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