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                         fa->fa_state |= FA_S_ACCESSED;
 339
 340                         if (fi == NULL) {
 341                                 if (next_fi != res->fi)
 342                                         break;
 343                         } else if (!fib_detect_death(fi, order, &last_resort,
 344                                                 &last_idx, tb->tb_default)) {
 345                                 fib_result_assign(res, fi);
 346                                 tb->tb_default = order;
 347                                 goto out;
 348                         }
 349                         fi = next_fi;
 350                         order++;
 351                 }
 352         }
 353
 354         if (order <= 0 || fi == NULL) {
 355                 tb->tb_default = -1;
 356                 goto out;
 357         }
 358
 359         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
 360                                 tb->tb_default)) {
 361                 fib_result_assign(res, fi);
 362                 tb->tb_default = order;
 363                 goto out;
 364         }
 365
 366         if (last_idx >= 0)
 367                 fib_result_assign(res, last_resort);
 368         tb->tb_default = last_idx;
 369 out:
 370         rcu_read_unlock();
 371 }
 372
 373 /* Insert node F to FZ. */
 374 static inline void fib_insert_node(struct fn_zone *fz, struct fib_node *f)
 375 {
 376         struct hlist_head *head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 377
 378         hlist_add_head_rcu(&f->fn_hash, head);
 379 }
 380
 381 /* Return the node in FZ matching KEY. */
 382 static struct fib_node *fib_find_node(struct fn_zone *fz, __be32 key)
 383 {
 384         struct hlist_head *head = &fz->fz_hash[fn_hash(key, fz)];
 385         struct hlist_node *node;
 386         struct fib_node *f;
 387
 388         hlist_for_each_entry_rcu(f, node, head, fn_hash) {
 389                 if (f->fn_key == key)
 390                         return f;
 391         }
 392
 393         return NULL;
 394 }
 395
 396
 397 static struct fib_alias *fib_fast_alloc(struct fib_node *f)
 398 {
 399         struct fib_alias *fa = &f->fn_embedded_alias;
 400
 401         if (fa->fa_info != NULL)
 402                 fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
 403         return fa;
 404 }
 405
 406 /* Caller must hold RTNL. */
 407 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
 408 {
 409         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 410         struct fib_node *new_f = NULL;
 411         struct fib_node *f;
 412         struct fib_alias *fa, *new_fa;
 413         struct fn_zone *fz;
 414         struct fib_info *fi;
 415         u8 tos = cfg->fc_tos;
 416         __be32 key;
 417         int err;
 418
 419         if (cfg->fc_dst_len > 32)
 420                 return -EINVAL;
 421
 422         fz = table->fn_zones[cfg->fc_dst_len];
 423         if (!fz && !(fz = fn_new_zone(table, cfg->fc_dst_len)))
 424                 return -ENOBUFS;
 425
 426         key = 0;
 427         if (cfg->fc_dst) {
 428                 if (cfg->fc_dst & ~FZ_MASK(fz))
 429                         return -EINVAL;
 430                 key = fz_key(cfg->fc_dst, fz);
 431         }
 432
 433         fi = fib_create_info(cfg);
 434         if (IS_ERR(fi))
 435                 return PTR_ERR(fi);
 436
 437         if (fz->fz_nent > (fz->fz_divisor<<1) &&
 438             fz->fz_divisor < FZ_MAX_DIVISOR &&
 439             (cfg->fc_dst_len == 32 ||
 440              (1 << cfg->fc_dst_len) > fz->fz_divisor))
 441                 fn_rehash_zone(fz);
 442
 443         f = fib_find_node(fz, key);
 444
 445         if (!f)
 446                 fa = NULL;
 447         else
 448                 fa = fib_find_alias(&f->fn_alias, tos, fi->fib_priority);
 449
 450         /* Now fa, if non-NULL, points to the first fib alias
 451          * with the same keys [prefix,tos,priority], if such key already
 452          * exists or to the node before which we will insert new one.
 453          *
 454          * If fa is NULL, we will need to allocate a new one and
 455          * insert to the head of f.
 456          *
 457          * If f is NULL, no fib node matched the destination key
 458          * and we need to allocate a new one of those as well.
 459          */
 460
 461         if (fa && fa->fa_tos == tos &&
 462             fa->fa_info->fib_priority == fi->fib_priority) {
 463                 struct fib_alias *fa_first, *fa_match;
 464
 465                 err = -EEXIST;
 466                 if (cfg->fc_nlflags & NLM_F_EXCL)
 467                         goto out;
 468
 469                 /* We have 2 goals:
 470                  * 1. Find exact match for type, scope, fib_info to avoid
 471                  * duplicate routes
 472                  * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
 473                  */
 474                 fa_match = NULL;
 475                 fa_first = fa;
 476                 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 477                 list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 478                         if (fa->fa_tos != tos)
 479                                 break;
 480                         if (fa->fa_info->fib_priority != fi->fib_priority)
 481                                 break;
 482                         if (fa->fa_type == cfg->fc_type &&
 483                             fa->fa_scope == cfg->fc_scope &&
 484                             fa->fa_info == fi) {
 485                                 fa_match = fa;
 486                                 break;
 487                         }
 488                 }
 489
 490                 if (cfg->fc_nlflags & NLM_F_REPLACE) {
 491                         u8 state;
 492
 493                         fa = fa_first;
 494                         if (fa_match) {
 495                                 if (fa == fa_match)
 496                                         err = 0;
 497                                 goto out;
 498                         }
 499                         err = -ENOBUFS;
 500                         new_fa = fib_fast_alloc(f);
 501                         if (new_fa == NULL)
 502                                 goto out;
 503
 504                         new_fa->fa_tos = fa->fa_tos;
 505                         new_fa->fa_info = fi;
 506                         new_fa->fa_type = cfg->fc_type;
 507                         new_fa->fa_scope = cfg->fc_scope;
 508                         state = fa->fa_state;
 509                         new_fa->fa_state = state & ~FA_S_ACCESSED;
 510                         fib_hash_genid++;
 511                         list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
 512
 513                         fn_free_alias(fa, f);
 514                         if (state & FA_S_ACCESSED)
 515                                 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
 516                         rtmsg_fib(RTM_NEWROUTE, key, new_fa, cfg->fc_dst_len,
 517                                   tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
 518                         return 0;
 519                 }
 520
 521                 /* Error if we find a perfect match which
 522                  * uses the same scope, type, and nexthop
 523                  * information.
 524                  */
 525                 if (fa_match)
 526                         goto out;
 527
 528                 if (!(cfg->fc_nlflags & NLM_F_APPEND))
 529                         fa = fa_first;
 530         }
 531
 532         err = -ENOENT;
 533         if (!(cfg->fc_nlflags & NLM_F_CREATE))
 534                 goto out;
 535
 536         err = -ENOBUFS;
 537
 538         if (!f) {
 539                 new_f = kmem_cache_zalloc(fn_hash_kmem, GFP_KERNEL);
 540                 if (new_f == NULL)
 541                         goto out;
 542
 543                 INIT_HLIST_NODE(&new_f->fn_hash);
 544                 INIT_LIST_HEAD(&new_f->fn_alias);
 545                 new_f->fn_key = key;
 546                 f = new_f;
 547         }
 548
 549         new_fa = fib_fast_alloc(f);
 550         if (new_fa == NULL)
 551                 goto out;
 552
 553         new_fa->fa_info = fi;
 554         new_fa->fa_tos = tos;
 555         new_fa->fa_type = cfg->fc_type;
 556         new_fa->fa_scope = cfg->fc_scope;
 557         new_fa->fa_state = 0;
 558
 559         /*
 560          * Insert new entry to the list.
 561          */
 562
 563         if (new_f)
 564                 fib_insert_node(fz, new_f);
 565         list_add_tail_rcu(&new_fa->fa_list,
 566                  (fa ? &fa->fa_list : &f->fn_alias));
 567         fib_hash_genid++;
 568
 569         if (new_f)
 570                 fz->fz_nent++;
 571         rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
 572
 573         rtmsg_fib(RTM_NEWROUTE, key, new_fa, cfg->fc_dst_len, tb->tb_id,
 574                   &cfg->fc_nlinfo, 0);
 575         return 0;
 576
 577 out:
 578         if (new_f)
 579                 kmem_cache_free(fn_hash_kmem, new_f);
 580         fib_release_info(fi);
 581         return err;
 582 }
 583
 584 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
 585 {
 586         struct fn_hash *table = (struct fn_hash *)tb->tb_data;
 587         struct fib_node *f;
 588         struct fib_alias *fa, *fa_to_delete;
 589         struct fn_zone *fz;
 590         __be32 key;
 591
 592         if (cfg->fc_dst_len > 32)
 593                 return -EINVAL;
 594
 595         if ((fz  = table->fn_zones[cfg->fc_dst_len]) == NULL)
 596                 return -ESRCH;
 597
 598         key = 0;
 599         if (cfg->fc_dst) {
 600                 if (cfg->fc_dst & ~FZ_MASK(fz))
 601                         return -EINVAL;
 602                 key = fz_key(cfg->fc_dst, fz);
 603         }
 604
 605         f = fib_find_node(fz, key);
 606
 607         if (!f)
 608                 fa = NULL;
 609         else
 610                 fa = fib_find_alias(&f->fn_alias, cfg->fc_tos, 0);
 611         if (!fa)
 612                 return -ESRCH;
 613
 614         fa_to_delete = NULL;
 615         fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 616         list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 617                 struct fib_info *fi = fa->fa_info;
 618
 619                 if (fa->fa_tos != cfg->fc_tos)
 620                         break;
 621
 622                 if ((!cfg->fc_type ||
 623                      fa->fa_type == cfg->fc_type) &&
 624                     (cfg->fc_scope == RT_SCOPE_NOWHERE ||
 625                      fa->fa_scope == cfg->fc_scope) &&
 626                     (!cfg->fc_protocol ||
 627                      fi->fib_protocol == cfg->fc_protocol) &&
 628                     fib_nh_match(cfg, fi) == 0) {
 629                         fa_to_delete = fa;
 630                         break;
 631                 }
 632         }
 633
 634         if (fa_to_delete) {
 635                 int kill_fn;
 636
 637                 fa = fa_to_delete;
 638                 rtmsg_fib(RTM_DELROUTE, key, fa, cfg->fc_dst_len,
 639                           tb->tb_id, &cfg->fc_nlinfo, 0);
 640
 641                 kill_fn = 0;
 642                 list_del_rcu(&fa->fa_list);
 643                 if (list_empty(&f->fn_alias)) {
 644                         hlist_del_rcu(&f->fn_hash);
 645                         kill_fn = 1;
 646                 }
 647                 fib_hash_genid++;
 648
 649                 if (fa->fa_state & FA_S_ACCESSED)
 650                         rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
 651                 fn_free_alias(fa, f);
 652                 if (kill_fn) {
 653                         fn_free_node(f);
 654                         fz->fz_nent--;
 655                 }
 656
 657                 return 0;
 658         }
 659         return -ESRCH;
 660 }
 661
 662 static int fn_flush_list(struct fn_zone *fz, int idx)
 663 {
 664         struct hlist_head *head = &fz->fz_hash[idx];
 665         struct hlist_node *node, *n;
 666         struct fib_node *f;
 667         int found = 0;
 668
 669         hlist_for_each_entry_safe(f, node, n, head, fn_hash) {
 670                 struct fib_alias *fa, *fa_node;
 671                 int kill_f;
 672
 673                 kill_f = 0;
 674                 list_for_each_entry_safe(fa, fa_node, &f->fn_alias, fa_list) {
 675                         struct fib_info *fi = fa->fa_info;
 676
 677                         if (fi && (fi->fib_flags&RTNH_F_DEAD)) {
 678                                 list_del_rcu(&fa->fa_list);
 679                                 if (list_empty(&f->fn_alias)) {
 680                                         hlist_del_rcu(&f->fn_hash);
 681                                         kill_f = 1;
 682                                 }
 683                                 fib_hash_genid++;
 684
 685                                 fn_free_alias(fa, f);
 686                                 found++;
 687                         }
 688                 }
 689                 if (kill_f) {
 690                         fn_free_node(f);
 691                         fz->fz_nent--;
 692                 }
 693         }
 694         return found;
 695 }
 696
 697 /* caller must hold RTNL. */
 698 int fib_table_flush(struct fib_table *tb)
 699 {
 700         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 701         struct fn_zone *fz;
 702         int found = 0;
 703
 704         for (fz = rtnl_dereference(table->fn_zone_list);
 705              fz != NULL;
 706              fz = rtnl_dereference(fz->fz_next)) {
 707                 int i;
 708
 709                 for (i = fz->fz_divisor - 1; i >= 0; i--)
 710                         found += fn_flush_list(fz, i);
 711         }
 712         return found;
 713 }
 714
 715
 716 static inline int
 717 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
 718                      struct fib_table *tb,
 719                      struct fn_zone *fz,
 720                      struct hlist_head *head)
 721 {
 722         struct hlist_node *node;
 723         struct fib_node *f;
 724         int i, s_i;
 725
 726         s_i = cb->args[4];
 727         i = 0;
 728         hlist_for_each_entry_rcu(f, node, head, fn_hash) {
 729                 struct fib_alias *fa;
 730
 731                 list_for_each_entry_rcu(fa, &f->fn_alias, fa_list) {
 732                         if (i < s_i)
 733                                 goto next;
 734
 735                         if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
 736                                           cb->nlh->nlmsg_seq,
 737                                           RTM_NEWROUTE,
 738                                           tb->tb_id,
 739                                           fa->fa_type,
 740                                           fa->fa_scope,
 741                                           f->fn_key,
 742                                           fz->fz_order,
 743                                           fa->fa_tos,
 744                                           fa->fa_info,
 745                                           NLM_F_MULTI) < 0) {
 746                                 cb->args[4] = i;
 747                                 return -1;
 748                         }
 749 next:
 750                         i++;
 751                 }
 752         }
 753         cb->args[4] = i;
 754         return skb->len;
 755 }
 756
 757 static inline int
 758 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
 759                    struct fib_table *tb,
 760                    struct fn_zone *fz)
 761 {
 762         int h, s_h;
 763
 764         if (fz->fz_hash == NULL)
 765                 return skb->len;
 766         s_h = cb->args[3];
 767         for (h = s_h; h < fz->fz_divisor; h++) {
 768                 if (hlist_empty(&fz->fz_hash[h]))
 769                         continue;
 770                 if (fn_hash_dump_bucket(skb, cb, tb, fz, &fz->fz_hash[h]) < 0) {
 771                         cb->args[3] = h;
 772                         return -1;
 773                 }
 774                 memset(&cb->args[4], 0,
 775                        sizeof(cb->args) - 4*sizeof(cb->args[0]));
 776         }
 777         cb->args[3] = h;
 778         return skb->len;
 779 }
 780
 781 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
 782                    struct netlink_callback *cb)
 783 {
 784         int m = 0, s_m;
 785         struct fn_zone *fz;
 786         struct fn_hash *table = (struct fn_hash *)tb->tb_data;
 787
 788         s_m = cb->args[2];
 789         rcu_read_lock();
 790         for (fz = rcu_dereference(table->fn_zone_list);
 791              fz != NULL;
 792              fz = rcu_dereference(fz->fz_next), m++) {
 793                 if (m < s_m)
 794                         continue;
 795                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
 796                         cb->args[2] = m;
 797                         rcu_read_unlock();
 798                         return -1;
 799                 }
 800                 memset(&cb->args[3], 0,
 801                        sizeof(cb->args) - 3*sizeof(cb->args[0]));
 802         }
 803         rcu_read_unlock();
 804         cb->args[2] = m;
 805         return skb->len;
 806 }
 807
 808 void __init fib_hash_init(void)
 809 {
 810         fn_hash_kmem = kmem_cache_create("ip_fib_hash", sizeof(struct fib_node),
 811                                          0, SLAB_PANIC, NULL);
 812
 813         fn_alias_kmem = kmem_cache_create("ip_fib_alias", sizeof(struct fib_alias),
 814                                           0, SLAB_PANIC, NULL);
 815
 816 }
 817
 818 struct fib_table *fib_hash_table(u32 id)
 819 {
 820         struct fib_table *tb;
 821
 822         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash),
 823                      GFP_KERNEL);
 824         if (tb == NULL)
 825                 return NULL;
 826
 827         tb->tb_id = id;
 828         tb->tb_default = -1;
 829
 830         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 831         return tb;
 832 }
 833
 834 /* ------------------------------------------------------------------------ */
 835 #ifdef CONFIG_PROC_FS
 836
 837 struct fib_iter_state {
 838         struct seq_net_private p;
 839         struct fn_zone  *zone;
 840         int             bucket;
 841         struct hlist_head *hash_head;
 842         struct fib_node *fn;
 843         struct fib_alias *fa;
 844         loff_t pos;
 845         unsigned int genid;
 846         int valid;
 847 };
 848
 849 static struct fib_alias *fib_get_first(struct seq_file *seq)
 850 {
 851         struct fib_iter_state *iter = seq->private;
 852         struct fib_table *main_table;
 853         struct fn_hash *table;
 854
 855         main_table = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
 856         table = (struct fn_hash *)main_table->tb_data;
 857
 858         iter->bucket    = 0;
 859         iter->hash_head = NULL;
 860         iter->fn        = NULL;
 861         iter->fa        = NULL;
 862         iter->pos       = 0;
 863         iter->genid     = fib_hash_genid;
 864         iter->valid     = 1;
 865
 866         for (iter->zone = rcu_dereference(table->fn_zone_list);
 867              iter->zone != NULL;
 868              iter->zone = rcu_dereference(iter->zone->fz_next)) {
 869                 int maxslot;
 870
 871                 if (!iter->zone->fz_nent)
 872                         continue;
 873
 874                 iter->hash_head = iter->zone->fz_hash;
 875                 maxslot = iter->zone->fz_divisor;
 876
 877                 for (iter->bucket = 0; iter->bucket < maxslot;
 878                      ++iter->bucket, ++iter->hash_head) {
 879                         struct hlist_node *node;
 880                         struct fib_node *fn;
 881
 882                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 883                                 struct fib_alias *fa;
 884
 885                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 886                                         iter->fn = fn;
 887                                         iter->fa = fa;
 888                                         goto out;
 889                                 }
 890                         }
 891                 }
 892         }
 893 out:
 894         return iter->fa;
 895 }
 896
 897 static struct fib_alias *fib_get_next(struct seq_file *seq)
 898 {
 899         struct fib_iter_state *iter = seq->private;
 900         struct fib_node *fn;
 901         struct fib_alias *fa;
 902
 903         /* Advance FA, if any. */
 904         fn = iter->fn;
 905         fa = iter->fa;
 906         if (fa) {
 907                 BUG_ON(!fn);
 908                 list_for_each_entry_continue(fa, &fn->fn_alias, fa_list) {
 909                         iter->fa = fa;
 910                         goto out;
 911                 }
 912         }
 913
 914         fa = iter->fa = NULL;
 915
 916         /* Advance FN. */
 917         if (fn) {
 918                 struct hlist_node *node = &fn->fn_hash;
 919                 hlist_for_each_entry_continue(fn, node, fn_hash) {
 920                         iter->fn = fn;
 921
 922                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 923                                 iter->fa = fa;
 924                                 goto out;
 925                         }
 926                 }
 927         }
 928
 929         fn = iter->fn = NULL;
 930
 931         /* Advance hash chain. */
 932         if (!iter->zone)
 933                 goto out;
 934
 935         for (;;) {
 936                 struct hlist_node *node;
 937                 int maxslot;
 938
 939                 maxslot = iter->zone->fz_divisor;
 940
 941                 while (++iter->bucket < maxslot) {
 942                         iter->hash_head++;
 943
 944                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 945                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 946                                         iter->fn = fn;
 947                                         iter->fa = fa;
 948                                         goto out;
 949                                 }
 950                         }
 951                 }
 952
 953                 iter->zone = rcu_dereference(iter->zone->fz_next);
 954
 955                 if (!iter->zone)
 956                         goto out;
 957
 958                 iter->bucket = 0;
 959                 iter->hash_head = iter->zone->fz_hash;
 960
 961                 hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 962                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 963                                 iter->fn = fn;
 964                                 iter->fa = fa;
 965                                 goto out;
 966                         }
 967                 }
 968         }
 969 out:
 970         iter->pos++;
 971         return fa;
 972 }
 973
 974 static struct fib_alias *fib_get_idx(struct seq_file *seq, loff_t pos)
 975 {
 976         struct fib_iter_state *iter = seq->private;
 977         struct fib_alias *fa;
 978
 979         if (iter->valid && pos >= iter->pos && iter->genid == fib_hash_genid) {
 980                 fa   = iter->fa;
 981                 pos -= iter->pos;
 982         } else
 983                 fa = fib_get_first(seq);
 984
 985         if (fa)
 986                 while (pos && (fa = fib_get_next(seq)))
 987                         --pos;
 988         return pos ? NULL : fa;
 989 }
 990
 991 static void *fib_seq_start(struct seq_file *seq, loff_t *pos)
 992         __acquires(RCU)
 993 {
 994         void *v = NULL;
 995
 996         rcu_read_lock();
 997         if (fib_get_table(seq_file_net(seq), RT_TABLE_MAIN))
 998                 v = *pos ? fib_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 999         return v;
1000 }
1001
1002 static void *fib_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1003 {
1004         ++*pos;
1005         return v == SEQ_START_TOKEN ? fib_get_first(seq) : fib_get_next(seq);
1006 }
1007
1008 static void fib_seq_stop(struct seq_file *seq, void *v)
1009         __releases(RCU)
1010 {
1011         rcu_read_unlock();
1012 }
1013
1014 static unsigned fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
1015 {
1016         static const unsigned type2flags[RTN_MAX + 1] = {
1017                 [7] = RTF_REJECT,
1018                 [8] = RTF_REJECT,
1019         };
1020         unsigned flags = type2flags[type];
1021
1022         if (fi && fi->fib_nh->nh_gw)
1023                 flags |= RTF_GATEWAY;
1024         if (mask == htonl(0xFFFFFFFF))
1025                 flags |= RTF_HOST;
1026         flags |= RTF_UP;
1027         return flags;
1028 }
1029
1030 /*
1031  *      This outputs /proc/net/route.
1032  *
1033  *      It always works in backward compatibility mode.
1034  *      The format of the file is not supposed to be changed.
1035  */
1036 static int fib_seq_show(struct seq_file *seq, void *v)
1037 {
1038         struct fib_iter_state *iter;
1039         int len;
1040         __be32 prefix, mask;
1041         unsigned flags;
1042         struct fib_node *f;
1043         struct fib_alias *fa;
1044         struct fib_info *fi;
1045
1046         if (v == SEQ_START_TOKEN) {
1047                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
1048                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
1049                            "\tWindow\tIRTT");
1050                 goto out;
1051         }
1052
1053         iter    = seq->private;
1054         f       = iter->fn;
1055         fa      = iter->fa;
1056         fi      = fa->fa_info;
1057         prefix  = f->fn_key;
1058         mask    = FZ_MASK(iter->zone);
1059         flags   = fib_flag_trans(fa->fa_type, mask, fi);
1060         if (fi)
1061                 seq_printf(seq,
1062                          "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u%n",
1063                          fi->fib_dev ? fi->fib_dev->name : "*", prefix,
1064                          fi->fib_nh->nh_gw, flags, 0, 0, fi->fib_priority,
1065                          mask, (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
1066                          fi->fib_window,
1067                          fi->fib_rtt >> 3, &len);
1068         else
1069                 seq_printf(seq,
1070                          "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u%n",
1071                          prefix, 0, flags, 0, 0, 0, mask, 0, 0, 0, &len);
1072
1073         seq_printf(seq, "%*s\n", 127 - len, "");
1074 out:
1075         return 0;
1076 }
1077
1078 static const struct seq_operations fib_seq_ops = {
1079         .start  = fib_seq_start,
1080         .next   = fib_seq_next,
1081         .stop   = fib_seq_stop,
1082         .show   = fib_seq_show,
1083 };
1084
1085 static int fib_seq_open(struct inode *inode, struct file *file)
1086 {
1087         return seq_open_net(inode, file, &fib_seq_ops,
1088                             sizeof(struct fib_iter_state));
1089 }
1090
1091 static const struct file_operations fib_seq_fops = {
1092         .owner          = THIS_MODULE,
1093         .open           = fib_seq_open,
1094         .read           = seq_read,
1095         .llseek         = seq_lseek,
1096         .release        = seq_release_net,
1097 };
1098
1099 int __net_init fib_proc_init(struct net *net)
1100 {
1101         if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_seq_fops))
1102                 return -ENOMEM;
1103         return 0;
1104 }
1105
1106 void __net_exit fib_proc_exit(struct net *net)
1107 {
1108         proc_net_remove(net, "route");
1109 }
1110 #endif /* CONFIG_PROC_FS */