                 atomic_set(&new->__refcnt, 1);
2771                 new->__use = 1;
2772                 new->input = dst_discard;
2773                 new->output = dst_discard;
2774                 memcpy(new->metrics, ort->u.dst.metrics, RTAX_MAX*sizeof(u32));
2775
2776                 new->dev = ort->u.dst.dev;
2777                 if (new->dev)
2778                         dev_hold(new->dev);
2779
2780                 rt->fl = ort->fl;
2781
2782                 rt->idev = ort->idev;
2783                 if (rt->idev)
2784                         in_dev_hold(rt->idev);
2785                 rt->rt_genid = rt_genid(net);
2786                 rt->rt_flags = ort->rt_flags;
2787                 rt->rt_type = ort->rt_type;
2788                 rt->rt_dst = ort->rt_dst;
2789                 rt->rt_src = ort->rt_src;
2790                 rt->rt_iif = ort->rt_iif;
2791                 rt->rt_gateway = ort->rt_gateway;
2792                 rt->rt_spec_dst = ort->rt_spec_dst;
2793                 rt->peer = ort->peer;
2794                 if (rt->peer)
2795                         atomic_inc(&rt->peer->refcnt);
2796
2797                 dst_free(new);
2798         }
2799
2800         dst_release(&(*rp)->u.dst);
2801         *rp = rt;
2802         return (rt ? 0 : -ENOMEM);
2803 }
2804
2805 int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2806                          struct sock *sk, int flags)
2807 {
2808         int err;
2809
2810         if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2811                 return err;
2812
2813         if (flp->proto) {
2814                 if (!flp->fl4_src)
2815                         flp->fl4_src = (*rp)->rt_src;
2816                 if (!flp->fl4_dst)
2817                         flp->fl4_dst = (*rp)->rt_dst;
2818                 err = __xfrm_lookup(net, (struct dst_entry **)rp, flp, sk,
2819                                     flags ? XFRM_LOOKUP_WAIT : 0);
2820                 if (err == -EREMOTE)
2821                         err = ipv4_dst_blackhole(net, rp, flp);
2822
2823                 return err;
2824         }
2825
2826         return 0;
2827 }
2828
2829 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2830
2831 int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2832 {
2833         return ip_route_output_flow(net, rp, flp, NULL, 0);
2834 }
2835
2836 static int rt_fill_info(struct net *net,
2837                         struct sk_buff *skb, u32 pid, u32 seq, int event,
2838                         int nowait, unsigned int flags)
2839 {
2840         struct rtable *rt = skb_rtable(skb);
2841         struct rtmsg *r;
2842         struct nlmsghdr *nlh;
2843         long expires;
2844         u32 id = 0, ts = 0, tsage = 0, error;
2845
2846         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2847         if (nlh == NULL)
2848                 return -EMSGSIZE;
2849
2850         r = nlmsg_data(nlh);
2851         r->rtm_family    = AF_INET;
2852         r->rtm_dst_len  = 32;
2853         r->rtm_src_len  = 0;
2854         r->rtm_tos      = rt->fl.fl4_tos;
2855         r->rtm_table    = RT_TABLE_MAIN;
2856         NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2857         r->rtm_type     = rt->rt_type;
2858         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2859         r->rtm_protocol = RTPROT_UNSPEC;
2860         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2861         if (rt->rt_flags & RTCF_NOTIFY)
2862                 r->rtm_flags |= RTM_F_NOTIFY;
2863
2864         NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2865
2866         if (rt->fl.fl4_src) {
2867                 r->rtm_src_len = 32;
2868                 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2869         }
2870         if (rt->u.dst.dev)
2871                 NLA_PUT_U32(skb, RTA_OIF, rt->u.dst.dev->ifindex);
2872 #ifdef CONFIG_NET_CLS_ROUTE
2873         if (rt->u.dst.tclassid)
2874                 NLA_PUT_U32(skb, RTA_FLOW, rt->u.dst.tclassid);
2875 #endif
2876         if (rt->fl.iif)
2877                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2878         else if (rt->rt_src != rt->fl.fl4_src)
2879                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2880
2881         if (rt->rt_dst != rt->rt_gateway)
2882                 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2883
2884         if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
2885                 goto nla_put_failure;
2886
2887         error = rt->u.dst.error;
2888         expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
2889         if (rt->peer) {
2890                 id = atomic_read(&rt->peer->ip_id_count) & 0xffff;
2891                 if (rt->peer->tcp_ts_stamp) {
2892                         ts = rt->peer->tcp_ts;
2893                         tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2894                 }
2895         }
2896
2897         if (rt->fl.iif) {
2898 #ifdef CONFIG_IP_MROUTE
2899                 __be32 dst = rt->rt_dst;
2900
2901                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2902                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2903                         int err = ipmr_get_route(net, skb, r, nowait);
2904                         if (err <= 0) {
2905                                 if (!nowait) {
2906                                         if (err == 0)
2907                                                 return 0;
2908                                         goto nla_put_failure;
2909                                 } else {
2910                                         if (err == -EMSGSIZE)
2911                                                 goto nla_put_failure;
2912                                         error = err;
2913                                 }
2914                         }
2915                 } else
2916 #endif
2917                         NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2918         }
2919
2920         if (rtnl_put_cacheinfo(skb, &rt->u.dst, id, ts, tsage,
2921                                expires, error) < 0)
2922                 goto nla_put_failure;
2923
2924         return nlmsg_end(skb, nlh);
2925
2926 nla_put_failure:
2927         nlmsg_cancel(skb, nlh);
2928         return -EMSGSIZE;
2929 }
2930
2931 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2932 {
2933         struct net *net = sock_net(in_skb->sk);
2934         struct rtmsg *rtm;
2935         struct nlattr *tb[RTA_MAX+1];
2936         struct rtable *rt = NULL;
2937         __be32 dst = 0;
2938         __be32 src = 0;
2939         u32 iif;
2940         int err;
2941         struct sk_buff *skb;
2942
2943         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2944         if (err < 0)
2945                 goto errout;
2946
2947         rtm = nlmsg_data(nlh);
2948
2949         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2950         if (skb == NULL) {
2951                 err = -ENOBUFS;
2952                 goto errout;
2953         }
2954
2955         /* Reserve room for dummy headers, this skb can pass
2956            through good chunk of routing engine.
2957          */
2958         skb_reset_mac_header(skb);
2959         skb_reset_network_header(skb);
2960
2961         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2962         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2963         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2964
2965         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2966         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2967         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2968
2969         if (iif) {
2970                 struct net_device *dev;
2971
2972                 dev = __dev_get_by_index(net, iif);
2973                 if (dev == NULL) {
2974                         err = -ENODEV;
2975                         goto errout_free;
2976                 }
2977
2978                 skb->protocol   = htons(ETH_P_IP);
2979                 skb->dev        = dev;
2980                 local_bh_disable();
2981                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2982                 local_bh_enable();
2983
2984                 rt = skb_rtable(skb);
2985                 if (err == 0 && rt->u.dst.error)
2986                         err = -rt->u.dst.error;
2987         } else {
2988                 struct flowi fl = {
2989                         .nl_u = {
2990                                 .ip4_u = {
2991                                         .daddr = dst,
2992                                         .saddr = src,
2993                                         .tos = rtm->rtm_tos,
2994                                 },
2995                         },
2996                         .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2997                 };
2998                 err = ip_route_output_key(net, &rt, &fl);
2999         }
3000
3001         if (err)
3002                 goto errout_free;
3003
3004         skb_dst_set(skb, &rt->u.dst);
3005         if (rtm->rtm_flags & RTM_F_NOTIFY)
3006                 rt->rt_flags |= RTCF_NOTIFY;
3007
3008         err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3009                            RTM_NEWROUTE, 0, 0);
3010         if (err <= 0)
3011                 goto errout_free;
3012
3013         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3014 errout:
3015         return err;
3016
3017 errout_free:
3018         kfree_skb(skb);
3019         goto errout;
3020 }
3021
3022 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
3023 {
3024         struct rtable *rt;
3025         int h, s_h;
3026         int idx, s_idx;
3027         struct net *net;
3028
3029         net = sock_net(skb->sk);
3030
3031         s_h = cb->args[0];
3032         if (s_h < 0)
3033                 s_h = 0;
3034         s_idx = idx = cb->args[1];
3035         for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3036                 if (!rt_hash_table[h].chain)
3037                         continue;
3038                 rcu_read_lock_bh();
3039                 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3040                      rt = rcu_dereference_bh(rt->u.dst.rt_next), idx++) {
3041                         if (!net_eq(dev_net(rt->u.dst.dev), net) || idx < s_idx)
3042                                 continue;
3043                         if (rt_is_expired(rt))
3044                                 continue;
3045                         skb_dst_set_noref(skb, &rt->u.dst);
3046                         if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3047                                          cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3048                                          1, NLM_F_MULTI) <= 0) {
3049                                 skb_dst_drop(skb);
3050                                 rcu_read_unlock_bh();
3051                                 goto done;
3052                         }
3053                         skb_dst_drop(skb);
3054                 }
3055                 rcu_read_unlock_bh();
3056         }
3057
3058 done:
3059         cb->args[0] = h;
3060         cb->args[1] = idx;
3061         return skb->len;
3062 }
3063
3064 void ip_rt_multicast_event(struct in_device *in_dev)
3065 {
3066         rt_cache_flush(dev_net(in_dev->dev), 0);
3067 }
3068
3069 #ifdef CONFIG_SYSCTL
3070 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3071                                         void __user *buffer,
3072                                         size_t *lenp, loff_t *ppos)
3073 {
3074         if (write) {
3075                 int flush_delay;
3076                 ctl_table ctl;
3077                 struct net *net;
3078
3079                 memcpy(&ctl, __ctl, sizeof(ctl));
3080                 ctl.data = &flush_delay;
3081                 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3082
3083                 net = (struct net *)__ctl->extra1;
3084                 rt_cache_flush(net, flush_delay);
3085                 return 0;
3086         }
3087
3088         return -EINVAL;
3089 }
3090
3091 static ctl_table ipv4_route_table[] = {
3092         {
3093                 .procname       = "gc_thresh",
3094                 .data           = &ipv4_dst_ops.gc_thresh,
3095                 .maxlen         = sizeof(int),
3096                 .mode           = 0644,
3097                 .proc_handler   = proc_dointvec,
3098         },
3099         {
3100                 .procname       = "max_size",
3101                 .data           = &ip_rt_max_size,
3102                 .maxlen         = sizeof(int),
3103                 .mode           = 0644,
3104                 .proc_handler   = proc_dointvec,
3105         },
3106         {
3107                 /*  Deprecated. Use gc_min_interval_ms */
3108
3109                 .procname       = "gc_min_interval",
3110                 .data           = &ip_rt_gc_min_interval,
3111                 .maxlen         = sizeof(int),
3112                 .mode           = 0644,
3113                 .proc_handler   = proc_dointvec_jiffies,
3114         },
3115         {
3116                 .procname       = "gc_min_interval_ms",
3117                 .data           = &ip_rt_gc_min_interval,
3118                 .maxlen         = sizeof(int),
3119                 .mode           = 0644,
3120                 .proc_handler   = proc_dointvec_ms_jiffies,
3121         },
3122         {
3123                 .procname       = "gc_timeout",
3124                 .data           = &ip_rt_gc_timeout,
3125                 .maxlen         = sizeof(int),
3126                 .mode           = 0644,
3127                 .proc_handler   = proc_dointvec_jiffies,
3128         },
3129         {
3130                 .procname       = "gc_interval",
3131                 .data           = &ip_rt_gc_interval,
3132                 .maxlen         = sizeof(int),
3133                 .mode           = 0644,
3134                 .proc_handler   = proc_dointvec_jiffies,
3135         },
3136         {
3137                 .procname       = "redirect_load",
3138                 .data           = &ip_rt_redirect_load,
3139                 .maxlen         = sizeof(int),
3140                 .mode           = 0644,
3141                 .proc_handler   = proc_dointvec,
3142         },
3143         {
3144                 .procname       = "redirect_number",
3145                 .data           = &ip_rt_redirect_number,
3146                 .maxlen         = sizeof(int),
3147                 .mode           = 0644,
3148                 .proc_handler   = proc_dointvec,
3149         },
3150         {
3151                 .procname       = "redirect_silence",
3152                 .data           = &ip_rt_redirect_silence,
3153                 .maxlen         = sizeof(int),
3154                 .mode           = 0644,
3155                 .proc_handler   = proc_dointvec,
3156         },
3157         {
3158                 .procname       = "error_cost",
3159                 .data           = &ip_rt_error_cost,
3160                 .maxlen         = sizeof(int),
3161                 .mode           = 0644,
3162                 .proc_handler   = proc_dointvec,
3163         },
3164         {
3165                 .procname       = "error_burst",
3166                 .data           = &ip_rt_error_burst,
3167                 .maxlen         = sizeof(int),
3168                 .mode           = 0644,
3169                 .proc_handler   = proc_dointvec,
3170         },
3171         {
3172                 .procname       = "gc_elasticity",
3173                 .data           = &ip_rt_gc_elasticity,
3174                 .maxlen         = sizeof(int),
3175                 .mode           = 0644,
3176                 .proc_handler   = proc_dointvec,
3177         },
3178         {
3179                 .procname       = "mtu_expires",
3180                 .data           = &ip_rt_mtu_expires,
3181                 .maxlen         = sizeof(int),
3182                 .mode           = 0644,
3183                 .proc_handler   = proc_dointvec_jiffies,
3184         },
3185         {
3186                 .procname       = "min_pmtu",
3187                 .data           = &ip_rt_min_pmtu,
3188                 .maxlen         = sizeof(int),
3189                 .mode           = 0644,
3190                 .proc_handler   = proc_dointvec,
3191         },
3192         {
3193                 .procname       = "min_adv_mss",
3194                 .data           = &ip_rt_min_advmss,
3195                 .maxlen         = sizeof(int),
3196                 .mode           = 0644,
3197                 .proc_handler   = proc_dointvec,
3198         },
3199         { }
3200 };
3201
3202 static struct ctl_table empty[1];
3203
3204 static struct ctl_table ipv4_skeleton[] =
3205 {
3206         { .procname = "route",
3207           .mode = 0555, .child = ipv4_route_table},
3208         { .procname = "neigh",
3209           .mode = 0555, .child = empty},
3210         { }
3211 };
3212
3213 static __net_initdata struct ctl_path ipv4_path[] = {
3214         { .procname = "net", },
3215         { .procname = "ipv4", },
3216         { },
3217 };
3218
3219 static struct ctl_table ipv4_route_flush_table[] = {
3220         {
3221                 .procname       = "flush",
3222                 .maxlen         = sizeof(int),
3223                 .mode           = 0200,
3224                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3225         },
3226         { },
3227 };
3228
3229 static __net_initdata struct ctl_path ipv4_route_path[] = {
3230         { .procname = "net", },
3231         { .procname = "ipv4", },
3232         { .procname = "route", },
3233         { },
3234 };
3235
3236 static __net_init int sysctl_route_net_init(struct net *net)
3237 {
3238         struct ctl_table *tbl;
3239
3240         tbl = ipv4_route_flush_table;
3241         if (!net_eq(net, &init_net)) {
3242                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3243                 if (tbl == NULL)
3244                         goto err_dup;
3245         }
3246         tbl[0].extra1 = net;
3247
3248         net->ipv4.route_hdr =
3249                 register_net_sysctl_table(net, ipv4_route_path, tbl);
3250         if (net->ipv4.route_hdr == NULL)
3251                 goto err_reg;
3252         return 0;
3253
3254 err_reg:
3255         if (tbl != ipv4_route_flush_table)
3256                 kfree(tbl);
3257 err_dup:
3258         return -ENOMEM;
3259 }
3260
3261 static __net_exit void sysctl_route_net_exit(struct net *net)
3262 {
3263         struct ctl_table *tbl;
3264
3265         tbl = net->ipv4.route_hdr->ctl_table_arg;
3266         unregister_net_sysctl_table(net->ipv4.route_hdr);
3267         BUG_ON(tbl == ipv4_route_flush_table);
3268         kfree(tbl);
3269 }
3270
3271 static __net_initdata struct pernet_operations sysctl_route_ops = {
3272         .init = sysctl_route_net_init,
3273         .exit = sysctl_route_net_exit,
3274 };
3275 #endif
3276
3277 static __net_init int rt_genid_init(struct net *net)
3278 {
3279         get_random_bytes(&net->ipv4.rt_genid,
3280                          sizeof(net->ipv4.rt_genid));
3281         return 0;
3282 }
3283
3284 static __net_initdata struct pernet_operations rt_genid_ops = {
3285         .init = rt_genid_init,
3286 };
3287
3288
3289 #ifdef CONFIG_NET_CLS_ROUTE
3290 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3291 #endif /* CONFIG_NET_CLS_ROUTE */
3292
3293 static __initdata unsigned long rhash_entries;
3294 static int __init set_rhash_entries(char *str)
3295 {
3296         if (!str)
3297                 return 0;
3298         rhash_entries = simple_strtoul(str, &str, 0);
3299         return 1;
3300 }
3301 __setup("rhash_entries=", set_rhash_entries);
3302
3303 int __init ip_rt_init(void)
3304 {
3305         int rc = 0;
3306
3307 #ifdef CONFIG_NET_CLS_ROUTE
3308         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3309         if (!ip_rt_acct)
3310                 panic("IP: failed to allocate ip_rt_acct\n");
3311 #endif
3312
3313         ipv4_dst_ops.kmem_cachep =
3314                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3315                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3316
3317         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3318
3319         rt_hash_table = (struct rt_hash_bucket *)
3320                 alloc_large_system_hash("IP route cache",
3321                                         sizeof(struct rt_hash_bucket),
3322                                         rhash_entries,
3323                                         (totalram_pages >= 128 * 1024) ?
3324                                         15 : 17,
3325                                         0,
3326                                         &rt_hash_log,
3327                                         &rt_hash_mask,
3328                                         rhash_entries ? 0 : 512 * 1024);
3329         memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3330         rt_hash_lock_init();
3331
3332         ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3333         ip_rt_max_size = (rt_hash_mask + 1) * 16;
3334
3335         devinet_init();
3336         ip_fib_init();
3337
3338         /* All the timers, started at system startup tend
3339            to synchronize. Perturb it a bit.
3340          */
3341         INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3342         expires_ljiffies = jiffies;
3343         schedule_delayed_work(&expires_work,
3344                 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3345
3346         if (ip_rt_proc_init())
3347                 printk(KERN_ERR "Unable to create route proc files\n");
3348 #ifdef CONFIG_XFRM
3349         xfrm_init();
3350         xfrm4_init(ip_rt_max_size);
3351 #endif
3352         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3353
3354 #ifdef CONFIG_SYSCTL
3355         register_pernet_subsys(&sysctl_route_ops);
3356 #endif
3357         register_pernet_subsys(&rt_genid_ops);
3358         return rc;
3359 }
3360
3361 #ifdef CONFIG_SYSCTL
3362 /*
3363  * We really need to sanitize the damn ipv4 init order, then all
3364  * this nonsense will go away.
3365  */
3366 void __init ip_static_sysctl_init(void)
3367 {
3368         register_sysctl_paths(ipv4_path, ipv4_skeleton);
3369 }
3370 #endif
3371
3372 EXPORT_SYMBOL(__ip_select_ident);
3373 EXPORT_SYMBOL(ip_route_output_key);