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         struct fib_alias        fn_embedded_alias;
  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 kzalloc(size, GFP_KERNEL);
 107         } else {
 108                 return (struct hlist_head *)
 109                         __get_free_pages(GFP_KERNEL | __GFP_ZERO, 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(KERN_DEBUG "fn_rehash_zone: hash for zone %d grows from %d\n",
 173                fz->fz_order, old_divisor);
 174 #endif
 175
 176         ht = fz_hash_alloc(new_divisor);
 177
 178         if (ht) {
 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, struct fib_node *f)
 198 {
 199         fib_release_info(fa->fa_info);
 200         if (fa == &f->fn_embedded_alias)
 201                 fa->fa_info = NULL;
 202         else
 203                 kmem_cache_free(fn_alias_kmem, fa);
 204 }
 205
 206 static struct fn_zone *
 207 fn_new_zone(struct fn_hash *table, int z)
 208 {
 209         int i;
 210         struct fn_zone *fz = kzalloc(sizeof(struct fn_zone), GFP_KERNEL);
 211         if (!fz)
 212                 return NULL;
 213
 214         if (z) {
 215                 fz->fz_divisor = 16;
 216         } else {
 217                 fz->fz_divisor = 1;
 218         }
 219         fz->fz_hashmask = (fz->fz_divisor - 1);
 220         fz->fz_hash = fz_hash_alloc(fz->fz_divisor);
 221         if (!fz->fz_hash) {
 222                 kfree(fz);
 223                 return NULL;
 224         }
 225         fz->fz_order = z;
 226         fz->fz_mask = inet_make_mask(z);
 227
 228         /* Find the first not empty zone with more specific mask */
 229         for (i=z+1; i<=32; i++)
 230                 if (table->fn_zones[i])
 231                         break;
 232         write_lock_bh(&fib_hash_lock);
 233         if (i>32) {
 234                 /* No more specific masks, we are the first. */
 235                 fz->fz_next = table->fn_zone_list;
 236                 table->fn_zone_list = fz;
 237         } else {
 238                 fz->fz_next = table->fn_zones[i]->fz_next;
 239                 table->fn_zones[i]->fz_next = fz;
 240         }
 241         table->fn_zones[z] = fz;
 242         fib_hash_genid++;
 243         write_unlock_bh(&fib_hash_lock);
 244         return fz;
 245 }
 246
 247 static int
 248 fn_hash_lookup(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
 249 {
 250         int err;
 251         struct fn_zone *fz;
 252         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 253
 254         read_lock(&fib_hash_lock);
 255         for (fz = t->fn_zone_list; fz; fz = fz->fz_next) {
 256                 struct hlist_head *head;
 257                 struct hlist_node *node;
 258                 struct fib_node *f;
 259                 __be32 k = fz_key(flp->fl4_dst, fz);
 260
 261                 head = &fz->fz_hash[fn_hash(k, fz)];
 262                 hlist_for_each_entry(f, node, head, fn_hash) {
 263                         if (f->fn_key != k)
 264                                 continue;
 265
 266                         err = fib_semantic_match(&f->fn_alias,
 267                                                  flp, res,
 268                                                  f->fn_key, fz->fz_mask,
 269                                                  fz->fz_order);
 270                         if (err <= 0)
 271                                 goto out;
 272                 }
 273         }
 274         err = 1;
 275 out:
 276         read_unlock(&fib_hash_lock);
 277         return err;
 278 }
 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, tb->tb_default)) {
 321                                 fib_result_assign(res, fi);
 322                                 tb->tb_default = order;
 323                                 goto out;
 324                         }
 325                         fi = next_fi;
 326                         order++;
 327                 }
 328         }
 329
 330         if (order <= 0 || fi == NULL) {
 331                 tb->tb_default = -1;
 332                 goto out;
 333         }
 334
 335         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
 336                                 tb->tb_default)) {
 337                 fib_result_assign(res, fi);
 338                 tb->tb_default = order;
 339                 goto out;
 340         }
 341
 342         if (last_idx >= 0)
 343                 fib_result_assign(res, last_resort);
 344         tb->tb_default = last_idx;
 345 out:
 346         read_unlock(&fib_hash_lock);
 347 }
 348
 349 /* Insert node F to FZ. */
 350 static inline void fib_insert_node(struct fn_zone *fz, struct fib_node *f)
 351 {
 352         struct hlist_head *head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 353
 354         hlist_add_head(&f->fn_hash, head);
 355 }
 356
 357 /* Return the node in FZ matching KEY. */
 358 static struct fib_node *fib_find_node(struct fn_zone *fz, __be32 key)
 359 {
 360         struct hlist_head *head = &fz->fz_hash[fn_hash(key, fz)];
 361         struct hlist_node *node;
 362         struct fib_node *f;
 363
 364         hlist_for_each_entry(f, node, head, fn_hash) {
 365                 if (f->fn_key == key)
 366                         return f;
 367         }
 368
 369         return NULL;
 370 }
 371
 372 static int fn_hash_insert(struct fib_table *tb, struct fib_config *cfg)
 373 {
 374         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 375         struct fib_node *new_f, *f;
 376         struct fib_alias *fa, *new_fa;
 377         struct fn_zone *fz;
 378         struct fib_info *fi;
 379         u8 tos = cfg->fc_tos;
 380         __be32 key;
 381         int err;
 382
 383         if (cfg->fc_dst_len > 32)
 384                 return -EINVAL;
 385
 386         fz = table->fn_zones[cfg->fc_dst_len];
 387         if (!fz && !(fz = fn_new_zone(table, cfg->fc_dst_len)))
 388                 return -ENOBUFS;
 389
 390         key = 0;
 391         if (cfg->fc_dst) {
 392                 if (cfg->fc_dst & ~FZ_MASK(fz))
 393                         return -EINVAL;
 394                 key = fz_key(cfg->fc_dst, fz);
 395         }
 396
 397         fi = fib_create_info(cfg);
 398         if (IS_ERR(fi))
 399                 return PTR_ERR(fi);
 400
 401         if (fz->fz_nent > (fz->fz_divisor<<1) &&
 402             fz->fz_divisor < FZ_MAX_DIVISOR &&
 403             (cfg->fc_dst_len == 32 ||
 404              (1 << cfg->fc_dst_len) > fz->fz_divisor))
 405                 fn_rehash_zone(fz);
 406
 407         f = fib_find_node(fz, key);
 408
 409         if (!f)
 410                 fa = NULL;
 411         else
 412                 fa = fib_find_alias(&f->fn_alias, tos, fi->fib_priority);
 413
 414         /* Now fa, if non-NULL, points to the first fib alias
 415          * with the same keys [prefix,tos,priority], if such key already
 416          * exists or to the node before which we will insert new one.
 417          *
 418          * If fa is NULL, we will need to allocate a new one and
 419          * insert to the head of f.
 420          *
 421          * If f is NULL, no fib node matched the destination key
 422          * and we need to allocate a new one of those as well.
 423          */
 424
 425         if (fa && fa->fa_tos == tos &&
 426             fa->fa_info->fib_priority == fi->fib_priority) {
 427                 struct fib_alias *fa_first, *fa_match;
 428
 429                 err = -EEXIST;
 430                 if (cfg->fc_nlflags & NLM_F_EXCL)
 431                         goto out;
 432
 433                 /* We have 2 goals:
 434                  * 1. Find exact match for type, scope, fib_info to avoid
 435                  * duplicate routes
 436                  * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
 437                  */
 438                 fa_match = NULL;
 439                 fa_first = fa;
 440                 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 441                 list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 442                         if (fa->fa_tos != tos)
 443                                 break;
 444                         if (fa->fa_info->fib_priority != fi->fib_priority)
 445                                 break;
 446                         if (fa->fa_type == cfg->fc_type &&
 447                             fa->fa_scope == cfg->fc_scope &&
 448                             fa->fa_info == fi) {
 449                                 fa_match = fa;
 450                                 break;
 451                         }
 452                 }
 453
 454                 if (cfg->fc_nlflags & NLM_F_REPLACE) {
 455                         struct fib_info *fi_drop;
 456                         u8 state;
 457
 458                         fa = fa_first;
 459                         if (fa_match) {
 460                                 if (fa == fa_match)
 461                                         err = 0;
 462                                 goto out;
 463                         }
 464                         write_lock_bh(&fib_hash_lock);
 465                         fi_drop = fa->fa_info;
 466                         fa->fa_info = fi;
 467                         fa->fa_type = cfg->fc_type;
 468                         fa->fa_scope = cfg->fc_scope;
 469                         state = fa->fa_state;
 470                         fa->fa_state &= ~FA_S_ACCESSED;
 471                         fib_hash_genid++;
 472                         write_unlock_bh(&fib_hash_lock);
 473
 474                         fib_release_info(fi_drop);
 475                         if (state & FA_S_ACCESSED)
 476                                 rt_cache_flush(-1);
 477                         rtmsg_fib(RTM_NEWROUTE, key, fa, cfg->fc_dst_len, tb->tb_id,
 478                                   &cfg->fc_nlinfo, NLM_F_REPLACE);
 479                         return 0;
 480                 }
 481
 482                 /* Error if we find a perfect match which
 483                  * uses the same scope, type, and nexthop
 484                  * information.
 485                  */
 486                 if (fa_match)
 487                         goto out;
 488
 489                 if (!(cfg->fc_nlflags & NLM_F_APPEND))
 490                         fa = fa_first;
 491         }
 492
 493         err = -ENOENT;
 494         if (!(cfg->fc_nlflags & NLM_F_CREATE))
 495                 goto out;
 496
 497         err = -ENOBUFS;
 498
 499         new_f = NULL;
 500         if (!f) {
 501                 new_f = kmem_cache_zalloc(fn_hash_kmem, GFP_KERNEL);
 502                 if (new_f == NULL)
 503                         goto out;
 504
 505                 INIT_HLIST_NODE(&new_f->fn_hash);
 506                 INIT_LIST_HEAD(&new_f->fn_alias);
 507                 new_f->fn_key = key;
 508                 f = new_f;
 509         }
 510
 511         new_fa = &f->fn_embedded_alias;
 512         if (new_fa->fa_info != NULL) {
 513                 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
 514                 if (new_fa == NULL)
 515                         goto out_free_new_f;
 516         }
 517         new_fa->fa_info = fi;
 518         new_fa->fa_tos = tos;
 519         new_fa->fa_type = cfg->fc_type;
 520         new_fa->fa_scope = cfg->fc_scope;
 521         new_fa->fa_state = 0;
 522
 523         /*
 524          * Insert new entry to the list.
 525          */
 526
 527         write_lock_bh(&fib_hash_lock);
 528         if (new_f)
 529                 fib_insert_node(fz, new_f);
 530         list_add_tail(&new_fa->fa_list,
 531                  (fa ? &fa->fa_list : &f->fn_alias));
 532         fib_hash_genid++;
 533         write_unlock_bh(&fib_hash_lock);
 534
 535         if (new_f)
 536                 fz->fz_nent++;
 537         rt_cache_flush(-1);
 538
 539         rtmsg_fib(RTM_NEWROUTE, key, new_fa, cfg->fc_dst_len, tb->tb_id,
 540                   &cfg->fc_nlinfo, 0);
 541         return 0;
 542
 543 out_free_new_f:
 544         kmem_cache_free(fn_hash_kmem, new_f);
 545 out:
 546         fib_release_info(fi);
 547         return err;
 548 }
 549
 550
 551 static int fn_hash_delete(struct fib_table *tb, struct fib_config *cfg)
 552 {
 553         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 554         struct fib_node *f;
 555         struct fib_alias *fa, *fa_to_delete;
 556         struct fn_zone *fz;
 557         __be32 key;
 558
 559         if (cfg->fc_dst_len > 32)
 560                 return -EINVAL;
 561
 562         if ((fz  = table->fn_zones[cfg->fc_dst_len]) == NULL)
 563                 return -ESRCH;
 564
 565         key = 0;
 566         if (cfg->fc_dst) {
 567                 if (cfg->fc_dst & ~FZ_MASK(fz))
 568                         return -EINVAL;
 569                 key = fz_key(cfg->fc_dst, fz);
 570         }
 571
 572         f = fib_find_node(fz, key);
 573
 574         if (!f)
 575                 fa = NULL;
 576         else
 577                 fa = fib_find_alias(&f->fn_alias, cfg->fc_tos, 0);
 578         if (!fa)
 579                 return -ESRCH;
 580
 581         fa_to_delete = NULL;
 582         fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 583         list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 584                 struct fib_info *fi = fa->fa_info;
 585
 586                 if (fa->fa_tos != cfg->fc_tos)
 587                         break;
 588
 589                 if ((!cfg->fc_type ||
 590                      fa->fa_type == cfg->fc_type) &&
 591                     (cfg->fc_scope == RT_SCOPE_NOWHERE ||
 592                      fa->fa_scope == cfg->fc_scope) &&
 593                     (!cfg->fc_protocol ||
 594                      fi->fib_protocol == cfg->fc_protocol) &&
 595                     fib_nh_match(cfg, fi) == 0) {
 596                         fa_to_delete = fa;
 597                         break;
 598                 }
 599         }
 600
 601         if (fa_to_delete) {
 602                 int kill_fn;
 603
 604                 fa = fa_to_delete;
 605                 rtmsg_fib(RTM_DELROUTE, key, fa, cfg->fc_dst_len,
 606                           tb->tb_id, &cfg->fc_nlinfo, 0);
 607
 608                 kill_fn = 0;
 609                 write_lock_bh(&fib_hash_lock);
 610                 list_del(&fa->fa_list);
 611                 if (list_empty(&f->fn_alias)) {
 612                         hlist_del(&f->fn_hash);
 613                         kill_fn = 1;
 614                 }
 615                 fib_hash_genid++;
 616                 write_unlock_bh(&fib_hash_lock);
 617
 618                 if (fa->fa_state & FA_S_ACCESSED)
 619                         rt_cache_flush(-1);
 620                 fn_free_alias(fa, f);
 621                 if (kill_fn) {
 622                         fn_free_node(f);
 623                         fz->fz_nent--;
 624                 }
 625
 626                 return 0;
 627         }
 628         return -ESRCH;
 629 }
 630
 631 static int fn_flush_list(struct fn_zone *fz, int idx)
 632 {
 633         struct hlist_head *head = &fz->fz_hash[idx];
 634         struct hlist_node *node, *n;
 635         struct fib_node *f;
 636         int found = 0;
 637
 638         hlist_for_each_entry_safe(f, node, n, head, fn_hash) {
 639                 struct fib_alias *fa, *fa_node;
 640                 int kill_f;
 641
 642                 kill_f = 0;
 643                 list_for_each_entry_safe(fa, fa_node, &f->fn_alias, fa_list) {
 644                         struct fib_info *fi = fa->fa_info;
 645
 646                         if (fi && (fi->fib_flags&RTNH_F_DEAD)) {
 647                                 write_lock_bh(&fib_hash_lock);
 648                                 list_del(&fa->fa_list);
 649                                 if (list_empty(&f->fn_alias)) {
 650                                         hlist_del(&f->fn_hash);
 651                                         kill_f = 1;
 652                                 }
 653                                 fib_hash_genid++;
 654                                 write_unlock_bh(&fib_hash_lock);
 655
 656                                 fn_free_alias(fa, f);
 657                                 found++;
 658                         }
 659                 }
 660                 if (kill_f) {
 661                         fn_free_node(f);
 662                         fz->fz_nent--;
 663                 }
 664         }
 665         return found;
 666 }
 667
 668 static int fn_hash_flush(struct fib_table *tb)
 669 {
 670         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 671         struct fn_zone *fz;
 672         int found = 0;
 673
 674         for (fz = table->fn_zone_list; fz; fz = fz->fz_next) {
 675                 int i;
 676
 677                 for (i = fz->fz_divisor - 1; i >= 0; i--)
 678                         found += fn_flush_list(fz, i);
 679         }
 680         return found;
 681 }
 682
 683
 684 static inline int
 685 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
 686                      struct fib_table *tb,
 687                      struct fn_zone *fz,
 688                      struct hlist_head *head)
 689 {
 690         struct hlist_node *node;
 691         struct fib_node *f;
 692         int i, s_i;
 693
 694         s_i = cb->args[4];
 695         i = 0;
 696         hlist_for_each_entry(f, node, head, fn_hash) {
 697                 struct fib_alias *fa;
 698
 699                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 700                         if (i < s_i)
 701                                 goto next;
 702
 703                         if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
 704                                           cb->nlh->nlmsg_seq,
 705                                           RTM_NEWROUTE,
 706                                           tb->tb_id,
 707                                           fa->fa_type,
 708                                           fa->fa_scope,
 709                                           f->fn_key,
 710                                           fz->fz_order,
 711                                           fa->fa_tos,
 712                                           fa->fa_info,
 713                                           NLM_F_MULTI) < 0) {
 714                                 cb->args[4] = i;
 715                                 return -1;
 716                         }
 717                 next:
 718                         i++;
 719                 }
 720         }
 721         cb->args[4] = i;
 722         return skb->len;
 723 }
 724
 725 static inline int
 726 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
 727                    struct fib_table *tb,
 728                    struct fn_zone *fz)
 729 {
 730         int h, s_h;
 731
 732         if (fz->fz_hash == NULL)
 733                 return skb->len;
 734         s_h = cb->args[3];
 735         for (h = s_h; h < fz->fz_divisor; h++) {
 736                 if (hlist_empty(&fz->fz_hash[h]))
 737                         continue;
 738                 if (fn_hash_dump_bucket(skb, cb, tb, fz, &fz->fz_hash[h]) < 0) {
 739                         cb->args[3] = h;
 740                         return -1;
 741                 }
 742                 memset(&cb->args[4], 0,
 743                        sizeof(cb->args) - 4*sizeof(cb->args[0]));
 744         }
 745         cb->args[3] = h;
 746         return skb->len;
 747 }
 748
 749 static int fn_hash_dump(struct fib_table *tb, struct sk_buff *skb, struct netlink_callback *cb)
 750 {
 751         int m, s_m;
 752         struct fn_zone *fz;
 753         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 754
 755         s_m = cb->args[2];
 756         read_lock(&fib_hash_lock);
 757         for (fz = table->fn_zone_list, m=0; fz; fz = fz->fz_next, m++) {
 758                 if (m < s_m) continue;
 759                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
 760                         cb->args[2] = m;
 761                         read_unlock(&fib_hash_lock);
 762                         return -1;
 763                 }
 764                 memset(&cb->args[3], 0,
 765                        sizeof(cb->args) - 3*sizeof(cb->args[0]));
 766         }
 767         read_unlock(&fib_hash_lock);
 768         cb->args[2] = m;
 769         return skb->len;
 770 }
 771
 772 void __init fib_hash_init(void)
 773 {
 774         fn_hash_kmem = kmem_cache_create("ip_fib_hash", sizeof(struct fib_node),
 775                                          0, SLAB_PANIC, NULL);
 776
 777         fn_alias_kmem = kmem_cache_create("ip_fib_alias", sizeof(struct fib_alias),
 778                                           0, SLAB_PANIC, NULL);
 779
 780 }
 781
 782 struct fib_table *fib_hash_table(u32 id)
 783 {
 784         struct fib_table *tb;
 785
 786         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash),
 787                      GFP_KERNEL);
 788         if (tb == NULL)
 789                 return NULL;
 790
 791         tb->tb_id = id;
 792         tb->tb_default = -1;
 793         tb->tb_lookup = fn_hash_lookup;
 794         tb->tb_insert = fn_hash_insert;
 795         tb->tb_delete = fn_hash_delete;
 796         tb->tb_flush = fn_hash_flush;
 797         tb->tb_select_default = fn_hash_select_default;
 798         tb->tb_dump = fn_hash_dump;
 799         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 800         return tb;
 801 }
 802
 803 /* ------------------------------------------------------------------------ */
 804 #ifdef CONFIG_PROC_FS
 805
 806 struct fib_iter_state {
 807         struct seq_net_private p;
 808         struct fn_zone  *zone;
 809         int             bucket;
 810         struct hlist_head *hash_head;
 811         struct fib_node *fn;
 812         struct fib_alias *fa;
 813         loff_t pos;
 814         unsigned int genid;
 815         int valid;
 816 };
 817
 818 static struct fib_alias *fib_get_first(struct seq_file *seq)
 819 {
 820         struct fib_iter_state *iter = seq->private;
 821         struct fib_table *main_table;
 822         struct fn_hash *table;
 823
 824         main_table = fib_get_table(iter->p.net, RT_TABLE_MAIN);
 825         table = (struct fn_hash *)main_table->tb_data;
 826
 827         iter->bucket    = 0;
 828         iter->hash_head = NULL;
 829         iter->fn        = NULL;
 830         iter->fa        = NULL;
 831         iter->pos       = 0;
 832         iter->genid     = fib_hash_genid;
 833         iter->valid     = 1;
 834
 835         for (iter->zone = table->fn_zone_list; iter->zone;
 836              iter->zone = iter->zone->fz_next) {
 837                 int maxslot;
 838
 839                 if (!iter->zone->fz_nent)
 840                         continue;
 841
 842                 iter->hash_head = iter->zone->fz_hash;
 843                 maxslot = iter->zone->fz_divisor;
 844
 845                 for (iter->bucket = 0; iter->bucket < maxslot;
 846                      ++iter->bucket, ++iter->hash_head) {
 847                         struct hlist_node *node;
 848                         struct fib_node *fn;
 849
 850                         hlist_for_each_entry(fn,node,iter->hash_head,fn_hash) {
 851                                 struct fib_alias *fa;
 852
 853                                 list_for_each_entry(fa,&fn->fn_alias,fa_list) {
 854                                         iter->fn = fn;
 855                                         iter->fa = fa;
 856                                         goto out;
 857                                 }
 858                         }
 859                 }
 860         }
 861 out:
 862         return iter->fa;
 863 }
 864
 865 static struct fib_alias *fib_get_next(struct seq_file *seq)
 866 {
 867         struct fib_iter_state *iter = seq->private;
 868         struct fib_node *fn;
 869         struct fib_alias *fa;
 870
 871         /* Advance FA, if any. */
 872         fn = iter->fn;
 873         fa = iter->fa;
 874         if (fa) {
 875                 BUG_ON(!fn);
 876                 list_for_each_entry_continue(fa, &fn->fn_alias, fa_list) {
 877                         iter->fa = fa;
 878                         goto out;
 879                 }
 880         }
 881
 882         fa = iter->fa = NULL;
 883
 884         /* Advance FN. */
 885         if (fn) {
 886                 struct hlist_node *node = &fn->fn_hash;
 887                 hlist_for_each_entry_continue(fn, node, fn_hash) {
 888                         iter->fn = fn;
 889
 890                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 891                                 iter->fa = fa;
 892                                 goto out;
 893                         }
 894                 }
 895         }
 896
 897         fn = iter->fn = NULL;
 898
 899         /* Advance hash chain. */
 900         if (!iter->zone)
 901                 goto out;
 902
 903         for (;;) {
 904                 struct hlist_node *node;
 905                 int maxslot;
 906
 907                 maxslot = iter->zone->fz_divisor;
 908
 909                 while (++iter->bucket < maxslot) {
 910                         iter->hash_head++;
 911
 912                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 913                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 914                                         iter->fn = fn;
 915                                         iter->fa = fa;
 916                                         goto out;
 917                                 }
 918                         }
 919                 }
 920
 921                 iter->zone = iter->zone->fz_next;
 922
 923                 if (!iter->zone)
 924                         goto out;
 925
 926                 iter->bucket = 0;
 927                 iter->hash_head = iter->zone->fz_hash;
 928
 929                 hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 930                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 931                                 iter->fn = fn;
 932                                 iter->fa = fa;
 933                                 goto out;
 934                         }
 935                 }
 936         }
 937 out:
 938         iter->pos++;
 939         return fa;
 940 }
 941
 942 static struct fib_alias *fib_get_idx(struct seq_file *seq, loff_t pos)
 943 {
 944         struct fib_iter_state *iter = seq->private;
 945         struct fib_alias *fa;
 946
 947         if (iter->valid && pos >= iter->pos && iter->genid == fib_hash_genid) {
 948                 fa   = iter->fa;
 949                 pos -= iter->pos;
 950         } else
 951                 fa = fib_get_first(seq);
 952
 953         if (fa)
 954                 while (pos && (fa = fib_get_next(seq)))
 955                         --pos;
 956         return pos ? NULL : fa;
 957 }
 958
 959 static void *fib_seq_start(struct seq_file *seq, loff_t *pos)
 960         __acquires(fib_hash_lock)
 961 {
 962         struct fib_iter_state *iter = seq->private;
 963         void *v = NULL;
 964
 965         read_lock(&fib_hash_lock);
 966         if (fib_get_table(iter->p.net, RT_TABLE_MAIN))
 967                 v = *pos ? fib_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 968         return v;
 969 }
 970
 971 static void *fib_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 972 {
 973         ++*pos;
 974         return v == SEQ_START_TOKEN ? fib_get_first(seq) : fib_get_next(seq);
 975 }
 976
 977 static void fib_seq_stop(struct seq_file *seq, void *v)
 978         __releases(fib_hash_lock)
 979 {
 980         read_unlock(&fib_hash_lock);
 981 }
 982
 983 static unsigned fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
 984 {
 985         static const unsigned type2flags[RTN_MAX + 1] = {
 986                 [7] = RTF_REJECT, [8] = RTF_REJECT,
 987         };
 988         unsigned flags = type2flags[type];
 989
 990         if (fi && fi->fib_nh->nh_gw)
 991                 flags |= RTF_GATEWAY;
 992         if (mask == htonl(0xFFFFFFFF))
 993                 flags |= RTF_HOST;
 994         flags |= RTF_UP;
 995         return flags;
 996 }
 997
 998 /*
 999  *      This outputs /proc/net/route.
1000  *
1001  *      It always works in backward compatibility mode.
1002  *      The format of the file is not supposed to be changed.
1003  */
1004 static int fib_seq_show(struct seq_file *seq, void *v)
1005 {
1006         struct fib_iter_state *iter;
1007         char bf[128];
1008         __be32 prefix, mask;
1009         unsigned flags;
1010         struct fib_node *f;
1011         struct fib_alias *fa;
1012         struct fib_info *fi;
1013
1014         if (v == SEQ_START_TOKEN) {
1015                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
1016                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
1017                            "\tWindow\tIRTT");
1018                 goto out;
1019         }
1020
1021         iter    = seq->private;
1022         f       = iter->fn;
1023         fa      = iter->fa;
1024         fi      = fa->fa_info;
1025         prefix  = f->fn_key;
1026         mask    = FZ_MASK(iter->zone);
1027         flags   = fib_flag_trans(fa->fa_type, mask, fi);
1028         if (fi)
1029                 snprintf(bf, sizeof(bf),
1030                          "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1031                          fi->fib_dev ? fi->fib_dev->name : "*", prefix,
1032                          fi->fib_nh->nh_gw, flags, 0, 0, fi->fib_priority,
1033                          mask, (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
1034                          fi->fib_window,
1035                          fi->fib_rtt >> 3);
1036         else
1037                 snprintf(bf, sizeof(bf),
1038                          "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1039                          prefix, 0, flags, 0, 0, 0, mask, 0, 0, 0);
1040         seq_printf(seq, "%-127s\n", bf);
1041 out:
1042         return 0;
1043 }
1044
1045 static const struct seq_operations fib_seq_ops = {
1046         .start  = fib_seq_start,
1047         .next   = fib_seq_next,
1048         .stop   = fib_seq_stop,
1049         .show   = fib_seq_show,
1050 };
1051
1052 static int fib_seq_open(struct inode *inode, struct file *file)
1053 {
1054         return seq_open_net(inode, file, &fib_seq_ops,
1055                             sizeof(struct fib_iter_state));
1056 }
1057
1058 static const struct file_operations fib_seq_fops = {
1059         .owner          = THIS_MODULE,
1060         .open           = fib_seq_open,
1061         .read           = seq_read,
1062         .llseek         = seq_lseek,
1063         .release        = seq_release_net,
1064 };
1065
1066 int __net_init fib_proc_init(struct net *net)
1067 {
1068         if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_seq_fops))
1069                 return -ENOMEM;
1070         return 0;
1071 }
1072
1073 void __net_exit fib_proc_exit(struct net *net)
1074 {
1075         proc_net_remove(net, "route");
1076 }
1077 #endif /* CONFIG_PROC_FS */