net/ipv4/fib_hash.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              IPv4 FIB: lookup engine and maintenance routines.
   7  *
   8  * Version:     $Id: fib_hash.c,v 1.13 2001/10/31 21:55:54 davem Exp $
   9  *
  10  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  11  *
  12  *              This program is free software; you can redistribute it and/or
  13  *              modify it under the terms of the GNU General Public License
  14  *              as published by the Free Software Foundation; either version
  15  *              2 of the License, or (at your option) any later version.
  16  */
  17
  18 #include <asm/uaccess.h>
  19 #include <asm/system.h>
  20 #include <linux/bitops.h>
  21 #include <linux/types.h>
  22 #include <linux/kernel.h>
  23 #include <linux/sched.h>
  24 #include <linux/mm.h>
  25 #include <linux/string.h>
  26 #include <linux/socket.h>
  27 #include <linux/sockios.h>
  28 #include <linux/errno.h>
  29 #include <linux/in.h>
  30 #include <linux/inet.h>
  31 #include <linux/inetdevice.h>
  32 #include <linux/netdevice.h>
  33 #include <linux/if_arp.h>
  34 #include <linux/proc_fs.h>
  35 #include <linux/skbuff.h>
  36 #include <linux/netlink.h>
  37 #include <linux/init.h>
  38 #include <linux/vs_context.h>
  39
  40 #include <net/ip.h>
  41 #include <net/protocol.h>
  42 #include <net/route.h>
  43 #include <net/tcp.h>
  44 #include <net/sock.h>
  45 #include <net/ip_fib.h>
  46
  47 #include "fib_lookup.h"
  48
  49 static struct kmem_cache *fn_hash_kmem __read_mostly;
  50 static struct kmem_cache *fn_alias_kmem __read_mostly;
  51
  52 struct fib_node {
  53         struct hlist_node       fn_hash;
  54         struct list_head        fn_alias;
  55         __be32                  fn_key;
  56 };
  57
  58 struct fn_zone {
  59         struct fn_zone          *fz_next;       /* Next not empty zone  */
  60         struct hlist_head       *fz_hash;       /* Hash table pointer   */
  61         int                     fz_nent;        /* Number of entries    */
  62
  63         int                     fz_divisor;     /* Hash divisor         */
  64         u32                     fz_hashmask;    /* (fz_divisor - 1)     */
  65 #define FZ_HASHMASK(fz)         ((fz)->fz_hashmask)
  66
  67         int                     fz_order;       /* Zone order           */
  68         __be32                  fz_mask;
  69 #define FZ_MASK(fz)             ((fz)->fz_mask)
  70 };
  71
  72 /* NOTE. On fast computers evaluation of fz_hashmask and fz_mask
  73  * can be cheaper than memory lookup, so that FZ_* macros are used.
  74  */
  75
  76 struct fn_hash {
  77         struct fn_zone  *fn_zones[33];
  78         struct fn_zone  *fn_zone_list;
  79 };
  80
  81 static inline u32 fn_hash(__be32 key, struct fn_zone *fz)
  82 {
  83         u32 h = ntohl(key)>>(32 - fz->fz_order);
  84         h ^= (h>>20);
  85         h ^= (h>>10);
  86         h ^= (h>>5);
  87         h &= FZ_HASHMASK(fz);
  88         return h;
  89 }
  90
  91 static inline __be32 fz_key(__be32 dst, struct fn_zone *fz)
  92 {
  93         return dst & FZ_MASK(fz);
  94 }
  95
  96 static DEFINE_RWLOCK(fib_hash_lock);
  97 static unsigned int fib_hash_genid;
  98
  99 #define FZ_MAX_DIVISOR ((PAGE_SIZE<<MAX_ORDER) / sizeof(struct hlist_head))
 100
 101 static struct hlist_head *fz_hash_alloc(int divisor)
 102 {
 103         unsigned long size = divisor * sizeof(struct hlist_head);
 104
 105         if (size <= PAGE_SIZE) {
 106                 return kmalloc(size, GFP_KERNEL);
 107         } else {
 108                 return (struct hlist_head *)
 109                         __get_free_pages(GFP_KERNEL, get_order(size));
 110         }
 111 }
 112
 113 /* The fib hash lock must be held when this is called. */
 114 static inline void fn_rebuild_zone(struct fn_zone *fz,
 115                                    struct hlist_head *old_ht,
 116                                    int old_divisor)
 117 {
 118         int i;
 119
 120         for (i = 0; i < old_divisor; i++) {
 121                 struct hlist_node *node, *n;
 122                 struct fib_node *f;
 123
 124                 hlist_for_each_entry_safe(f, node, n, &old_ht[i], fn_hash) {
 125                         struct hlist_head *new_head;
 126
 127                         hlist_del(&f->fn_hash);
 128
 129                         new_head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 130                         hlist_add_head(&f->fn_hash, new_head);
 131                 }
 132         }
 133 }
 134
 135 static void fz_hash_free(struct hlist_head *hash, int divisor)
 136 {
 137         unsigned long size = divisor * sizeof(struct hlist_head);
 138
 139         if (size <= PAGE_SIZE)
 140                 kfree(hash);
 141         else
 142                 free_pages((unsigned long)hash, get_order(size));
 143 }
 144
 145 static void fn_rehash_zone(struct fn_zone *fz)
 146 {
 147         struct hlist_head *ht, *old_ht;
 148         int old_divisor, new_divisor;
 149         u32 new_hashmask;
 150
 151         old_divisor = fz->fz_divisor;
 152
 153         switch (old_divisor) {
 154         case 16:
 155                 new_divisor = 256;
 156                 break;
 157         case 256:
 158                 new_divisor = 1024;
 159                 break;
 160         default:
 161                 if ((old_divisor << 1) > FZ_MAX_DIVISOR) {
 162                         printk(KERN_CRIT "route.c: bad divisor %d!\n", old_divisor);
 163                         return;
 164                 }
 165                 new_divisor = (old_divisor << 1);
 166                 break;
 167         }
 168
 169         new_hashmask = (new_divisor - 1);
 170
 171 #if RT_CACHE_DEBUG >= 2
 172         printk("fn_rehash_zone: hash for zone %d grows from %d\n", fz->fz_order, old_divisor);
 173 #endif
 174
 175         ht = fz_hash_alloc(new_divisor);
 176
 177         if (ht) {
 178                 memset(ht, 0, new_divisor * sizeof(struct hlist_head));
 179
 180                 write_lock_bh(&fib_hash_lock);
 181                 old_ht = fz->fz_hash;
 182                 fz->fz_hash = ht;
 183                 fz->fz_hashmask = new_hashmask;
 184                 fz->fz_divisor = new_divisor;
 185                 fn_rebuild_zone(fz, old_ht, old_divisor);
 186                 fib_hash_genid++;
 187                 write_unlock_bh(&fib_hash_lock);
 188
 189                 fz_hash_free(old_ht, old_divisor);
 190         }
 191 }
 192
 193 static inline void fn_free_node(struct fib_node * f)
 194 {
 195         kmem_cache_free(fn_hash_kmem, f);
 196 }
 197
 198 static inline void fn_free_alias(struct fib_alias *fa)
 199 {
 200         fib_release_info(fa->fa_info);
 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         memset(fz->fz_hash, 0, fz->fz_divisor * sizeof(struct hlist_head *));
 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 static int
 247 fn_hash_lookup(struct fib_table *tb, 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                                                  f->fn_key, fz->fz_mask,
 268                                                  fz->fz_order);
 269                         if (err <= 0)
 270                                 goto out;
 271                 }
 272         }
 273         err = 1;
 274 out:
 275         read_unlock(&fib_hash_lock);
 276         return err;
 277 }
 278
 279 static int fn_hash_last_dflt=-1;
 280
 281 static void
 282 fn_hash_select_default(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
 283 {
 284         int order, last_idx;
 285         struct hlist_node *node;
 286         struct fib_node *f;
 287         struct fib_info *fi = NULL;
 288         struct fib_info *last_resort;
 289         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 290         struct fn_zone *fz = t->fn_zones[0];
 291
 292         if (fz == NULL)
 293                 return;
 294
 295         last_idx = -1;
 296         last_resort = NULL;
 297         order = -1;
 298
 299         read_lock(&fib_hash_lock);
 300         hlist_for_each_entry(f, node, &fz->fz_hash[0], fn_hash) {
 301                 struct fib_alias *fa;
 302
 303                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 304                         struct fib_info *next_fi = fa->fa_info;
 305
 306                         if (fa->fa_scope != res->scope ||
 307                             fa->fa_type != RTN_UNICAST)
 308                                 continue;
 309
 310                         if (next_fi->fib_priority > res->fi->fib_priority)
 311                                 break;
 312                         if (!next_fi->fib_nh[0].nh_gw ||
 313                             next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
 314                                 continue;
 315                         fa->fa_state |= FA_S_ACCESSED;
 316
 317                         if (fi == NULL) {
 318                                 if (next_fi != res->fi)
 319                                         break;
 320                         } else if (!fib_detect_death(fi, order, &last_resort,
 321                                                      &last_idx, &fn_hash_last_dflt)) {
 322                                 if (res->fi)
 323                                         fib_info_put(res->fi);
 324                                 res->fi = fi;
 325                                 atomic_inc(&fi->fib_clntref);
 326                                 fn_hash_last_dflt = order;
 327                                 goto out;
 328                         }
 329                         fi = next_fi;
 330                         order++;
 331                 }
 332         }
 333
 334         if (order <= 0 || fi == NULL) {
 335                 fn_hash_last_dflt = -1;
 336                 goto out;
 337         }
 338
 339         if (!fib_detect_death(fi, order, &last_resort, &last_idx, &fn_hash_last_dflt)) {
 340                 if (res->fi)
 341                         fib_info_put(res->fi);
 342                 res->fi = fi;
 343                 atomic_inc(&fi->fib_clntref);
 344                 fn_hash_last_dflt = order;
 345                 goto out;
 346         }
 347
 348         if (last_idx >= 0) {
 349                 if (res->fi)
 350                         fib_info_put(res->fi);
 351                 res->fi = last_resort;
 352                 if (last_resort)
 353                         atomic_inc(&last_resort->fib_clntref);
 354         }
 355         fn_hash_last_dflt = last_idx;
 356 out:
 357         read_unlock(&fib_hash_lock);
 358 }
 359
 360 /* Insert node F to FZ. */
 361 static inline void fib_insert_node(struct fn_zone *fz, struct fib_node *f)
 362 {
 363         struct hlist_head *head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 364
 365         hlist_add_head(&f->fn_hash, head);
 366 }
 367
 368 /* Return the node in FZ matching KEY. */
 369 static struct fib_node *fib_find_node(struct fn_zone *fz, __be32 key)
 370 {
 371         struct hlist_head *head = &fz->fz_hash[fn_hash(key, fz)];
 372         struct hlist_node *node;
 373         struct fib_node *f;
 374
 375         hlist_for_each_entry(f, node, head, fn_hash) {
 376                 if (f->fn_key == key)
 377                         return f;
 378         }
 379
 380         return NULL;
 381 }
 382
 383 static int fn_hash_insert(struct fib_table *tb, struct fib_config *cfg)
 384 {
 385         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 386         struct fib_node *new_f, *f;
 387         struct fib_alias *fa, *new_fa;
 388         struct fn_zone *fz;
 389         struct fib_info *fi;
 390         u8 tos = cfg->fc_tos;
 391         __be32 key;
 392         int err;
 393
 394         if (cfg->fc_dst_len > 32)
 395                 return -EINVAL;
 396
 397         fz = table->fn_zones[cfg->fc_dst_len];
 398         if (!fz && !(fz = fn_new_zone(table, cfg->fc_dst_len)))
 399                 return -ENOBUFS;
 400
 401         key = 0;
 402         if (cfg->fc_dst) {
 403                 if (cfg->fc_dst & ~FZ_MASK(fz))
 404                         return -EINVAL;
 405                 key = fz_key(cfg->fc_dst, fz);
 406         }
 407
 408         fi = fib_create_info(cfg);
 409         if (IS_ERR(fi))
 410                 return PTR_ERR(fi);
 411
 412         if (fz->fz_nent > (fz->fz_divisor<<1) &&
 413             fz->fz_divisor < FZ_MAX_DIVISOR &&
 414             (cfg->fc_dst_len == 32 ||
 415              (1 << cfg->fc_dst_len) > fz->fz_divisor))
 416                 fn_rehash_zone(fz);
 417
 418         f = fib_find_node(fz, key);
 419
 420         if (!f)
 421                 fa = NULL;
 422         else
 423                 fa = fib_find_alias(&f->fn_alias, tos, fi->fib_priority);
 424
 425         /* Now fa, if non-NULL, points to the first fib alias
 426          * with the same keys [prefix,tos,priority], if such key already
 427          * exists or to the node before which we will insert new one.
 428          *
 429          * If fa is NULL, we will need to allocate a new one and
 430          * insert to the head of f.
 431          *
 432          * If f is NULL, no fib node matched the destination key
 433          * and we need to allocate a new one of those as well.
 434          */
 435
 436         if (fa && fa->fa_tos == tos &&
 437             fa->fa_info->fib_priority == fi->fib_priority) {
 438                 struct fib_alias *fa_orig;
 439
 440                 err = -EEXIST;
 441                 if (cfg->fc_nlflags & NLM_F_EXCL)
 442                         goto out;
 443
 444                 if (cfg->fc_nlflags & NLM_F_REPLACE) {
 445                         struct fib_info *fi_drop;
 446                         u8 state;
 447
 448                         write_lock_bh(&fib_hash_lock);
 449                         fi_drop = fa->fa_info;
 450                         fa->fa_info = fi;
 451                         fa->fa_type = cfg->fc_type;
 452                         fa->fa_scope = cfg->fc_scope;
 453                         state = fa->fa_state;
 454                         fa->fa_state &= ~FA_S_ACCESSED;
 455                         fib_hash_genid++;
 456                         write_unlock_bh(&fib_hash_lock);
 457
 458                         fib_release_info(fi_drop);
 459                         if (state & FA_S_ACCESSED)
 460                                 rt_cache_flush(-1);
 461                         return 0;
 462                 }
 463
 464                 /* Error if we find a perfect match which
 465                  * uses the same scope, type, and nexthop
 466                  * information.
 467                  */
 468                 fa_orig = fa;
 469                 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 470                 list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 471                         if (fa->fa_tos != tos)
 472                                 break;
 473                         if (fa->fa_info->fib_priority != fi->fib_priority)
 474                                 break;
 475                         if (fa->fa_type == cfg->fc_type &&
 476                             fa->fa_scope == cfg->fc_scope &&
 477                             fa->fa_info == fi)
 478                                 goto out;
 479                 }
 480                 if (!(cfg->fc_nlflags & NLM_F_APPEND))
 481                         fa = fa_orig;
 482         }
 483
 484         err = -ENOENT;
 485         if (!(cfg->fc_nlflags & NLM_F_CREATE))
 486                 goto out;
 487
 488         err = -ENOBUFS;
 489         new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
 490         if (new_fa == NULL)
 491                 goto out;
 492
 493         new_f = NULL;
 494         if (!f) {
 495                 new_f = kmem_cache_alloc(fn_hash_kmem, GFP_KERNEL);
 496                 if (new_f == NULL)
 497                         goto out_free_new_fa;
 498
 499                 INIT_HLIST_NODE(&new_f->fn_hash);
 500                 INIT_LIST_HEAD(&new_f->fn_alias);
 501                 new_f->fn_key = key;
 502                 f = new_f;
 503         }
 504
 505         new_fa->fa_info = fi;
 506         new_fa->fa_tos = tos;
 507         new_fa->fa_type = cfg->fc_type;
 508         new_fa->fa_scope = cfg->fc_scope;
 509         new_fa->fa_state = 0;
 510
 511         /*
 512          * Insert new entry to the list.
 513          */
 514
 515         write_lock_bh(&fib_hash_lock);
 516         if (new_f)
 517                 fib_insert_node(fz, new_f);
 518         list_add_tail(&new_fa->fa_list,
 519                  (fa ? &fa->fa_list : &f->fn_alias));
 520         fib_hash_genid++;
 521         write_unlock_bh(&fib_hash_lock);
 522
 523         if (new_f)
 524                 fz->fz_nent++;
 525         rt_cache_flush(-1);
 526
 527         rtmsg_fib(RTM_NEWROUTE, key, new_fa, cfg->fc_dst_len, tb->tb_id,
 528                   &cfg->fc_nlinfo);
 529         return 0;
 530
 531 out_free_new_fa:
 532         kmem_cache_free(fn_alias_kmem, new_fa);
 533 out:
 534         fib_release_info(fi);
 535         return err;
 536 }
 537
 538
 539 static int fn_hash_delete(struct fib_table *tb, struct fib_config *cfg)
 540 {
 541         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 542         struct fib_node *f;
 543         struct fib_alias *fa, *fa_to_delete;
 544         struct fn_zone *fz;
 545         __be32 key;
 546
 547         if (cfg->fc_dst_len > 32)
 548                 return -EINVAL;
 549
 550         if ((fz  = table->fn_zones[cfg->fc_dst_len]) == NULL)
 551                 return -ESRCH;
 552
 553         key = 0;
 554         if (cfg->fc_dst) {
 555                 if (cfg->fc_dst & ~FZ_MASK(fz))
 556                         return -EINVAL;
 557                 key = fz_key(cfg->fc_dst, fz);
 558         }
 559
 560         f = fib_find_node(fz, key);
 561
 562         if (!f)
 563                 fa = NULL;
 564         else
 565                 fa = fib_find_alias(&f->fn_alias, cfg->fc_tos, 0);
 566         if (!fa)
 567                 return -ESRCH;
 568
 569         fa_to_delete = NULL;
 570         fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 571         list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 572                 struct fib_info *fi = fa->fa_info;
 573
 574                 if (fa->fa_tos != cfg->fc_tos)
 575                         break;
 576
 577                 if ((!cfg->fc_type ||
 578                      fa->fa_type == cfg->fc_type) &&
 579                     (cfg->fc_scope == RT_SCOPE_NOWHERE ||
 580                      fa->fa_scope == cfg->fc_scope) &&
 581                     (!cfg->fc_protocol ||
 582                      fi->fib_protocol == cfg->fc_protocol) &&
 583                     fib_nh_match(cfg, fi) == 0) {
 584                         fa_to_delete = fa;
 585                         break;
 586                 }
 587         }
 588
 589         if (fa_to_delete) {
 590                 int kill_fn;
 591
 592                 fa = fa_to_delete;
 593                 rtmsg_fib(RTM_DELROUTE, key, fa, cfg->fc_dst_len,
 594                           tb->tb_id, &cfg->fc_nlinfo);
 595
 596                 kill_fn = 0;
 597                 write_lock_bh(&fib_hash_lock);
 598                 list_del(&fa->fa_list);
 599                 if (list_empty(&f->fn_alias)) {
 600                         hlist_del(&f->fn_hash);
 601                         kill_fn = 1;
 602                 }
 603                 fib_hash_genid++;
 604                 write_unlock_bh(&fib_hash_lock);
 605
 606                 if (fa->fa_state & FA_S_ACCESSED)
 607                         rt_cache_flush(-1);
 608                 fn_free_alias(fa);
 609                 if (kill_fn) {
 610                         fn_free_node(f);
 611                         fz->fz_nent--;
 612                 }
 613
 614                 return 0;
 615         }
 616         return -ESRCH;
 617 }
 618
 619 static int fn_flush_list(struct fn_zone *fz, int idx)
 620 {
 621         struct hlist_head *head = &fz->fz_hash[idx];
 622         struct hlist_node *node, *n;
 623         struct fib_node *f;
 624         int found = 0;
 625
 626         hlist_for_each_entry_safe(f, node, n, head, fn_hash) {
 627                 struct fib_alias *fa, *fa_node;
 628                 int kill_f;
 629
 630                 kill_f = 0;
 631                 list_for_each_entry_safe(fa, fa_node, &f->fn_alias, fa_list) {
 632                         struct fib_info *fi = fa->fa_info;
 633
 634                         if (fi && (fi->fib_flags&RTNH_F_DEAD)) {
 635                                 write_lock_bh(&fib_hash_lock);
 636                                 list_del(&fa->fa_list);
 637                                 if (list_empty(&f->fn_alias)) {
 638                                         hlist_del(&f->fn_hash);
 639                                         kill_f = 1;
 640                                 }
 641                                 fib_hash_genid++;
 642                                 write_unlock_bh(&fib_hash_lock);
 643
 644                                 fn_free_alias(fa);
 645                                 found++;
 646                         }
 647                 }
 648                 if (kill_f) {
 649                         fn_free_node(f);
 650                         fz->fz_nent--;
 651                 }
 652         }
 653         return found;
 654 }
 655
 656 static int fn_hash_flush(struct fib_table *tb)
 657 {
 658         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 659         struct fn_zone *fz;
 660         int found = 0;
 661
 662         for (fz = table->fn_zone_list; fz; fz = fz->fz_next) {
 663                 int i;
 664
 665                 for (i = fz->fz_divisor - 1; i >= 0; i--)
 666                         found += fn_flush_list(fz, i);
 667         }
 668         return found;
 669 }
 670
 671
 672 static inline int
 673 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
 674                      struct fib_table *tb,
 675                      struct fn_zone *fz,
 676                      struct hlist_head *head)
 677 {
 678         struct hlist_node *node;
 679         struct fib_node *f;
 680         int i, s_i;
 681
 682         s_i = cb->args[4];
 683         i = 0;
 684         hlist_for_each_entry(f, node, head, fn_hash) {
 685                 struct fib_alias *fa;
 686
 687                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 688                         if (i < s_i)
 689                                 goto next;
 690
 691                         if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
 692                                           cb->nlh->nlmsg_seq,
 693                                           RTM_NEWROUTE,
 694                                           tb->tb_id,
 695                                           fa->fa_type,
 696                                           fa->fa_scope,
 697                                           f->fn_key,
 698                                           fz->fz_order,
 699                                           fa->fa_tos,
 700                                           fa->fa_info,
 701                                           NLM_F_MULTI) < 0) {
 702                                 cb->args[4] = i;
 703                                 return -1;
 704                         }
 705                 next:
 706                         i++;
 707                 }
 708         }
 709         cb->args[4] = i;
 710         return skb->len;
 711 }
 712
 713 static inline int
 714 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
 715                    struct fib_table *tb,
 716                    struct fn_zone *fz)
 717 {
 718         int h, s_h;
 719
 720         s_h = cb->args[3];
 721         for (h=0; h < fz->fz_divisor; h++) {
 722                 if (h < s_h) continue;
 723                 if (h > s_h)
 724                         memset(&cb->args[4], 0,
 725                                sizeof(cb->args) - 4*sizeof(cb->args[0]));
 726                 if (fz->fz_hash == NULL ||
 727                     hlist_empty(&fz->fz_hash[h]))
 728                         continue;
 729                 if (fn_hash_dump_bucket(skb, cb, tb, fz, &fz->fz_hash[h])<0) {
 730                         cb->args[3] = h;
 731                         return -1;
 732                 }
 733         }
 734         cb->args[3] = h;
 735         return skb->len;
 736 }
 737
 738 static int fn_hash_dump(struct fib_table *tb, struct sk_buff *skb, struct netlink_callback *cb)
 739 {
 740         int m, s_m;
 741         struct fn_zone *fz;
 742         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 743
 744         s_m = cb->args[2];
 745         read_lock(&fib_hash_lock);
 746         for (fz = table->fn_zone_list, m=0; fz; fz = fz->fz_next, m++) {
 747                 if (m < s_m) continue;
 748                 if (m > s_m)
 749                         memset(&cb->args[3], 0,
 750                                sizeof(cb->args) - 3*sizeof(cb->args[0]));
 751                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
 752                         cb->args[2] = m;
 753                         read_unlock(&fib_hash_lock);
 754                         return -1;
 755                 }
 756         }
 757         read_unlock(&fib_hash_lock);
 758         cb->args[2] = m;
 759         return skb->len;
 760 }
 761
 762 #ifdef CONFIG_IP_MULTIPLE_TABLES
 763 struct fib_table * fib_hash_init(u32 id)
 764 #else
 765 struct fib_table * __init fib_hash_init(u32 id)
 766 #endif
 767 {
 768         struct fib_table *tb;
 769
 770         if (fn_hash_kmem == NULL)
 771                 fn_hash_kmem = kmem_cache_create("ip_fib_hash",
 772                                                  sizeof(struct fib_node),
 773                                                  0, SLAB_HWCACHE_ALIGN,
 774                                                  NULL, NULL);
 775
 776         if (fn_alias_kmem == NULL)
 777                 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
 778                                                   sizeof(struct fib_alias),
 779                                                   0, SLAB_HWCACHE_ALIGN,
 780                                                   NULL, NULL);
 781
 782         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash),
 783                      GFP_KERNEL);
 784         if (tb == NULL)
 785                 return NULL;
 786
 787         tb->tb_id = id;
 788         tb->tb_lookup = fn_hash_lookup;
 789         tb->tb_insert = fn_hash_insert;
 790         tb->tb_delete = fn_hash_delete;
 791         tb->tb_flush = fn_hash_flush;
 792         tb->tb_select_default = fn_hash_select_default;
 793         tb->tb_dump = fn_hash_dump;
 794         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 795         return tb;
 796 }
 797
 798 /* ------------------------------------------------------------------------ */
 799 #ifdef CONFIG_PROC_FS
 800
 801 struct fib_iter_state {
 802         struct fn_zone  *zone;
 803         int             bucket;
 804         struct hlist_head *hash_head;
 805         struct fib_node *fn;
 806         struct fib_alias *fa;
 807         loff_t pos;
 808         unsigned int genid;
 809         int valid;
 810 };
 811
 812 static struct fib_alias *fib_get_first(struct seq_file *seq)
 813 {
 814         struct fib_iter_state *iter = seq->private;
 815         struct fn_hash *table = (struct fn_hash *) ip_fib_main_table->tb_data;
 816
 817         iter->bucket    = 0;
 818         iter->hash_head = NULL;
 819         iter->fn        = NULL;
 820         iter->fa        = NULL;
 821         iter->pos       = 0;
 822         iter->genid     = fib_hash_genid;
 823         iter->valid     = 1;
 824
 825         for (iter->zone = table->fn_zone_list; iter->zone;
 826              iter->zone = iter->zone->fz_next) {
 827                 int maxslot;
 828
 829                 if (!iter->zone->fz_nent)
 830                         continue;
 831
 832                 iter->hash_head = iter->zone->fz_hash;
 833                 maxslot = iter->zone->fz_divisor;
 834
 835                 for (iter->bucket = 0; iter->bucket < maxslot;
 836                      ++iter->bucket, ++iter->hash_head) {
 837                         struct hlist_node *node;
 838                         struct fib_node *fn;
 839
 840                         hlist_for_each_entry(fn,node,iter->hash_head,fn_hash) {
 841                                 struct fib_alias *fa;
 842
 843                                 list_for_each_entry(fa,&fn->fn_alias,fa_list) {
 844                                         iter->fn = fn;
 845                                         iter->fa = fa;
 846                                         goto out;
 847                                 }
 848                         }
 849                 }
 850         }
 851 out:
 852         return iter->fa;
 853 }
 854
 855 static struct fib_alias *fib_get_next(struct seq_file *seq)
 856 {
 857         struct fib_iter_state *iter = seq->private;
 858         struct fib_node *fn;
 859         struct fib_alias *fa;
 860
 861         /* Advance FA, if any. */
 862         fn = iter->fn;
 863         fa = iter->fa;
 864         if (fa) {
 865                 BUG_ON(!fn);
 866                 list_for_each_entry_continue(fa, &fn->fn_alias, fa_list) {
 867                         iter->fa = fa;
 868                         goto out;
 869                 }
 870         }
 871
 872         fa = iter->fa = NULL;
 873
 874         /* Advance FN. */
 875         if (fn) {
 876                 struct hlist_node *node = &fn->fn_hash;
 877                 hlist_for_each_entry_continue(fn, node, fn_hash) {
 878                         iter->fn = fn;
 879
 880                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 881                                 iter->fa = fa;
 882                                 goto out;
 883                         }
 884                 }
 885         }
 886
 887         fn = iter->fn = NULL;
 888
 889         /* Advance hash chain. */
 890         if (!iter->zone)
 891                 goto out;
 892
 893         for (;;) {
 894                 struct hlist_node *node;
 895                 int maxslot;
 896
 897                 maxslot = iter->zone->fz_divisor;
 898
 899                 while (++iter->bucket < maxslot) {
 900                         iter->hash_head++;
 901
 902                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 903                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 904                                         iter->fn = fn;
 905                                         iter->fa = fa;
 906                                         goto out;
 907                                 }
 908                         }
 909                 }
 910
 911                 iter->zone = iter->zone->fz_next;
 912
 913                 if (!iter->zone)
 914                         goto out;
 915
 916                 iter->bucket = 0;
 917                 iter->hash_head = iter->zone->fz_hash;
 918
 919                 hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 920                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 921                                 iter->fn = fn;
 922                                 iter->fa = fa;
 923                                 goto out;
 924                         }
 925                 }
 926         }
 927 out:
 928         iter->pos++;
 929         return fa;
 930 }
 931
 932 static struct fib_alias *fib_get_idx(struct seq_file *seq, loff_t pos)
 933 {
 934         struct fib_iter_state *iter = seq->private;
 935         struct fib_alias *fa;
 936
 937         if (iter->valid && pos >= iter->pos && iter->genid == fib_hash_genid) {
 938                 fa   = iter->fa;
 939                 pos -= iter->pos;
 940         } else
 941                 fa = fib_get_first(seq);
 942
 943         if (fa)
 944                 while (pos && (fa = fib_get_next(seq)))
 945                         --pos;
 946         return pos ? NULL : fa;
 947 }
 948
 949 static void *fib_seq_start(struct seq_file *seq, loff_t *pos)
 950 {
 951         void *v = NULL;
 952
 953         read_lock(&fib_hash_lock);
 954         if (ip_fib_main_table)
 955                 v = *pos ? fib_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 956         return v;
 957 }
 958
 959 static void *fib_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 960 {
 961         ++*pos;
 962         return v == SEQ_START_TOKEN ? fib_get_first(seq) : fib_get_next(seq);
 963 }
 964
 965 static void fib_seq_stop(struct seq_file *seq, void *v)
 966 {
 967         read_unlock(&fib_hash_lock);
 968 }
 969
 970 static unsigned fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
 971 {
 972         static const unsigned type2flags[RTN_MAX + 1] = {
 973                 [7] = RTF_REJECT, [8] = RTF_REJECT,
 974         };
 975         unsigned flags = type2flags[type];
 976
 977         if (fi && fi->fib_nh->nh_gw)
 978                 flags |= RTF_GATEWAY;
 979         if (mask == htonl(0xFFFFFFFF))
 980                 flags |= RTF_HOST;
 981         flags |= RTF_UP;
 982         return flags;
 983 }
 984
 985 extern int dev_in_nx_info(struct net_device *, struct nx_info *);
 986
 987 /*
 988  *      This outputs /proc/net/route.
 989  *
 990  *      It always works in backward compatibility mode.
 991  *      The format of the file is not supposed to be changed.
 992  */
 993 static int fib_seq_show(struct seq_file *seq, void *v)
 994 {
 995         struct fib_iter_state *iter;
 996         char bf[128];
 997         __be32 prefix, mask;
 998         unsigned flags;
 999         struct fib_node *f;
1000         struct fib_alias *fa;
1001         struct fib_info *fi;
1002
1003         if (v == SEQ_START_TOKEN) {
1004                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
1005                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
1006                            "\tWindow\tIRTT");
1007                 goto out;
1008         }
1009
1010         iter    = seq->private;
1011         f       = iter->fn;
1012         fa      = iter->fa;
1013         fi      = fa->fa_info;
1014         prefix  = f->fn_key;
1015         mask    = FZ_MASK(iter->zone);
1016         flags   = fib_flag_trans(fa->fa_type, mask, fi);
1017         if (fi && (!vx_flags(VXF_HIDE_NETIF, 0) ||
1018                 dev_in_nx_info(fi->fib_dev, current->nx_info)))
1019                 snprintf(bf, sizeof(bf),
1020                          "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1021                          fi->fib_dev ? fi->fib_dev->name : "*", prefix,
1022                          fi->fib_nh->nh_gw, flags, 0, 0, fi->fib_priority,
1023                          mask, (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
1024                          fi->fib_window,
1025                          fi->fib_rtt >> 3);
1026         else
1027                 snprintf(bf, sizeof(bf),
1028                          "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1029                          prefix, 0, flags, 0, 0, 0, mask, 0, 0, 0);
1030         seq_printf(seq, "%-127s\n", bf);
1031 out:
1032         return 0;
1033 }
1034
1035 static struct seq_operations fib_seq_ops = {
1036         .start  = fib_seq_start,
1037         .next   = fib_seq_next,
1038         .stop   = fib_seq_stop,
1039         .show   = fib_seq_show,
1040 };
1041
1042 static int fib_seq_open(struct inode *inode, struct file *file)
1043 {
1044         struct seq_file *seq;
1045         int rc = -ENOMEM;
1046         struct fib_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1047
1048         if (!s)
1049                 goto out;
1050
1051         rc = seq_open(file, &fib_seq_ops);
1052         if (rc)
1053                 goto out_kfree;
1054
1055         seq          = file->private_data;
1056         seq->private = s;
1057 out:
1058         return rc;
1059 out_kfree:
1060         kfree(s);
1061         goto out;
1062 }
1063
1064 static struct file_operations fib_seq_fops = {
1065         .owner          = THIS_MODULE,
1066         .open           = fib_seq_open,
1067         .read           = seq_read,
1068         .llseek         = seq_lseek,
1069         .release        = seq_release_private,
1070 };
1071
1072 int __init fib_proc_init(void)
1073 {
1074         if (!proc_net_fops_create("route", S_IRUGO, &fib_seq_fops))
1075                 return -ENOMEM;
1076         return 0;
1077 }
1078
1079 void __init fib_proc_exit(void)
1080 {
1081         proc_net_remove("route");
1082 }
1083 #endif /* CONFIG_PROC_FS */