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