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  * Version:     $Id: fib_hash.c,v 1.13 2001/10/31 21:55:54 davem Exp $
   9  *
  10  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  11  *
  12  *              This program is free software; you can redistribute it and/or
  13  *              modify it under the terms of the GNU General Public License
  14  *              as published by the Free Software Foundation; either version
  15  *              2 of the License, or (at your option) any later version.
  16  */
  17
  18 #include <asm/uaccess.h>
  19 #include <asm/system.h>
  20 #include <linux/bitops.h>
  21 #include <linux/types.h>
  22 #include <linux/kernel.h>
  23 #include <linux/mm.h>
  24 #include <linux/string.h>
  25 #include <linux/socket.h>
  26 #include <linux/sockios.h>
  27 #include <linux/errno.h>
  28 #include <linux/in.h>
  29 #include <linux/inet.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/netdevice.h>
  32 #include <linux/if_arp.h>
  33 #include <linux/proc_fs.h>
  34 #include <linux/skbuff.h>
  35 #include <linux/netlink.h>
  36 #include <linux/init.h>
  37
  38 #include <net/net_namespace.h>
  39 #include <net/ip.h>
  40 #include <net/protocol.h>
  41 #include <net/route.h>
  42 #include <net/tcp.h>
  43 #include <net/sock.h>
  44 #include <net/ip_fib.h>
  45
  46 #include "fib_lookup.h"
  47
  48 static struct kmem_cache *fn_hash_kmem __read_mostly;
  49 static struct kmem_cache *fn_alias_kmem __read_mostly;
  50
  51 struct fib_node {
  52         struct hlist_node       fn_hash;
  53         struct list_head        fn_alias;
  54         __be32                  fn_key;
  55 };
  56
  57 struct fn_zone {
  58         struct fn_zone          *fz_next;       /* Next not empty zone  */
  59         struct hlist_head       *fz_hash;       /* Hash table pointer   */
  60         int                     fz_nent;        /* Number of entries    */
  61
  62         int                     fz_divisor;     /* Hash divisor         */
  63         u32                     fz_hashmask;    /* (fz_divisor - 1)     */
  64 #define FZ_HASHMASK(fz)         ((fz)->fz_hashmask)
  65
  66         int                     fz_order;       /* Zone order           */
  67         __be32                  fz_mask;
  68 #define FZ_MASK(fz)             ((fz)->fz_mask)
  69 };
  70
  71 /* NOTE. On fast computers evaluation of fz_hashmask and fz_mask
  72  * can be cheaper than memory lookup, so that FZ_* macros are used.
  73  */
  74
  75 struct fn_hash {
  76         struct fn_zone  *fn_zones[33];
  77         struct fn_zone  *fn_zone_list;
  78 };
  79
  80 static inline u32 fn_hash(__be32 key, struct fn_zone *fz)
  81 {
  82         u32 h = ntohl(key)>>(32 - fz->fz_order);
  83         h ^= (h>>20);
  84         h ^= (h>>10);
  85         h ^= (h>>5);
  86         h &= FZ_HASHMASK(fz);
  87         return h;
  88 }
  89
  90 static inline __be32 fz_key(__be32 dst, struct fn_zone *fz)
  91 {
  92         return dst & FZ_MASK(fz);
  93 }
  94
  95 static DEFINE_RWLOCK(fib_hash_lock);
  96 static unsigned int fib_hash_genid;
  97
  98 #define FZ_MAX_DIVISOR ((PAGE_SIZE<<MAX_ORDER) / sizeof(struct hlist_head))
  99
 100 static struct hlist_head *fz_hash_alloc(int divisor)
 101 {
 102         unsigned long size = divisor * sizeof(struct hlist_head);
 103
 104         if (size <= PAGE_SIZE) {
 105                 return kzalloc(size, GFP_KERNEL);
 106         } else {
 107                 return (struct hlist_head *)
 108                         __get_free_pages(GFP_KERNEL | __GFP_ZERO, get_order(size));
 109         }
 110 }
 111
 112 /* The fib hash lock must be held when this is called. */
 113 static inline void fn_rebuild_zone(struct fn_zone *fz,
 114                                    struct hlist_head *old_ht,
 115                                    int old_divisor)
 116 {
 117         int i;
 118
 119         for (i = 0; i < old_divisor; i++) {
 120                 struct hlist_node *node, *n;
 121                 struct fib_node *f;
 122
 123                 hlist_for_each_entry_safe(f, node, n, &old_ht[i], fn_hash) {
 124                         struct hlist_head *new_head;
 125
 126                         hlist_del(&f->fn_hash);
 127
 128                         new_head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 129                         hlist_add_head(&f->fn_hash, new_head);
 130                 }
 131         }
 132 }
 133
 134 static void fz_hash_free(struct hlist_head *hash, int divisor)
 135 {
 136         unsigned long size = divisor * sizeof(struct hlist_head);
 137
 138         if (size <= PAGE_SIZE)
 139                 kfree(hash);
 140         else
 141                 free_pages((unsigned long)hash, get_order(size));
 142 }
 143
 144 static void fn_rehash_zone(struct fn_zone *fz)
 145 {
 146         struct hlist_head *ht, *old_ht;
 147         int old_divisor, new_divisor;
 148         u32 new_hashmask;
 149
 150         old_divisor = fz->fz_divisor;
 151
 152         switch (old_divisor) {
 153         case 16:
 154                 new_divisor = 256;
 155                 break;
 156         case 256:
 157                 new_divisor = 1024;
 158                 break;
 159         default:
 160                 if ((old_divisor << 1) > FZ_MAX_DIVISOR) {
 161                         printk(KERN_CRIT "route.c: bad divisor %d!\n", old_divisor);
 162                         return;
 163                 }
 164                 new_divisor = (old_divisor << 1);
 165                 break;
 166         }
 167
 168         new_hashmask = (new_divisor - 1);
 169
 170 #if RT_CACHE_DEBUG >= 2
 171         printk(KERN_DEBUG "fn_rehash_zone: hash for zone %d grows from %d\n",
 172                fz->fz_order, old_divisor);
 173 #endif
 174
 175         ht = fz_hash_alloc(new_divisor);
 176
 177         if (ht) {
 178                 write_lock_bh(&fib_hash_lock);
 179                 old_ht = fz->fz_hash;
 180                 fz->fz_hash = ht;
 181                 fz->fz_hashmask = new_hashmask;
 182                 fz->fz_divisor = new_divisor;
 183                 fn_rebuild_zone(fz, old_ht, old_divisor);
 184                 fib_hash_genid++;
 185                 write_unlock_bh(&fib_hash_lock);
 186
 187                 fz_hash_free(old_ht, old_divisor);
 188         }
 189 }
 190
 191 static inline void fn_free_node(struct fib_node * f)
 192 {
 193         kmem_cache_free(fn_hash_kmem, f);
 194 }
 195
 196 static inline void fn_free_alias(struct fib_alias *fa)
 197 {
 198         fib_release_info(fa->fa_info);
 199         kmem_cache_free(fn_alias_kmem, fa);
 200 }
 201
 202 static struct fn_zone *
 203 fn_new_zone(struct fn_hash *table, int z)
 204 {
 205         int i;
 206         struct fn_zone *fz = kzalloc(sizeof(struct fn_zone), GFP_KERNEL);
 207         if (!fz)
 208                 return NULL;
 209
 210         if (z) {
 211                 fz->fz_divisor = 16;
 212         } else {
 213                 fz->fz_divisor = 1;
 214         }
 215         fz->fz_hashmask = (fz->fz_divisor - 1);
 216         fz->fz_hash = fz_hash_alloc(fz->fz_divisor);
 217         if (!fz->fz_hash) {
 218                 kfree(fz);
 219                 return NULL;
 220         }
 221         fz->fz_order = z;
 222         fz->fz_mask = inet_make_mask(z);
 223
 224         /* Find the first not empty zone with more specific mask */
 225         for (i=z+1; i<=32; i++)
 226                 if (table->fn_zones[i])
 227                         break;
 228         write_lock_bh(&fib_hash_lock);
 229         if (i>32) {
 230                 /* No more specific masks, we are the first. */
 231                 fz->fz_next = table->fn_zone_list;
 232                 table->fn_zone_list = fz;
 233         } else {
 234                 fz->fz_next = table->fn_zones[i]->fz_next;
 235                 table->fn_zones[i]->fz_next = fz;
 236         }
 237         table->fn_zones[z] = fz;
 238         fib_hash_genid++;
 239         write_unlock_bh(&fib_hash_lock);
 240         return fz;
 241 }
 242
 243 static int
 244 fn_hash_lookup(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
 245 {
 246         int err;
 247         struct fn_zone *fz;
 248         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 249
 250         read_lock(&fib_hash_lock);
 251         for (fz = t->fn_zone_list; fz; fz = fz->fz_next) {
 252                 struct hlist_head *head;
 253                 struct hlist_node *node;
 254                 struct fib_node *f;
 255                 __be32 k = fz_key(flp->fl4_dst, fz);
 256
 257                 head = &fz->fz_hash[fn_hash(k, fz)];
 258                 hlist_for_each_entry(f, node, head, fn_hash) {
 259                         if (f->fn_key != k)
 260                                 continue;
 261
 262                         err = fib_semantic_match(&f->fn_alias,
 263                                                  flp, res,
 264                                                  f->fn_key, fz->fz_mask,
 265                                                  fz->fz_order);
 266                         if (err <= 0)
 267                                 goto out;
 268                 }
 269         }
 270         err = 1;
 271 out:
 272         read_unlock(&fib_hash_lock);
 273         return err;
 274 }
 275
 276 static void
 277 fn_hash_select_default(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
 278 {
 279         int order, last_idx;
 280         struct hlist_node *node;
 281         struct fib_node *f;
 282         struct fib_info *fi = NULL;
 283         struct fib_info *last_resort;
 284         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 285         struct fn_zone *fz = t->fn_zones[0];
 286
 287         if (fz == NULL)
 288                 return;
 289
 290         last_idx = -1;
 291         last_resort = NULL;
 292         order = -1;
 293
 294         read_lock(&fib_hash_lock);
 295         hlist_for_each_entry(f, node, &fz->fz_hash[0], fn_hash) {
 296                 struct fib_alias *fa;
 297
 298                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 299                         struct fib_info *next_fi = fa->fa_info;
 300
 301                         if (fa->fa_scope != res->scope ||
 302                             fa->fa_type != RTN_UNICAST)
 303                                 continue;
 304
 305                         if (next_fi->fib_priority > res->fi->fib_priority)
 306                                 break;
 307                         if (!next_fi->fib_nh[0].nh_gw ||
 308                             next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
 309                                 continue;
 310                         fa->fa_state |= FA_S_ACCESSED;
 311
 312                         if (fi == NULL) {
 313                                 if (next_fi != res->fi)
 314                                         break;
 315                         } else if (!fib_detect_death(fi, order, &last_resort,
 316                                                 &last_idx, tb->tb_default)) {
 317                                 fib_result_assign(res, fi);
 318                                 tb->tb_default = order;
 319                                 goto out;
 320                         }
 321                         fi = next_fi;
 322                         order++;
 323                 }
 324         }
 325
 326         if (order <= 0 || fi == NULL) {
 327                 tb->tb_default = -1;
 328                 goto out;
 329         }
 330
 331         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
 332                                 tb->tb_default)) {
 333                 fib_result_assign(res, fi);
 334                 tb->tb_default = order;
 335                 goto out;
 336         }
 337
 338         if (last_idx >= 0)
 339                 fib_result_assign(res, last_resort);
 340         tb->tb_default = last_idx;
 341 out:
 342         read_unlock(&fib_hash_lock);
 343 }
 344
 345 /* Insert node F to FZ. */
 346 static inline void fib_insert_node(struct fn_zone *fz, struct fib_node *f)
 347 {
 348         struct hlist_head *head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 349
 350         hlist_add_head(&f->fn_hash, head);
 351 }
 352
 353 /* Return the node in FZ matching KEY. */
 354 static struct fib_node *fib_find_node(struct fn_zone *fz, __be32 key)
 355 {
 356         struct hlist_head *head = &fz->fz_hash[fn_hash(key, fz)];
 357         struct hlist_node *node;
 358         struct fib_node *f;
 359
 360         hlist_for_each_entry(f, node, head, fn_hash) {
 361                 if (f->fn_key == key)
 362                         return f;
 363         }
 364
 365         return NULL;
 366 }
 367
 368 static int fn_hash_insert(struct fib_table *tb, struct fib_config *cfg)
 369 {
 370         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 371         struct fib_node *new_f, *f;
 372         struct fib_alias *fa, *new_fa;
 373         struct fn_zone *fz;
 374         struct fib_info *fi;
 375         u8 tos = cfg->fc_tos;
 376         __be32 key;
 377         int err;
 378
 379         if (cfg->fc_dst_len > 32)
 380                 return -EINVAL;
 381
 382         fz = table->fn_zones[cfg->fc_dst_len];
 383         if (!fz && !(fz = fn_new_zone(table, cfg->fc_dst_len)))
 384                 return -ENOBUFS;
 385
 386         key = 0;
 387         if (cfg->fc_dst) {
 388                 if (cfg->fc_dst & ~FZ_MASK(fz))
 389                         return -EINVAL;
 390                 key = fz_key(cfg->fc_dst, fz);
 391         }
 392
 393         fi = fib_create_info(cfg);
 394         if (IS_ERR(fi))
 395                 return PTR_ERR(fi);
 396
 397         if (fz->fz_nent > (fz->fz_divisor<<1) &&
 398             fz->fz_divisor < FZ_MAX_DIVISOR &&
 399             (cfg->fc_dst_len == 32 ||
 400              (1 << cfg->fc_dst_len) > fz->fz_divisor))
 401                 fn_rehash_zone(fz);
 402
 403         f = fib_find_node(fz, key);
 404
 405         if (!f)
 406                 fa = NULL;
 407         else
 408                 fa = fib_find_alias(&f->fn_alias, tos, fi->fib_priority);
 409
 410         /* Now fa, if non-NULL, points to the first fib alias
 411          * with the same keys [prefix,tos,priority], if such key already
 412          * exists or to the node before which we will insert new one.
 413          *
 414          * If fa is NULL, we will need to allocate a new one and
 415          * insert to the head of f.
 416          *
 417          * If f is NULL, no fib node matched the destination key
 418          * and we need to allocate a new one of those as well.
 419          */
 420
 421         if (fa && fa->fa_tos == tos &&
 422             fa->fa_info->fib_priority == fi->fib_priority) {
 423                 struct fib_alias *fa_orig;
 424
 425                 err = -EEXIST;
 426                 if (cfg->fc_nlflags & NLM_F_EXCL)
 427                         goto out;
 428
 429                 if (cfg->fc_nlflags & NLM_F_REPLACE) {
 430                         struct fib_info *fi_drop;
 431                         u8 state;
 432
 433                         if (fi->fib_treeref > 1)
 434                                 goto out;
 435
 436                         write_lock_bh(&fib_hash_lock);
 437                         fi_drop = fa->fa_info;
 438                         fa->fa_info = fi;
 439                         fa->fa_type = cfg->fc_type;
 440                         fa->fa_scope = cfg->fc_scope;
 441                         state = fa->fa_state;
 442                         fa->fa_state &= ~FA_S_ACCESSED;
 443                         fib_hash_genid++;
 444                         write_unlock_bh(&fib_hash_lock);
 445
 446                         fib_release_info(fi_drop);
 447                         if (state & FA_S_ACCESSED)
 448                                 rt_cache_flush(-1);
 449                         rtmsg_fib(RTM_NEWROUTE, key, fa, cfg->fc_dst_len, tb->tb_id,
 450                                   &cfg->fc_nlinfo, NLM_F_REPLACE);
 451                         return 0;
 452                 }
 453
 454                 /* Error if we find a perfect match which
 455                  * uses the same scope, type, and nexthop
 456                  * information.
 457                  */
 458                 fa_orig = fa;
 459                 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 460                 list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 461                         if (fa->fa_tos != tos)
 462                                 break;
 463                         if (fa->fa_info->fib_priority != fi->fib_priority)
 464                                 break;
 465                         if (fa->fa_type == cfg->fc_type &&
 466                             fa->fa_scope == cfg->fc_scope &&
 467                             fa->fa_info == fi)
 468                                 goto out;
 469                 }
 470                 if (!(cfg->fc_nlflags & NLM_F_APPEND))
 471                         fa = fa_orig;
 472         }
 473
 474         err = -ENOENT;
 475         if (!(cfg->fc_nlflags & NLM_F_CREATE))
 476                 goto out;
 477
 478         err = -ENOBUFS;
 479         new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
 480         if (new_fa == NULL)
 481                 goto out;
 482
 483         new_f = NULL;
 484         if (!f) {
 485                 new_f = kmem_cache_alloc(fn_hash_kmem, GFP_KERNEL);
 486                 if (new_f == NULL)
 487                         goto out_free_new_fa;
 488
 489                 INIT_HLIST_NODE(&new_f->fn_hash);
 490                 INIT_LIST_HEAD(&new_f->fn_alias);
 491                 new_f->fn_key = key;
 492                 f = new_f;
 493         }
 494
 495         new_fa->fa_info = fi;
 496         new_fa->fa_tos = tos;
 497         new_fa->fa_type = cfg->fc_type;
 498         new_fa->fa_scope = cfg->fc_scope;
 499         new_fa->fa_state = 0;
 500
 501         /*
 502          * Insert new entry to the list.
 503          */
 504
 505         write_lock_bh(&fib_hash_lock);
 506         if (new_f)
 507                 fib_insert_node(fz, new_f);
 508         list_add_tail(&new_fa->fa_list,
 509                  (fa ? &fa->fa_list : &f->fn_alias));
 510         fib_hash_genid++;
 511         write_unlock_bh(&fib_hash_lock);
 512
 513         if (new_f)
 514                 fz->fz_nent++;
 515         rt_cache_flush(-1);
 516
 517         rtmsg_fib(RTM_NEWROUTE, key, new_fa, cfg->fc_dst_len, tb->tb_id,
 518                   &cfg->fc_nlinfo, 0);
 519         return 0;
 520
 521 out_free_new_fa:
 522         kmem_cache_free(fn_alias_kmem, new_fa);
 523 out:
 524         fib_release_info(fi);
 525         return err;
 526 }
 527
 528
 529 static int fn_hash_delete(struct fib_table *tb, struct fib_config *cfg)
 530 {
 531         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 532         struct fib_node *f;
 533         struct fib_alias *fa, *fa_to_delete;
 534         struct fn_zone *fz;
 535         __be32 key;
 536
 537         if (cfg->fc_dst_len > 32)
 538                 return -EINVAL;
 539
 540         if ((fz  = table->fn_zones[cfg->fc_dst_len]) == NULL)
 541                 return -ESRCH;
 542
 543         key = 0;
 544         if (cfg->fc_dst) {
 545                 if (cfg->fc_dst & ~FZ_MASK(fz))
 546                         return -EINVAL;
 547                 key = fz_key(cfg->fc_dst, fz);
 548         }
 549
 550         f = fib_find_node(fz, key);
 551
 552         if (!f)
 553                 fa = NULL;
 554         else
 555                 fa = fib_find_alias(&f->fn_alias, cfg->fc_tos, 0);
 556         if (!fa)
 557                 return -ESRCH;
 558
 559         fa_to_delete = NULL;
 560         fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 561         list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 562                 struct fib_info *fi = fa->fa_info;
 563
 564                 if (fa->fa_tos != cfg->fc_tos)
 565                         break;
 566
 567                 if ((!cfg->fc_type ||
 568                      fa->fa_type == cfg->fc_type) &&
 569                     (cfg->fc_scope == RT_SCOPE_NOWHERE ||
 570                      fa->fa_scope == cfg->fc_scope) &&
 571                     (!cfg->fc_protocol ||
 572                      fi->fib_protocol == cfg->fc_protocol) &&
 573                     fib_nh_match(cfg, fi) == 0) {
 574                         fa_to_delete = fa;
 575                         break;
 576                 }
 577         }
 578
 579         if (fa_to_delete) {
 580                 int kill_fn;
 581
 582                 fa = fa_to_delete;
 583                 rtmsg_fib(RTM_DELROUTE, key, fa, cfg->fc_dst_len,
 584                           tb->tb_id, &cfg->fc_nlinfo, 0);
 585
 586                 kill_fn = 0;
 587                 write_lock_bh(&fib_hash_lock);
 588                 list_del(&fa->fa_list);
 589                 if (list_empty(&f->fn_alias)) {
 590                         hlist_del(&f->fn_hash);
 591                         kill_fn = 1;
 592                 }
 593                 fib_hash_genid++;
 594                 write_unlock_bh(&fib_hash_lock);
 595
 596                 if (fa->fa_state & FA_S_ACCESSED)
 597                         rt_cache_flush(-1);
 598                 fn_free_alias(fa);
 599                 if (kill_fn) {
 600                         fn_free_node(f);
 601                         fz->fz_nent--;
 602                 }
 603
 604                 return 0;
 605         }
 606         return -ESRCH;
 607 }
 608
 609 static int fn_flush_list(struct fn_zone *fz, int idx)
 610 {
 611         struct hlist_head *head = &fz->fz_hash[idx];
 612         struct hlist_node *node, *n;
 613         struct fib_node *f;
 614         int found = 0;
 615
 616         hlist_for_each_entry_safe(f, node, n, head, fn_hash) {
 617                 struct fib_alias *fa, *fa_node;
 618                 int kill_f;
 619
 620                 kill_f = 0;
 621                 list_for_each_entry_safe(fa, fa_node, &f->fn_alias, fa_list) {
 622                         struct fib_info *fi = fa->fa_info;
 623
 624                         if (fi && (fi->fib_flags&RTNH_F_DEAD)) {
 625                                 write_lock_bh(&fib_hash_lock);
 626                                 list_del(&fa->fa_list);
 627                                 if (list_empty(&f->fn_alias)) {
 628                                         hlist_del(&f->fn_hash);
 629                                         kill_f = 1;
 630                                 }
 631                                 fib_hash_genid++;
 632                                 write_unlock_bh(&fib_hash_lock);
 633
 634                                 fn_free_alias(fa);
 635                                 found++;
 636                         }
 637                 }
 638                 if (kill_f) {
 639                         fn_free_node(f);
 640                         fz->fz_nent--;
 641                 }
 642         }
 643         return found;
 644 }
 645
 646 static int fn_hash_flush(struct fib_table *tb)
 647 {
 648         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 649         struct fn_zone *fz;
 650         int found = 0;
 651
 652         for (fz = table->fn_zone_list; fz; fz = fz->fz_next) {
 653                 int i;
 654
 655                 for (i = fz->fz_divisor - 1; i >= 0; i--)
 656                         found += fn_flush_list(fz, i);
 657         }
 658         return found;
 659 }
 660
 661
 662 static inline int
 663 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
 664                      struct fib_table *tb,
 665                      struct fn_zone *fz,
 666                      struct hlist_head *head)
 667 {
 668         struct hlist_node *node;
 669         struct fib_node *f;
 670         int i, s_i;
 671
 672         s_i = cb->args[4];
 673         i = 0;
 674         hlist_for_each_entry(f, node, head, fn_hash) {
 675                 struct fib_alias *fa;
 676
 677                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 678                         if (i < s_i)
 679                                 goto next;
 680
 681                         if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
 682                                           cb->nlh->nlmsg_seq,
 683                                           RTM_NEWROUTE,
 684                                           tb->tb_id,
 685                                           fa->fa_type,
 686                                           fa->fa_scope,
 687                                           f->fn_key,
 688                                           fz->fz_order,
 689                                           fa->fa_tos,
 690                                           fa->fa_info,
 691                                           NLM_F_MULTI) < 0) {
 692                                 cb->args[4] = i;
 693                                 return -1;
 694                         }
 695                 next:
 696                         i++;
 697                 }
 698         }
 699         cb->args[4] = i;
 700         return skb->len;
 701 }
 702
 703 static inline int
 704 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
 705                    struct fib_table *tb,
 706                    struct fn_zone *fz)
 707 {
 708         int h, s_h;
 709
 710         if (fz->fz_hash == NULL)
 711                 return skb->len;
 712         s_h = cb->args[3];
 713         for (h = s_h; h < fz->fz_divisor; h++) {
 714                 if (hlist_empty(&fz->fz_hash[h]))
 715                         continue;
 716                 if (fn_hash_dump_bucket(skb, cb, tb, fz, &fz->fz_hash[h]) < 0) {
 717                         cb->args[3] = h;
 718                         return -1;
 719                 }
 720                 memset(&cb->args[4], 0,
 721                        sizeof(cb->args) - 4*sizeof(cb->args[0]));
 722         }
 723         cb->args[3] = h;
 724         return skb->len;
 725 }
 726
 727 static int fn_hash_dump(struct fib_table *tb, struct sk_buff *skb, struct netlink_callback *cb)
 728 {
 729         int m, s_m;
 730         struct fn_zone *fz;
 731         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 732
 733         s_m = cb->args[2];
 734         read_lock(&fib_hash_lock);
 735         for (fz = table->fn_zone_list, m=0; fz; fz = fz->fz_next, m++) {
 736                 if (m < s_m) continue;
 737                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
 738                         cb->args[2] = m;
 739                         read_unlock(&fib_hash_lock);
 740                         return -1;
 741                 }
 742                 memset(&cb->args[3], 0,
 743                        sizeof(cb->args) - 3*sizeof(cb->args[0]));
 744         }
 745         read_unlock(&fib_hash_lock);
 746         cb->args[2] = m;
 747         return skb->len;
 748 }
 749
 750 struct fib_table *fib_hash_init(u32 id)
 751 {
 752         struct fib_table *tb;
 753
 754         if (fn_hash_kmem == NULL)
 755                 fn_hash_kmem = kmem_cache_create("ip_fib_hash",
 756                                                  sizeof(struct fib_node),
 757                                                  0, SLAB_HWCACHE_ALIGN,
 758                                                  NULL);
 759
 760         if (fn_alias_kmem == NULL)
 761                 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
 762                                                   sizeof(struct fib_alias),
 763                                                   0, SLAB_HWCACHE_ALIGN,
 764                                                   NULL);
 765
 766         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash),
 767                      GFP_KERNEL);
 768         if (tb == NULL)
 769                 return NULL;
 770
 771         tb->tb_id = id;
 772         tb->tb_default = -1;
 773         tb->tb_lookup = fn_hash_lookup;
 774         tb->tb_insert = fn_hash_insert;
 775         tb->tb_delete = fn_hash_delete;
 776         tb->tb_flush = fn_hash_flush;
 777         tb->tb_select_default = fn_hash_select_default;
 778         tb->tb_dump = fn_hash_dump;
 779         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 780         return tb;
 781 }
 782
 783 /* ------------------------------------------------------------------------ */
 784 #ifdef CONFIG_PROC_FS
 785
 786 struct fib_iter_state {
 787         struct seq_net_private p;
 788         struct fn_zone  *zone;
 789         int             bucket;
 790         struct hlist_head *hash_head;
 791         struct fib_node *fn;
 792         struct fib_alias *fa;
 793         loff_t pos;
 794         unsigned int genid;
 795         int valid;
 796 };
 797
 798 static struct fib_alias *fib_get_first(struct seq_file *seq)
 799 {
 800         struct fib_iter_state *iter = seq->private;
 801         struct fib_table *main_table;
 802         struct fn_hash *table;
 803
 804         main_table = fib_get_table(iter->p.net, RT_TABLE_MAIN);
 805         table = (struct fn_hash *)main_table->tb_data;
 806
 807         iter->bucket    = 0;
 808         iter->hash_head = NULL;
 809         iter->fn        = NULL;
 810         iter->fa        = NULL;
 811         iter->pos       = 0;
 812         iter->genid     = fib_hash_genid;
 813         iter->valid     = 1;
 814
 815         for (iter->zone = table->fn_zone_list; iter->zone;
 816              iter->zone = iter->zone->fz_next) {
 817                 int maxslot;
 818
 819                 if (!iter->zone->fz_nent)
 820                         continue;
 821
 822                 iter->hash_head = iter->zone->fz_hash;
 823                 maxslot = iter->zone->fz_divisor;
 824
 825                 for (iter->bucket = 0; iter->bucket < maxslot;
 826                      ++iter->bucket, ++iter->hash_head) {
 827                         struct hlist_node *node;
 828                         struct fib_node *fn;
 829
 830                         hlist_for_each_entry(fn,node,iter->hash_head,fn_hash) {
 831                                 struct fib_alias *fa;
 832
 833                                 list_for_each_entry(fa,&fn->fn_alias,fa_list) {
 834                                         iter->fn = fn;
 835                                         iter->fa = fa;
 836                                         goto out;
 837                                 }
 838                         }
 839                 }
 840         }
 841 out:
 842         return iter->fa;
 843 }
 844
 845 static struct fib_alias *fib_get_next(struct seq_file *seq)
 846 {
 847         struct fib_iter_state *iter = seq->private;
 848         struct fib_node *fn;
 849         struct fib_alias *fa;
 850
 851         /* Advance FA, if any. */
 852         fn = iter->fn;
 853         fa = iter->fa;
 854         if (fa) {
 855                 BUG_ON(!fn);
 856                 list_for_each_entry_continue(fa, &fn->fn_alias, fa_list) {
 857                         iter->fa = fa;
 858                         goto out;
 859                 }
 860         }
 861
 862         fa = iter->fa = NULL;
 863
 864         /* Advance FN. */
 865         if (fn) {
 866                 struct hlist_node *node = &fn->fn_hash;
 867                 hlist_for_each_entry_continue(fn, node, fn_hash) {
 868                         iter->fn = fn;
 869
 870                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 871                                 iter->fa = fa;
 872                                 goto out;
 873                         }
 874                 }
 875         }
 876
 877         fn = iter->fn = NULL;
 878
 879         /* Advance hash chain. */
 880         if (!iter->zone)
 881                 goto out;
 882
 883         for (;;) {
 884                 struct hlist_node *node;
 885                 int maxslot;
 886
 887                 maxslot = iter->zone->fz_divisor;
 888
 889                 while (++iter->bucket < maxslot) {
 890                         iter->hash_head++;
 891
 892                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 893                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 894                                         iter->fn = fn;
 895                                         iter->fa = fa;
 896                                         goto out;
 897                                 }
 898                         }
 899                 }
 900
 901                 iter->zone = iter->zone->fz_next;
 902
 903                 if (!iter->zone)
 904                         goto out;
 905
 906                 iter->bucket = 0;
 907                 iter->hash_head = iter->zone->fz_hash;
 908
 909                 hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 910                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 911                                 iter->fn = fn;
 912                                 iter->fa = fa;
 913                                 goto out;
 914                         }
 915                 }
 916         }
 917 out:
 918         iter->pos++;
 919         return fa;
 920 }
 921
 922 static struct fib_alias *fib_get_idx(struct seq_file *seq, loff_t pos)
 923 {
 924         struct fib_iter_state *iter = seq->private;
 925         struct fib_alias *fa;
 926
 927         if (iter->valid && pos >= iter->pos && iter->genid == fib_hash_genid) {
 928                 fa   = iter->fa;
 929                 pos -= iter->pos;
 930         } else
 931                 fa = fib_get_first(seq);
 932
 933         if (fa)
 934                 while (pos && (fa = fib_get_next(seq)))
 935                         --pos;
 936         return pos ? NULL : fa;
 937 }
 938
 939 static void *fib_seq_start(struct seq_file *seq, loff_t *pos)
 940         __acquires(fib_hash_lock)
 941 {
 942         struct fib_iter_state *iter = seq->private;
 943         void *v = NULL;
 944
 945         read_lock(&fib_hash_lock);
 946         if (fib_get_table(iter->p.net, RT_TABLE_MAIN))
 947                 v = *pos ? fib_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 948         return v;
 949 }
 950
 951 static void *fib_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 952 {
 953         ++*pos;
 954         return v == SEQ_START_TOKEN ? fib_get_first(seq) : fib_get_next(seq);
 955 }
 956
 957 static void fib_seq_stop(struct seq_file *seq, void *v)
 958         __releases(fib_hash_lock)
 959 {
 960         read_unlock(&fib_hash_lock);
 961 }
 962
 963 static unsigned fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
 964 {
 965         static const unsigned type2flags[RTN_MAX + 1] = {
 966                 [7] = RTF_REJECT, [8] = RTF_REJECT,
 967         };
 968         unsigned flags = type2flags[type];
 969
 970         if (fi && fi->fib_nh->nh_gw)
 971                 flags |= RTF_GATEWAY;
 972         if (mask == htonl(0xFFFFFFFF))
 973                 flags |= RTF_HOST;
 974         flags |= RTF_UP;
 975         return flags;
 976 }
 977
 978 /*
 979  *      This outputs /proc/net/route.
 980  *
 981  *      It always works in backward compatibility mode.
 982  *      The format of the file is not supposed to be changed.
 983  */
 984 static int fib_seq_show(struct seq_file *seq, void *v)
 985 {
 986         struct fib_iter_state *iter;
 987         char bf[128];
 988         __be32 prefix, mask;
 989         unsigned flags;
 990         struct fib_node *f;
 991         struct fib_alias *fa;
 992         struct fib_info *fi;
 993
 994         if (v == SEQ_START_TOKEN) {
 995                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
 996                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
 997                            "\tWindow\tIRTT");
 998                 goto out;
 999         }
1000
1001         iter    = seq->private;
1002         f       = iter->fn;
1003         fa      = iter->fa;
1004         fi      = fa->fa_info;
1005         prefix  = f->fn_key;
1006         mask    = FZ_MASK(iter->zone);
1007         flags   = fib_flag_trans(fa->fa_type, mask, fi);
1008         if (fi)
1009                 snprintf(bf, sizeof(bf),
1010                          "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1011                          fi->fib_dev ? fi->fib_dev->name : "*", prefix,
1012                          fi->fib_nh->nh_gw, flags, 0, 0, fi->fib_priority,
1013                          mask, (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
1014                          fi->fib_window,
1015                          fi->fib_rtt >> 3);
1016         else
1017                 snprintf(bf, sizeof(bf),
1018                          "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1019                          prefix, 0, flags, 0, 0, 0, mask, 0, 0, 0);
1020         seq_printf(seq, "%-127s\n", bf);
1021 out:
1022         return 0;
1023 }
1024
1025 static const struct seq_operations fib_seq_ops = {
1026         .start  = fib_seq_start,
1027         .next   = fib_seq_next,
1028         .stop   = fib_seq_stop,
1029         .show   = fib_seq_show,
1030 };
1031
1032 static int fib_seq_open(struct inode *inode, struct file *file)
1033 {
1034         return seq_open_net(inode, file, &fib_seq_ops,
1035                             sizeof(struct fib_iter_state));
1036 }
1037
1038 static const struct file_operations fib_seq_fops = {
1039         .owner          = THIS_MODULE,
1040         .open           = fib_seq_open,
1041         .read           = seq_read,
1042         .llseek         = seq_lseek,
1043         .release        = seq_release_net,
1044 };
1045
1046 int __net_init fib_proc_init(struct net *net)
1047 {
1048         if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_seq_fops))
1049                 return -ENOMEM;
1050         return 0;
1051 }
1052
1053 void __net_exit fib_proc_exit(struct net *net)
1054 {
1055         proc_net_remove(net, "route");
1056 }
1057 #endif /* CONFIG_PROC_FS */