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/mm.h>
  24 #include <linux/string.h>
  25 #include <linux/socket.h>
  26 #include <linux/sockios.h>
  27 #include <linux/errno.h>
  28 #include <linux/in.h>
  29 #include <linux/inet.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/netdevice.h>
  32 #include <linux/if_arp.h>
  33 #include <linux/proc_fs.h>
  34 #include <linux/skbuff.h>
  35 #include <linux/netlink.h>
  36 #include <linux/init.h>
  37
  38 #include <net/net_namespace.h>
  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 struct kmem_cache *fn_hash_kmem __read_mostly;
  49 static struct kmem_cache *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                         rtmsg_fib(RTM_NEWROUTE, key, fa, cfg->fc_dst_len, tb->tb_id,
 461                                   &cfg->fc_nlinfo, NLM_F_REPLACE);
 462                         return 0;
 463                 }
 464
 465                 /* Error if we find a perfect match which
 466                  * uses the same scope, type, and nexthop
 467                  * information.
 468                  */
 469                 fa_orig = fa;
 470                 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 471                 list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 472                         if (fa->fa_tos != tos)
 473                                 break;
 474                         if (fa->fa_info->fib_priority != fi->fib_priority)
 475                                 break;
 476                         if (fa->fa_type == cfg->fc_type &&
 477                             fa->fa_scope == cfg->fc_scope &&
 478                             fa->fa_info == fi)
 479                                 goto out;
 480                 }
 481                 if (!(cfg->fc_nlflags & NLM_F_APPEND))
 482                         fa = fa_orig;
 483         }
 484
 485         err = -ENOENT;
 486         if (!(cfg->fc_nlflags & NLM_F_CREATE))
 487                 goto out;
 488
 489         err = -ENOBUFS;
 490         new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
 491         if (new_fa == NULL)
 492                 goto out;
 493
 494         new_f = NULL;
 495         if (!f) {
 496                 new_f = kmem_cache_alloc(fn_hash_kmem, GFP_KERNEL);
 497                 if (new_f == NULL)
 498                         goto out_free_new_fa;
 499
 500                 INIT_HLIST_NODE(&new_f->fn_hash);
 501                 INIT_LIST_HEAD(&new_f->fn_alias);
 502                 new_f->fn_key = key;
 503                 f = new_f;
 504         }
 505
 506         new_fa->fa_info = fi;
 507         new_fa->fa_tos = tos;
 508         new_fa->fa_type = cfg->fc_type;
 509         new_fa->fa_scope = cfg->fc_scope;
 510         new_fa->fa_state = 0;
 511
 512         /*
 513          * Insert new entry to the list.
 514          */
 515
 516         write_lock_bh(&fib_hash_lock);
 517         if (new_f)
 518                 fib_insert_node(fz, new_f);
 519         list_add_tail(&new_fa->fa_list,
 520                  (fa ? &fa->fa_list : &f->fn_alias));
 521         fib_hash_genid++;
 522         write_unlock_bh(&fib_hash_lock);
 523
 524         if (new_f)
 525                 fz->fz_nent++;
 526         rt_cache_flush(-1);
 527
 528         rtmsg_fib(RTM_NEWROUTE, key, new_fa, cfg->fc_dst_len, tb->tb_id,
 529                   &cfg->fc_nlinfo, 0);
 530         return 0;
 531
 532 out_free_new_fa:
 533         kmem_cache_free(fn_alias_kmem, new_fa);
 534 out:
 535         fib_release_info(fi);
 536         return err;
 537 }
 538
 539
 540 static int fn_hash_delete(struct fib_table *tb, struct fib_config *cfg)
 541 {
 542         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 543         struct fib_node *f;
 544         struct fib_alias *fa, *fa_to_delete;
 545         struct fn_zone *fz;
 546         __be32 key;
 547
 548         if (cfg->fc_dst_len > 32)
 549                 return -EINVAL;
 550
 551         if ((fz  = table->fn_zones[cfg->fc_dst_len]) == NULL)
 552                 return -ESRCH;
 553
 554         key = 0;
 555         if (cfg->fc_dst) {
 556                 if (cfg->fc_dst & ~FZ_MASK(fz))
 557                         return -EINVAL;
 558                 key = fz_key(cfg->fc_dst, fz);
 559         }
 560
 561         f = fib_find_node(fz, key);
 562
 563         if (!f)
 564                 fa = NULL;
 565         else
 566                 fa = fib_find_alias(&f->fn_alias, cfg->fc_tos, 0);
 567         if (!fa)
 568                 return -ESRCH;
 569
 570         fa_to_delete = NULL;
 571         fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 572         list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 573                 struct fib_info *fi = fa->fa_info;
 574
 575                 if (fa->fa_tos != cfg->fc_tos)
 576                         break;
 577
 578                 if ((!cfg->fc_type ||
 579                      fa->fa_type == cfg->fc_type) &&
 580                     (cfg->fc_scope == RT_SCOPE_NOWHERE ||
 581                      fa->fa_scope == cfg->fc_scope) &&
 582                     (!cfg->fc_protocol ||
 583                      fi->fib_protocol == cfg->fc_protocol) &&
 584                     fib_nh_match(cfg, fi) == 0) {
 585                         fa_to_delete = fa;
 586                         break;
 587                 }
 588         }
 589
 590         if (fa_to_delete) {
 591                 int kill_fn;
 592
 593                 fa = fa_to_delete;
 594                 rtmsg_fib(RTM_DELROUTE, key, fa, cfg->fc_dst_len,
 595                           tb->tb_id, &cfg->fc_nlinfo, 0);
 596
 597                 kill_fn = 0;
 598                 write_lock_bh(&fib_hash_lock);
 599                 list_del(&fa->fa_list);
 600                 if (list_empty(&f->fn_alias)) {
 601                         hlist_del(&f->fn_hash);
 602                         kill_fn = 1;
 603                 }
 604                 fib_hash_genid++;
 605                 write_unlock_bh(&fib_hash_lock);
 606
 607                 if (fa->fa_state & FA_S_ACCESSED)
 608                         rt_cache_flush(-1);
 609                 fn_free_alias(fa);
 610                 if (kill_fn) {
 611                         fn_free_node(f);
 612                         fz->fz_nent--;
 613                 }
 614
 615                 return 0;
 616         }
 617         return -ESRCH;
 618 }
 619
 620 static int fn_flush_list(struct fn_zone *fz, int idx)
 621 {
 622         struct hlist_head *head = &fz->fz_hash[idx];
 623         struct hlist_node *node, *n;
 624         struct fib_node *f;
 625         int found = 0;
 626
 627         hlist_for_each_entry_safe(f, node, n, head, fn_hash) {
 628                 struct fib_alias *fa, *fa_node;
 629                 int kill_f;
 630
 631                 kill_f = 0;
 632                 list_for_each_entry_safe(fa, fa_node, &f->fn_alias, fa_list) {
 633                         struct fib_info *fi = fa->fa_info;
 634
 635                         if (fi && (fi->fib_flags&RTNH_F_DEAD)) {
 636                                 write_lock_bh(&fib_hash_lock);
 637                                 list_del(&fa->fa_list);
 638                                 if (list_empty(&f->fn_alias)) {
 639                                         hlist_del(&f->fn_hash);
 640                                         kill_f = 1;
 641                                 }
 642                                 fib_hash_genid++;
 643                                 write_unlock_bh(&fib_hash_lock);
 644
 645                                 fn_free_alias(fa);
 646                                 found++;
 647                         }
 648                 }
 649                 if (kill_f) {
 650                         fn_free_node(f);
 651                         fz->fz_nent--;
 652                 }
 653         }
 654         return found;
 655 }
 656
 657 static int fn_hash_flush(struct fib_table *tb)
 658 {
 659         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 660         struct fn_zone *fz;
 661         int found = 0;
 662
 663         for (fz = table->fn_zone_list; fz; fz = fz->fz_next) {
 664                 int i;
 665
 666                 for (i = fz->fz_divisor - 1; i >= 0; i--)
 667                         found += fn_flush_list(fz, i);
 668         }
 669         return found;
 670 }
 671
 672
 673 static inline int
 674 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
 675                      struct fib_table *tb,
 676                      struct fn_zone *fz,
 677                      struct hlist_head *head)
 678 {
 679         struct hlist_node *node;
 680         struct fib_node *f;
 681         int i, s_i;
 682
 683         s_i = cb->args[4];
 684         i = 0;
 685         hlist_for_each_entry(f, node, head, fn_hash) {
 686                 struct fib_alias *fa;
 687
 688                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 689                         if (i < s_i)
 690                                 goto next;
 691
 692                         if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
 693                                           cb->nlh->nlmsg_seq,
 694                                           RTM_NEWROUTE,
 695                                           tb->tb_id,
 696                                           fa->fa_type,
 697                                           fa->fa_scope,
 698                                           f->fn_key,
 699                                           fz->fz_order,
 700                                           fa->fa_tos,
 701                                           fa->fa_info,
 702                                           NLM_F_MULTI) < 0) {
 703                                 cb->args[4] = i;
 704                                 return -1;
 705                         }
 706                 next:
 707                         i++;
 708                 }
 709         }
 710         cb->args[4] = i;
 711         return skb->len;
 712 }
 713
 714 static inline int
 715 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
 716                    struct fib_table *tb,
 717                    struct fn_zone *fz)
 718 {
 719         int h, s_h;
 720
 721         s_h = cb->args[3];
 722         for (h=0; h < fz->fz_divisor; h++) {
 723                 if (h < s_h) continue;
 724                 if (h > s_h)
 725                         memset(&cb->args[4], 0,
 726                                sizeof(cb->args) - 4*sizeof(cb->args[0]));
 727                 if (fz->fz_hash == NULL ||
 728                     hlist_empty(&fz->fz_hash[h]))
 729                         continue;
 730                 if (fn_hash_dump_bucket(skb, cb, tb, fz, &fz->fz_hash[h])<0) {
 731                         cb->args[3] = h;
 732                         return -1;
 733                 }
 734         }
 735         cb->args[3] = h;
 736         return skb->len;
 737 }
 738
 739 static int fn_hash_dump(struct fib_table *tb, struct sk_buff *skb, struct netlink_callback *cb)
 740 {
 741         int m, s_m;
 742         struct fn_zone *fz;
 743         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 744
 745         s_m = cb->args[2];
 746         read_lock(&fib_hash_lock);
 747         for (fz = table->fn_zone_list, m=0; fz; fz = fz->fz_next, m++) {
 748                 if (m < s_m) continue;
 749                 if (m > s_m)
 750                         memset(&cb->args[3], 0,
 751                                sizeof(cb->args) - 3*sizeof(cb->args[0]));
 752                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
 753                         cb->args[2] = m;
 754                         read_unlock(&fib_hash_lock);
 755                         return -1;
 756                 }
 757         }
 758         read_unlock(&fib_hash_lock);
 759         cb->args[2] = m;
 760         return skb->len;
 761 }
 762
 763 #ifdef CONFIG_IP_MULTIPLE_TABLES
 764 struct fib_table * fib_hash_init(u32 id)
 765 #else
 766 struct fib_table * __init fib_hash_init(u32 id)
 767 #endif
 768 {
 769         struct fib_table *tb;
 770
 771         if (fn_hash_kmem == NULL)
 772                 fn_hash_kmem = kmem_cache_create("ip_fib_hash",
 773                                                  sizeof(struct fib_node),
 774                                                  0, SLAB_HWCACHE_ALIGN,
 775                                                  NULL);
 776
 777         if (fn_alias_kmem == NULL)
 778                 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
 779                                                   sizeof(struct fib_alias),
 780                                                   0, SLAB_HWCACHE_ALIGN,
 781                                                   NULL);
 782
 783         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash),
 784                      GFP_KERNEL);
 785         if (tb == NULL)
 786                 return NULL;
 787
 788         tb->tb_id = id;
 789         tb->tb_lookup = fn_hash_lookup;
 790         tb->tb_insert = fn_hash_insert;
 791         tb->tb_delete = fn_hash_delete;
 792         tb->tb_flush = fn_hash_flush;
 793         tb->tb_select_default = fn_hash_select_default;
 794         tb->tb_dump = fn_hash_dump;
 795         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 796         return tb;
 797 }
 798
 799 /* ------------------------------------------------------------------------ */
 800 #ifdef CONFIG_PROC_FS
 801
 802 struct fib_iter_state {
 803         struct fn_zone  *zone;
 804         int             bucket;
 805         struct hlist_head *hash_head;
 806         struct fib_node *fn;
 807         struct fib_alias *fa;
 808         loff_t pos;
 809         unsigned int genid;
 810         int valid;
 811 };
 812
 813 static struct fib_alias *fib_get_first(struct seq_file *seq)
 814 {
 815         struct fib_iter_state *iter = seq->private;
 816         struct fn_hash *table = (struct fn_hash *) ip_fib_main_table->tb_data;
 817
 818         iter->bucket    = 0;
 819         iter->hash_head = NULL;
 820         iter->fn        = NULL;
 821         iter->fa        = NULL;
 822         iter->pos       = 0;
 823         iter->genid     = fib_hash_genid;
 824         iter->valid     = 1;
 825
 826         for (iter->zone = table->fn_zone_list; iter->zone;
 827              iter->zone = iter->zone->fz_next) {
 828                 int maxslot;
 829
 830                 if (!iter->zone->fz_nent)
 831                         continue;
 832
 833                 iter->hash_head = iter->zone->fz_hash;
 834                 maxslot = iter->zone->fz_divisor;
 835
 836                 for (iter->bucket = 0; iter->bucket < maxslot;
 837                      ++iter->bucket, ++iter->hash_head) {
 838                         struct hlist_node *node;
 839                         struct fib_node *fn;
 840
 841                         hlist_for_each_entry(fn,node,iter->hash_head,fn_hash) {
 842                                 struct fib_alias *fa;
 843
 844                                 list_for_each_entry(fa,&fn->fn_alias,fa_list) {
 845                                         iter->fn = fn;
 846                                         iter->fa = fa;
 847                                         goto out;
 848                                 }
 849                         }
 850                 }
 851         }
 852 out:
 853         return iter->fa;
 854 }
 855
 856 static struct fib_alias *fib_get_next(struct seq_file *seq)
 857 {
 858         struct fib_iter_state *iter = seq->private;
 859         struct fib_node *fn;
 860         struct fib_alias *fa;
 861
 862         /* Advance FA, if any. */
 863         fn = iter->fn;
 864         fa = iter->fa;
 865         if (fa) {
 866                 BUG_ON(!fn);
 867                 list_for_each_entry_continue(fa, &fn->fn_alias, fa_list) {
 868                         iter->fa = fa;
 869                         goto out;
 870                 }
 871         }
 872
 873         fa = iter->fa = NULL;
 874
 875         /* Advance FN. */
 876         if (fn) {
 877                 struct hlist_node *node = &fn->fn_hash;
 878                 hlist_for_each_entry_continue(fn, node, fn_hash) {
 879                         iter->fn = fn;
 880
 881                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 882                                 iter->fa = fa;
 883                                 goto out;
 884                         }
 885                 }
 886         }
 887
 888         fn = iter->fn = NULL;
 889
 890         /* Advance hash chain. */
 891         if (!iter->zone)
 892                 goto out;
 893
 894         for (;;) {
 895                 struct hlist_node *node;
 896                 int maxslot;
 897
 898                 maxslot = iter->zone->fz_divisor;
 899
 900                 while (++iter->bucket < maxslot) {
 901                         iter->hash_head++;
 902
 903                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 904                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 905                                         iter->fn = fn;
 906                                         iter->fa = fa;
 907                                         goto out;
 908                                 }
 909                         }
 910                 }
 911
 912                 iter->zone = iter->zone->fz_next;
 913
 914                 if (!iter->zone)
 915                         goto out;
 916
 917                 iter->bucket = 0;
 918                 iter->hash_head = iter->zone->fz_hash;
 919
 920                 hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 921                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 922                                 iter->fn = fn;
 923                                 iter->fa = fa;
 924                                 goto out;
 925                         }
 926                 }
 927         }
 928 out:
 929         iter->pos++;
 930         return fa;
 931 }
 932
 933 static struct fib_alias *fib_get_idx(struct seq_file *seq, loff_t pos)
 934 {
 935         struct fib_iter_state *iter = seq->private;
 936         struct fib_alias *fa;
 937
 938         if (iter->valid && pos >= iter->pos && iter->genid == fib_hash_genid) {
 939                 fa   = iter->fa;
 940                 pos -= iter->pos;
 941         } else
 942                 fa = fib_get_first(seq);
 943
 944         if (fa)
 945                 while (pos && (fa = fib_get_next(seq)))
 946                         --pos;
 947         return pos ? NULL : fa;
 948 }
 949
 950 static void *fib_seq_start(struct seq_file *seq, loff_t *pos)
 951 {
 952         void *v = NULL;
 953
 954         read_lock(&fib_hash_lock);
 955         if (ip_fib_main_table)
 956                 v = *pos ? fib_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 957         return v;
 958 }
 959
 960 static void *fib_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 961 {
 962         ++*pos;
 963         return v == SEQ_START_TOKEN ? fib_get_first(seq) : fib_get_next(seq);
 964 }
 965
 966 static void fib_seq_stop(struct seq_file *seq, void *v)
 967 {
 968         read_unlock(&fib_hash_lock);
 969 }
 970
 971 static unsigned fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
 972 {
 973         static const unsigned type2flags[RTN_MAX + 1] = {
 974                 [7] = RTF_REJECT, [8] = RTF_REJECT,
 975         };
 976         unsigned flags = type2flags[type];
 977
 978         if (fi && fi->fib_nh->nh_gw)
 979                 flags |= RTF_GATEWAY;
 980         if (mask == htonl(0xFFFFFFFF))
 981                 flags |= RTF_HOST;
 982         flags |= RTF_UP;
 983         return flags;
 984 }
 985
 986 /*
 987  *      This outputs /proc/net/route.
 988  *
 989  *      It always works in backward compatibility mode.
 990  *      The format of the file is not supposed to be changed.
 991  */
 992 static int fib_seq_show(struct seq_file *seq, void *v)
 993 {
 994         struct fib_iter_state *iter;
 995         char bf[128];
 996         __be32 prefix, mask;
 997         unsigned flags;
 998         struct fib_node *f;
 999         struct fib_alias *fa;
1000         struct fib_info *fi;
1001
1002         if (v == SEQ_START_TOKEN) {
1003                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
1004                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
1005                            "\tWindow\tIRTT");
1006                 goto out;
1007         }
1008
1009         iter    = seq->private;
1010         f       = iter->fn;
1011         fa      = iter->fa;
1012         fi      = fa->fa_info;
1013         prefix  = f->fn_key;
1014         mask    = FZ_MASK(iter->zone);
1015         flags   = fib_flag_trans(fa->fa_type, mask, fi);
1016         if (fi)
1017                 snprintf(bf, sizeof(bf),
1018                          "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1019                          fi->fib_dev ? fi->fib_dev->name : "*", prefix,
1020                          fi->fib_nh->nh_gw, flags, 0, 0, fi->fib_priority,
1021                          mask, (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
1022                          fi->fib_window,
1023                          fi->fib_rtt >> 3);
1024         else
1025                 snprintf(bf, sizeof(bf),
1026                          "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1027                          prefix, 0, flags, 0, 0, 0, mask, 0, 0, 0);
1028         seq_printf(seq, "%-127s\n", bf);
1029 out:
1030         return 0;
1031 }
1032
1033 static const struct seq_operations fib_seq_ops = {
1034         .start  = fib_seq_start,
1035         .next   = fib_seq_next,
1036         .stop   = fib_seq_stop,
1037         .show   = fib_seq_show,
1038 };
1039
1040 static int fib_seq_open(struct inode *inode, struct file *file)
1041 {
1042         return seq_open_private(file, &fib_seq_ops,
1043                         sizeof(struct fib_iter_state));
1044 }
1045
1046 static const struct file_operations fib_seq_fops = {
1047         .owner          = THIS_MODULE,
1048         .open           = fib_seq_open,
1049         .read           = seq_read,
1050         .llseek         = seq_lseek,
1051         .release        = seq_release_private,
1052 };
1053
1054 int __init fib_proc_init(void)
1055 {
1056         if (!proc_net_fops_create(&init_net, "route", S_IRUGO, &fib_seq_fops))
1057                 return -ENOMEM;
1058         return 0;
1059 }
1060
1061 void __init fib_proc_exit(void)
1062 {
1063         proc_net_remove(&init_net, "route");
1064 }
1065 #endif /* CONFIG_PROC_FS */