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 #include <linux/slab.h>
  36
  37 #include <net/net_namespace.h>
  38 #include <net/ip.h>
  39 #include <net/protocol.h>
  40 #include <net/route.h>
  41 #include <net/tcp.h>
  42 #include <net/sock.h>
  43 #include <net/ip_fib.h>
  44
  45 #include "fib_lookup.h"
  46
  47 static struct kmem_cache *fn_hash_kmem __read_mostly;
  48 static struct kmem_cache *fn_alias_kmem __read_mostly;
  49
  50 struct fib_node {
  51         struct hlist_node       fn_hash;
  52         struct list_head        fn_alias;
  53         __be32                  fn_key;
  54         struct fib_alias        fn_embedded_alias;
  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 kzalloc(size, GFP_KERNEL);
 106         } else {
 107                 return (struct hlist_head *)
 108                         __get_free_pages(GFP_KERNEL | __GFP_ZERO, 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(KERN_DEBUG "fn_rehash_zone: hash for zone %d grows from %d\n",
 172                fz->fz_order, old_divisor);
 173 #endif
 174
 175         ht = fz_hash_alloc(new_divisor);
 176
 177         if (ht) {
 178                 write_lock_bh(&fib_hash_lock);
 179                 old_ht = fz->fz_hash;
 180                 fz->fz_hash = ht;
 181                 fz->fz_hashmask = new_hashmask;
 182                 fz->fz_divisor = new_divisor;
 183                 fn_rebuild_zone(fz, old_ht, old_divisor);
 184                 fib_hash_genid++;
 185                 write_unlock_bh(&fib_hash_lock);
 186
 187                 fz_hash_free(old_ht, old_divisor);
 188         }
 189 }
 190
 191 static inline void fn_free_node(struct fib_node * f)
 192 {
 193         kmem_cache_free(fn_hash_kmem, f);
 194 }
 195
 196 static inline void fn_free_alias(struct fib_alias *fa, struct fib_node *f)
 197 {
 198         fib_release_info(fa->fa_info);
 199         if (fa == &f->fn_embedded_alias)
 200                 fa->fa_info = NULL;
 201         else
 202                 kmem_cache_free(fn_alias_kmem, fa);
 203 }
 204
 205 static struct fn_zone *
 206 fn_new_zone(struct fn_hash *table, int z)
 207 {
 208         int i;
 209         struct fn_zone *fz = kzalloc(sizeof(struct fn_zone), GFP_KERNEL);
 210         if (!fz)
 211                 return NULL;
 212
 213         if (z) {
 214                 fz->fz_divisor = 16;
 215         } else {
 216                 fz->fz_divisor = 1;
 217         }
 218         fz->fz_hashmask = (fz->fz_divisor - 1);
 219         fz->fz_hash = fz_hash_alloc(fz->fz_divisor);
 220         if (!fz->fz_hash) {
 221                 kfree(fz);
 222                 return NULL;
 223         }
 224         fz->fz_order = z;
 225         fz->fz_mask = inet_make_mask(z);
 226
 227         /* Find the first not empty zone with more specific mask */
 228         for (i=z+1; i<=32; i++)
 229                 if (table->fn_zones[i])
 230                         break;
 231         write_lock_bh(&fib_hash_lock);
 232         if (i>32) {
 233                 /* No more specific masks, we are the first. */
 234                 fz->fz_next = table->fn_zone_list;
 235                 table->fn_zone_list = fz;
 236         } else {
 237                 fz->fz_next = table->fn_zones[i]->fz_next;
 238                 table->fn_zones[i]->fz_next = fz;
 239         }
 240         table->fn_zones[z] = fz;
 241         fib_hash_genid++;
 242         write_unlock_bh(&fib_hash_lock);
 243         return fz;
 244 }
 245
 246 int fib_table_lookup(struct fib_table *tb,
 247                      const struct flowi *flp, struct fib_result *res)
 248 {
 249         int err;
 250         struct fn_zone *fz;
 251         struct fn_hash *t = (struct fn_hash *)tb->tb_data;
 252
 253         read_lock(&fib_hash_lock);
 254         for (fz = t->fn_zone_list; fz; fz = fz->fz_next) {
 255                 struct hlist_head *head;
 256                 struct hlist_node *node;
 257                 struct fib_node *f;
 258                 __be32 k = fz_key(flp->fl4_dst, fz);
 259
 260                 head = &fz->fz_hash[fn_hash(k, fz)];
 261                 hlist_for_each_entry(f, node, head, fn_hash) {
 262                         if (f->fn_key != k)
 263                                 continue;
 264
 265                         err = fib_semantic_match(&f->fn_alias,
 266                                                  flp, res,
 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 void fib_table_select_default(struct fib_table *tb,
 279                               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 int fib_table_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(cfg->fc_nlinfo.nl_net, -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(cfg->fc_nlinfo.nl_net, -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 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
 549 {
 550         struct fn_hash *table = (struct fn_hash *)tb->tb_data;
 551         struct fib_node *f;
 552         struct fib_alias *fa, *fa_to_delete;
 553         struct fn_zone *fz;
 554         __be32 key;
 555
 556         if (cfg->fc_dst_len > 32)
 557                 return -EINVAL;
 558
 559         if ((fz  = table->fn_zones[cfg->fc_dst_len]) == NULL)
 560                 return -ESRCH;
 561
 562         key = 0;
 563         if (cfg->fc_dst) {
 564                 if (cfg->fc_dst & ~FZ_MASK(fz))
 565                         return -EINVAL;
 566                 key = fz_key(cfg->fc_dst, fz);
 567         }
 568
 569         f = fib_find_node(fz, key);
 570
 571         if (!f)
 572                 fa = NULL;
 573         else
 574                 fa = fib_find_alias(&f->fn_alias, cfg->fc_tos, 0);
 575         if (!fa)
 576                 return -ESRCH;
 577
 578         fa_to_delete = NULL;
 579         fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 580         list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 581                 struct fib_info *fi = fa->fa_info;
 582
 583                 if (fa->fa_tos != cfg->fc_tos)
 584                         break;
 585
 586                 if ((!cfg->fc_type ||
 587                      fa->fa_type == cfg->fc_type) &&
 588                     (cfg->fc_scope == RT_SCOPE_NOWHERE ||
 589                      fa->fa_scope == cfg->fc_scope) &&
 590                     (!cfg->fc_protocol ||
 591                      fi->fib_protocol == cfg->fc_protocol) &&
 592                     fib_nh_match(cfg, fi) == 0) {
 593                         fa_to_delete = fa;
 594                         break;
 595                 }
 596         }
 597
 598         if (fa_to_delete) {
 599                 int kill_fn;
 600
 601                 fa = fa_to_delete;
 602                 rtmsg_fib(RTM_DELROUTE, key, fa, cfg->fc_dst_len,
 603                           tb->tb_id, &cfg->fc_nlinfo, 0);
 604
 605                 kill_fn = 0;
 606                 write_lock_bh(&fib_hash_lock);
 607                 list_del(&fa->fa_list);
 608                 if (list_empty(&f->fn_alias)) {
 609                         hlist_del(&f->fn_hash);
 610                         kill_fn = 1;
 611                 }
 612                 fib_hash_genid++;
 613                 write_unlock_bh(&fib_hash_lock);
 614
 615                 if (fa->fa_state & FA_S_ACCESSED)
 616                         rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
 617                 fn_free_alias(fa, f);
 618                 if (kill_fn) {
 619                         fn_free_node(f);
 620                         fz->fz_nent--;
 621                 }
 622
 623                 return 0;
 624         }
 625         return -ESRCH;
 626 }
 627
 628 static int fn_flush_list(struct fn_zone *fz, int idx)
 629 {
 630         struct hlist_head *head = &fz->fz_hash[idx];
 631         struct hlist_node *node, *n;
 632         struct fib_node *f;
 633         int found = 0;
 634
 635         hlist_for_each_entry_safe(f, node, n, head, fn_hash) {
 636                 struct fib_alias *fa, *fa_node;
 637                 int kill_f;
 638
 639                 kill_f = 0;
 640                 list_for_each_entry_safe(fa, fa_node, &f->fn_alias, fa_list) {
 641                         struct fib_info *fi = fa->fa_info;
 642
 643                         if (fi && (fi->fib_flags&RTNH_F_DEAD)) {
 644                                 write_lock_bh(&fib_hash_lock);
 645                                 list_del(&fa->fa_list);
 646                                 if (list_empty(&f->fn_alias)) {
 647                                         hlist_del(&f->fn_hash);
 648                                         kill_f = 1;
 649                                 }
 650                                 fib_hash_genid++;
 651                                 write_unlock_bh(&fib_hash_lock);
 652
 653                                 fn_free_alias(fa, f);
 654                                 found++;
 655                         }
 656                 }
 657                 if (kill_f) {
 658                         fn_free_node(f);
 659                         fz->fz_nent--;
 660                 }
 661         }
 662         return found;
 663 }
 664
 665 int fib_table_flush(struct fib_table *tb)
 666 {
 667         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 668         struct fn_zone *fz;
 669         int found = 0;
 670
 671         for (fz = table->fn_zone_list; fz; fz = fz->fz_next) {
 672                 int i;
 673
 674                 for (i = fz->fz_divisor - 1; i >= 0; i--)
 675                         found += fn_flush_list(fz, i);
 676         }
 677         return found;
 678 }
 679
 680
 681 static inline int
 682 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
 683                      struct fib_table *tb,
 684                      struct fn_zone *fz,
 685                      struct hlist_head *head)
 686 {
 687         struct hlist_node *node;
 688         struct fib_node *f;
 689         int i, s_i;
 690
 691         s_i = cb->args[4];
 692         i = 0;
 693         hlist_for_each_entry(f, node, head, fn_hash) {
 694                 struct fib_alias *fa;
 695
 696                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 697                         if (i < s_i)
 698                                 goto next;
 699
 700                         if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
 701                                           cb->nlh->nlmsg_seq,
 702                                           RTM_NEWROUTE,
 703                                           tb->tb_id,
 704                                           fa->fa_type,
 705                                           fa->fa_scope,
 706                                           f->fn_key,
 707                                           fz->fz_order,
 708                                           fa->fa_tos,
 709                                           fa->fa_info,
 710                                           NLM_F_MULTI) < 0) {
 711                                 cb->args[4] = i;
 712                                 return -1;
 713                         }
 714                 next:
 715                         i++;
 716                 }
 717         }
 718         cb->args[4] = i;
 719         return skb->len;
 720 }
 721
 722 static inline int
 723 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
 724                    struct fib_table *tb,
 725                    struct fn_zone *fz)
 726 {
 727         int h, s_h;
 728
 729         if (fz->fz_hash == NULL)
 730                 return skb->len;
 731         s_h = cb->args[3];
 732         for (h = s_h; h < fz->fz_divisor; h++) {
 733                 if (hlist_empty(&fz->fz_hash[h]))
 734                         continue;
 735                 if (fn_hash_dump_bucket(skb, cb, tb, fz, &fz->fz_hash[h]) < 0) {
 736                         cb->args[3] = h;
 737                         return -1;
 738                 }
 739                 memset(&cb->args[4], 0,
 740                        sizeof(cb->args) - 4*sizeof(cb->args[0]));
 741         }
 742         cb->args[3] = h;
 743         return skb->len;
 744 }
 745
 746 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
 747                    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
 792         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 793         return tb;
 794 }
 795
 796 /* ------------------------------------------------------------------------ */
 797 #ifdef CONFIG_PROC_FS
 798
 799 struct fib_iter_state {
 800         struct seq_net_private p;
 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 fib_table *main_table;
 815         struct fn_hash *table;
 816
 817         main_table = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
 818         table = (struct fn_hash *)main_table->tb_data;
 819
 820         iter->bucket    = 0;
 821         iter->hash_head = NULL;
 822         iter->fn        = NULL;
 823         iter->fa        = NULL;
 824         iter->pos       = 0;
 825         iter->genid     = fib_hash_genid;
 826         iter->valid     = 1;
 827
 828         for (iter->zone = table->fn_zone_list; iter->zone;
 829              iter->zone = iter->zone->fz_next) {
 830                 int maxslot;
 831
 832                 if (!iter->zone->fz_nent)
 833                         continue;
 834
 835                 iter->hash_head = iter->zone->fz_hash;
 836                 maxslot = iter->zone->fz_divisor;
 837
 838                 for (iter->bucket = 0; iter->bucket < maxslot;
 839                      ++iter->bucket, ++iter->hash_head) {
 840                         struct hlist_node *node;
 841                         struct fib_node *fn;
 842
 843                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 844                                 struct fib_alias *fa;
 845
 846                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 847                                         iter->fn = fn;
 848                                         iter->fa = fa;
 849                                         goto out;
 850                                 }
 851                         }
 852                 }
 853         }
 854 out:
 855         return iter->fa;
 856 }
 857
 858 static struct fib_alias *fib_get_next(struct seq_file *seq)
 859 {
 860         struct fib_iter_state *iter = seq->private;
 861         struct fib_node *fn;
 862         struct fib_alias *fa;
 863
 864         /* Advance FA, if any. */
 865         fn = iter->fn;
 866         fa = iter->fa;
 867         if (fa) {
 868                 BUG_ON(!fn);
 869                 list_for_each_entry_continue(fa, &fn->fn_alias, fa_list) {
 870                         iter->fa = fa;
 871                         goto out;
 872                 }
 873         }
 874
 875         fa = iter->fa = NULL;
 876
 877         /* Advance FN. */
 878         if (fn) {
 879                 struct hlist_node *node = &fn->fn_hash;
 880                 hlist_for_each_entry_continue(fn, node, fn_hash) {
 881                         iter->fn = fn;
 882
 883                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 884                                 iter->fa = fa;
 885                                 goto out;
 886                         }
 887                 }
 888         }
 889
 890         fn = iter->fn = NULL;
 891
 892         /* Advance hash chain. */
 893         if (!iter->zone)
 894                 goto out;
 895
 896         for (;;) {
 897                 struct hlist_node *node;
 898                 int maxslot;
 899
 900                 maxslot = iter->zone->fz_divisor;
 901
 902                 while (++iter->bucket < maxslot) {
 903                         iter->hash_head++;
 904
 905                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 906                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 907                                         iter->fn = fn;
 908                                         iter->fa = fa;
 909                                         goto out;
 910                                 }
 911                         }
 912                 }
 913
 914                 iter->zone = iter->zone->fz_next;
 915
 916                 if (!iter->zone)
 917                         goto out;
 918
 919                 iter->bucket = 0;
 920                 iter->hash_head = iter->zone->fz_hash;
 921
 922                 hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 923                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 924                                 iter->fn = fn;
 925                                 iter->fa = fa;
 926                                 goto out;
 927                         }
 928                 }
 929         }
 930 out:
 931         iter->pos++;
 932         return fa;
 933 }
 934
 935 static struct fib_alias *fib_get_idx(struct seq_file *seq, loff_t pos)
 936 {
 937         struct fib_iter_state *iter = seq->private;
 938         struct fib_alias *fa;
 939
 940         if (iter->valid && pos >= iter->pos && iter->genid == fib_hash_genid) {
 941                 fa   = iter->fa;
 942                 pos -= iter->pos;
 943         } else
 944                 fa = fib_get_first(seq);
 945
 946         if (fa)
 947                 while (pos && (fa = fib_get_next(seq)))
 948                         --pos;
 949         return pos ? NULL : fa;
 950 }
 951
 952 static void *fib_seq_start(struct seq_file *seq, loff_t *pos)
 953         __acquires(fib_hash_lock)
 954 {
 955         void *v = NULL;
 956
 957         read_lock(&fib_hash_lock);
 958         if (fib_get_table(seq_file_net(seq), RT_TABLE_MAIN))
 959                 v = *pos ? fib_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 960         return v;
 961 }
 962
 963 static void *fib_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 964 {
 965         ++*pos;
 966         return v == SEQ_START_TOKEN ? fib_get_first(seq) : fib_get_next(seq);
 967 }
 968
 969 static void fib_seq_stop(struct seq_file *seq, void *v)
 970         __releases(fib_hash_lock)
 971 {
 972         read_unlock(&fib_hash_lock);
 973 }
 974
 975 static unsigned fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
 976 {
 977         static const unsigned type2flags[RTN_MAX + 1] = {
 978                 [7] = RTF_REJECT, [8] = RTF_REJECT,
 979         };
 980         unsigned flags = type2flags[type];
 981
 982         if (fi && fi->fib_nh->nh_gw)
 983                 flags |= RTF_GATEWAY;
 984         if (mask == htonl(0xFFFFFFFF))
 985                 flags |= RTF_HOST;
 986         flags |= RTF_UP;
 987         return flags;
 988 }
 989
 990 /*
 991  *      This outputs /proc/net/route.
 992  *
 993  *      It always works in backward compatibility mode.
 994  *      The format of the file is not supposed to be changed.
 995  */
 996 static int fib_seq_show(struct seq_file *seq, void *v)
 997 {
 998         struct fib_iter_state *iter;
 999         int len;
1000         __be32 prefix, mask;
1001         unsigned flags;
1002         struct fib_node *f;
1003         struct fib_alias *fa;
1004         struct fib_info *fi;
1005
1006         if (v == SEQ_START_TOKEN) {
1007                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
1008                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
1009                            "\tWindow\tIRTT");
1010                 goto out;
1011         }
1012
1013         iter    = seq->private;
1014         f       = iter->fn;
1015         fa      = iter->fa;
1016         fi      = fa->fa_info;
1017         prefix  = f->fn_key;
1018         mask    = FZ_MASK(iter->zone);
1019         flags   = fib_flag_trans(fa->fa_type, mask, fi);
1020         if (fi)
1021                 seq_printf(seq,
1022                          "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u%n",
1023                          fi->fib_dev ? fi->fib_dev->name : "*", prefix,
1024                          fi->fib_nh->nh_gw, flags, 0, 0, fi->fib_priority,
1025                          mask, (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
1026                          fi->fib_window,
1027                          fi->fib_rtt >> 3, &len);
1028         else
1029                 seq_printf(seq,
1030                          "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u%n",
1031                          prefix, 0, flags, 0, 0, 0, mask, 0, 0, 0, &len);
1032
1033         seq_printf(seq, "%*s\n", 127 - len, "");
1034 out:
1035         return 0;
1036 }
1037
1038 static const struct seq_operations fib_seq_ops = {
1039         .start  = fib_seq_start,
1040         .next   = fib_seq_next,
1041         .stop   = fib_seq_stop,
1042         .show   = fib_seq_show,
1043 };
1044
1045 static int fib_seq_open(struct inode *inode, struct file *file)
1046 {
1047         return seq_open_net(inode, file, &fib_seq_ops,
1048                             sizeof(struct fib_iter_state));
1049 }
1050
1051 static const struct file_operations fib_seq_fops = {
1052         .owner          = THIS_MODULE,
1053         .open           = fib_seq_open,
1054         .read           = seq_read,
1055         .llseek         = seq_lseek,
1056         .release        = seq_release_net,
1057 };
1058
1059 int __net_init fib_proc_init(struct net *net)
1060 {
1061         if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_seq_fops))
1062                 return -ENOMEM;
1063         return 0;
1064 }
1065
1066 void __net_exit fib_proc_exit(struct net *net)
1067 {
1068         proc_net_remove(net, "route");
1069 }
1070 #endif /* CONFIG_PROC_FS */