net/ipv4/fib_hash.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              IPv4 FIB: lookup engine and maintenance routines.
   7  *
   8  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
   9  *
  10  *              This program is free software; you can redistribute it and/or
  11  *              modify it under the terms of the GNU General Public License
  12  *              as published by the Free Software Foundation; either version
  13  *              2 of the License, or (at your option) any later version.
  14  */
  15
  16 #include <asm/uaccess.h>
  17 #include <asm/system.h>
  18 #include <linux/bitops.h>
  19 #include <linux/types.h>
  20 #include <linux/kernel.h>
  21 #include <linux/mm.h>
  22 #include <linux/string.h>
  23 #include <linux/socket.h>
  24 #include <linux/sockios.h>
  25 #include <linux/errno.h>
  26 #include <linux/in.h>
  27 #include <linux/inet.h>
  28 #include <linux/inetdevice.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/if_arp.h>
  31 #include <linux/proc_fs.h>
  32 #include <linux/skbuff.h>
  33 #include <linux/netlink.h>
  34 #include <linux/init.h>
  35 #include <linux/slab.h>
  36
  37 #include <net/net_namespace.h>
  38 #include <net/ip.h>
  39 #include <net/protocol.h>
  40 #include <net/route.h>
  41 #include <net/tcp.h>
  42 #include <net/sock.h>
  43 #include <net/ip_fib.h>
  44
  45 #include "fib_lookup.h"
  46
  47 static struct kmem_cache *fn_hash_kmem __read_mostly;
  48 static struct kmem_cache *fn_alias_kmem __read_mostly;
  49
  50 struct fib_node {
  51         struct hlist_node       fn_hash;
  52         struct list_head        fn_alias;
  53         __be32                  fn_key;
  54         struct fib_alias        fn_embedded_alias;
  55 };
  56
  57 struct fn_zone {
  58         struct fn_zone          *fz_next;       /* Next not empty zone  */
  59         struct hlist_head       *fz_hash;       /* Hash table pointer   */
  60         int                     fz_nent;        /* Number of entries    */
  61
  62         int                     fz_divisor;     /* Hash divisor         */
  63         u32                     fz_hashmask;    /* (fz_divisor - 1)     */
  64 #define FZ_HASHMASK(fz)         ((fz)->fz_hashmask)
  65
  66         int                     fz_order;       /* Zone order           */
  67         __be32                  fz_mask;
  68 #define FZ_MASK(fz)             ((fz)->fz_mask)
  69 };
  70
  71 /* NOTE. On fast computers evaluation of fz_hashmask and fz_mask
  72  * can be cheaper than memory lookup, so that FZ_* macros are used.
  73  */
  74
  75 struct fn_hash {
  76         struct fn_zone  *fn_zones[33];
  77         struct fn_zone  *fn_zone_list;
  78 };
  79
  80 static inline u32 fn_hash(__be32 key, struct fn_zone *fz)
  81 {
  82         u32 h = ntohl(key)>>(32 - fz->fz_order);
  83         h ^= (h>>20);
  84         h ^= (h>>10);
  85         h ^= (h>>5);
  86         h &= FZ_HASHMASK(fz);
  87         return h;
  88 }
  89
  90 static inline __be32 fz_key(__be32 dst, struct fn_zone *fz)
  91 {
  92         return dst & FZ_MASK(fz);
  93 }
  94
  95 static DEFINE_RWLOCK(fib_hash_lock);
  96 static unsigned int fib_hash_genid;
  97
  98 #define FZ_MAX_DIVISOR ((PAGE_SIZE<<MAX_ORDER) / sizeof(struct hlist_head))
  99
 100 static struct hlist_head *fz_hash_alloc(int divisor)
 101 {
 102         unsigned long size = divisor * sizeof(struct hlist_head);
 103
 104         if (size <= PAGE_SIZE) {
 105                 return kzalloc(size, GFP_KERNEL);
 106         } else {
 107                 return (struct hlist_head *)
 108                         __get_free_pages(GFP_KERNEL | __GFP_ZERO, get_order(size));
 109         }
 110 }
 111
 112 /* The fib hash lock must be held when this is called. */
 113 static inline void fn_rebuild_zone(struct fn_zone *fz,
 114                                    struct hlist_head *old_ht,
 115                                    int old_divisor)
 116 {
 117         int i;
 118
 119         for (i = 0; i < old_divisor; i++) {
 120                 struct hlist_node *node, *n;
 121                 struct fib_node *f;
 122
 123                 hlist_for_each_entry_safe(f, node, n, &old_ht[i], fn_hash) {
 124                         struct hlist_head *new_head;
 125
 126                         hlist_del(&f->fn_hash);
 127
 128                         new_head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 129                         hlist_add_head(&f->fn_hash, new_head);
 130                 }
 131         }
 132 }
 133
 134 static void fz_hash_free(struct hlist_head *hash, int divisor)
 135 {
 136         unsigned long size = divisor * sizeof(struct hlist_head);
 137
 138         if (size <= PAGE_SIZE)
 139                 kfree(hash);
 140         else
 141                 free_pages((unsigned long)hash, get_order(size));
 142 }
 143
 144 static void fn_rehash_zone(struct fn_zone *fz)
 145 {
 146         struct hlist_head *ht, *old_ht;
 147         int old_divisor, new_divisor;
 148         u32 new_hashmask;
 149
 150         old_divisor = fz->fz_divisor;
 151
 152         switch (old_divisor) {
 153         case 16:
 154                 new_divisor = 256;
 155                 break;
 156         case 256:
 157                 new_divisor = 1024;
 158                 break;
 159         default:
 160                 if ((old_divisor << 1) > FZ_MAX_DIVISOR) {
 161                         printk(KERN_CRIT "route.c: bad divisor %d!\n", old_divisor);
 162                         return;
 163                 }
 164                 new_divisor = (old_divisor << 1);
 165                 break;
 166         }
 167
 168         new_hashmask = (new_divisor - 1);
 169
 170 #if RT_CACHE_DEBUG >= 2
 171         printk(KERN_DEBUG "fn_rehash_zone: hash for zone %d grows from %d\n",
 172                fz->fz_order, old_divisor);
 173 #endif
 174
 175         ht = fz_hash_alloc(new_divisor);
 176
 177         if (ht) {
 178                 write_lock_bh(&fib_hash_lock);
 179                 old_ht = fz->fz_hash;
 180                 fz->fz_hash = ht;
 181                 fz->fz_hashmask = new_hashmask;
 182                 fz->fz_divisor = new_divisor;
 183                 fn_rebuild_zone(fz, old_ht, old_divisor);
 184                 fib_hash_genid++;
 185                 write_unlock_bh(&fib_hash_lock);
 186
 187                 fz_hash_free(old_ht, old_divisor);
 188         }
 189 }
 190
 191 static inline void fn_free_node(struct fib_node * f)
 192 {
 193         kmem_cache_free(fn_hash_kmem, f);
 194 }
 195
 196 static inline void fn_free_alias(struct fib_alias *fa, struct fib_node *f)
 197 {
 198         fib_release_info(fa->fa_info);
 199         if (fa == &f->fn_embedded_alias)
 200                 fa->fa_info = NULL;
 201         else
 202                 kmem_cache_free(fn_alias_kmem, fa);
 203 }
 204
 205 static struct fn_zone *
 206 fn_new_zone(struct fn_hash *table, int z)
 207 {
 208         int i;
 209         struct fn_zone *fz = kzalloc(sizeof(struct fn_zone), GFP_KERNEL);
 210         if (!fz)
 211                 return NULL;
 212
 213         if (z) {
 214                 fz->fz_divisor = 16;
 215         } else {
 216                 fz->fz_divisor = 1;
 217         }
 218         fz->fz_hashmask = (fz->fz_divisor - 1);
 219         fz->fz_hash = fz_hash_alloc(fz->fz_divisor);
 220         if (!fz->fz_hash) {
 221                 kfree(fz);
 222                 return NULL;
 223         }
 224         fz->fz_order = z;
 225         fz->fz_mask = inet_make_mask(z);
 226
 227         /* Find the first not empty zone with more specific mask */
 228         for (i=z+1; i<=32; i++)
 229                 if (table->fn_zones[i])
 230                         break;
 231         write_lock_bh(&fib_hash_lock);
 232         if (i>32) {
 233                 /* No more specific masks, we are the first. */
 234                 fz->fz_next = table->fn_zone_list;
 235                 table->fn_zone_list = fz;
 236         } else {
 237                 fz->fz_next = table->fn_zones[i]->fz_next;
 238                 table->fn_zones[i]->fz_next = fz;
 239         }
 240         table->fn_zones[z] = fz;
 241         fib_hash_genid++;
 242         write_unlock_bh(&fib_hash_lock);
 243         return fz;
 244 }
 245
 246 int fib_table_lookup(struct fib_table *tb,
 247                      const struct flowi *flp, struct fib_result *res,
 248                      int fib_flags)
 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                                                  fz->fz_order, fib_flags);
 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 void fib_table_select_default(struct fib_table *tb,
 280                               const struct flowi *flp, struct fib_result *res)
 281 {
 282         int order, last_idx;
 283         struct hlist_node *node;
 284         struct fib_node *f;
 285         struct fib_info *fi = NULL;
 286         struct fib_info *last_resort;
 287         struct fn_hash *t = (struct fn_hash *)tb->tb_data;
 288         struct fn_zone *fz = t->fn_zones[0];
 289
 290         if (fz == NULL)
 291                 return;
 292
 293         last_idx = -1;
 294         last_resort = NULL;
 295         order = -1;
 296
 297         read_lock(&fib_hash_lock);
 298         hlist_for_each_entry(f, node, &fz->fz_hash[0], fn_hash) {
 299                 struct fib_alias *fa;
 300
 301                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 302                         struct fib_info *next_fi = fa->fa_info;
 303
 304                         if (fa->fa_scope != res->scope ||
 305                             fa->fa_type != RTN_UNICAST)
 306                                 continue;
 307
 308                         if (next_fi->fib_priority > res->fi->fib_priority)
 309                                 break;
 310                         if (!next_fi->fib_nh[0].nh_gw ||
 311                             next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
 312                                 continue;
 313                         fa->fa_state |= FA_S_ACCESSED;
 314
 315                         if (fi == NULL) {
 316                                 if (next_fi != res->fi)
 317                                         break;
 318                         } else if (!fib_detect_death(fi, order, &last_resort,
 319                                                 &last_idx, tb->tb_default)) {
 320                                 fib_result_assign(res, fi);
 321                                 tb->tb_default = order;
 322                                 goto out;
 323                         }
 324                         fi = next_fi;
 325                         order++;
 326                 }
 327         }
 328
 329         if (order <= 0 || fi == NULL) {
 330                 tb->tb_default = -1;
 331                 goto out;
 332         }
 333
 334         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
 335                                 tb->tb_default)) {
 336                 fib_result_assign(res, fi);
 337                 tb->tb_default = order;
 338                 goto out;
 339         }
 340
 341         if (last_idx >= 0)
 342                 fib_result_assign(res, last_resort);
 343         tb->tb_default = last_idx;
 344 out:
 345         read_unlock(&fib_hash_lock);
 346 }
 347
 348 /* Insert node F to FZ. */
 349 static inline void fib_insert_node(struct fn_zone *fz, struct fib_node *f)
 350 {
 351         struct hlist_head *head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 352
 353         hlist_add_head(&f->fn_hash, head);
 354 }
 355
 356 /* Return the node in FZ matching KEY. */
 357 static struct fib_node *fib_find_node(struct fn_zone *fz, __be32 key)
 358 {
 359         struct hlist_head *head = &fz->fz_hash[fn_hash(key, fz)];
 360         struct hlist_node *node;
 361         struct fib_node *f;
 362
 363         hlist_for_each_entry(f, node, head, fn_hash) {
 364                 if (f->fn_key == key)
 365                         return f;
 366         }
 367
 368         return NULL;
 369 }
 370
 371 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
 372 {
 373         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 374         struct fib_node *new_f = NULL;
 375         struct fib_node *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(cfg->fc_nlinfo.nl_net, -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         if (!f) {
 500                 new_f = kmem_cache_zalloc(fn_hash_kmem, GFP_KERNEL);
 501                 if (new_f == NULL)
 502                         goto out;
 503
 504                 INIT_HLIST_NODE(&new_f->fn_hash);
 505                 INIT_LIST_HEAD(&new_f->fn_alias);
 506                 new_f->fn_key = key;
 507                 f = new_f;
 508         }
 509
 510         new_fa = &f->fn_embedded_alias;
 511         if (new_fa->fa_info != NULL) {
 512                 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
 513                 if (new_fa == NULL)
 514                         goto out;
 515         }
 516         new_fa->fa_info = fi;
 517         new_fa->fa_tos = tos;
 518         new_fa->fa_type = cfg->fc_type;
 519         new_fa->fa_scope = cfg->fc_scope;
 520         new_fa->fa_state = 0;
 521
 522         /*
 523          * Insert new entry to the list.
 524          */
 525
 526         write_lock_bh(&fib_hash_lock);
 527         if (new_f)
 528                 fib_insert_node(fz, new_f);
 529         list_add_tail(&new_fa->fa_list,
 530                  (fa ? &fa->fa_list : &f->fn_alias));
 531         fib_hash_genid++;
 532         write_unlock_bh(&fib_hash_lock);
 533
 534         if (new_f)
 535                 fz->fz_nent++;
 536         rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
 537
 538         rtmsg_fib(RTM_NEWROUTE, key, new_fa, cfg->fc_dst_len, tb->tb_id,
 539                   &cfg->fc_nlinfo, 0);
 540         return 0;
 541
 542 out:
 543         if (new_f)
 544                 kmem_cache_free(fn_hash_kmem, new_f);
 545         fib_release_info(fi);
 546         return err;
 547 }
 548
 549 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
 550 {
 551         struct fn_hash *table = (struct fn_hash *)tb->tb_data;
 552         struct fib_node *f;
 553         struct fib_alias *fa, *fa_to_delete;
 554         struct fn_zone *fz;
 555         __be32 key;
 556
 557         if (cfg->fc_dst_len > 32)
 558                 return -EINVAL;
 559
 560         if ((fz  = table->fn_zones[cfg->fc_dst_len]) == NULL)
 561                 return -ESRCH;
 562
 563         key = 0;
 564         if (cfg->fc_dst) {
 565                 if (cfg->fc_dst & ~FZ_MASK(fz))
 566                         return -EINVAL;
 567                 key = fz_key(cfg->fc_dst, fz);
 568         }
 569
 570         f = fib_find_node(fz, key);
 571
 572         if (!f)
 573                 fa = NULL;
 574         else
 575                 fa = fib_find_alias(&f->fn_alias, cfg->fc_tos, 0);
 576         if (!fa)
 577                 return -ESRCH;
 578
 579         fa_to_delete = NULL;
 580         fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 581         list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 582                 struct fib_info *fi = fa->fa_info;
 583
 584                 if (fa->fa_tos != cfg->fc_tos)
 585                         break;
 586
 587                 if ((!cfg->fc_type ||
 588                      fa->fa_type == cfg->fc_type) &&
 589                     (cfg->fc_scope == RT_SCOPE_NOWHERE ||
 590                      fa->fa_scope == cfg->fc_scope) &&
 591                     (!cfg->fc_protocol ||
 592                      fi->fib_protocol == cfg->fc_protocol) &&
 593                     fib_nh_match(cfg, fi) == 0) {
 594                         fa_to_delete = fa;
 595                         break;
 596                 }
 597         }
 598
 599         if (fa_to_delete) {
 600                 int kill_fn;
 601
 602                 fa = fa_to_delete;
 603                 rtmsg_fib(RTM_DELROUTE, key, fa, cfg->fc_dst_len,
 604                           tb->tb_id, &cfg->fc_nlinfo, 0);
 605
 606                 kill_fn = 0;
 607                 write_lock_bh(&fib_hash_lock);
 608                 list_del(&fa->fa_list);
 609                 if (list_empty(&f->fn_alias)) {
 610                         hlist_del(&f->fn_hash);
 611                         kill_fn = 1;
 612                 }
 613                 fib_hash_genid++;
 614                 write_unlock_bh(&fib_hash_lock);
 615
 616                 if (fa->fa_state & FA_S_ACCESSED)
 617                         rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
 618                 fn_free_alias(fa, f);
 619                 if (kill_fn) {
 620                         fn_free_node(f);
 621                         fz->fz_nent--;
 622                 }
 623
 624                 return 0;
 625         }
 626         return -ESRCH;
 627 }
 628
 629 static int fn_flush_list(struct fn_zone *fz, int idx)
 630 {
 631         struct hlist_head *head = &fz->fz_hash[idx];
 632         struct hlist_node *node, *n;
 633         struct fib_node *f;
 634         int found = 0;
 635
 636         hlist_for_each_entry_safe(f, node, n, head, fn_hash) {
 637                 struct fib_alias *fa, *fa_node;
 638                 int kill_f;
 639
 640                 kill_f = 0;
 641                 list_for_each_entry_safe(fa, fa_node, &f->fn_alias, fa_list) {
 642                         struct fib_info *fi = fa->fa_info;
 643
 644                         if (fi && (fi->fib_flags&RTNH_F_DEAD)) {
 645                                 write_lock_bh(&fib_hash_lock);
 646                                 list_del(&fa->fa_list);
 647                                 if (list_empty(&f->fn_alias)) {
 648                                         hlist_del(&f->fn_hash);
 649                                         kill_f = 1;
 650                                 }
 651                                 fib_hash_genid++;
 652                                 write_unlock_bh(&fib_hash_lock);
 653
 654                                 fn_free_alias(fa, f);
 655                                 found++;
 656                         }
 657                 }
 658                 if (kill_f) {
 659                         fn_free_node(f);
 660                         fz->fz_nent--;
 661                 }
 662         }
 663         return found;
 664 }
 665
 666 int fib_table_flush(struct fib_table *tb)
 667 {
 668         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 669         struct fn_zone *fz;
 670         int found = 0;
 671
 672         for (fz = table->fn_zone_list; fz; fz = fz->fz_next) {
 673                 int i;
 674
 675                 for (i = fz->fz_divisor - 1; i >= 0; i--)
 676                         found += fn_flush_list(fz, i);
 677         }
 678         return found;
 679 }
 680
 681
 682 static inline int
 683 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
 684                      struct fib_table *tb,
 685                      struct fn_zone *fz,
 686                      struct hlist_head *head)
 687 {
 688         struct hlist_node *node;
 689         struct fib_node *f;
 690         int i, s_i;
 691
 692         s_i = cb->args[4];
 693         i = 0;
 694         hlist_for_each_entry(f, node, head, fn_hash) {
 695                 struct fib_alias *fa;
 696
 697                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 698                         if (i < s_i)
 699                                 goto next;
 700
 701                         if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
 702                                           cb->nlh->nlmsg_seq,
 703                                           RTM_NEWROUTE,
 704                                           tb->tb_id,
 705                                           fa->fa_type,
 706                                           fa->fa_scope,
 707                                           f->fn_key,
 708                                           fz->fz_order,
 709                                           fa->fa_tos,
 710                                           fa->fa_info,
 711                                           NLM_F_MULTI) < 0) {
 712                                 cb->args[4] = i;
 713                                 return -1;
 714                         }
 715                 next:
 716                         i++;
 717                 }
 718         }
 719         cb->args[4] = i;
 720         return skb->len;
 721 }
 722
 723 static inline int
 724 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
 725                    struct fib_table *tb,
 726                    struct fn_zone *fz)
 727 {
 728         int h, s_h;
 729
 730         if (fz->fz_hash == NULL)
 731                 return skb->len;
 732         s_h = cb->args[3];
 733         for (h = s_h; h < fz->fz_divisor; h++) {
 734                 if (hlist_empty(&fz->fz_hash[h]))
 735                         continue;
 736                 if (fn_hash_dump_bucket(skb, cb, tb, fz, &fz->fz_hash[h]) < 0) {
 737                         cb->args[3] = h;
 738                         return -1;
 739                 }
 740                 memset(&cb->args[4], 0,
 741                        sizeof(cb->args) - 4*sizeof(cb->args[0]));
 742         }
 743         cb->args[3] = h;
 744         return skb->len;
 745 }
 746
 747 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
 748                    struct netlink_callback *cb)
 749 {
 750         int m, s_m;
 751         struct fn_zone *fz;
 752         struct fn_hash *table = (struct fn_hash *)tb->tb_data;
 753
 754         s_m = cb->args[2];
 755         read_lock(&fib_hash_lock);
 756         for (fz = table->fn_zone_list, m=0; fz; fz = fz->fz_next, m++) {
 757                 if (m < s_m) continue;
 758                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
 759                         cb->args[2] = m;
 760                         read_unlock(&fib_hash_lock);
 761                         return -1;
 762                 }
 763                 memset(&cb->args[3], 0,
 764                        sizeof(cb->args) - 3*sizeof(cb->args[0]));
 765         }
 766         read_unlock(&fib_hash_lock);
 767         cb->args[2] = m;
 768         return skb->len;
 769 }
 770
 771 void __init fib_hash_init(void)
 772 {
 773         fn_hash_kmem = kmem_cache_create("ip_fib_hash", sizeof(struct fib_node),
 774                                          0, SLAB_PANIC, NULL);
 775
 776         fn_alias_kmem = kmem_cache_create("ip_fib_alias", sizeof(struct fib_alias),
 777                                           0, SLAB_PANIC, NULL);
 778
 779 }
 780
 781 struct fib_table *fib_hash_table(u32 id)
 782 {
 783         struct fib_table *tb;
 784
 785         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash),
 786                      GFP_KERNEL);
 787         if (tb == NULL)
 788                 return NULL;
 789
 790         tb->tb_id = id;
 791         tb->tb_default = -1;
 792
 793         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 794         return tb;
 795 }
 796
 797 /* ------------------------------------------------------------------------ */
 798 #ifdef CONFIG_PROC_FS
 799
 800 struct fib_iter_state {
 801         struct seq_net_private p;
 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 fib_table *main_table;
 816         struct fn_hash *table;
 817
 818         main_table = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
 819         table = (struct fn_hash *)main_table->tb_data;
 820
 821         iter->bucket    = 0;
 822         iter->hash_head = NULL;
 823         iter->fn        = NULL;
 824         iter->fa        = NULL;
 825         iter->pos       = 0;
 826         iter->genid     = fib_hash_genid;
 827         iter->valid     = 1;
 828
 829         for (iter->zone = table->fn_zone_list; iter->zone;
 830              iter->zone = iter->zone->fz_next) {
 831                 int maxslot;
 832
 833                 if (!iter->zone->fz_nent)
 834                         continue;
 835
 836                 iter->hash_head = iter->zone->fz_hash;
 837                 maxslot = iter->zone->fz_divisor;
 838
 839                 for (iter->bucket = 0; iter->bucket < maxslot;
 840                      ++iter->bucket, ++iter->hash_head) {
 841                         struct hlist_node *node;
 842                         struct fib_node *fn;
 843
 844                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 845                                 struct fib_alias *fa;
 846
 847                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 848                                         iter->fn = fn;
 849                                         iter->fa = fa;
 850                                         goto out;
 851                                 }
 852                         }
 853                 }
 854         }
 855 out:
 856         return iter->fa;
 857 }
 858
 859 static struct fib_alias *fib_get_next(struct seq_file *seq)
 860 {
 861         struct fib_iter_state *iter = seq->private;
 862         struct fib_node *fn;
 863         struct fib_alias *fa;
 864
 865         /* Advance FA, if any. */
 866         fn = iter->fn;
 867         fa = iter->fa;
 868         if (fa) {
 869                 BUG_ON(!fn);
 870                 list_for_each_entry_continue(fa, &fn->fn_alias, fa_list) {
 871                         iter->fa = fa;
 872                         goto out;
 873                 }
 874         }
 875
 876         fa = iter->fa = NULL;
 877
 878         /* Advance FN. */
 879         if (fn) {
 880                 struct hlist_node *node = &fn->fn_hash;
 881                 hlist_for_each_entry_continue(fn, node, fn_hash) {
 882                         iter->fn = fn;
 883
 884                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 885                                 iter->fa = fa;
 886                                 goto out;
 887                         }
 888                 }
 889         }
 890
 891         fn = iter->fn = NULL;
 892
 893         /* Advance hash chain. */
 894         if (!iter->zone)
 895                 goto out;
 896
 897         for (;;) {
 898                 struct hlist_node *node;
 899                 int maxslot;
 900
 901                 maxslot = iter->zone->fz_divisor;
 902
 903                 while (++iter->bucket < maxslot) {
 904                         iter->hash_head++;
 905
 906                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 907                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 908                                         iter->fn = fn;
 909                                         iter->fa = fa;
 910                                         goto out;
 911                                 }
 912                         }
 913                 }
 914
 915                 iter->zone = iter->zone->fz_next;
 916
 917                 if (!iter->zone)
 918                         goto out;
 919
 920                 iter->bucket = 0;
 921                 iter->hash_head = iter->zone->fz_hash;
 922
 923                 hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 924                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 925                                 iter->fn = fn;
 926                                 iter->fa = fa;
 927                                 goto out;
 928                         }
 929                 }
 930         }
 931 out:
 932         iter->pos++;
 933         return fa;
 934 }
 935
 936 static struct fib_alias *fib_get_idx(struct seq_file *seq, loff_t pos)
 937 {
 938         struct fib_iter_state *iter = seq->private;
 939         struct fib_alias *fa;
 940
 941         if (iter->valid && pos >= iter->pos && iter->genid == fib_hash_genid) {
 942                 fa   = iter->fa;
 943                 pos -= iter->pos;
 944         } else
 945                 fa = fib_get_first(seq);
 946
 947         if (fa)
 948                 while (pos && (fa = fib_get_next(seq)))
 949                         --pos;
 950         return pos ? NULL : fa;
 951 }
 952
 953 static void *fib_seq_start(struct seq_file *seq, loff_t *pos)
 954         __acquires(fib_hash_lock)
 955 {
 956         void *v = NULL;
 957
 958         read_lock(&fib_hash_lock);
 959         if (fib_get_table(seq_file_net(seq), RT_TABLE_MAIN))
 960                 v = *pos ? fib_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 961         return v;
 962 }
 963
 964 static void *fib_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 965 {
 966         ++*pos;
 967         return v == SEQ_START_TOKEN ? fib_get_first(seq) : fib_get_next(seq);
 968 }
 969
 970 static void fib_seq_stop(struct seq_file *seq, void *v)
 971         __releases(fib_hash_lock)
 972 {
 973         read_unlock(&fib_hash_lock);
 974 }
 975
 976 static unsigned fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
 977 {
 978         static const unsigned type2flags[RTN_MAX + 1] = {
 979                 [7] = RTF_REJECT, [8] = RTF_REJECT,
 980         };
 981         unsigned flags = type2flags[type];
 982
 983         if (fi && fi->fib_nh->nh_gw)
 984                 flags |= RTF_GATEWAY;
 985         if (mask == htonl(0xFFFFFFFF))
 986                 flags |= RTF_HOST;
 987         flags |= RTF_UP;
 988         return flags;
 989 }
 990
 991 /*
 992  *      This outputs /proc/net/route.
 993  *
 994  *      It always works in backward compatibility mode.
 995  *      The format of the file is not supposed to be changed.
 996  */
 997 static int fib_seq_show(struct seq_file *seq, void *v)
 998 {
 999         struct fib_iter_state *iter;
1000         int len;
1001         __be32 prefix, mask;
1002         unsigned flags;
1003         struct fib_node *f;
1004         struct fib_alias *fa;
1005         struct fib_info *fi;
1006
1007         if (v == SEQ_START_TOKEN) {
1008                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
1009                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
1010                            "\tWindow\tIRTT");
1011                 goto out;
1012         }
1013
1014         iter    = seq->private;
1015         f       = iter->fn;
1016         fa      = iter->fa;
1017         fi      = fa->fa_info;
1018         prefix  = f->fn_key;
1019         mask    = FZ_MASK(iter->zone);
1020         flags   = fib_flag_trans(fa->fa_type, mask, fi);
1021         if (fi)
1022                 seq_printf(seq,
1023                          "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u%n",
1024                          fi->fib_dev ? fi->fib_dev->name : "*", prefix,
1025                          fi->fib_nh->nh_gw, flags, 0, 0, fi->fib_priority,
1026                          mask, (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
1027                          fi->fib_window,
1028                          fi->fib_rtt >> 3, &len);
1029         else
1030                 seq_printf(seq,
1031                          "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u%n",
1032                          prefix, 0, flags, 0, 0, 0, mask, 0, 0, 0, &len);
1033
1034         seq_printf(seq, "%*s\n", 127 - len, "");
1035 out:
1036         return 0;
1037 }
1038
1039 static const struct seq_operations fib_seq_ops = {
1040         .start  = fib_seq_start,
1041         .next   = fib_seq_next,
1042         .stop   = fib_seq_stop,
1043         .show   = fib_seq_show,
1044 };
1045
1046 static int fib_seq_open(struct inode *inode, struct file *file)
1047 {
1048         return seq_open_net(inode, file, &fib_seq_ops,
1049                             sizeof(struct fib_iter_state));
1050 }
1051
1052 static const struct file_operations fib_seq_fops = {
1053         .owner          = THIS_MODULE,
1054         .open           = fib_seq_open,
1055         .read           = seq_read,
1056         .llseek         = seq_lseek,
1057         .release        = seq_release_net,
1058 };
1059
1060 int __net_init fib_proc_init(struct net *net)
1061 {
1062         if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_seq_fops))
1063                 return -ENOMEM;
1064         return 0;
1065 }
1066
1067 void __net_exit fib_proc_exit(struct net *net)
1068 {
1069         proc_net_remove(net, "route");
1070 }
1071 #endif /* CONFIG_PROC_FS */