git.oblomov.eu Git - linux-2.6/blob - net/ipv4/fib_hash.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  *              IPv4 FIB: lookup engine and maintenance routines.
   7  *
   8  * Version:     $Id: fib_hash.c,v 1.13 2001/10/31 21:55:54 davem Exp $
   9  *
  10  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  11  *
  12  *              This program is free software; you can redistribute it and/or
  13  *              modify it under the terms of the GNU General Public License
  14  *              as published by the Free Software Foundation; either version
  15  *              2 of the License, or (at your option) any later version.
  16  */
  17
  18 #include <asm/uaccess.h>
  19 #include <asm/system.h>
  20 #include <linux/bitops.h>
  21 #include <linux/types.h>
  22 #include <linux/kernel.h>
  23 #include <linux/sched.h>
  24 #include <linux/mm.h>
  25 #include <linux/string.h>
  26 #include <linux/socket.h>
  27 #include <linux/sockios.h>
  28 #include <linux/errno.h>
  29 #include <linux/in.h>
  30 #include <linux/inet.h>
  31 #include <linux/inetdevice.h>
  32 #include <linux/netdevice.h>
  33 #include <linux/if_arp.h>
  34 #include <linux/proc_fs.h>
  35 #include <linux/skbuff.h>
  36 #include <linux/netlink.h>
  37 #include <linux/init.h>
  38
  39 #include <net/ip.h>
  40 #include <net/protocol.h>
  41 #include <net/route.h>
  42 #include <net/tcp.h>
  43 #include <net/sock.h>
  44 #include <net/ip_fib.h>
  45
  46 #include "fib_lookup.h"
  47
  48 static kmem_cache_t *fn_hash_kmem __read_mostly;
  49 static kmem_cache_t *fn_alias_kmem __read_mostly;
  50
  51 struct fib_node {
  52         struct hlist_node       fn_hash;
  53         struct list_head        fn_alias;
  54         __be32                  fn_key;
  55 };
  56
  57 struct fn_zone {
  58         struct fn_zone          *fz_next;       /* Next not empty zone  */
  59         struct hlist_head       *fz_hash;       /* Hash table pointer   */
  60         int                     fz_nent;        /* Number of entries    */
  61
  62         int                     fz_divisor;     /* Hash divisor         */
  63         u32                     fz_hashmask;    /* (fz_divisor - 1)     */
  64 #define FZ_HASHMASK(fz)         ((fz)->fz_hashmask)
  65
  66         int                     fz_order;       /* Zone order           */
  67         __be32                  fz_mask;
  68 #define FZ_MASK(fz)             ((fz)->fz_mask)
  69 };
  70
  71 /* NOTE. On fast computers evaluation of fz_hashmask and fz_mask
  72  * can be cheaper than memory lookup, so that FZ_* macros are used.
  73  */
  74
  75 struct fn_hash {
  76         struct fn_zone  *fn_zones[33];
  77         struct fn_zone  *fn_zone_list;
  78 };
  79
  80 static inline u32 fn_hash(__be32 key, struct fn_zone *fz)
  81 {
  82         u32 h = ntohl(key)>>(32 - fz->fz_order);
  83         h ^= (h>>20);
  84         h ^= (h>>10);
  85         h ^= (h>>5);
  86         h &= FZ_HASHMASK(fz);
  87         return h;
  88 }
  89
  90 static inline __be32 fz_key(__be32 dst, struct fn_zone *fz)
  91 {
  92         return dst & FZ_MASK(fz);
  93 }
  94
  95 static DEFINE_RWLOCK(fib_hash_lock);
  96 static unsigned int fib_hash_genid;
  97
  98 #define FZ_MAX_DIVISOR ((PAGE_SIZE<<MAX_ORDER) / sizeof(struct hlist_head))
  99
 100 static struct hlist_head *fz_hash_alloc(int divisor)
 101 {
 102         unsigned long size = divisor * sizeof(struct hlist_head);
 103
 104         if (size <= PAGE_SIZE) {
 105                 return kmalloc(size, GFP_KERNEL);
 106         } else {
 107                 return (struct hlist_head *)
 108                         __get_free_pages(GFP_KERNEL, get_order(size));
 109         }
 110 }
 111
 112 /* The fib hash lock must be held when this is called. */
 113 static inline void fn_rebuild_zone(struct fn_zone *fz,
 114                                    struct hlist_head *old_ht,
 115                                    int old_divisor)
 116 {
 117         int i;
 118
 119         for (i = 0; i < old_divisor; i++) {
 120                 struct hlist_node *node, *n;
 121                 struct fib_node *f;
 122
 123                 hlist_for_each_entry_safe(f, node, n, &old_ht[i], fn_hash) {
 124                         struct hlist_head *new_head;
 125
 126                         hlist_del(&f->fn_hash);
 127
 128                         new_head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 129                         hlist_add_head(&f->fn_hash, new_head);
 130                 }
 131         }
 132 }
 133
 134 static void fz_hash_free(struct hlist_head *hash, int divisor)
 135 {
 136         unsigned long size = divisor * sizeof(struct hlist_head);
 137
 138         if (size <= PAGE_SIZE)
 139                 kfree(hash);
 140         else
 141                 free_pages((unsigned long)hash, get_order(size));
 142 }
 143
 144 static void fn_rehash_zone(struct fn_zone *fz)
 145 {
 146         struct hlist_head *ht, *old_ht;
 147         int old_divisor, new_divisor;
 148         u32 new_hashmask;
 149
 150         old_divisor = fz->fz_divisor;
 151
 152         switch (old_divisor) {
 153         case 16:
 154                 new_divisor = 256;
 155                 break;
 156         case 256:
 157                 new_divisor = 1024;
 158                 break;
 159         default:
 160                 if ((old_divisor << 1) > FZ_MAX_DIVISOR) {
 161                         printk(KERN_CRIT "route.c: bad divisor %d!\n", old_divisor);
 162                         return;
 163                 }
 164                 new_divisor = (old_divisor << 1);
 165                 break;
 166         }
 167
 168         new_hashmask = (new_divisor - 1);
 169
 170 #if RT_CACHE_DEBUG >= 2
 171         printk("fn_rehash_zone: hash for zone %d grows from %d\n", fz->fz_order, old_divisor);
 172 #endif
 173
 174         ht = fz_hash_alloc(new_divisor);
 175
 176         if (ht) {
 177                 memset(ht, 0, new_divisor * sizeof(struct hlist_head));
 178
 179                 write_lock_bh(&fib_hash_lock);
 180                 old_ht = fz->fz_hash;
 181                 fz->fz_hash = ht;
 182                 fz->fz_hashmask = new_hashmask;
 183                 fz->fz_divisor = new_divisor;
 184                 fn_rebuild_zone(fz, old_ht, old_divisor);
 185                 fib_hash_genid++;
 186                 write_unlock_bh(&fib_hash_lock);
 187
 188                 fz_hash_free(old_ht, old_divisor);
 189         }
 190 }
 191
 192 static inline void fn_free_node(struct fib_node * f)
 193 {
 194         kmem_cache_free(fn_hash_kmem, f);
 195 }
 196
 197 static inline void fn_free_alias(struct fib_alias *fa)
 198 {
 199         fib_release_info(fa->fa_info);
 200         kmem_cache_free(fn_alias_kmem, fa);
 201 }
 202
 203 static struct fn_zone *
 204 fn_new_zone(struct fn_hash *table, int z)
 205 {
 206         int i;
 207         struct fn_zone *fz = kzalloc(sizeof(struct fn_zone), GFP_KERNEL);
 208         if (!fz)
 209                 return NULL;
 210
 211         if (z) {
 212                 fz->fz_divisor = 16;
 213         } else {
 214                 fz->fz_divisor = 1;
 215         }
 216         fz->fz_hashmask = (fz->fz_divisor - 1);
 217         fz->fz_hash = fz_hash_alloc(fz->fz_divisor);
 218         if (!fz->fz_hash) {
 219                 kfree(fz);
 220                 return NULL;
 221         }
 222         memset(fz->fz_hash, 0, fz->fz_divisor * sizeof(struct hlist_head *));
 223         fz->fz_order = z;
 224         fz->fz_mask = inet_make_mask(z);
 225
 226         /* Find the first not empty zone with more specific mask */
 227         for (i=z+1; i<=32; i++)
 228                 if (table->fn_zones[i])
 229                         break;
 230         write_lock_bh(&fib_hash_lock);
 231         if (i>32) {
 232                 /* No more specific masks, we are the first. */
 233                 fz->fz_next = table->fn_zone_list;
 234                 table->fn_zone_list = fz;
 235         } else {
 236                 fz->fz_next = table->fn_zones[i]->fz_next;
 237                 table->fn_zones[i]->fz_next = fz;
 238         }
 239         table->fn_zones[z] = fz;
 240         fib_hash_genid++;
 241         write_unlock_bh(&fib_hash_lock);
 242         return fz;
 243 }
 244
 245 static int
 246 fn_hash_lookup(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
 247 {
 248         int err;
 249         struct fn_zone *fz;
 250         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 251
 252         read_lock(&fib_hash_lock);
 253         for (fz = t->fn_zone_list; fz; fz = fz->fz_next) {
 254                 struct hlist_head *head;
 255                 struct hlist_node *node;
 256                 struct fib_node *f;
 257                 __be32 k = fz_key(flp->fl4_dst, fz);
 258
 259                 head = &fz->fz_hash[fn_hash(k, fz)];
 260                 hlist_for_each_entry(f, node, head, fn_hash) {
 261                         if (f->fn_key != k)
 262                                 continue;
 263
 264                         err = fib_semantic_match(&f->fn_alias,
 265                                                  flp, res,
 266                                                  f->fn_key, fz->fz_mask,
 267                                                  fz->fz_order);
 268                         if (err <= 0)
 269                                 goto out;
 270                 }
 271         }
 272         err = 1;
 273 out:
 274         read_unlock(&fib_hash_lock);
 275         return err;
 276 }
 277
 278 static int fn_hash_last_dflt=-1;
 279
 280 static void
 281 fn_hash_select_default(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
 282 {
 283         int order, last_idx;
 284         struct hlist_node *node;
 285         struct fib_node *f;
 286         struct fib_info *fi = NULL;
 287         struct fib_info *last_resort;
 288         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 289         struct fn_zone *fz = t->fn_zones[0];
 290
 291         if (fz == NULL)
 292                 return;
 293
 294         last_idx = -1;
 295         last_resort = NULL;
 296         order = -1;
 297
 298         read_lock(&fib_hash_lock);
 299         hlist_for_each_entry(f, node, &fz->fz_hash[0], fn_hash) {
 300                 struct fib_alias *fa;
 301
 302                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 303                         struct fib_info *next_fi = fa->fa_info;
 304
 305                         if (fa->fa_scope != res->scope ||
 306                             fa->fa_type != RTN_UNICAST)
 307                                 continue;
 308
 309                         if (next_fi->fib_priority > res->fi->fib_priority)
 310                                 break;
 311                         if (!next_fi->fib_nh[0].nh_gw ||
 312                             next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
 313                                 continue;
 314                         fa->fa_state |= FA_S_ACCESSED;
 315
 316                         if (fi == NULL) {
 317                                 if (next_fi != res->fi)
 318                                         break;
 319                         } else if (!fib_detect_death(fi, order, &last_resort,
 320                                                      &last_idx, &fn_hash_last_dflt)) {
 321                                 if (res->fi)
 322                                         fib_info_put(res->fi);
 323                                 res->fi = fi;
 324                                 atomic_inc(&fi->fib_clntref);
 325                                 fn_hash_last_dflt = order;
 326                                 goto out;
 327                         }
 328                         fi = next_fi;
 329                         order++;
 330                 }
 331         }
 332
 333         if (order <= 0 || fi == NULL) {
 334                 fn_hash_last_dflt = -1;
 335                 goto out;
 336         }
 337
 338         if (!fib_detect_death(fi, order, &last_resort, &last_idx, &fn_hash_last_dflt)) {
 339                 if (res->fi)
 340                         fib_info_put(res->fi);
 341                 res->fi = fi;
 342                 atomic_inc(&fi->fib_clntref);
 343                 fn_hash_last_dflt = order;
 344                 goto out;
 345         }
 346
 347         if (last_idx >= 0) {
 348                 if (res->fi)
 349                         fib_info_put(res->fi);
 350                 res->fi = last_resort;
 351                 if (last_resort)
 352                         atomic_inc(&last_resort->fib_clntref);
 353         }
 354         fn_hash_last_dflt = last_idx;
 355 out:
 356         read_unlock(&fib_hash_lock);
 357 }
 358
 359 /* Insert node F to FZ. */
 360 static inline void fib_insert_node(struct fn_zone *fz, struct fib_node *f)
 361 {
 362         struct hlist_head *head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 363
 364         hlist_add_head(&f->fn_hash, head);
 365 }
 366
 367 /* Return the node in FZ matching KEY. */
 368 static struct fib_node *fib_find_node(struct fn_zone *fz, __be32 key)
 369 {
 370         struct hlist_head *head = &fz->fz_hash[fn_hash(key, fz)];
 371         struct hlist_node *node;
 372         struct fib_node *f;
 373
 374         hlist_for_each_entry(f, node, head, fn_hash) {
 375                 if (f->fn_key == key)
 376                         return f;
 377         }
 378
 379         return NULL;
 380 }
 381
 382 static int fn_hash_insert(struct fib_table *tb, struct fib_config *cfg)
 383 {
 384         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 385         struct fib_node *new_f, *f;
 386         struct fib_alias *fa, *new_fa;
 387         struct fn_zone *fz;
 388         struct fib_info *fi;
 389         u8 tos = cfg->fc_tos;
 390         __be32 key;
 391         int err;
 392
 393         if (cfg->fc_dst_len > 32)
 394                 return -EINVAL;
 395
 396         fz = table->fn_zones[cfg->fc_dst_len];
 397         if (!fz && !(fz = fn_new_zone(table, cfg->fc_dst_len)))
 398                 return -ENOBUFS;
 399
 400         key = 0;
 401         if (cfg->fc_dst) {
 402                 if (cfg->fc_dst & ~FZ_MASK(fz))
 403                         return -EINVAL;
 404                 key = fz_key(cfg->fc_dst, fz);
 405         }
 406
 407         fi = fib_create_info(cfg);
 408         if (IS_ERR(fi))
 409                 return PTR_ERR(fi);
 410
 411         if (fz->fz_nent > (fz->fz_divisor<<1) &&
 412             fz->fz_divisor < FZ_MAX_DIVISOR &&
 413             (cfg->fc_dst_len == 32 ||
 414              (1 << cfg->fc_dst_len) > fz->fz_divisor))
 415                 fn_rehash_zone(fz);
 416
 417         f = fib_find_node(fz, key);
 418
 419         if (!f)
 420                 fa = NULL;
 421         else
 422                 fa = fib_find_alias(&f->fn_alias, tos, fi->fib_priority);
 423
 424         /* Now fa, if non-NULL, points to the first fib alias
 425          * with the same keys [prefix,tos,priority], if such key already
 426          * exists or to the node before which we will insert new one.
 427          *
 428          * If fa is NULL, we will need to allocate a new one and
 429          * insert to the head of f.
 430          *
 431          * If f is NULL, no fib node matched the destination key
 432          * and we need to allocate a new one of those as well.
 433          */
 434
 435         if (fa && fa->fa_tos == tos &&
 436             fa->fa_info->fib_priority == fi->fib_priority) {
 437                 struct fib_alias *fa_orig;
 438
 439                 err = -EEXIST;
 440                 if (cfg->fc_nlflags & NLM_F_EXCL)
 441                         goto out;
 442
 443                 if (cfg->fc_nlflags & NLM_F_REPLACE) {
 444                         struct fib_info *fi_drop;
 445                         u8 state;
 446
 447                         write_lock_bh(&fib_hash_lock);
 448                         fi_drop = fa->fa_info;
 449                         fa->fa_info = fi;
 450                         fa->fa_type = cfg->fc_type;
 451                         fa->fa_scope = cfg->fc_scope;
 452                         state = fa->fa_state;
 453                         fa->fa_state &= ~FA_S_ACCESSED;
 454                         fib_hash_genid++;
 455                         write_unlock_bh(&fib_hash_lock);
 456
 457                         fib_release_info(fi_drop);
 458                         if (state & FA_S_ACCESSED)
 459                                 rt_cache_flush(-1);
 460                         return 0;
 461                 }
 462
 463                 /* Error if we find a perfect match which
 464                  * uses the same scope, type, and nexthop
 465                  * information.
 466                  */
 467                 fa_orig = fa;
 468                 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 469                 list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 470                         if (fa->fa_tos != tos)
 471                                 break;
 472                         if (fa->fa_info->fib_priority != fi->fib_priority)
 473                                 break;
 474                         if (fa->fa_type == cfg->fc_type &&
 475                             fa->fa_scope == cfg->fc_scope &&
 476                             fa->fa_info == fi)
 477                                 goto out;
 478                 }
 479                 if (!(cfg->fc_nlflags & NLM_F_APPEND))
 480                         fa = fa_orig;
 481         }
 482
 483         err = -ENOENT;
 484         if (!(cfg->fc_nlflags & NLM_F_CREATE))
 485                 goto out;
 486
 487         err = -ENOBUFS;
 488         new_fa = kmem_cache_alloc(fn_alias_kmem, SLAB_KERNEL);
 489         if (new_fa == NULL)
 490                 goto out;
 491
 492         new_f = NULL;
 493         if (!f) {
 494                 new_f = kmem_cache_alloc(fn_hash_kmem, SLAB_KERNEL);
 495                 if (new_f == NULL)
 496                         goto out_free_new_fa;
 497
 498                 INIT_HLIST_NODE(&new_f->fn_hash);
 499                 INIT_LIST_HEAD(&new_f->fn_alias);
 500                 new_f->fn_key = key;
 501                 f = new_f;
 502         }
 503
 504         new_fa->fa_info = fi;
 505         new_fa->fa_tos = tos;
 506         new_fa->fa_type = cfg->fc_type;
 507         new_fa->fa_scope = cfg->fc_scope;
 508         new_fa->fa_state = 0;
 509
 510         /*
 511          * Insert new entry to the list.
 512          */
 513
 514         write_lock_bh(&fib_hash_lock);
 515         if (new_f)
 516                 fib_insert_node(fz, new_f);
 517         list_add_tail(&new_fa->fa_list,
 518                  (fa ? &fa->fa_list : &f->fn_alias));
 519         fib_hash_genid++;
 520         write_unlock_bh(&fib_hash_lock);
 521
 522         if (new_f)
 523                 fz->fz_nent++;
 524         rt_cache_flush(-1);
 525
 526         rtmsg_fib(RTM_NEWROUTE, key, new_fa, cfg->fc_dst_len, tb->tb_id,
 527                   &cfg->fc_nlinfo);
 528         return 0;
 529
 530 out_free_new_fa:
 531         kmem_cache_free(fn_alias_kmem, new_fa);
 532 out:
 533         fib_release_info(fi);
 534         return err;
 535 }
 536
 537
 538 static int fn_hash_delete(struct fib_table *tb, struct fib_config *cfg)
 539 {
 540         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 541         struct fib_node *f;
 542         struct fib_alias *fa, *fa_to_delete;
 543         struct fn_zone *fz;
 544         __be32 key;
 545
 546         if (cfg->fc_dst_len > 32)
 547                 return -EINVAL;
 548
 549         if ((fz  = table->fn_zones[cfg->fc_dst_len]) == NULL)
 550                 return -ESRCH;
 551
 552         key = 0;
 553         if (cfg->fc_dst) {
 554                 if (cfg->fc_dst & ~FZ_MASK(fz))
 555                         return -EINVAL;
 556                 key = fz_key(cfg->fc_dst, fz);
 557         }
 558
 559         f = fib_find_node(fz, key);
 560
 561         if (!f)
 562                 fa = NULL;
 563         else
 564                 fa = fib_find_alias(&f->fn_alias, cfg->fc_tos, 0);
 565         if (!fa)
 566                 return -ESRCH;
 567
 568         fa_to_delete = NULL;
 569         fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 570         list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 571                 struct fib_info *fi = fa->fa_info;
 572
 573                 if (fa->fa_tos != cfg->fc_tos)
 574                         break;
 575
 576                 if ((!cfg->fc_type ||
 577                      fa->fa_type == cfg->fc_type) &&
 578                     (cfg->fc_scope == RT_SCOPE_NOWHERE ||
 579                      fa->fa_scope == cfg->fc_scope) &&
 580                     (!cfg->fc_protocol ||
 581                      fi->fib_protocol == cfg->fc_protocol) &&
 582                     fib_nh_match(cfg, fi) == 0) {
 583                         fa_to_delete = fa;
 584                         break;
 585                 }
 586         }
 587
 588         if (fa_to_delete) {
 589                 int kill_fn;
 590
 591                 fa = fa_to_delete;
 592                 rtmsg_fib(RTM_DELROUTE, key, fa, cfg->fc_dst_len,
 593                           tb->tb_id, &cfg->fc_nlinfo);
 594
 595                 kill_fn = 0;
 596                 write_lock_bh(&fib_hash_lock);
 597                 list_del(&fa->fa_list);
 598                 if (list_empty(&f->fn_alias)) {
 599                         hlist_del(&f->fn_hash);
 600                         kill_fn = 1;
 601                 }
 602                 fib_hash_genid++;
 603                 write_unlock_bh(&fib_hash_lock);
 604
 605                 if (fa->fa_state & FA_S_ACCESSED)
 606                         rt_cache_flush(-1);
 607                 fn_free_alias(fa);
 608                 if (kill_fn) {
 609                         fn_free_node(f);
 610                         fz->fz_nent--;
 611                 }
 612
 613                 return 0;
 614         }
 615         return -ESRCH;
 616 }
 617
 618 static int fn_flush_list(struct fn_zone *fz, int idx)
 619 {
 620         struct hlist_head *head = &fz->fz_hash[idx];
 621         struct hlist_node *node, *n;
 622         struct fib_node *f;
 623         int found = 0;
 624
 625         hlist_for_each_entry_safe(f, node, n, head, fn_hash) {
 626                 struct fib_alias *fa, *fa_node;
 627                 int kill_f;
 628
 629                 kill_f = 0;
 630                 list_for_each_entry_safe(fa, fa_node, &f->fn_alias, fa_list) {
 631                         struct fib_info *fi = fa->fa_info;
 632
 633                         if (fi && (fi->fib_flags&RTNH_F_DEAD)) {
 634                                 write_lock_bh(&fib_hash_lock);
 635                                 list_del(&fa->fa_list);
 636                                 if (list_empty(&f->fn_alias)) {
 637                                         hlist_del(&f->fn_hash);
 638                                         kill_f = 1;
 639                                 }
 640                                 fib_hash_genid++;
 641                                 write_unlock_bh(&fib_hash_lock);
 642
 643                                 fn_free_alias(fa);
 644                                 found++;
 645                         }
 646                 }
 647                 if (kill_f) {
 648                         fn_free_node(f);
 649                         fz->fz_nent--;
 650                 }
 651         }
 652         return found;
 653 }
 654
 655 static int fn_hash_flush(struct fib_table *tb)
 656 {
 657         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 658         struct fn_zone *fz;
 659         int found = 0;
 660
 661         for (fz = table->fn_zone_list; fz; fz = fz->fz_next) {
 662                 int i;
 663
 664                 for (i = fz->fz_divisor - 1; i >= 0; i--)
 665                         found += fn_flush_list(fz, i);
 666         }
 667         return found;
 668 }
 669
 670
 671 static inline int
 672 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
 673                      struct fib_table *tb,
 674                      struct fn_zone *fz,
 675                      struct hlist_head *head)
 676 {
 677         struct hlist_node *node;
 678         struct fib_node *f;
 679         int i, s_i;
 680
 681         s_i = cb->args[4];
 682         i = 0;
 683         hlist_for_each_entry(f, node, head, fn_hash) {
 684                 struct fib_alias *fa;
 685
 686                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 687                         if (i < s_i)
 688                                 goto next;
 689
 690                         if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
 691                                           cb->nlh->nlmsg_seq,
 692                                           RTM_NEWROUTE,
 693                                           tb->tb_id,
 694                                           fa->fa_type,
 695                                           fa->fa_scope,
 696                                           f->fn_key,
 697                                           fz->fz_order,
 698                                           fa->fa_tos,
 699                                           fa->fa_info,
 700                                           NLM_F_MULTI) < 0) {
 701                                 cb->args[4] = i;
 702                                 return -1;
 703                         }
 704                 next:
 705                         i++;
 706                 }
 707         }
 708         cb->args[4] = i;
 709         return skb->len;
 710 }
 711
 712 static inline int
 713 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
 714                    struct fib_table *tb,
 715                    struct fn_zone *fz)
 716 {
 717         int h, s_h;
 718
 719         s_h = cb->args[3];
 720         for (h=0; h < fz->fz_divisor; h++) {
 721                 if (h < s_h) continue;
 722                 if (h > s_h)
 723                         memset(&cb->args[4], 0,
 724                                sizeof(cb->args) - 4*sizeof(cb->args[0]));
 725                 if (fz->fz_hash == NULL ||
 726                     hlist_empty(&fz->fz_hash[h]))
 727                         continue;
 728                 if (fn_hash_dump_bucket(skb, cb, tb, fz, &fz->fz_hash[h])<0) {
 729                         cb->args[3] = h;
 730                         return -1;
 731                 }
 732         }
 733         cb->args[3] = h;
 734         return skb->len;
 735 }
 736
 737 static int fn_hash_dump(struct fib_table *tb, struct sk_buff *skb, struct netlink_callback *cb)
 738 {
 739         int m, s_m;
 740         struct fn_zone *fz;
 741         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 742
 743         s_m = cb->args[2];
 744         read_lock(&fib_hash_lock);
 745         for (fz = table->fn_zone_list, m=0; fz; fz = fz->fz_next, m++) {
 746                 if (m < s_m) continue;
 747                 if (m > s_m)
 748                         memset(&cb->args[3], 0,
 749                                sizeof(cb->args) - 3*sizeof(cb->args[0]));
 750                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
 751                         cb->args[2] = m;
 752                         read_unlock(&fib_hash_lock);
 753                         return -1;
 754                 }
 755         }
 756         read_unlock(&fib_hash_lock);
 757         cb->args[2] = m;
 758         return skb->len;
 759 }
 760
 761 #ifdef CONFIG_IP_MULTIPLE_TABLES
 762 struct fib_table * fib_hash_init(u32 id)
 763 #else
 764 struct fib_table * __init fib_hash_init(u32 id)
 765 #endif
 766 {
 767         struct fib_table *tb;
 768
 769         if (fn_hash_kmem == NULL)
 770                 fn_hash_kmem = kmem_cache_create("ip_fib_hash",
 771                                                  sizeof(struct fib_node),
 772                                                  0, SLAB_HWCACHE_ALIGN,
 773                                                  NULL, NULL);
 774
 775         if (fn_alias_kmem == NULL)
 776                 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
 777                                                   sizeof(struct fib_alias),
 778                                                   0, SLAB_HWCACHE_ALIGN,
 779                                                   NULL, NULL);
 780
 781         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash),
 782                      GFP_KERNEL);
 783         if (tb == NULL)
 784                 return NULL;
 785
 786         tb->tb_id = id;
 787         tb->tb_lookup = fn_hash_lookup;
 788         tb->tb_insert = fn_hash_insert;
 789         tb->tb_delete = fn_hash_delete;
 790         tb->tb_flush = fn_hash_flush;
 791         tb->tb_select_default = fn_hash_select_default;
 792         tb->tb_dump = fn_hash_dump;
 793         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 794         return tb;
 795 }
 796
 797 /* ------------------------------------------------------------------------ */
 798 #ifdef CONFIG_PROC_FS
 799
 800 struct fib_iter_state {
 801         struct fn_zone  *zone;
 802         int             bucket;
 803         struct hlist_head *hash_head;
 804         struct fib_node *fn;
 805         struct fib_alias *fa;
 806         loff_t pos;
 807         unsigned int genid;
 808         int valid;
 809 };
 810
 811 static struct fib_alias *fib_get_first(struct seq_file *seq)
 812 {
 813         struct fib_iter_state *iter = seq->private;
 814         struct fn_hash *table = (struct fn_hash *) ip_fib_main_table->tb_data;
 815
 816         iter->bucket    = 0;
 817         iter->hash_head = NULL;
 818         iter->fn        = NULL;
 819         iter->fa        = NULL;
 820         iter->pos       = 0;
 821         iter->genid     = fib_hash_genid;
 822         iter->valid     = 1;
 823
 824         for (iter->zone = table->fn_zone_list; iter->zone;
 825              iter->zone = iter->zone->fz_next) {
 826                 int maxslot;
 827
 828                 if (!iter->zone->fz_nent)
 829                         continue;
 830
 831                 iter->hash_head = iter->zone->fz_hash;
 832                 maxslot = iter->zone->fz_divisor;
 833
 834                 for (iter->bucket = 0; iter->bucket < maxslot;
 835                      ++iter->bucket, ++iter->hash_head) {
 836                         struct hlist_node *node;
 837                         struct fib_node *fn;
 838
 839                         hlist_for_each_entry(fn,node,iter->hash_head,fn_hash) {
 840                                 struct fib_alias *fa;
 841
 842                                 list_for_each_entry(fa,&fn->fn_alias,fa_list) {
 843                                         iter->fn = fn;
 844                                         iter->fa = fa;
 845                                         goto out;
 846                                 }
 847                         }
 848                 }
 849         }
 850 out:
 851         return iter->fa;
 852 }
 853
 854 static struct fib_alias *fib_get_next(struct seq_file *seq)
 855 {
 856         struct fib_iter_state *iter = seq->private;
 857         struct fib_node *fn;
 858         struct fib_alias *fa;
 859
 860         /* Advance FA, if any. */
 861         fn = iter->fn;
 862         fa = iter->fa;
 863         if (fa) {
 864                 BUG_ON(!fn);
 865                 list_for_each_entry_continue(fa, &fn->fn_alias, fa_list) {
 866                         iter->fa = fa;
 867                         goto out;
 868                 }
 869         }
 870
 871         fa = iter->fa = NULL;
 872
 873         /* Advance FN. */
 874         if (fn) {
 875                 struct hlist_node *node = &fn->fn_hash;
 876                 hlist_for_each_entry_continue(fn, node, fn_hash) {
 877                         iter->fn = fn;
 878
 879                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 880                                 iter->fa = fa;
 881                                 goto out;
 882                         }
 883                 }
 884         }
 885
 886         fn = iter->fn = NULL;
 887
 888         /* Advance hash chain. */
 889         if (!iter->zone)
 890                 goto out;
 891
 892         for (;;) {
 893                 struct hlist_node *node;
 894                 int maxslot;
 895
 896                 maxslot = iter->zone->fz_divisor;
 897
 898                 while (++iter->bucket < maxslot) {
 899                         iter->hash_head++;
 900
 901                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 902                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 903                                         iter->fn = fn;
 904                                         iter->fa = fa;
 905                                         goto out;
 906                                 }
 907                         }
 908                 }
 909
 910                 iter->zone = iter->zone->fz_next;
 911
 912                 if (!iter->zone)
 913                         goto out;
 914
 915                 iter->bucket = 0;
 916                 iter->hash_head = iter->zone->fz_hash;
 917
 918                 hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 919                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 920                                 iter->fn = fn;
 921                                 iter->fa = fa;
 922                                 goto out;
 923                         }
 924                 }
 925         }
 926 out:
 927         iter->pos++;
 928         return fa;
 929 }
 930
 931 static struct fib_alias *fib_get_idx(struct seq_file *seq, loff_t pos)
 932 {
 933         struct fib_iter_state *iter = seq->private;
 934         struct fib_alias *fa;
 935
 936         if (iter->valid && pos >= iter->pos && iter->genid == fib_hash_genid) {
 937                 fa   = iter->fa;
 938                 pos -= iter->pos;
 939         } else
 940                 fa = fib_get_first(seq);
 941
 942         if (fa)
 943                 while (pos && (fa = fib_get_next(seq)))
 944                         --pos;
 945         return pos ? NULL : fa;
 946 }
 947
 948 static void *fib_seq_start(struct seq_file *seq, loff_t *pos)
 949 {
 950         void *v = NULL;
 951
 952         read_lock(&fib_hash_lock);
 953         if (ip_fib_main_table)
 954                 v = *pos ? fib_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 955         return v;
 956 }
 957
 958 static void *fib_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 959 {
 960         ++*pos;
 961         return v == SEQ_START_TOKEN ? fib_get_first(seq) : fib_get_next(seq);
 962 }
 963
 964 static void fib_seq_stop(struct seq_file *seq, void *v)
 965 {
 966         read_unlock(&fib_hash_lock);
 967 }
 968
 969 static unsigned fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
 970 {
 971         static const unsigned type2flags[RTN_MAX + 1] = {
 972                 [7] = RTF_REJECT, [8] = RTF_REJECT,
 973         };
 974         unsigned flags = type2flags[type];
 975
 976         if (fi && fi->fib_nh->nh_gw)
 977                 flags |= RTF_GATEWAY;
 978         if (mask == htonl(0xFFFFFFFF))
 979                 flags |= RTF_HOST;
 980         flags |= RTF_UP;
 981         return flags;
 982 }
 983
 984 /*
 985  *      This outputs /proc/net/route.
 986  *
 987  *      It always works in backward compatibility mode.
 988  *      The format of the file is not supposed to be changed.
 989  */
 990 static int fib_seq_show(struct seq_file *seq, void *v)
 991 {
 992         struct fib_iter_state *iter;
 993         char bf[128];
 994         __be32 prefix, mask;
 995         unsigned flags;
 996         struct fib_node *f;
 997         struct fib_alias *fa;
 998         struct fib_info *fi;
 999
1000         if (v == SEQ_START_TOKEN) {
1001                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
1002                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
1003                            "\tWindow\tIRTT");
1004                 goto out;
1005         }
1006
1007         iter    = seq->private;
1008         f       = iter->fn;
1009         fa      = iter->fa;
1010         fi      = fa->fa_info;
1011         prefix  = f->fn_key;
1012         mask    = FZ_MASK(iter->zone);
1013         flags   = fib_flag_trans(fa->fa_type, mask, fi);
1014         if (fi)
1015                 snprintf(bf, sizeof(bf),
1016                          "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1017                          fi->fib_dev ? fi->fib_dev->name : "*", prefix,
1018                          fi->fib_nh->nh_gw, flags, 0, 0, fi->fib_priority,
1019                          mask, (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
1020                          fi->fib_window,
1021                          fi->fib_rtt >> 3);
1022         else
1023                 snprintf(bf, sizeof(bf),
1024                          "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1025                          prefix, 0, flags, 0, 0, 0, mask, 0, 0, 0);
1026         seq_printf(seq, "%-127s\n", bf);
1027 out:
1028         return 0;
1029 }
1030
1031 static struct seq_operations fib_seq_ops = {
1032         .start  = fib_seq_start,
1033         .next   = fib_seq_next,
1034         .stop   = fib_seq_stop,
1035         .show   = fib_seq_show,
1036 };
1037
1038 static int fib_seq_open(struct inode *inode, struct file *file)
1039 {
1040         struct seq_file *seq;
1041         int rc = -ENOMEM;
1042         struct fib_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1043
1044         if (!s)
1045                 goto out;
1046
1047         rc = seq_open(file, &fib_seq_ops);
1048         if (rc)
1049                 goto out_kfree;
1050
1051         seq          = file->private_data;
1052         seq->private = s;
1053 out:
1054         return rc;
1055 out_kfree:
1056         kfree(s);
1057         goto out;
1058 }
1059
1060 static struct file_operations fib_seq_fops = {
1061         .owner          = THIS_MODULE,
1062         .open           = fib_seq_open,
1063         .read           = seq_read,
1064         .llseek         = seq_lseek,
1065         .release        = seq_release_private,
1066 };
1067
1068 int __init fib_proc_init(void)
1069 {
1070         if (!proc_net_fops_create("route", S_IRUGO, &fib_seq_fops))
1071                 return -ENOMEM;
1072         return 0;
1073 }
1074
1075 void __init fib_proc_exit(void)
1076 {
1077         proc_net_remove("route");
1078 }
1079 #endif /* CONFIG_PROC_FS */