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 __rcu *fz_hash;       /* Hash table pointer   */
  62         seqlock_t               fz_lock;
  63         u32                     fz_hashmask;    /* (fz_divisor - 1)     */
  64
  65         u8                      fz_order;       /* Zone order (0..32)   */
  66         u8                      fz_revorder;    /* 32 - fz_order        */
  67         __be32                  fz_mask;        /* inet_make_mask(order) */
  68 #define FZ_MASK(fz)             ((fz)->fz_mask)
  69
  70         struct hlist_head       fz_embedded_hash[EMBEDDED_HASH_SIZE];
  71
  72         int                     fz_nent;        /* Number of entries    */
  73         int                     fz_divisor;     /* Hash size (mask+1)   */
  74 };
  75
  76 struct fn_hash {
  77         struct fn_zone          *fn_zones[33];
  78         struct fn_zone __rcu    *fn_zone_list;
  79 };
  80
  81 static inline u32 fn_hash(__be32 key, struct fn_zone *fz)
  82 {
  83         u32 h = ntohl(key) >> fz->fz_revorder;
  84         h ^= (h>>20);
  85         h ^= (h>>10);
  86         h ^= (h>>5);
  87         h &= fz->fz_hashmask;
  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 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
 107         return (struct hlist_head *)
 108                 __get_free_pages(GFP_KERNEL | __GFP_ZERO, get_order(size));
 109 }
 110
 111 /* The fib hash lock must be held when this is called. */
 112 static inline void fn_rebuild_zone(struct fn_zone *fz,
 113                                    struct hlist_head *old_ht,
 114                                    int old_divisor)
 115 {
 116         int i;
 117
 118         for (i = 0; i < old_divisor; i++) {
 119                 struct hlist_node *node, *n;
 120                 struct fib_node *f;
 121
 122                 hlist_for_each_entry_safe(f, node, n, &old_ht[i], fn_hash) {
 123                         struct hlist_head __rcu *new_head;
 124
 125                         hlist_del_rcu(&f->fn_hash);
 126
 127                         new_head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 128                         hlist_add_head_rcu(&f->fn_hash, new_head);
 129                 }
 130         }
 131 }
 132
 133 static void fz_hash_free(struct hlist_head *hash, int divisor)
 134 {
 135         unsigned long size = divisor * sizeof(struct hlist_head);
 136
 137         if (size <= PAGE_SIZE)
 138                 kfree(hash);
 139         else
 140                 free_pages((unsigned long)hash, get_order(size));
 141 }
 142
 143 static void fn_rehash_zone(struct fn_zone *fz)
 144 {
 145         struct hlist_head *ht, *old_ht;
 146         int old_divisor, new_divisor;
 147         u32 new_hashmask;
 148
 149         new_divisor = old_divisor = fz->fz_divisor;
 150
 151         switch (old_divisor) {
 152         case EMBEDDED_HASH_SIZE:
 153                 new_divisor *= EMBEDDED_HASH_SIZE;
 154                 break;
 155         case EMBEDDED_HASH_SIZE*EMBEDDED_HASH_SIZE:
 156                 new_divisor *= (EMBEDDED_HASH_SIZE/2);
 157                 break;
 158         default:
 159                 if ((old_divisor << 1) > FZ_MAX_DIVISOR) {
 160                         printk(KERN_CRIT "route.c: bad divisor %d!\n", old_divisor);
 161                         return;
 162                 }
 163                 new_divisor = (old_divisor << 1);
 164                 break;
 165         }
 166
 167         new_hashmask = (new_divisor - 1);
 168
 169 #if RT_CACHE_DEBUG >= 2
 170         printk(KERN_DEBUG "fn_rehash_zone: hash for zone %d grows from %d\n",
 171                fz->fz_order, old_divisor);
 172 #endif
 173
 174         ht = fz_hash_alloc(new_divisor);
 175
 176         if (ht) {
 177                 struct fn_zone nfz;
 178
 179                 memcpy(&nfz, fz, sizeof(nfz));
 180
 181                 write_seqlock_bh(&fz->fz_lock);
 182                 old_ht = fz->fz_hash;
 183                 nfz.fz_hash = ht;
 184                 nfz.fz_hashmask = new_hashmask;
 185                 nfz.fz_divisor = new_divisor;
 186                 fn_rebuild_zone(&nfz, old_ht, old_divisor);
 187                 fib_hash_genid++;
 188                 rcu_assign_pointer(fz->fz_hash, ht);
 189                 fz->fz_hashmask = new_hashmask;
 190                 fz->fz_divisor = new_divisor;
 191                 write_sequnlock_bh(&fz->fz_lock);
 192
 193                 if (old_ht != fz->fz_embedded_hash) {
 194                         synchronize_rcu();
 195                         fz_hash_free(old_ht, old_divisor);
 196                 }
 197         }
 198 }
 199
 200 static void fn_free_node_rcu(struct rcu_head *head)
 201 {
 202         struct fib_node *f = container_of(head, struct fib_node, fn_embedded_alias.rcu);
 203
 204         kmem_cache_free(fn_hash_kmem, f);
 205 }
 206
 207 static inline void fn_free_node(struct fib_node *f)
 208 {
 209         call_rcu(&f->fn_embedded_alias.rcu, fn_free_node_rcu);
 210 }
 211
 212 static void fn_free_alias_rcu(struct rcu_head *head)
 213 {
 214         struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
 215
 216         kmem_cache_free(fn_alias_kmem, fa);
 217 }
 218
 219 static inline void fn_free_alias(struct fib_alias *fa, struct fib_node *f)
 220 {
 221         fib_release_info(fa->fa_info);
 222         if (fa == &f->fn_embedded_alias)
 223                 fa->fa_info = NULL;
 224         else
 225                 call_rcu(&fa->rcu, fn_free_alias_rcu);
 226 }
 227
 228 static struct fn_zone *
 229 fn_new_zone(struct fn_hash *table, int z)
 230 {
 231         int i;
 232         struct fn_zone *fz = kzalloc(sizeof(struct fn_zone), GFP_KERNEL);
 233         if (!fz)
 234                 return NULL;
 235
 236         seqlock_init(&fz->fz_lock);
 237         fz->fz_divisor = z ? EMBEDDED_HASH_SIZE : 1;
 238         fz->fz_hashmask = fz->fz_divisor - 1;
 239         fz->fz_hash = fz->fz_embedded_hash;
 240         fz->fz_order = z;
 241         fz->fz_revorder = 32 - z;
 242         fz->fz_mask = inet_make_mask(z);
 243
 244         /* Find the first not empty zone with more specific mask */
 245         for (i = z + 1; i <= 32; i++)
 246                 if (table->fn_zones[i])
 247                         break;
 248         if (i > 32) {
 249                 /* No more specific masks, we are the first. */
 250                 rcu_assign_pointer(fz->fz_next,
 251                                    rtnl_dereference(table->fn_zone_list));
 252                 rcu_assign_pointer(table->fn_zone_list, fz);
 253         } else {
 254                 rcu_assign_pointer(fz->fz_next,
 255                                    rtnl_dereference(table->fn_zones[i]->fz_next));
 256                 rcu_assign_pointer(table->fn_zones[i]->fz_next, fz);
 257         }
 258         table->fn_zones[z] = fz;
 259         fib_hash_genid++;
 260         return fz;
 261 }
 262
 263 int fib_table_lookup(struct fib_table *tb,
 264                      const struct flowi *flp, struct fib_result *res,
 265                      int fib_flags)
 266 {
 267         int err;
 268         struct fn_zone *fz;
 269         struct fn_hash *t = (struct fn_hash *)tb->tb_data;
 270
 271         rcu_read_lock();
 272         for (fz = rcu_dereference(t->fn_zone_list);
 273              fz != NULL;
 274              fz = rcu_dereference(fz->fz_next)) {
 275                 struct hlist_head __rcu *head;
 276                 struct hlist_node *node;
 277                 struct fib_node *f;
 278                 __be32 k;
 279                 unsigned int seq;
 280
 281                 do {
 282                         seq = read_seqbegin(&fz->fz_lock);
 283                         k = fz_key(flp->fl4_dst, fz);
 284
 285                         head = &fz->fz_hash[fn_hash(k, fz)];
 286                         hlist_for_each_entry_rcu(f, node, head, fn_hash) {
 287                                 if (f->fn_key != k)
 288                                         continue;
 289
 290                                 err = fib_semantic_match(&f->fn_alias,
 291                                                  flp, res,
 292                                                  fz->fz_order, fib_flags);
 293                                 if (err <= 0)
 294                                         goto out;
 295                         }
 296                 } while (read_seqretry(&fz->fz_lock, seq));
 297         }
 298         err = 1;
 299 out:
 300         rcu_read_unlock();
 301         return err;
 302 }
 303
 304 void fib_table_select_default(struct fib_table *tb,
 305                               const struct flowi *flp, struct fib_result *res)
 306 {
 307         int order, last_idx;
 308         struct hlist_node *node;
 309         struct fib_node *f;
 310         struct fib_info *fi = NULL;
 311         struct fib_info *last_resort;
 312         struct fn_hash *t = (struct fn_hash *)tb->tb_data;
 313         struct fn_zone *fz = t->fn_zones[0];
 314
 315         if (fz == NULL)
 316                 return;
 317
 318         last_idx = -1;
 319         last_resort = NULL;
 320         order = -1;
 321
 322         rcu_read_lock();
 323         hlist_for_each_entry_rcu(f, node, &fz->fz_hash[0], fn_hash) {
 324                 struct fib_alias *fa;
 325
 326                 list_for_each_entry_rcu(fa, &f->fn_alias, fa_list) {
 327                         struct fib_info *next_fi = fa->fa_info;
 328
 329                         if (fa->fa_scope != res->scope ||
 330                             fa->fa_type != RTN_UNICAST)
 331                                 continue;
 332
 333                         if (next_fi->fib_priority > res->fi->fib_priority)
 334                                 break;
 335                         if (!next_fi->fib_nh[0].nh_gw ||
 336                             next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
 337                                 continue;
 338
 339                         fib_alias_accessed(fa);
 340
 341                         if (fi == NULL) {
 342                                 if (next_fi != res->fi)
 343                                         break;
 344                         } else if (!fib_detect_death(fi, order, &last_resort,
 345                                                 &last_idx, tb->tb_default)) {
 346                                 fib_result_assign(res, fi);
 347                                 tb->tb_default = order;
 348                                 goto out;
 349                         }
 350                         fi = next_fi;
 351                         order++;
 352                 }
 353         }
 354
 355         if (order <= 0 || fi == NULL) {
 356                 tb->tb_default = -1;
 357                 goto out;
 358         }
 359
 360         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
 361                                 tb->tb_default)) {
 362                 fib_result_assign(res, fi);
 363                 tb->tb_default = order;
 364                 goto out;
 365         }
 366
 367         if (last_idx >= 0)
 368                 fib_result_assign(res, last_resort);
 369         tb->tb_default = last_idx;
 370 out:
 371         rcu_read_unlock();
 372 }
 373
 374 /* Insert node F to FZ. */
 375 static inline void fib_insert_node(struct fn_zone *fz, struct fib_node *f)
 376 {
 377         struct hlist_head *head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 378
 379         hlist_add_head_rcu(&f->fn_hash, head);
 380 }
 381
 382 /* Return the node in FZ matching KEY. */
 383 static struct fib_node *fib_find_node(struct fn_zone *fz, __be32 key)
 384 {
 385         struct hlist_head *head = &fz->fz_hash[fn_hash(key, fz)];
 386         struct hlist_node *node;
 387         struct fib_node *f;
 388
 389         hlist_for_each_entry_rcu(f, node, head, fn_hash) {
 390                 if (f->fn_key == key)
 391                         return f;
 392         }
 393
 394         return NULL;
 395 }
 396
 397
 398 static struct fib_alias *fib_fast_alloc(struct fib_node *f)
 399 {
 400         struct fib_alias *fa = &f->fn_embedded_alias;
 401
 402         if (fa->fa_info != NULL)
 403                 fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
 404         return fa;
 405 }
 406
 407 /* Caller must hold RTNL. */
 408 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
 409 {
 410         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 411         struct fib_node *new_f = NULL;
 412         struct fib_node *f;
 413         struct fib_alias *fa, *new_fa;
 414         struct fn_zone *fz;
 415         struct fib_info *fi;
 416         u8 tos = cfg->fc_tos;
 417         __be32 key;
 418         int err;
 419
 420         if (cfg->fc_dst_len > 32)
 421                 return -EINVAL;
 422
 423         fz = table->fn_zones[cfg->fc_dst_len];
 424         if (!fz && !(fz = fn_new_zone(table, cfg->fc_dst_len)))
 425                 return -ENOBUFS;
 426
 427         key = 0;
 428         if (cfg->fc_dst) {
 429                 if (cfg->fc_dst & ~FZ_MASK(fz))
 430                         return -EINVAL;
 431                 key = fz_key(cfg->fc_dst, fz);
 432         }
 433
 434         fi = fib_create_info(cfg);
 435         if (IS_ERR(fi))
 436                 return PTR_ERR(fi);
 437
 438         if (fz->fz_nent > (fz->fz_divisor<<1) &&
 439             fz->fz_divisor < FZ_MAX_DIVISOR &&
 440             (cfg->fc_dst_len == 32 ||
 441              (1 << cfg->fc_dst_len) > fz->fz_divisor))
 442                 fn_rehash_zone(fz);
 443
 444         f = fib_find_node(fz, key);
 445
 446         if (!f)
 447                 fa = NULL;
 448         else
 449                 fa = fib_find_alias(&f->fn_alias, tos, fi->fib_priority);
 450
 451         /* Now fa, if non-NULL, points to the first fib alias
 452          * with the same keys [prefix,tos,priority], if such key already
 453          * exists or to the node before which we will insert new one.
 454          *
 455          * If fa is NULL, we will need to allocate a new one and
 456          * insert to the head of f.
 457          *
 458          * If f is NULL, no fib node matched the destination key
 459          * and we need to allocate a new one of those as well.
 460          */
 461
 462         if (fa && fa->fa_tos == tos &&
 463             fa->fa_info->fib_priority == fi->fib_priority) {
 464                 struct fib_alias *fa_first, *fa_match;
 465
 466                 err = -EEXIST;
 467                 if (cfg->fc_nlflags & NLM_F_EXCL)
 468                         goto out;
 469
 470                 /* We have 2 goals:
 471                  * 1. Find exact match for type, scope, fib_info to avoid
 472                  * duplicate routes
 473                  * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
 474                  */
 475                 fa_match = NULL;
 476                 fa_first = fa;
 477                 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 478                 list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 479                         if (fa->fa_tos != tos)
 480                                 break;
 481                         if (fa->fa_info->fib_priority != fi->fib_priority)
 482                                 break;
 483                         if (fa->fa_type == cfg->fc_type &&
 484                             fa->fa_scope == cfg->fc_scope &&
 485                             fa->fa_info == fi) {
 486                                 fa_match = fa;
 487                                 break;
 488                         }
 489                 }
 490
 491                 if (cfg->fc_nlflags & NLM_F_REPLACE) {
 492                         u8 state;
 493
 494                         fa = fa_first;
 495                         if (fa_match) {
 496                                 if (fa == fa_match)
 497                                         err = 0;
 498                                 goto out;
 499                         }
 500                         err = -ENOBUFS;
 501                         new_fa = fib_fast_alloc(f);
 502                         if (new_fa == NULL)
 503                                 goto out;
 504
 505                         new_fa->fa_tos = fa->fa_tos;
 506                         new_fa->fa_info = fi;
 507                         new_fa->fa_type = cfg->fc_type;
 508                         new_fa->fa_scope = cfg->fc_scope;
 509                         state = fa->fa_state;
 510                         new_fa->fa_state = state & ~FA_S_ACCESSED;
 511                         fib_hash_genid++;
 512                         list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
 513
 514                         fn_free_alias(fa, f);
 515                         if (state & FA_S_ACCESSED)
 516                                 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
 517                         rtmsg_fib(RTM_NEWROUTE, key, new_fa, cfg->fc_dst_len,
 518                                   tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
 519                         return 0;
 520                 }
 521
 522                 /* Error if we find a perfect match which
 523                  * uses the same scope, type, and nexthop
 524                  * information.
 525                  */
 526                 if (fa_match)
 527                         goto out;
 528
 529                 if (!(cfg->fc_nlflags & NLM_F_APPEND))
 530                         fa = fa_first;
 531         }
 532
 533         err = -ENOENT;
 534         if (!(cfg->fc_nlflags & NLM_F_CREATE))
 535                 goto out;
 536
 537         err = -ENOBUFS;
 538
 539         if (!f) {
 540                 new_f = kmem_cache_zalloc(fn_hash_kmem, GFP_KERNEL);
 541                 if (new_f == NULL)
 542                         goto out;
 543
 544                 INIT_HLIST_NODE(&new_f->fn_hash);
 545                 INIT_LIST_HEAD(&new_f->fn_alias);
 546                 new_f->fn_key = key;
 547                 f = new_f;
 548         }
 549
 550         new_fa = fib_fast_alloc(f);
 551         if (new_fa == NULL)
 552                 goto out;
 553
 554         new_fa->fa_info = fi;
 555         new_fa->fa_tos = tos;
 556         new_fa->fa_type = cfg->fc_type;
 557         new_fa->fa_scope = cfg->fc_scope;
 558         new_fa->fa_state = 0;
 559
 560         /*
 561          * Insert new entry to the list.
 562          */
 563
 564         if (new_f)
 565                 fib_insert_node(fz, new_f);
 566         list_add_tail_rcu(&new_fa->fa_list,
 567                  (fa ? &fa->fa_list : &f->fn_alias));
 568         fib_hash_genid++;
 569
 570         if (new_f)
 571                 fz->fz_nent++;
 572         rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
 573
 574         rtmsg_fib(RTM_NEWROUTE, key, new_fa, cfg->fc_dst_len, tb->tb_id,
 575                   &cfg->fc_nlinfo, 0);
 576         return 0;
 577
 578 out:
 579         if (new_f)
 580                 kmem_cache_free(fn_hash_kmem, new_f);
 581         fib_release_info(fi);
 582         return err;
 583 }
 584
 585 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
 586 {
 587         struct fn_hash *table = (struct fn_hash *)tb->tb_data;
 588         struct fib_node *f;
 589         struct fib_alias *fa, *fa_to_delete;
 590         struct fn_zone *fz;
 591         __be32 key;
 592
 593         if (cfg->fc_dst_len > 32)
 594                 return -EINVAL;
 595
 596         if ((fz  = table->fn_zones[cfg->fc_dst_len]) == NULL)
 597                 return -ESRCH;
 598
 599         key = 0;
 600         if (cfg->fc_dst) {
 601                 if (cfg->fc_dst & ~FZ_MASK(fz))
 602                         return -EINVAL;
 603                 key = fz_key(cfg->fc_dst, fz);
 604         }
 605
 606         f = fib_find_node(fz, key);
 607
 608         if (!f)
 609                 fa = NULL;
 610         else
 611                 fa = fib_find_alias(&f->fn_alias, cfg->fc_tos, 0);
 612         if (!fa)
 613                 return -ESRCH;
 614
 615         fa_to_delete = NULL;
 616         fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 617         list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 618                 struct fib_info *fi = fa->fa_info;
 619
 620                 if (fa->fa_tos != cfg->fc_tos)
 621                         break;
 622
 623                 if ((!cfg->fc_type ||
 624                      fa->fa_type == cfg->fc_type) &&
 625                     (cfg->fc_scope == RT_SCOPE_NOWHERE ||
 626                      fa->fa_scope == cfg->fc_scope) &&
 627                     (!cfg->fc_protocol ||
 628                      fi->fib_protocol == cfg->fc_protocol) &&
 629                     fib_nh_match(cfg, fi) == 0) {
 630                         fa_to_delete = fa;
 631                         break;
 632                 }
 633         }
 634
 635         if (fa_to_delete) {
 636                 int kill_fn;
 637
 638                 fa = fa_to_delete;
 639                 rtmsg_fib(RTM_DELROUTE, key, fa, cfg->fc_dst_len,
 640                           tb->tb_id, &cfg->fc_nlinfo, 0);
 641
 642                 kill_fn = 0;
 643                 list_del_rcu(&fa->fa_list);
 644                 if (list_empty(&f->fn_alias)) {
 645                         hlist_del_rcu(&f->fn_hash);
 646                         kill_fn = 1;
 647                 }
 648                 fib_hash_genid++;
 649
 650                 if (fa->fa_state & FA_S_ACCESSED)
 651                         rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
 652                 fn_free_alias(fa, f);
 653                 if (kill_fn) {
 654                         fn_free_node(f);
 655                         fz->fz_nent--;
 656                 }
 657
 658                 return 0;
 659         }
 660         return -ESRCH;
 661 }
 662
 663 static int fn_flush_list(struct fn_zone *fz, int idx)
 664 {
 665         struct hlist_head *head = &fz->fz_hash[idx];
 666         struct hlist_node *node, *n;
 667         struct fib_node *f;
 668         int found = 0;
 669
 670         hlist_for_each_entry_safe(f, node, n, head, fn_hash) {
 671                 struct fib_alias *fa, *fa_node;
 672                 int kill_f;
 673
 674                 kill_f = 0;
 675                 list_for_each_entry_safe(fa, fa_node, &f->fn_alias, fa_list) {
 676                         struct fib_info *fi = fa->fa_info;
 677
 678                         if (fi && (fi->fib_flags&RTNH_F_DEAD)) {
 679                                 list_del_rcu(&fa->fa_list);
 680                                 if (list_empty(&f->fn_alias)) {
 681                                         hlist_del_rcu(&f->fn_hash);
 682                                         kill_f = 1;
 683                                 }
 684                                 fib_hash_genid++;
 685
 686                                 fn_free_alias(fa, f);
 687                                 found++;
 688                         }
 689                 }
 690                 if (kill_f) {
 691                         fn_free_node(f);
 692                         fz->fz_nent--;
 693                 }
 694         }
 695         return found;
 696 }
 697
 698 /* caller must hold RTNL. */
 699 int fib_table_flush(struct fib_table *tb)
 700 {
 701         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 702         struct fn_zone *fz;
 703         int found = 0;
 704
 705         for (fz = rtnl_dereference(table->fn_zone_list);
 706              fz != NULL;
 707              fz = rtnl_dereference(fz->fz_next)) {
 708                 int i;
 709
 710                 for (i = fz->fz_divisor - 1; i >= 0; i--)
 711                         found += fn_flush_list(fz, i);
 712         }
 713         return found;
 714 }
 715
 716
 717 static inline int
 718 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
 719                      struct fib_table *tb,
 720                      struct fn_zone *fz,
 721                      struct hlist_head *head)
 722 {
 723         struct hlist_node *node;
 724         struct fib_node *f;
 725         int i, s_i;
 726
 727         s_i = cb->args[4];
 728         i = 0;
 729         hlist_for_each_entry_rcu(f, node, head, fn_hash) {
 730                 struct fib_alias *fa;
 731
 732                 list_for_each_entry_rcu(fa, &f->fn_alias, fa_list) {
 733                         if (i < s_i)
 734                                 goto next;
 735
 736                         if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
 737                                           cb->nlh->nlmsg_seq,
 738                                           RTM_NEWROUTE,
 739                                           tb->tb_id,
 740                                           fa->fa_type,
 741                                           fa->fa_scope,
 742                                           f->fn_key,
 743                                           fz->fz_order,
 744                                           fa->fa_tos,
 745                                           fa->fa_info,
 746                                           NLM_F_MULTI) < 0) {
 747                                 cb->args[4] = i;
 748                                 return -1;
 749                         }
 750 next:
 751                         i++;
 752                 }
 753         }
 754         cb->args[4] = i;
 755         return skb->len;
 756 }
 757
 758 static inline int
 759 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
 760                    struct fib_table *tb,
 761                    struct fn_zone *fz)
 762 {
 763         int h, s_h;
 764
 765         if (fz->fz_hash == NULL)
 766                 return skb->len;
 767         s_h = cb->args[3];
 768         for (h = s_h; h < fz->fz_divisor; h++) {
 769                 if (hlist_empty(&fz->fz_hash[h]))
 770                         continue;
 771                 if (fn_hash_dump_bucket(skb, cb, tb, fz, &fz->fz_hash[h]) < 0) {
 772                         cb->args[3] = h;
 773                         return -1;
 774                 }
 775                 memset(&cb->args[4], 0,
 776                        sizeof(cb->args) - 4*sizeof(cb->args[0]));
 777         }
 778         cb->args[3] = h;
 779         return skb->len;
 780 }
 781
 782 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
 783                    struct netlink_callback *cb)
 784 {
 785         int m = 0, s_m;
 786         struct fn_zone *fz;
 787         struct fn_hash *table = (struct fn_hash *)tb->tb_data;
 788
 789         s_m = cb->args[2];
 790         rcu_read_lock();
 791         for (fz = rcu_dereference(table->fn_zone_list);
 792              fz != NULL;
 793              fz = rcu_dereference(fz->fz_next), m++) {
 794                 if (m < s_m)
 795                         continue;
 796                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
 797                         cb->args[2] = m;
 798                         rcu_read_unlock();
 799                         return -1;
 800                 }
 801                 memset(&cb->args[3], 0,
 802                        sizeof(cb->args) - 3*sizeof(cb->args[0]));
 803         }
 804         rcu_read_unlock();
 805         cb->args[2] = m;
 806         return skb->len;
 807 }
 808
 809 void __init fib_hash_init(void)
 810 {
 811         fn_hash_kmem = kmem_cache_create("ip_fib_hash", sizeof(struct fib_node),
 812                                          0, SLAB_PANIC, NULL);
 813
 814         fn_alias_kmem = kmem_cache_create("ip_fib_alias", sizeof(struct fib_alias),
 815                                           0, SLAB_PANIC, NULL);
 816
 817 }
 818
 819 struct fib_table *fib_hash_table(u32 id)
 820 {
 821         struct fib_table *tb;
 822
 823         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash),
 824                      GFP_KERNEL);
 825         if (tb == NULL)
 826                 return NULL;
 827
 828         tb->tb_id = id;
 829         tb->tb_default = -1;
 830
 831         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 832         return tb;
 833 }
 834
 835 /* ------------------------------------------------------------------------ */
 836 #ifdef CONFIG_PROC_FS
 837
 838 struct fib_iter_state {
 839         struct seq_net_private p;
 840         struct fn_zone  *zone;
 841         int             bucket;
 842         struct hlist_head *hash_head;
 843         struct fib_node *fn;
 844         struct fib_alias *fa;
 845         loff_t pos;
 846         unsigned int genid;
 847         int valid;
 848 };
 849
 850 static struct fib_alias *fib_get_first(struct seq_file *seq)
 851 {
 852         struct fib_iter_state *iter = seq->private;
 853         struct fib_table *main_table;
 854         struct fn_hash *table;
 855
 856         main_table = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
 857         table = (struct fn_hash *)main_table->tb_data;
 858
 859         iter->bucket    = 0;
 860         iter->hash_head = NULL;
 861         iter->fn        = NULL;
 862         iter->fa        = NULL;
 863         iter->pos       = 0;
 864         iter->genid     = fib_hash_genid;
 865         iter->valid     = 1;
 866
 867         for (iter->zone = rcu_dereference(table->fn_zone_list);
 868              iter->zone != NULL;
 869              iter->zone = rcu_dereference(iter->zone->fz_next)) {
 870                 int maxslot;
 871
 872                 if (!iter->zone->fz_nent)
 873                         continue;
 874
 875                 iter->hash_head = iter->zone->fz_hash;
 876                 maxslot = iter->zone->fz_divisor;
 877
 878                 for (iter->bucket = 0; iter->bucket < maxslot;
 879                      ++iter->bucket, ++iter->hash_head) {
 880                         struct hlist_node *node;
 881                         struct fib_node *fn;
 882
 883                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 884                                 struct fib_alias *fa;
 885
 886                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 887                                         iter->fn = fn;
 888                                         iter->fa = fa;
 889                                         goto out;
 890                                 }
 891                         }
 892                 }
 893         }
 894 out:
 895         return iter->fa;
 896 }
 897
 898 static struct fib_alias *fib_get_next(struct seq_file *seq)
 899 {
 900         struct fib_iter_state *iter = seq->private;
 901         struct fib_node *fn;
 902         struct fib_alias *fa;
 903
 904         /* Advance FA, if any. */
 905         fn = iter->fn;
 906         fa = iter->fa;
 907         if (fa) {
 908                 BUG_ON(!fn);
 909                 list_for_each_entry_continue(fa, &fn->fn_alias, fa_list) {
 910                         iter->fa = fa;
 911                         goto out;
 912                 }
 913         }
 914
 915         fa = iter->fa = NULL;
 916
 917         /* Advance FN. */
 918         if (fn) {
 919                 struct hlist_node *node = &fn->fn_hash;
 920                 hlist_for_each_entry_continue(fn, node, fn_hash) {
 921                         iter->fn = fn;
 922
 923                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 924                                 iter->fa = fa;
 925                                 goto out;
 926                         }
 927                 }
 928         }
 929
 930         fn = iter->fn = NULL;
 931
 932         /* Advance hash chain. */
 933         if (!iter->zone)
 934                 goto out;
 935
 936         for (;;) {
 937                 struct hlist_node *node;
 938                 int maxslot;
 939
 940                 maxslot = iter->zone->fz_divisor;
 941
 942                 while (++iter->bucket < maxslot) {
 943                         iter->hash_head++;
 944
 945                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 946                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 947                                         iter->fn = fn;
 948                                         iter->fa = fa;
 949                                         goto out;
 950                                 }
 951                         }
 952                 }
 953
 954                 iter->zone = rcu_dereference(iter->zone->fz_next);
 955
 956                 if (!iter->zone)
 957                         goto out;
 958
 959                 iter->bucket = 0;
 960                 iter->hash_head = iter->zone->fz_hash;
 961
 962                 hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 963                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 964                                 iter->fn = fn;
 965                                 iter->fa = fa;
 966                                 goto out;
 967                         }
 968                 }
 969         }
 970 out:
 971         iter->pos++;
 972         return fa;
 973 }
 974
 975 static struct fib_alias *fib_get_idx(struct seq_file *seq, loff_t pos)
 976 {
 977         struct fib_iter_state *iter = seq->private;
 978         struct fib_alias *fa;
 979
 980         if (iter->valid && pos >= iter->pos && iter->genid == fib_hash_genid) {
 981                 fa   = iter->fa;
 982                 pos -= iter->pos;
 983         } else
 984                 fa = fib_get_first(seq);
 985
 986         if (fa)
 987                 while (pos && (fa = fib_get_next(seq)))
 988                         --pos;
 989         return pos ? NULL : fa;
 990 }
 991
 992 static void *fib_seq_start(struct seq_file *seq, loff_t *pos)
 993         __acquires(RCU)
 994 {
 995         void *v = NULL;
 996
 997         rcu_read_lock();
 998         if (fib_get_table(seq_file_net(seq), RT_TABLE_MAIN))
 999                 v = *pos ? fib_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1000         return v;
1001 }
1002
1003 static void *fib_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1004 {
1005         ++*pos;
1006         return v == SEQ_START_TOKEN ? fib_get_first(seq) : fib_get_next(seq);
1007 }
1008
1009 static void fib_seq_stop(struct seq_file *seq, void *v)
1010         __releases(RCU)
1011 {
1012         rcu_read_unlock();
1013 }
1014
1015 static unsigned fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
1016 {
1017         static const unsigned type2flags[RTN_MAX + 1] = {
1018                 [7] = RTF_REJECT,
1019                 [8] = RTF_REJECT,
1020         };
1021         unsigned flags = type2flags[type];
1022
1023         if (fi && fi->fib_nh->nh_gw)
1024                 flags |= RTF_GATEWAY;
1025         if (mask == htonl(0xFFFFFFFF))
1026                 flags |= RTF_HOST;
1027         flags |= RTF_UP;
1028         return flags;
1029 }
1030
1031 /*
1032  *      This outputs /proc/net/route.
1033  *
1034  *      It always works in backward compatibility mode.
1035  *      The format of the file is not supposed to be changed.
1036  */
1037 static int fib_seq_show(struct seq_file *seq, void *v)
1038 {
1039         struct fib_iter_state *iter;
1040         int len;
1041         __be32 prefix, mask;
1042         unsigned flags;
1043         struct fib_node *f;
1044         struct fib_alias *fa;
1045         struct fib_info *fi;
1046
1047         if (v == SEQ_START_TOKEN) {
1048                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
1049                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
1050                            "\tWindow\tIRTT");
1051                 goto out;
1052         }
1053
1054         iter    = seq->private;
1055         f       = iter->fn;
1056         fa      = iter->fa;
1057         fi      = fa->fa_info;
1058         prefix  = f->fn_key;
1059         mask    = FZ_MASK(iter->zone);
1060         flags   = fib_flag_trans(fa->fa_type, mask, fi);
1061         if (fi)
1062                 seq_printf(seq,
1063                          "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u%n",
1064                          fi->fib_dev ? fi->fib_dev->name : "*", prefix,
1065                          fi->fib_nh->nh_gw, flags, 0, 0, fi->fib_priority,
1066                          mask, (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
1067                          fi->fib_window,
1068                          fi->fib_rtt >> 3, &len);
1069         else
1070                 seq_printf(seq,
1071                          "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u%n",
1072                          prefix, 0, flags, 0, 0, 0, mask, 0, 0, 0, &len);
1073
1074         seq_printf(seq, "%*s\n", 127 - len, "");
1075 out:
1076         return 0;
1077 }
1078
1079 static const struct seq_operations fib_seq_ops = {
1080         .start  = fib_seq_start,
1081         .next   = fib_seq_next,
1082         .stop   = fib_seq_stop,
1083         .show   = fib_seq_show,
1084 };
1085
1086 static int fib_seq_open(struct inode *inode, struct file *file)
1087 {
1088         return seq_open_net(inode, file, &fib_seq_ops,
1089                             sizeof(struct fib_iter_state));
1090 }
1091
1092 static const struct file_operations fib_seq_fops = {
1093         .owner          = THIS_MODULE,
1094         .open           = fib_seq_open,
1095         .read           = seq_read,
1096         .llseek         = seq_lseek,
1097         .release        = seq_release_net,
1098 };
1099
1100 int __net_init fib_proc_init(struct net *net)
1101 {
1102         if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_seq_fops))
1103                 return -ENOMEM;
1104         return 0;
1105 }
1106
1107 void __net_exit fib_proc_exit(struct net *net)
1108 {
1109         proc_net_remove(net, "route");
1110 }
1111 #endif /* CONFIG_PROC_FS */