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 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("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                 write_lock_bh(&fib_hash_lock);
 178                 old_ht = fz->fz_hash;
 179                 fz->fz_hash = ht;
 180                 fz->fz_hashmask = new_hashmask;
 181                 fz->fz_divisor = new_divisor;
 182                 fn_rebuild_zone(fz, old_ht, old_divisor);
 183                 fib_hash_genid++;
 184                 write_unlock_bh(&fib_hash_lock);
 185
 186                 fz_hash_free(old_ht, old_divisor);
 187         }
 188 }
 189
 190 static inline void fn_free_node(struct fib_node * f)
 191 {
 192         kmem_cache_free(fn_hash_kmem, f);
 193 }
 194
 195 static inline void fn_free_alias(struct fib_alias *fa)
 196 {
 197         fib_release_info(fa->fa_info);
 198         kmem_cache_free(fn_alias_kmem, fa);
 199 }
 200
 201 static struct fn_zone *
 202 fn_new_zone(struct fn_hash *table, int z)
 203 {
 204         int i;
 205         struct fn_zone *fz = kzalloc(sizeof(struct fn_zone), GFP_KERNEL);
 206         if (!fz)
 207                 return NULL;
 208
 209         if (z) {
 210                 fz->fz_divisor = 16;
 211         } else {
 212                 fz->fz_divisor = 1;
 213         }
 214         fz->fz_hashmask = (fz->fz_divisor - 1);
 215         fz->fz_hash = fz_hash_alloc(fz->fz_divisor);
 216         if (!fz->fz_hash) {
 217                 kfree(fz);
 218                 return NULL;
 219         }
 220         fz->fz_order = z;
 221         fz->fz_mask = inet_make_mask(z);
 222
 223         /* Find the first not empty zone with more specific mask */
 224         for (i=z+1; i<=32; i++)
 225                 if (table->fn_zones[i])
 226                         break;
 227         write_lock_bh(&fib_hash_lock);
 228         if (i>32) {
 229                 /* No more specific masks, we are the first. */
 230                 fz->fz_next = table->fn_zone_list;
 231                 table->fn_zone_list = fz;
 232         } else {
 233                 fz->fz_next = table->fn_zones[i]->fz_next;
 234                 table->fn_zones[i]->fz_next = fz;
 235         }
 236         table->fn_zones[z] = fz;
 237         fib_hash_genid++;
 238         write_unlock_bh(&fib_hash_lock);
 239         return fz;
 240 }
 241
 242 static int
 243 fn_hash_lookup(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
 244 {
 245         int err;
 246         struct fn_zone *fz;
 247         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 248
 249         read_lock(&fib_hash_lock);
 250         for (fz = t->fn_zone_list; fz; fz = fz->fz_next) {
 251                 struct hlist_head *head;
 252                 struct hlist_node *node;
 253                 struct fib_node *f;
 254                 __be32 k = fz_key(flp->fl4_dst, fz);
 255
 256                 head = &fz->fz_hash[fn_hash(k, fz)];
 257                 hlist_for_each_entry(f, node, head, fn_hash) {
 258                         if (f->fn_key != k)
 259                                 continue;
 260
 261                         err = fib_semantic_match(&f->fn_alias,
 262                                                  flp, res,
 263                                                  f->fn_key, fz->fz_mask,
 264                                                  fz->fz_order);
 265                         if (err <= 0)
 266                                 goto out;
 267                 }
 268         }
 269         err = 1;
 270 out:
 271         read_unlock(&fib_hash_lock);
 272         return err;
 273 }
 274
 275 static void
 276 fn_hash_select_default(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
 277 {
 278         int order, last_idx;
 279         struct hlist_node *node;
 280         struct fib_node *f;
 281         struct fib_info *fi = NULL;
 282         struct fib_info *last_resort;
 283         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 284         struct fn_zone *fz = t->fn_zones[0];
 285
 286         if (fz == NULL)
 287                 return;
 288
 289         last_idx = -1;
 290         last_resort = NULL;
 291         order = -1;
 292
 293         read_lock(&fib_hash_lock);
 294         hlist_for_each_entry(f, node, &fz->fz_hash[0], fn_hash) {
 295                 struct fib_alias *fa;
 296
 297                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 298                         struct fib_info *next_fi = fa->fa_info;
 299
 300                         if (fa->fa_scope != res->scope ||
 301                             fa->fa_type != RTN_UNICAST)
 302                                 continue;
 303
 304                         if (next_fi->fib_priority > res->fi->fib_priority)
 305                                 break;
 306                         if (!next_fi->fib_nh[0].nh_gw ||
 307                             next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
 308                                 continue;
 309                         fa->fa_state |= FA_S_ACCESSED;
 310
 311                         if (fi == NULL) {
 312                                 if (next_fi != res->fi)
 313                                         break;
 314                         } else if (!fib_detect_death(fi, order, &last_resort,
 315                                                 &last_idx, tb->tb_default)) {
 316                                 fib_result_assign(res, fi);
 317                                 tb->tb_default = order;
 318                                 goto out;
 319                         }
 320                         fi = next_fi;
 321                         order++;
 322                 }
 323         }
 324
 325         if (order <= 0 || fi == NULL) {
 326                 tb->tb_default = -1;
 327                 goto out;
 328         }
 329
 330         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
 331                                 tb->tb_default)) {
 332                 fib_result_assign(res, fi);
 333                 tb->tb_default = order;
 334                 goto out;
 335         }
 336
 337         if (last_idx >= 0)
 338                 fib_result_assign(res, last_resort);
 339         tb->tb_default = last_idx;
 340 out:
 341         read_unlock(&fib_hash_lock);
 342 }
 343
 344 /* Insert node F to FZ. */
 345 static inline void fib_insert_node(struct fn_zone *fz, struct fib_node *f)
 346 {
 347         struct hlist_head *head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 348
 349         hlist_add_head(&f->fn_hash, head);
 350 }
 351
 352 /* Return the node in FZ matching KEY. */
 353 static struct fib_node *fib_find_node(struct fn_zone *fz, __be32 key)
 354 {
 355         struct hlist_head *head = &fz->fz_hash[fn_hash(key, fz)];
 356         struct hlist_node *node;
 357         struct fib_node *f;
 358
 359         hlist_for_each_entry(f, node, head, fn_hash) {
 360                 if (f->fn_key == key)
 361                         return f;
 362         }
 363
 364         return NULL;
 365 }
 366
 367 static int fn_hash_insert(struct fib_table *tb, struct fib_config *cfg)
 368 {
 369         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 370         struct fib_node *new_f, *f;
 371         struct fib_alias *fa, *new_fa;
 372         struct fn_zone *fz;
 373         struct fib_info *fi;
 374         u8 tos = cfg->fc_tos;
 375         __be32 key;
 376         int err;
 377
 378         if (cfg->fc_dst_len > 32)
 379                 return -EINVAL;
 380
 381         fz = table->fn_zones[cfg->fc_dst_len];
 382         if (!fz && !(fz = fn_new_zone(table, cfg->fc_dst_len)))
 383                 return -ENOBUFS;
 384
 385         key = 0;
 386         if (cfg->fc_dst) {
 387                 if (cfg->fc_dst & ~FZ_MASK(fz))
 388                         return -EINVAL;
 389                 key = fz_key(cfg->fc_dst, fz);
 390         }
 391
 392         fi = fib_create_info(cfg);
 393         if (IS_ERR(fi))
 394                 return PTR_ERR(fi);
 395
 396         if (fz->fz_nent > (fz->fz_divisor<<1) &&
 397             fz->fz_divisor < FZ_MAX_DIVISOR &&
 398             (cfg->fc_dst_len == 32 ||
 399              (1 << cfg->fc_dst_len) > fz->fz_divisor))
 400                 fn_rehash_zone(fz);
 401
 402         f = fib_find_node(fz, key);
 403
 404         if (!f)
 405                 fa = NULL;
 406         else
 407                 fa = fib_find_alias(&f->fn_alias, tos, fi->fib_priority);
 408
 409         /* Now fa, if non-NULL, points to the first fib alias
 410          * with the same keys [prefix,tos,priority], if such key already
 411          * exists or to the node before which we will insert new one.
 412          *
 413          * If fa is NULL, we will need to allocate a new one and
 414          * insert to the head of f.
 415          *
 416          * If f is NULL, no fib node matched the destination key
 417          * and we need to allocate a new one of those as well.
 418          */
 419
 420         if (fa && fa->fa_tos == tos &&
 421             fa->fa_info->fib_priority == fi->fib_priority) {
 422                 struct fib_alias *fa_orig;
 423
 424                 err = -EEXIST;
 425                 if (cfg->fc_nlflags & NLM_F_EXCL)
 426                         goto out;
 427
 428                 if (cfg->fc_nlflags & NLM_F_REPLACE) {
 429                         struct fib_info *fi_drop;
 430                         u8 state;
 431
 432                         if (fi->fib_treeref > 1)
 433                                 goto out;
 434
 435                         write_lock_bh(&fib_hash_lock);
 436                         fi_drop = fa->fa_info;
 437                         fa->fa_info = fi;
 438                         fa->fa_type = cfg->fc_type;
 439                         fa->fa_scope = cfg->fc_scope;
 440                         state = fa->fa_state;
 441                         fa->fa_state &= ~FA_S_ACCESSED;
 442                         fib_hash_genid++;
 443                         write_unlock_bh(&fib_hash_lock);
 444
 445                         fib_release_info(fi_drop);
 446                         if (state & FA_S_ACCESSED)
 447                                 rt_cache_flush(-1);
 448                         rtmsg_fib(RTM_NEWROUTE, key, fa, cfg->fc_dst_len, tb->tb_id,
 449                                   &cfg->fc_nlinfo, NLM_F_REPLACE);
 450                         return 0;
 451                 }
 452
 453                 /* Error if we find a perfect match which
 454                  * uses the same scope, type, and nexthop
 455                  * information.
 456                  */
 457                 fa_orig = fa;
 458                 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 459                 list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 460                         if (fa->fa_tos != tos)
 461                                 break;
 462                         if (fa->fa_info->fib_priority != fi->fib_priority)
 463                                 break;
 464                         if (fa->fa_type == cfg->fc_type &&
 465                             fa->fa_scope == cfg->fc_scope &&
 466                             fa->fa_info == fi)
 467                                 goto out;
 468                 }
 469                 if (!(cfg->fc_nlflags & NLM_F_APPEND))
 470                         fa = fa_orig;
 471         }
 472
 473         err = -ENOENT;
 474         if (!(cfg->fc_nlflags & NLM_F_CREATE))
 475                 goto out;
 476
 477         err = -ENOBUFS;
 478         new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
 479         if (new_fa == NULL)
 480                 goto out;
 481
 482         new_f = NULL;
 483         if (!f) {
 484                 new_f = kmem_cache_alloc(fn_hash_kmem, GFP_KERNEL);
 485                 if (new_f == NULL)
 486                         goto out_free_new_fa;
 487
 488                 INIT_HLIST_NODE(&new_f->fn_hash);
 489                 INIT_LIST_HEAD(&new_f->fn_alias);
 490                 new_f->fn_key = key;
 491                 f = new_f;
 492         }
 493
 494         new_fa->fa_info = fi;
 495         new_fa->fa_tos = tos;
 496         new_fa->fa_type = cfg->fc_type;
 497         new_fa->fa_scope = cfg->fc_scope;
 498         new_fa->fa_state = 0;
 499
 500         /*
 501          * Insert new entry to the list.
 502          */
 503
 504         write_lock_bh(&fib_hash_lock);
 505         if (new_f)
 506                 fib_insert_node(fz, new_f);
 507         list_add_tail(&new_fa->fa_list,
 508                  (fa ? &fa->fa_list : &f->fn_alias));
 509         fib_hash_genid++;
 510         write_unlock_bh(&fib_hash_lock);
 511
 512         if (new_f)
 513                 fz->fz_nent++;
 514         rt_cache_flush(-1);
 515
 516         rtmsg_fib(RTM_NEWROUTE, key, new_fa, cfg->fc_dst_len, tb->tb_id,
 517                   &cfg->fc_nlinfo, 0);
 518         return 0;
 519
 520 out_free_new_fa:
 521         kmem_cache_free(fn_alias_kmem, new_fa);
 522 out:
 523         fib_release_info(fi);
 524         return err;
 525 }
 526
 527
 528 static int fn_hash_delete(struct fib_table *tb, struct fib_config *cfg)
 529 {
 530         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 531         struct fib_node *f;
 532         struct fib_alias *fa, *fa_to_delete;
 533         struct fn_zone *fz;
 534         __be32 key;
 535
 536         if (cfg->fc_dst_len > 32)
 537                 return -EINVAL;
 538
 539         if ((fz  = table->fn_zones[cfg->fc_dst_len]) == NULL)
 540                 return -ESRCH;
 541
 542         key = 0;
 543         if (cfg->fc_dst) {
 544                 if (cfg->fc_dst & ~FZ_MASK(fz))
 545                         return -EINVAL;
 546                 key = fz_key(cfg->fc_dst, fz);
 547         }
 548
 549         f = fib_find_node(fz, key);
 550
 551         if (!f)
 552                 fa = NULL;
 553         else
 554                 fa = fib_find_alias(&f->fn_alias, cfg->fc_tos, 0);
 555         if (!fa)
 556                 return -ESRCH;
 557
 558         fa_to_delete = NULL;
 559         fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 560         list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 561                 struct fib_info *fi = fa->fa_info;
 562
 563                 if (fa->fa_tos != cfg->fc_tos)
 564                         break;
 565
 566                 if ((!cfg->fc_type ||
 567                      fa->fa_type == cfg->fc_type) &&
 568                     (cfg->fc_scope == RT_SCOPE_NOWHERE ||
 569                      fa->fa_scope == cfg->fc_scope) &&
 570                     (!cfg->fc_protocol ||
 571                      fi->fib_protocol == cfg->fc_protocol) &&
 572                     fib_nh_match(cfg, fi) == 0) {
 573                         fa_to_delete = fa;
 574                         break;
 575                 }
 576         }
 577
 578         if (fa_to_delete) {
 579                 int kill_fn;
 580
 581                 fa = fa_to_delete;
 582                 rtmsg_fib(RTM_DELROUTE, key, fa, cfg->fc_dst_len,
 583                           tb->tb_id, &cfg->fc_nlinfo, 0);
 584
 585                 kill_fn = 0;
 586                 write_lock_bh(&fib_hash_lock);
 587                 list_del(&fa->fa_list);
 588                 if (list_empty(&f->fn_alias)) {
 589                         hlist_del(&f->fn_hash);
 590                         kill_fn = 1;
 591                 }
 592                 fib_hash_genid++;
 593                 write_unlock_bh(&fib_hash_lock);
 594
 595                 if (fa->fa_state & FA_S_ACCESSED)
 596                         rt_cache_flush(-1);
 597                 fn_free_alias(fa);
 598                 if (kill_fn) {
 599                         fn_free_node(f);
 600                         fz->fz_nent--;
 601                 }
 602
 603                 return 0;
 604         }
 605         return -ESRCH;
 606 }
 607
 608 static int fn_flush_list(struct fn_zone *fz, int idx)
 609 {
 610         struct hlist_head *head = &fz->fz_hash[idx];
 611         struct hlist_node *node, *n;
 612         struct fib_node *f;
 613         int found = 0;
 614
 615         hlist_for_each_entry_safe(f, node, n, head, fn_hash) {
 616                 struct fib_alias *fa, *fa_node;
 617                 int kill_f;
 618
 619                 kill_f = 0;
 620                 list_for_each_entry_safe(fa, fa_node, &f->fn_alias, fa_list) {
 621                         struct fib_info *fi = fa->fa_info;
 622
 623                         if (fi && (fi->fib_flags&RTNH_F_DEAD)) {
 624                                 write_lock_bh(&fib_hash_lock);
 625                                 list_del(&fa->fa_list);
 626                                 if (list_empty(&f->fn_alias)) {
 627                                         hlist_del(&f->fn_hash);
 628                                         kill_f = 1;
 629                                 }
 630                                 fib_hash_genid++;
 631                                 write_unlock_bh(&fib_hash_lock);
 632
 633                                 fn_free_alias(fa);
 634                                 found++;
 635                         }
 636                 }
 637                 if (kill_f) {
 638                         fn_free_node(f);
 639                         fz->fz_nent--;
 640                 }
 641         }
 642         return found;
 643 }
 644
 645 static int fn_hash_flush(struct fib_table *tb)
 646 {
 647         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 648         struct fn_zone *fz;
 649         int found = 0;
 650
 651         for (fz = table->fn_zone_list; fz; fz = fz->fz_next) {
 652                 int i;
 653
 654                 for (i = fz->fz_divisor - 1; i >= 0; i--)
 655                         found += fn_flush_list(fz, i);
 656         }
 657         return found;
 658 }
 659
 660
 661 static inline int
 662 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
 663                      struct fib_table *tb,
 664                      struct fn_zone *fz,
 665                      struct hlist_head *head)
 666 {
 667         struct hlist_node *node;
 668         struct fib_node *f;
 669         int i, s_i;
 670
 671         s_i = cb->args[4];
 672         i = 0;
 673         hlist_for_each_entry(f, node, head, fn_hash) {
 674                 struct fib_alias *fa;
 675
 676                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 677                         if (i < s_i)
 678                                 goto next;
 679
 680                         if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
 681                                           cb->nlh->nlmsg_seq,
 682                                           RTM_NEWROUTE,
 683                                           tb->tb_id,
 684                                           fa->fa_type,
 685                                           fa->fa_scope,
 686                                           f->fn_key,
 687                                           fz->fz_order,
 688                                           fa->fa_tos,
 689                                           fa->fa_info,
 690                                           NLM_F_MULTI) < 0) {
 691                                 cb->args[4] = i;
 692                                 return -1;
 693                         }
 694                 next:
 695                         i++;
 696                 }
 697         }
 698         cb->args[4] = i;
 699         return skb->len;
 700 }
 701
 702 static inline int
 703 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
 704                    struct fib_table *tb,
 705                    struct fn_zone *fz)
 706 {
 707         int h, s_h;
 708
 709         if (fz->fz_hash == NULL)
 710                 return skb->len;
 711         s_h = cb->args[3];
 712         for (h = s_h; h < fz->fz_divisor; h++) {
 713                 if (hlist_empty(&fz->fz_hash[h]))
 714                         continue;
 715                 if (fn_hash_dump_bucket(skb, cb, tb, fz, &fz->fz_hash[h]) < 0) {
 716                         cb->args[3] = h;
 717                         return -1;
 718                 }
 719                 memset(&cb->args[4], 0,
 720                        sizeof(cb->args) - 4*sizeof(cb->args[0]));
 721         }
 722         cb->args[3] = h;
 723         return skb->len;
 724 }
 725
 726 static int fn_hash_dump(struct fib_table *tb, struct sk_buff *skb, struct netlink_callback *cb)
 727 {
 728         int m, s_m;
 729         struct fn_zone *fz;
 730         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 731
 732         s_m = cb->args[2];
 733         read_lock(&fib_hash_lock);
 734         for (fz = table->fn_zone_list, m=0; fz; fz = fz->fz_next, m++) {
 735                 if (m < s_m) continue;
 736                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
 737                         cb->args[2] = m;
 738                         read_unlock(&fib_hash_lock);
 739                         return -1;
 740                 }
 741                 memset(&cb->args[3], 0,
 742                        sizeof(cb->args) - 3*sizeof(cb->args[0]));
 743         }
 744         read_unlock(&fib_hash_lock);
 745         cb->args[2] = m;
 746         return skb->len;
 747 }
 748
 749 #ifdef CONFIG_IP_MULTIPLE_TABLES
 750 struct fib_table * fib_hash_init(u32 id)
 751 #else
 752 struct fib_table * __init fib_hash_init(u32 id)
 753 #endif
 754 {
 755         struct fib_table *tb;
 756
 757         if (fn_hash_kmem == NULL)
 758                 fn_hash_kmem = kmem_cache_create("ip_fib_hash",
 759                                                  sizeof(struct fib_node),
 760                                                  0, SLAB_HWCACHE_ALIGN,
 761                                                  NULL);
 762
 763         if (fn_alias_kmem == NULL)
 764                 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
 765                                                   sizeof(struct fib_alias),
 766                                                   0, SLAB_HWCACHE_ALIGN,
 767                                                   NULL);
 768
 769         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash),
 770                      GFP_KERNEL);
 771         if (tb == NULL)
 772                 return NULL;
 773
 774         tb->tb_id = id;
 775         tb->tb_default = -1;
 776         tb->tb_lookup = fn_hash_lookup;
 777         tb->tb_insert = fn_hash_insert;
 778         tb->tb_delete = fn_hash_delete;
 779         tb->tb_flush = fn_hash_flush;
 780         tb->tb_select_default = fn_hash_select_default;
 781         tb->tb_dump = fn_hash_dump;
 782         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 783         return tb;
 784 }
 785
 786 /* ------------------------------------------------------------------------ */
 787 #ifdef CONFIG_PROC_FS
 788
 789 struct fib_iter_state {
 790         struct fn_zone  *zone;
 791         int             bucket;
 792         struct hlist_head *hash_head;
 793         struct fib_node *fn;
 794         struct fib_alias *fa;
 795         loff_t pos;
 796         unsigned int genid;
 797         int valid;
 798 };
 799
 800 static struct fib_alias *fib_get_first(struct seq_file *seq)
 801 {
 802         struct fib_iter_state *iter = seq->private;
 803         struct fib_table *main_table = fib_get_table(RT_TABLE_MAIN);
 804         struct fn_hash *table = (struct fn_hash *)main_table->tb_data;
 805
 806         iter->bucket    = 0;
 807         iter->hash_head = NULL;
 808         iter->fn        = NULL;
 809         iter->fa        = NULL;
 810         iter->pos       = 0;
 811         iter->genid     = fib_hash_genid;
 812         iter->valid     = 1;
 813
 814         for (iter->zone = table->fn_zone_list; iter->zone;
 815              iter->zone = iter->zone->fz_next) {
 816                 int maxslot;
 817
 818                 if (!iter->zone->fz_nent)
 819                         continue;
 820
 821                 iter->hash_head = iter->zone->fz_hash;
 822                 maxslot = iter->zone->fz_divisor;
 823
 824                 for (iter->bucket = 0; iter->bucket < maxslot;
 825                      ++iter->bucket, ++iter->hash_head) {
 826                         struct hlist_node *node;
 827                         struct fib_node *fn;
 828
 829                         hlist_for_each_entry(fn,node,iter->hash_head,fn_hash) {
 830                                 struct fib_alias *fa;
 831
 832                                 list_for_each_entry(fa,&fn->fn_alias,fa_list) {
 833                                         iter->fn = fn;
 834                                         iter->fa = fa;
 835                                         goto out;
 836                                 }
 837                         }
 838                 }
 839         }
 840 out:
 841         return iter->fa;
 842 }
 843
 844 static struct fib_alias *fib_get_next(struct seq_file *seq)
 845 {
 846         struct fib_iter_state *iter = seq->private;
 847         struct fib_node *fn;
 848         struct fib_alias *fa;
 849
 850         /* Advance FA, if any. */
 851         fn = iter->fn;
 852         fa = iter->fa;
 853         if (fa) {
 854                 BUG_ON(!fn);
 855                 list_for_each_entry_continue(fa, &fn->fn_alias, fa_list) {
 856                         iter->fa = fa;
 857                         goto out;
 858                 }
 859         }
 860
 861         fa = iter->fa = NULL;
 862
 863         /* Advance FN. */
 864         if (fn) {
 865                 struct hlist_node *node = &fn->fn_hash;
 866                 hlist_for_each_entry_continue(fn, node, fn_hash) {
 867                         iter->fn = fn;
 868
 869                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 870                                 iter->fa = fa;
 871                                 goto out;
 872                         }
 873                 }
 874         }
 875
 876         fn = iter->fn = NULL;
 877
 878         /* Advance hash chain. */
 879         if (!iter->zone)
 880                 goto out;
 881
 882         for (;;) {
 883                 struct hlist_node *node;
 884                 int maxslot;
 885
 886                 maxslot = iter->zone->fz_divisor;
 887
 888                 while (++iter->bucket < maxslot) {
 889                         iter->hash_head++;
 890
 891                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 892                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 893                                         iter->fn = fn;
 894                                         iter->fa = fa;
 895                                         goto out;
 896                                 }
 897                         }
 898                 }
 899
 900                 iter->zone = iter->zone->fz_next;
 901
 902                 if (!iter->zone)
 903                         goto out;
 904
 905                 iter->bucket = 0;
 906                 iter->hash_head = iter->zone->fz_hash;
 907
 908                 hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 909                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 910                                 iter->fn = fn;
 911                                 iter->fa = fa;
 912                                 goto out;
 913                         }
 914                 }
 915         }
 916 out:
 917         iter->pos++;
 918         return fa;
 919 }
 920
 921 static struct fib_alias *fib_get_idx(struct seq_file *seq, loff_t pos)
 922 {
 923         struct fib_iter_state *iter = seq->private;
 924         struct fib_alias *fa;
 925
 926         if (iter->valid && pos >= iter->pos && iter->genid == fib_hash_genid) {
 927                 fa   = iter->fa;
 928                 pos -= iter->pos;
 929         } else
 930                 fa = fib_get_first(seq);
 931
 932         if (fa)
 933                 while (pos && (fa = fib_get_next(seq)))
 934                         --pos;
 935         return pos ? NULL : fa;
 936 }
 937
 938 static void *fib_seq_start(struct seq_file *seq, loff_t *pos)
 939 {
 940         void *v = NULL;
 941
 942         read_lock(&fib_hash_lock);
 943         if (fib_get_table(RT_TABLE_MAIN))
 944                 v = *pos ? fib_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 945         return v;
 946 }
 947
 948 static void *fib_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 949 {
 950         ++*pos;
 951         return v == SEQ_START_TOKEN ? fib_get_first(seq) : fib_get_next(seq);
 952 }
 953
 954 static void fib_seq_stop(struct seq_file *seq, void *v)
 955 {
 956         read_unlock(&fib_hash_lock);
 957 }
 958
 959 static unsigned fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
 960 {
 961         static const unsigned type2flags[RTN_MAX + 1] = {
 962                 [7] = RTF_REJECT, [8] = RTF_REJECT,
 963         };
 964         unsigned flags = type2flags[type];
 965
 966         if (fi && fi->fib_nh->nh_gw)
 967                 flags |= RTF_GATEWAY;
 968         if (mask == htonl(0xFFFFFFFF))
 969                 flags |= RTF_HOST;
 970         flags |= RTF_UP;
 971         return flags;
 972 }
 973
 974 /*
 975  *      This outputs /proc/net/route.
 976  *
 977  *      It always works in backward compatibility mode.
 978  *      The format of the file is not supposed to be changed.
 979  */
 980 static int fib_seq_show(struct seq_file *seq, void *v)
 981 {
 982         struct fib_iter_state *iter;
 983         char bf[128];
 984         __be32 prefix, mask;
 985         unsigned flags;
 986         struct fib_node *f;
 987         struct fib_alias *fa;
 988         struct fib_info *fi;
 989
 990         if (v == SEQ_START_TOKEN) {
 991                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
 992                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
 993                            "\tWindow\tIRTT");
 994                 goto out;
 995         }
 996
 997         iter    = seq->private;
 998         f       = iter->fn;
 999         fa      = iter->fa;
1000         fi      = fa->fa_info;
1001         prefix  = f->fn_key;
1002         mask    = FZ_MASK(iter->zone);
1003         flags   = fib_flag_trans(fa->fa_type, mask, fi);
1004         if (fi)
1005                 snprintf(bf, sizeof(bf),
1006                          "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1007                          fi->fib_dev ? fi->fib_dev->name : "*", prefix,
1008                          fi->fib_nh->nh_gw, flags, 0, 0, fi->fib_priority,
1009                          mask, (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
1010                          fi->fib_window,
1011                          fi->fib_rtt >> 3);
1012         else
1013                 snprintf(bf, sizeof(bf),
1014                          "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1015                          prefix, 0, flags, 0, 0, 0, mask, 0, 0, 0);
1016         seq_printf(seq, "%-127s\n", bf);
1017 out:
1018         return 0;
1019 }
1020
1021 static const struct seq_operations fib_seq_ops = {
1022         .start  = fib_seq_start,
1023         .next   = fib_seq_next,
1024         .stop   = fib_seq_stop,
1025         .show   = fib_seq_show,
1026 };
1027
1028 static int fib_seq_open(struct inode *inode, struct file *file)
1029 {
1030         return seq_open_private(file, &fib_seq_ops,
1031                         sizeof(struct fib_iter_state));
1032 }
1033
1034 static const struct file_operations fib_seq_fops = {
1035         .owner          = THIS_MODULE,
1036         .open           = fib_seq_open,
1037         .read           = seq_read,
1038         .llseek         = seq_lseek,
1039         .release        = seq_release_private,
1040 };
1041
1042 int __init fib_proc_init(void)
1043 {
1044         if (!proc_net_fops_create(&init_net, "route", S_IRUGO, &fib_seq_fops))
1045                 return -ENOMEM;
1046         return 0;
1047 }
1048
1049 void __init fib_proc_exit(void)
1050 {
1051         proc_net_remove(&init_net, "route");
1052 }
1053 #endif /* CONFIG_PROC_FS */