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 #define EMBEDDED_HASH_SIZE (L1_CACHE_BYTES / sizeof(struct hlist_head))
  58
  59 struct fn_zone {
  60         struct fn_zone __rcu    *fz_next;       /* Next not empty zone  */
  61         struct hlist_head       *fz_hash;       /* Hash table pointer   */
  62         u32                     fz_hashmask;    /* (fz_divisor - 1)     */
  63
  64         u8                      fz_order;       /* Zone order (0..32)   */
  65         u8                      fz_revorder;    /* 32 - fz_order        */
  66         __be32                  fz_mask;        /* inet_make_mask(order) */
  67 #define FZ_MASK(fz)             ((fz)->fz_mask)
  68
  69         struct hlist_head       fz_embedded_hash[EMBEDDED_HASH_SIZE];
  70
  71         int                     fz_nent;        /* Number of entries    */
  72         int                     fz_divisor;     /* Hash size (mask+1)   */
  73 };
  74
  75 struct fn_hash {
  76         struct fn_zone          *fn_zones[33];
  77         struct fn_zone __rcu    *fn_zone_list;
  78 };
  79
  80 static inline u32 fn_hash(__be32 key, struct fn_zone *fz)
  81 {
  82         u32 h = ntohl(key) >> fz->fz_revorder;
  83         h ^= (h>>20);
  84         h ^= (h>>10);
  85         h ^= (h>>5);
  86         h &= fz->fz_hashmask;
  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         new_divisor = old_divisor = fz->fz_divisor;
 151
 152         switch (old_divisor) {
 153         case EMBEDDED_HASH_SIZE:
 154                 new_divisor *= EMBEDDED_HASH_SIZE;
 155                 break;
 156         case EMBEDDED_HASH_SIZE*EMBEDDED_HASH_SIZE:
 157                 new_divisor *= (EMBEDDED_HASH_SIZE/2);
 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                 if (old_ht != fz->fz_embedded_hash)
 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         fz->fz_divisor = z ? EMBEDDED_HASH_SIZE : 1;
 215         fz->fz_hashmask = fz->fz_divisor - 1;
 216         fz->fz_hash = fz->fz_embedded_hash;
 217         fz->fz_order = z;
 218         fz->fz_revorder = 32 - z;
 219         fz->fz_mask = inet_make_mask(z);
 220
 221         /* Find the first not empty zone with more specific mask */
 222         for (i = z + 1; i <= 32; i++)
 223                 if (table->fn_zones[i])
 224                         break;
 225         if (i > 32) {
 226                 /* No more specific masks, we are the first. */
 227                 rcu_assign_pointer(fz->fz_next,
 228                                    rtnl_dereference(table->fn_zone_list));
 229                 rcu_assign_pointer(table->fn_zone_list, fz);
 230         } else {
 231                 rcu_assign_pointer(fz->fz_next,
 232                                    rtnl_dereference(table->fn_zones[i]->fz_next));
 233                 rcu_assign_pointer(table->fn_zones[i]->fz_next, fz);
 234         }
 235         table->fn_zones[z] = fz;
 236         fib_hash_genid++;
 237         return fz;
 238 }
 239
 240 int fib_table_lookup(struct fib_table *tb,
 241                      const struct flowi *flp, struct fib_result *res,
 242                      int fib_flags)
 243 {
 244         int err;
 245         struct fn_zone *fz;
 246         struct fn_hash *t = (struct fn_hash *)tb->tb_data;
 247
 248         rcu_read_lock();
 249         read_lock(&fib_hash_lock);
 250         for (fz = rcu_dereference(t->fn_zone_list);
 251              fz != NULL;
 252              fz = rcu_dereference(fz->fz_next)) {
 253                 struct hlist_head *head;
 254                 struct hlist_node *node;
 255                 struct fib_node *f;
 256                 __be32 k = fz_key(flp->fl4_dst, fz);
 257
 258                 head = &fz->fz_hash[fn_hash(k, fz)];
 259                 hlist_for_each_entry(f, node, head, fn_hash) {
 260                         if (f->fn_key != k)
 261                                 continue;
 262
 263                         err = fib_semantic_match(&f->fn_alias,
 264                                                  flp, res,
 265                                                  fz->fz_order, fib_flags);
 266                         if (err <= 0)
 267                                 goto out;
 268                 }
 269         }
 270         err = 1;
 271 out:
 272         read_unlock(&fib_hash_lock);
 273         rcu_read_unlock();
 274         return err;
 275 }
 276
 277 void fib_table_select_default(struct fib_table *tb,
 278                               const struct flowi *flp, struct fib_result *res)
 279 {
 280         int order, last_idx;
 281         struct hlist_node *node;
 282         struct fib_node *f;
 283         struct fib_info *fi = NULL;
 284         struct fib_info *last_resort;
 285         struct fn_hash *t = (struct fn_hash *)tb->tb_data;
 286         struct fn_zone *fz = t->fn_zones[0];
 287
 288         if (fz == NULL)
 289                 return;
 290
 291         last_idx = -1;
 292         last_resort = NULL;
 293         order = -1;
 294
 295         read_lock(&fib_hash_lock);
 296         hlist_for_each_entry(f, node, &fz->fz_hash[0], fn_hash) {
 297                 struct fib_alias *fa;
 298
 299                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 300                         struct fib_info *next_fi = fa->fa_info;
 301
 302                         if (fa->fa_scope != res->scope ||
 303                             fa->fa_type != RTN_UNICAST)
 304                                 continue;
 305
 306                         if (next_fi->fib_priority > res->fi->fib_priority)
 307                                 break;
 308                         if (!next_fi->fib_nh[0].nh_gw ||
 309                             next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
 310                                 continue;
 311                         fa->fa_state |= FA_S_ACCESSED;
 312
 313                         if (fi == NULL) {
 314                                 if (next_fi != res->fi)
 315                                         break;
 316                         } else if (!fib_detect_death(fi, order, &last_resort,
 317                                                 &last_idx, tb->tb_default)) {
 318                                 fib_result_assign(res, fi);
 319                                 tb->tb_default = order;
 320                                 goto out;
 321                         }
 322                         fi = next_fi;
 323                         order++;
 324                 }
 325         }
 326
 327         if (order <= 0 || fi == NULL) {
 328                 tb->tb_default = -1;
 329                 goto out;
 330         }
 331
 332         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
 333                                 tb->tb_default)) {
 334                 fib_result_assign(res, fi);
 335                 tb->tb_default = order;
 336                 goto out;
 337         }
 338
 339         if (last_idx >= 0)
 340                 fib_result_assign(res, last_resort);
 341         tb->tb_default = last_idx;
 342 out:
 343         read_unlock(&fib_hash_lock);
 344 }
 345
 346 /* Insert node F to FZ. */
 347 static inline void fib_insert_node(struct fn_zone *fz, struct fib_node *f)
 348 {
 349         struct hlist_head *head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 350
 351         hlist_add_head(&f->fn_hash, head);
 352 }
 353
 354 /* Return the node in FZ matching KEY. */
 355 static struct fib_node *fib_find_node(struct fn_zone *fz, __be32 key)
 356 {
 357         struct hlist_head *head = &fz->fz_hash[fn_hash(key, fz)];
 358         struct hlist_node *node;
 359         struct fib_node *f;
 360
 361         hlist_for_each_entry(f, node, head, fn_hash) {
 362                 if (f->fn_key == key)
 363                         return f;
 364         }
 365
 366         return NULL;
 367 }
 368
 369 /* Caller must hold RTNL. */
 370 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
 371 {
 372         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 373         struct fib_node *new_f = NULL;
 374         struct fib_node *f;
 375         struct fib_alias *fa, *new_fa;
 376         struct fn_zone *fz;
 377         struct fib_info *fi;
 378         u8 tos = cfg->fc_tos;
 379         __be32 key;
 380         int err;
 381
 382         if (cfg->fc_dst_len > 32)
 383                 return -EINVAL;
 384
 385         fz = table->fn_zones[cfg->fc_dst_len];
 386         if (!fz && !(fz = fn_new_zone(table, cfg->fc_dst_len)))
 387                 return -ENOBUFS;
 388
 389         key = 0;
 390         if (cfg->fc_dst) {
 391                 if (cfg->fc_dst & ~FZ_MASK(fz))
 392                         return -EINVAL;
 393                 key = fz_key(cfg->fc_dst, fz);
 394         }
 395
 396         fi = fib_create_info(cfg);
 397         if (IS_ERR(fi))
 398                 return PTR_ERR(fi);
 399
 400         if (fz->fz_nent > (fz->fz_divisor<<1) &&
 401             fz->fz_divisor < FZ_MAX_DIVISOR &&
 402             (cfg->fc_dst_len == 32 ||
 403              (1 << cfg->fc_dst_len) > fz->fz_divisor))
 404                 fn_rehash_zone(fz);
 405
 406         f = fib_find_node(fz, key);
 407
 408         if (!f)
 409                 fa = NULL;
 410         else
 411                 fa = fib_find_alias(&f->fn_alias, tos, fi->fib_priority);
 412
 413         /* Now fa, if non-NULL, points to the first fib alias
 414          * with the same keys [prefix,tos,priority], if such key already
 415          * exists or to the node before which we will insert new one.
 416          *
 417          * If fa is NULL, we will need to allocate a new one and
 418          * insert to the head of f.
 419          *
 420          * If f is NULL, no fib node matched the destination key
 421          * and we need to allocate a new one of those as well.
 422          */
 423
 424         if (fa && fa->fa_tos == tos &&
 425             fa->fa_info->fib_priority == fi->fib_priority) {
 426                 struct fib_alias *fa_first, *fa_match;
 427
 428                 err = -EEXIST;
 429                 if (cfg->fc_nlflags & NLM_F_EXCL)
 430                         goto out;
 431
 432                 /* We have 2 goals:
 433                  * 1. Find exact match for type, scope, fib_info to avoid
 434                  * duplicate routes
 435                  * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
 436                  */
 437                 fa_match = NULL;
 438                 fa_first = fa;
 439                 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 440                 list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 441                         if (fa->fa_tos != tos)
 442                                 break;
 443                         if (fa->fa_info->fib_priority != fi->fib_priority)
 444                                 break;
 445                         if (fa->fa_type == cfg->fc_type &&
 446                             fa->fa_scope == cfg->fc_scope &&
 447                             fa->fa_info == fi) {
 448                                 fa_match = fa;
 449                                 break;
 450                         }
 451                 }
 452
 453                 if (cfg->fc_nlflags & NLM_F_REPLACE) {
 454                         struct fib_info *fi_drop;
 455                         u8 state;
 456
 457                         fa = fa_first;
 458                         if (fa_match) {
 459                                 if (fa == fa_match)
 460                                         err = 0;
 461                                 goto out;
 462                         }
 463                         write_lock_bh(&fib_hash_lock);
 464                         fi_drop = fa->fa_info;
 465                         fa->fa_info = fi;
 466                         fa->fa_type = cfg->fc_type;
 467                         fa->fa_scope = cfg->fc_scope;
 468                         state = fa->fa_state;
 469                         fa->fa_state &= ~FA_S_ACCESSED;
 470                         fib_hash_genid++;
 471                         write_unlock_bh(&fib_hash_lock);
 472
 473                         fib_release_info(fi_drop);
 474                         if (state & FA_S_ACCESSED)
 475                                 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
 476                         rtmsg_fib(RTM_NEWROUTE, key, fa, cfg->fc_dst_len, tb->tb_id,
 477                                   &cfg->fc_nlinfo, NLM_F_REPLACE);
 478                         return 0;
 479                 }
 480
 481                 /* Error if we find a perfect match which
 482                  * uses the same scope, type, and nexthop
 483                  * information.
 484                  */
 485                 if (fa_match)
 486                         goto out;
 487
 488                 if (!(cfg->fc_nlflags & NLM_F_APPEND))
 489                         fa = fa_first;
 490         }
 491
 492         err = -ENOENT;
 493         if (!(cfg->fc_nlflags & NLM_F_CREATE))
 494                 goto out;
 495
 496         err = -ENOBUFS;
 497
 498         if (!f) {
 499                 new_f = kmem_cache_zalloc(fn_hash_kmem, GFP_KERNEL);
 500                 if (new_f == NULL)
 501                         goto out;
 502
 503                 INIT_HLIST_NODE(&new_f->fn_hash);
 504                 INIT_LIST_HEAD(&new_f->fn_alias);
 505                 new_f->fn_key = key;
 506                 f = new_f;
 507         }
 508
 509         new_fa = &f->fn_embedded_alias;
 510         if (new_fa->fa_info != NULL) {
 511                 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
 512                 if (new_fa == NULL)
 513                         goto out;
 514         }
 515         new_fa->fa_info = fi;
 516         new_fa->fa_tos = tos;
 517         new_fa->fa_type = cfg->fc_type;
 518         new_fa->fa_scope = cfg->fc_scope;
 519         new_fa->fa_state = 0;
 520
 521         /*
 522          * Insert new entry to the list.
 523          */
 524
 525         write_lock_bh(&fib_hash_lock);
 526         if (new_f)
 527                 fib_insert_node(fz, new_f);
 528         list_add_tail(&new_fa->fa_list,
 529                  (fa ? &fa->fa_list : &f->fn_alias));
 530         fib_hash_genid++;
 531         write_unlock_bh(&fib_hash_lock);
 532
 533         if (new_f)
 534                 fz->fz_nent++;
 535         rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
 536
 537         rtmsg_fib(RTM_NEWROUTE, key, new_fa, cfg->fc_dst_len, tb->tb_id,
 538                   &cfg->fc_nlinfo, 0);
 539         return 0;
 540
 541 out:
 542         if (new_f)
 543                 kmem_cache_free(fn_hash_kmem, new_f);
 544         fib_release_info(fi);
 545         return err;
 546 }
 547
 548 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
 549 {
 550         struct fn_hash *table = (struct fn_hash *)tb->tb_data;
 551         struct fib_node *f;
 552         struct fib_alias *fa, *fa_to_delete;
 553         struct fn_zone *fz;
 554         __be32 key;
 555
 556         if (cfg->fc_dst_len > 32)
 557                 return -EINVAL;
 558
 559         if ((fz  = table->fn_zones[cfg->fc_dst_len]) == NULL)
 560                 return -ESRCH;
 561
 562         key = 0;
 563         if (cfg->fc_dst) {
 564                 if (cfg->fc_dst & ~FZ_MASK(fz))
 565                         return -EINVAL;
 566                 key = fz_key(cfg->fc_dst, fz);
 567         }
 568
 569         f = fib_find_node(fz, key);
 570
 571         if (!f)
 572                 fa = NULL;
 573         else
 574                 fa = fib_find_alias(&f->fn_alias, cfg->fc_tos, 0);
 575         if (!fa)
 576                 return -ESRCH;
 577
 578         fa_to_delete = NULL;
 579         fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 580         list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 581                 struct fib_info *fi = fa->fa_info;
 582
 583                 if (fa->fa_tos != cfg->fc_tos)
 584                         break;
 585
 586                 if ((!cfg->fc_type ||
 587                      fa->fa_type == cfg->fc_type) &&
 588                     (cfg->fc_scope == RT_SCOPE_NOWHERE ||
 589                      fa->fa_scope == cfg->fc_scope) &&
 590                     (!cfg->fc_protocol ||
 591                      fi->fib_protocol == cfg->fc_protocol) &&
 592                     fib_nh_match(cfg, fi) == 0) {
 593                         fa_to_delete = fa;
 594                         break;
 595                 }
 596         }
 597
 598         if (fa_to_delete) {
 599                 int kill_fn;
 600
 601                 fa = fa_to_delete;
 602                 rtmsg_fib(RTM_DELROUTE, key, fa, cfg->fc_dst_len,
 603                           tb->tb_id, &cfg->fc_nlinfo, 0);
 604
 605                 kill_fn = 0;
 606                 write_lock_bh(&fib_hash_lock);
 607                 list_del(&fa->fa_list);
 608                 if (list_empty(&f->fn_alias)) {
 609                         hlist_del(&f->fn_hash);
 610                         kill_fn = 1;
 611                 }
 612                 fib_hash_genid++;
 613                 write_unlock_bh(&fib_hash_lock);
 614
 615                 if (fa->fa_state & FA_S_ACCESSED)
 616                         rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
 617                 fn_free_alias(fa, f);
 618                 if (kill_fn) {
 619                         fn_free_node(f);
 620                         fz->fz_nent--;
 621                 }
 622
 623                 return 0;
 624         }
 625         return -ESRCH;
 626 }
 627
 628 static int fn_flush_list(struct fn_zone *fz, int idx)
 629 {
 630         struct hlist_head *head = &fz->fz_hash[idx];
 631         struct hlist_node *node, *n;
 632         struct fib_node *f;
 633         int found = 0;
 634
 635         hlist_for_each_entry_safe(f, node, n, head, fn_hash) {
 636                 struct fib_alias *fa, *fa_node;
 637                 int kill_f;
 638
 639                 kill_f = 0;
 640                 list_for_each_entry_safe(fa, fa_node, &f->fn_alias, fa_list) {
 641                         struct fib_info *fi = fa->fa_info;
 642
 643                         if (fi && (fi->fib_flags&RTNH_F_DEAD)) {
 644                                 write_lock_bh(&fib_hash_lock);
 645                                 list_del(&fa->fa_list);
 646                                 if (list_empty(&f->fn_alias)) {
 647                                         hlist_del(&f->fn_hash);
 648                                         kill_f = 1;
 649                                 }
 650                                 fib_hash_genid++;
 651                                 write_unlock_bh(&fib_hash_lock);
 652
 653                                 fn_free_alias(fa, f);
 654                                 found++;
 655                         }
 656                 }
 657                 if (kill_f) {
 658                         fn_free_node(f);
 659                         fz->fz_nent--;
 660                 }
 661         }
 662         return found;
 663 }
 664
 665 /* caller must hold RTNL. */
 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 = rtnl_dereference(table->fn_zone_list);
 673              fz != NULL;
 674              fz = rtnl_dereference(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 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
 750                    struct netlink_callback *cb)
 751 {
 752         int m = 0, s_m;
 753         struct fn_zone *fz;
 754         struct fn_hash *table = (struct fn_hash *)tb->tb_data;
 755
 756         s_m = cb->args[2];
 757         rcu_read_lock();
 758         read_lock(&fib_hash_lock);
 759         for (fz = rcu_dereference(table->fn_zone_list);
 760              fz != NULL;
 761              fz = rcu_dereference(fz->fz_next), m++) {
 762                 if (m < s_m)
 763                         continue;
 764                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
 765                         cb->args[2] = m;
 766                         read_unlock(&fib_hash_lock);
 767                         rcu_read_unlock();
 768                         return -1;
 769                 }
 770                 memset(&cb->args[3], 0,
 771                        sizeof(cb->args) - 3*sizeof(cb->args[0]));
 772         }
 773         read_unlock(&fib_hash_lock);
 774         rcu_read_unlock();
 775         cb->args[2] = m;
 776         return skb->len;
 777 }
 778
 779 void __init fib_hash_init(void)
 780 {
 781         fn_hash_kmem = kmem_cache_create("ip_fib_hash", sizeof(struct fib_node),
 782                                          0, SLAB_PANIC, NULL);
 783
 784         fn_alias_kmem = kmem_cache_create("ip_fib_alias", sizeof(struct fib_alias),
 785                                           0, SLAB_PANIC, NULL);
 786
 787 }
 788
 789 struct fib_table *fib_hash_table(u32 id)
 790 {
 791         struct fib_table *tb;
 792
 793         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash),
 794                      GFP_KERNEL);
 795         if (tb == NULL)
 796                 return NULL;
 797
 798         tb->tb_id = id;
 799         tb->tb_default = -1;
 800
 801         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 802         return tb;
 803 }
 804
 805 /* ------------------------------------------------------------------------ */
 806 #ifdef CONFIG_PROC_FS
 807
 808 struct fib_iter_state {
 809         struct seq_net_private p;
 810         struct fn_zone  *zone;
 811         int             bucket;
 812         struct hlist_head *hash_head;
 813         struct fib_node *fn;
 814         struct fib_alias *fa;
 815         loff_t pos;
 816         unsigned int genid;
 817         int valid;
 818 };
 819
 820 static struct fib_alias *fib_get_first(struct seq_file *seq)
 821 {
 822         struct fib_iter_state *iter = seq->private;
 823         struct fib_table *main_table;
 824         struct fn_hash *table;
 825
 826         main_table = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
 827         table = (struct fn_hash *)main_table->tb_data;
 828
 829         iter->bucket    = 0;
 830         iter->hash_head = NULL;
 831         iter->fn        = NULL;
 832         iter->fa        = NULL;
 833         iter->pos       = 0;
 834         iter->genid     = fib_hash_genid;
 835         iter->valid     = 1;
 836
 837         for (iter->zone = rcu_dereference(table->fn_zone_list);
 838              iter->zone != NULL;
 839              iter->zone = rcu_dereference(iter->zone->fz_next)) {
 840                 int maxslot;
 841
 842                 if (!iter->zone->fz_nent)
 843                         continue;
 844
 845                 iter->hash_head = iter->zone->fz_hash;
 846                 maxslot = iter->zone->fz_divisor;
 847
 848                 for (iter->bucket = 0; iter->bucket < maxslot;
 849                      ++iter->bucket, ++iter->hash_head) {
 850                         struct hlist_node *node;
 851                         struct fib_node *fn;
 852
 853                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 854                                 struct fib_alias *fa;
 855
 856                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 857                                         iter->fn = fn;
 858                                         iter->fa = fa;
 859                                         goto out;
 860                                 }
 861                         }
 862                 }
 863         }
 864 out:
 865         return iter->fa;
 866 }
 867
 868 static struct fib_alias *fib_get_next(struct seq_file *seq)
 869 {
 870         struct fib_iter_state *iter = seq->private;
 871         struct fib_node *fn;
 872         struct fib_alias *fa;
 873
 874         /* Advance FA, if any. */
 875         fn = iter->fn;
 876         fa = iter->fa;
 877         if (fa) {
 878                 BUG_ON(!fn);
 879                 list_for_each_entry_continue(fa, &fn->fn_alias, fa_list) {
 880                         iter->fa = fa;
 881                         goto out;
 882                 }
 883         }
 884
 885         fa = iter->fa = NULL;
 886
 887         /* Advance FN. */
 888         if (fn) {
 889                 struct hlist_node *node = &fn->fn_hash;
 890                 hlist_for_each_entry_continue(fn, node, fn_hash) {
 891                         iter->fn = fn;
 892
 893                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 894                                 iter->fa = fa;
 895                                 goto out;
 896                         }
 897                 }
 898         }
 899
 900         fn = iter->fn = NULL;
 901
 902         /* Advance hash chain. */
 903         if (!iter->zone)
 904                 goto out;
 905
 906         for (;;) {
 907                 struct hlist_node *node;
 908                 int maxslot;
 909
 910                 maxslot = iter->zone->fz_divisor;
 911
 912                 while (++iter->bucket < maxslot) {
 913                         iter->hash_head++;
 914
 915                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 916                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 917                                         iter->fn = fn;
 918                                         iter->fa = fa;
 919                                         goto out;
 920                                 }
 921                         }
 922                 }
 923
 924                 iter->zone = rcu_dereference(iter->zone->fz_next);
 925
 926                 if (!iter->zone)
 927                         goto out;
 928
 929                 iter->bucket = 0;
 930                 iter->hash_head = iter->zone->fz_hash;
 931
 932                 hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 933                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 934                                 iter->fn = fn;
 935                                 iter->fa = fa;
 936                                 goto out;
 937                         }
 938                 }
 939         }
 940 out:
 941         iter->pos++;
 942         return fa;
 943 }
 944
 945 static struct fib_alias *fib_get_idx(struct seq_file *seq, loff_t pos)
 946 {
 947         struct fib_iter_state *iter = seq->private;
 948         struct fib_alias *fa;
 949
 950         if (iter->valid && pos >= iter->pos && iter->genid == fib_hash_genid) {
 951                 fa   = iter->fa;
 952                 pos -= iter->pos;
 953         } else
 954                 fa = fib_get_first(seq);
 955
 956         if (fa)
 957                 while (pos && (fa = fib_get_next(seq)))
 958                         --pos;
 959         return pos ? NULL : fa;
 960 }
 961
 962 static void *fib_seq_start(struct seq_file *seq, loff_t *pos)
 963         __acquires(fib_hash_lock)
 964         __acquires(RCU)
 965 {
 966         void *v = NULL;
 967
 968         rcu_read_lock();
 969         read_lock(&fib_hash_lock);
 970         if (fib_get_table(seq_file_net(seq), RT_TABLE_MAIN))
 971                 v = *pos ? fib_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 972         return v;
 973 }
 974
 975 static void *fib_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 976 {
 977         ++*pos;
 978         return v == SEQ_START_TOKEN ? fib_get_first(seq) : fib_get_next(seq);
 979 }
 980
 981 static void fib_seq_stop(struct seq_file *seq, void *v)
 982         __releases(fib_hash_lock)
 983         __releases(RCU)
 984 {
 985         read_unlock(&fib_hash_lock);
 986         rcu_read_unlock();
 987 }
 988
 989 static unsigned fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
 990 {
 991         static const unsigned type2flags[RTN_MAX + 1] = {
 992                 [7] = RTF_REJECT, [8] = RTF_REJECT,
 993         };
 994         unsigned flags = type2flags[type];
 995
 996         if (fi && fi->fib_nh->nh_gw)
 997                 flags |= RTF_GATEWAY;
 998         if (mask == htonl(0xFFFFFFFF))
 999                 flags |= RTF_HOST;
1000         flags |= RTF_UP;
1001         return flags;
1002 }
1003
1004 /*
1005  *      This outputs /proc/net/route.
1006  *
1007  *      It always works in backward compatibility mode.
1008  *      The format of the file is not supposed to be changed.
1009  */
1010 static int fib_seq_show(struct seq_file *seq, void *v)
1011 {
1012         struct fib_iter_state *iter;
1013         int len;
1014         __be32 prefix, mask;
1015         unsigned flags;
1016         struct fib_node *f;
1017         struct fib_alias *fa;
1018         struct fib_info *fi;
1019
1020         if (v == SEQ_START_TOKEN) {
1021                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
1022                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
1023                            "\tWindow\tIRTT");
1024                 goto out;
1025         }
1026
1027         iter    = seq->private;
1028         f       = iter->fn;
1029         fa      = iter->fa;
1030         fi      = fa->fa_info;
1031         prefix  = f->fn_key;
1032         mask    = FZ_MASK(iter->zone);
1033         flags   = fib_flag_trans(fa->fa_type, mask, fi);
1034         if (fi)
1035                 seq_printf(seq,
1036                          "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u%n",
1037                          fi->fib_dev ? fi->fib_dev->name : "*", prefix,
1038                          fi->fib_nh->nh_gw, flags, 0, 0, fi->fib_priority,
1039                          mask, (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
1040                          fi->fib_window,
1041                          fi->fib_rtt >> 3, &len);
1042         else
1043                 seq_printf(seq,
1044                          "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u%n",
1045                          prefix, 0, flags, 0, 0, 0, mask, 0, 0, 0, &len);
1046
1047         seq_printf(seq, "%*s\n", 127 - len, "");
1048 out:
1049         return 0;
1050 }
1051
1052 static const struct seq_operations fib_seq_ops = {
1053         .start  = fib_seq_start,
1054         .next   = fib_seq_next,
1055         .stop   = fib_seq_stop,
1056         .show   = fib_seq_show,
1057 };
1058
1059 static int fib_seq_open(struct inode *inode, struct file *file)
1060 {
1061         return seq_open_net(inode, file, &fib_seq_ops,
1062                             sizeof(struct fib_iter_state));
1063 }
1064
1065 static const struct file_operations fib_seq_fops = {
1066         .owner          = THIS_MODULE,
1067         .open           = fib_seq_open,
1068         .read           = seq_read,
1069         .llseek         = seq_lseek,
1070         .release        = seq_release_net,
1071 };
1072
1073 int __net_init fib_proc_init(struct net *net)
1074 {
1075         if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_seq_fops))
1076                 return -ENOMEM;
1077         return 0;
1078 }
1079
1080 void __net_exit fib_proc_exit(struct net *net)
1081 {
1082         proc_net_remove(net, "route");
1083 }
1084 #endif /* CONFIG_PROC_FS */