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/sched.h>
  24 #include <linux/mm.h>
  25 #include <linux/string.h>
  26 #include <linux/socket.h>
  27 #include <linux/sockios.h>
  28 #include <linux/errno.h>
  29 #include <linux/in.h>
  30 #include <linux/inet.h>
  31 #include <linux/inetdevice.h>
  32 #include <linux/netdevice.h>
  33 #include <linux/if_arp.h>
  34 #include <linux/proc_fs.h>
  35 #include <linux/skbuff.h>
  36 #include <linux/netlink.h>
  37 #include <linux/init.h>
  38
  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 kmem_cache_t *fn_hash_kmem __read_mostly;
  49 static kmem_cache_t *fn_alias_kmem __read_mostly;
  50
  51 struct fib_node {
  52         struct hlist_node       fn_hash;
  53         struct list_head        fn_alias;
  54         u32                     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         u32                     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(u32 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 u32 fz_key(u32 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 kmalloc(size, GFP_KERNEL);
 106         } else {
 107                 return (struct hlist_head *)
 108                         __get_free_pages(GFP_KERNEL, 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("fn_rehash_zone: hash for zone %d grows from %d\n", fz->fz_order, old_divisor);
 172 #endif
 173
 174         ht = fz_hash_alloc(new_divisor);
 175
 176         if (ht) {
 177                 memset(ht, 0, new_divisor * sizeof(struct hlist_head));
 178
 179                 write_lock_bh(&fib_hash_lock);
 180                 old_ht = fz->fz_hash;
 181                 fz->fz_hash = ht;
 182                 fz->fz_hashmask = new_hashmask;
 183                 fz->fz_divisor = new_divisor;
 184                 fn_rebuild_zone(fz, old_ht, old_divisor);
 185                 fib_hash_genid++;
 186                 write_unlock_bh(&fib_hash_lock);
 187
 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)
 198 {
 199         fib_release_info(fa->fa_info);
 200         kmem_cache_free(fn_alias_kmem, fa);
 201 }
 202
 203 static struct fn_zone *
 204 fn_new_zone(struct fn_hash *table, int z)
 205 {
 206         int i;
 207         struct fn_zone *fz = kzalloc(sizeof(struct fn_zone), GFP_KERNEL);
 208         if (!fz)
 209                 return NULL;
 210
 211         if (z) {
 212                 fz->fz_divisor = 16;
 213         } else {
 214                 fz->fz_divisor = 1;
 215         }
 216         fz->fz_hashmask = (fz->fz_divisor - 1);
 217         fz->fz_hash = fz_hash_alloc(fz->fz_divisor);
 218         if (!fz->fz_hash) {
 219                 kfree(fz);
 220                 return NULL;
 221         }
 222         memset(fz->fz_hash, 0, fz->fz_divisor * sizeof(struct hlist_head *));
 223         fz->fz_order = z;
 224         fz->fz_mask = inet_make_mask(z);
 225
 226         /* Find the first not empty zone with more specific mask */
 227         for (i=z+1; i<=32; i++)
 228                 if (table->fn_zones[i])
 229                         break;
 230         write_lock_bh(&fib_hash_lock);
 231         if (i>32) {
 232                 /* No more specific masks, we are the first. */
 233                 fz->fz_next = table->fn_zone_list;
 234                 table->fn_zone_list = fz;
 235         } else {
 236                 fz->fz_next = table->fn_zones[i]->fz_next;
 237                 table->fn_zones[i]->fz_next = fz;
 238         }
 239         table->fn_zones[z] = fz;
 240         fib_hash_genid++;
 241         write_unlock_bh(&fib_hash_lock);
 242         return fz;
 243 }
 244
 245 static int
 246 fn_hash_lookup(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
 247 {
 248         int err;
 249         struct fn_zone *fz;
 250         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 251
 252         read_lock(&fib_hash_lock);
 253         for (fz = t->fn_zone_list; fz; fz = fz->fz_next) {
 254                 struct hlist_head *head;
 255                 struct hlist_node *node;
 256                 struct fib_node *f;
 257                 u32 k = fz_key(flp->fl4_dst, fz);
 258
 259                 head = &fz->fz_hash[fn_hash(k, fz)];
 260                 hlist_for_each_entry(f, node, head, fn_hash) {
 261                         if (f->fn_key != k)
 262                                 continue;
 263
 264                         err = fib_semantic_match(&f->fn_alias,
 265                                                  flp, res,
 266                                                  f->fn_key, fz->fz_mask,
 267                                                  fz->fz_order);
 268                         if (err <= 0)
 269                                 goto out;
 270                 }
 271         }
 272         err = 1;
 273 out:
 274         read_unlock(&fib_hash_lock);
 275         return err;
 276 }
 277
 278 static int fn_hash_last_dflt=-1;
 279
 280 static void
 281 fn_hash_select_default(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
 282 {
 283         int order, last_idx;
 284         struct hlist_node *node;
 285         struct fib_node *f;
 286         struct fib_info *fi = NULL;
 287         struct fib_info *last_resort;
 288         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 289         struct fn_zone *fz = t->fn_zones[0];
 290
 291         if (fz == NULL)
 292                 return;
 293
 294         last_idx = -1;
 295         last_resort = NULL;
 296         order = -1;
 297
 298         read_lock(&fib_hash_lock);
 299         hlist_for_each_entry(f, node, &fz->fz_hash[0], fn_hash) {
 300                 struct fib_alias *fa;
 301
 302                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 303                         struct fib_info *next_fi = fa->fa_info;
 304
 305                         if (fa->fa_scope != res->scope ||
 306                             fa->fa_type != RTN_UNICAST)
 307                                 continue;
 308
 309                         if (next_fi->fib_priority > res->fi->fib_priority)
 310                                 break;
 311                         if (!next_fi->fib_nh[0].nh_gw ||
 312                             next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
 313                                 continue;
 314                         fa->fa_state |= FA_S_ACCESSED;
 315
 316                         if (fi == NULL) {
 317                                 if (next_fi != res->fi)
 318                                         break;
 319                         } else if (!fib_detect_death(fi, order, &last_resort,
 320                                                      &last_idx, &fn_hash_last_dflt)) {
 321                                 if (res->fi)
 322                                         fib_info_put(res->fi);
 323                                 res->fi = fi;
 324                                 atomic_inc(&fi->fib_clntref);
 325                                 fn_hash_last_dflt = order;
 326                                 goto out;
 327                         }
 328                         fi = next_fi;
 329                         order++;
 330                 }
 331         }
 332
 333         if (order <= 0 || fi == NULL) {
 334                 fn_hash_last_dflt = -1;
 335                 goto out;
 336         }
 337
 338         if (!fib_detect_death(fi, order, &last_resort, &last_idx, &fn_hash_last_dflt)) {
 339                 if (res->fi)
 340                         fib_info_put(res->fi);
 341                 res->fi = fi;
 342                 atomic_inc(&fi->fib_clntref);
 343                 fn_hash_last_dflt = order;
 344                 goto out;
 345         }
 346
 347         if (last_idx >= 0) {
 348                 if (res->fi)
 349                         fib_info_put(res->fi);
 350                 res->fi = last_resort;
 351                 if (last_resort)
 352                         atomic_inc(&last_resort->fib_clntref);
 353         }
 354         fn_hash_last_dflt = last_idx;
 355 out:
 356         read_unlock(&fib_hash_lock);
 357 }
 358
 359 /* Insert node F to FZ. */
 360 static inline void fib_insert_node(struct fn_zone *fz, struct fib_node *f)
 361 {
 362         struct hlist_head *head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 363
 364         hlist_add_head(&f->fn_hash, head);
 365 }
 366
 367 /* Return the node in FZ matching KEY. */
 368 static struct fib_node *fib_find_node(struct fn_zone *fz, u32 key)
 369 {
 370         struct hlist_head *head = &fz->fz_hash[fn_hash(key, fz)];
 371         struct hlist_node *node;
 372         struct fib_node *f;
 373
 374         hlist_for_each_entry(f, node, head, fn_hash) {
 375                 if (f->fn_key == key)
 376                         return f;
 377         }
 378
 379         return NULL;
 380 }
 381
 382 static int
 383 fn_hash_insert(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
 384                struct nlmsghdr *n, struct netlink_skb_parms *req)
 385 {
 386         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 387         struct fib_node *new_f, *f;
 388         struct fib_alias *fa, *new_fa;
 389         struct fn_zone *fz;
 390         struct fib_info *fi;
 391         int z = r->rtm_dst_len;
 392         int type = r->rtm_type;
 393         u8 tos = r->rtm_tos;
 394         u32 key;
 395         int err;
 396
 397         if (z > 32)
 398                 return -EINVAL;
 399         fz = table->fn_zones[z];
 400         if (!fz && !(fz = fn_new_zone(table, z)))
 401                 return -ENOBUFS;
 402
 403         key = 0;
 404         if (rta->rta_dst) {
 405                 u32 dst;
 406                 memcpy(&dst, rta->rta_dst, 4);
 407                 if (dst & ~FZ_MASK(fz))
 408                         return -EINVAL;
 409                 key = fz_key(dst, fz);
 410         }
 411
 412         if  ((fi = fib_create_info(r, rta, n, &err)) == NULL)
 413                 return err;
 414
 415         if (fz->fz_nent > (fz->fz_divisor<<1) &&
 416             fz->fz_divisor < FZ_MAX_DIVISOR &&
 417             (z==32 || (1<<z) > fz->fz_divisor))
 418                 fn_rehash_zone(fz);
 419
 420         f = fib_find_node(fz, key);
 421
 422         if (!f)
 423                 fa = NULL;
 424         else
 425                 fa = fib_find_alias(&f->fn_alias, tos, fi->fib_priority);
 426
 427         /* Now fa, if non-NULL, points to the first fib alias
 428          * with the same keys [prefix,tos,priority], if such key already
 429          * exists or to the node before which we will insert new one.
 430          *
 431          * If fa is NULL, we will need to allocate a new one and
 432          * insert to the head of f.
 433          *
 434          * If f is NULL, no fib node matched the destination key
 435          * and we need to allocate a new one of those as well.
 436          */
 437
 438         if (fa && fa->fa_tos == tos &&
 439             fa->fa_info->fib_priority == fi->fib_priority) {
 440                 struct fib_alias *fa_orig;
 441
 442                 err = -EEXIST;
 443                 if (n->nlmsg_flags & NLM_F_EXCL)
 444                         goto out;
 445
 446                 if (n->nlmsg_flags & NLM_F_REPLACE) {
 447                         struct fib_info *fi_drop;
 448                         u8 state;
 449
 450                         write_lock_bh(&fib_hash_lock);
 451                         fi_drop = fa->fa_info;
 452                         fa->fa_info = fi;
 453                         fa->fa_type = type;
 454                         fa->fa_scope = r->rtm_scope;
 455                         state = fa->fa_state;
 456                         fa->fa_state &= ~FA_S_ACCESSED;
 457                         fib_hash_genid++;
 458                         write_unlock_bh(&fib_hash_lock);
 459
 460                         fib_release_info(fi_drop);
 461                         if (state & FA_S_ACCESSED)
 462                                 rt_cache_flush(-1);
 463                         return 0;
 464                 }
 465
 466                 /* Error if we find a perfect match which
 467                  * uses the same scope, type, and nexthop
 468                  * information.
 469                  */
 470                 fa_orig = fa;
 471                 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 472                 list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 473                         if (fa->fa_tos != tos)
 474                                 break;
 475                         if (fa->fa_info->fib_priority != fi->fib_priority)
 476                                 break;
 477                         if (fa->fa_type == type &&
 478                             fa->fa_scope == r->rtm_scope &&
 479                             fa->fa_info == fi)
 480                                 goto out;
 481                 }
 482                 if (!(n->nlmsg_flags & NLM_F_APPEND))
 483                         fa = fa_orig;
 484         }
 485
 486         err = -ENOENT;
 487         if (!(n->nlmsg_flags&NLM_F_CREATE))
 488                 goto out;
 489
 490         err = -ENOBUFS;
 491         new_fa = kmem_cache_alloc(fn_alias_kmem, SLAB_KERNEL);
 492         if (new_fa == NULL)
 493                 goto out;
 494
 495         new_f = NULL;
 496         if (!f) {
 497                 new_f = kmem_cache_alloc(fn_hash_kmem, SLAB_KERNEL);
 498                 if (new_f == NULL)
 499                         goto out_free_new_fa;
 500
 501                 INIT_HLIST_NODE(&new_f->fn_hash);
 502                 INIT_LIST_HEAD(&new_f->fn_alias);
 503                 new_f->fn_key = key;
 504                 f = new_f;
 505         }
 506
 507         new_fa->fa_info = fi;
 508         new_fa->fa_tos = tos;
 509         new_fa->fa_type = type;
 510         new_fa->fa_scope = r->rtm_scope;
 511         new_fa->fa_state = 0;
 512
 513         /*
 514          * Insert new entry to the list.
 515          */
 516
 517         write_lock_bh(&fib_hash_lock);
 518         if (new_f)
 519                 fib_insert_node(fz, new_f);
 520         list_add_tail(&new_fa->fa_list,
 521                  (fa ? &fa->fa_list : &f->fn_alias));
 522         fib_hash_genid++;
 523         write_unlock_bh(&fib_hash_lock);
 524
 525         if (new_f)
 526                 fz->fz_nent++;
 527         rt_cache_flush(-1);
 528
 529         rtmsg_fib(RTM_NEWROUTE, key, new_fa, z, tb->tb_id, n, req);
 530         return 0;
 531
 532 out_free_new_fa:
 533         kmem_cache_free(fn_alias_kmem, new_fa);
 534 out:
 535         fib_release_info(fi);
 536         return err;
 537 }
 538
 539
 540 static int
 541 fn_hash_delete(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
 542                struct nlmsghdr *n, struct netlink_skb_parms *req)
 543 {
 544         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 545         struct fib_node *f;
 546         struct fib_alias *fa, *fa_to_delete;
 547         int z = r->rtm_dst_len;
 548         struct fn_zone *fz;
 549         u32 key;
 550         u8 tos = r->rtm_tos;
 551
 552         if (z > 32)
 553                 return -EINVAL;
 554         if ((fz  = table->fn_zones[z]) == NULL)
 555                 return -ESRCH;
 556
 557         key = 0;
 558         if (rta->rta_dst) {
 559                 u32 dst;
 560                 memcpy(&dst, rta->rta_dst, 4);
 561                 if (dst & ~FZ_MASK(fz))
 562                         return -EINVAL;
 563                 key = fz_key(dst, fz);
 564         }
 565
 566         f = fib_find_node(fz, key);
 567
 568         if (!f)
 569                 fa = NULL;
 570         else
 571                 fa = fib_find_alias(&f->fn_alias, tos, 0);
 572         if (!fa)
 573                 return -ESRCH;
 574
 575         fa_to_delete = NULL;
 576         fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 577         list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 578                 struct fib_info *fi = fa->fa_info;
 579
 580                 if (fa->fa_tos != tos)
 581                         break;
 582
 583                 if ((!r->rtm_type ||
 584                      fa->fa_type == r->rtm_type) &&
 585                     (r->rtm_scope == RT_SCOPE_NOWHERE ||
 586                      fa->fa_scope == r->rtm_scope) &&
 587                     (!r->rtm_protocol ||
 588                      fi->fib_protocol == r->rtm_protocol) &&
 589                     fib_nh_match(r, n, rta, fi) == 0) {
 590                         fa_to_delete = fa;
 591                         break;
 592                 }
 593         }
 594
 595         if (fa_to_delete) {
 596                 int kill_fn;
 597
 598                 fa = fa_to_delete;
 599                 rtmsg_fib(RTM_DELROUTE, key, fa, z, tb->tb_id, n, req);
 600
 601                 kill_fn = 0;
 602                 write_lock_bh(&fib_hash_lock);
 603                 list_del(&fa->fa_list);
 604                 if (list_empty(&f->fn_alias)) {
 605                         hlist_del(&f->fn_hash);
 606                         kill_fn = 1;
 607                 }
 608                 fib_hash_genid++;
 609                 write_unlock_bh(&fib_hash_lock);
 610
 611                 if (fa->fa_state & FA_S_ACCESSED)
 612                         rt_cache_flush(-1);
 613                 fn_free_alias(fa);
 614                 if (kill_fn) {
 615                         fn_free_node(f);
 616                         fz->fz_nent--;
 617                 }
 618
 619                 return 0;
 620         }
 621         return -ESRCH;
 622 }
 623
 624 static int fn_flush_list(struct fn_zone *fz, int idx)
 625 {
 626         struct hlist_head *head = &fz->fz_hash[idx];
 627         struct hlist_node *node, *n;
 628         struct fib_node *f;
 629         int found = 0;
 630
 631         hlist_for_each_entry_safe(f, node, n, head, fn_hash) {
 632                 struct fib_alias *fa, *fa_node;
 633                 int kill_f;
 634
 635                 kill_f = 0;
 636                 list_for_each_entry_safe(fa, fa_node, &f->fn_alias, fa_list) {
 637                         struct fib_info *fi = fa->fa_info;
 638
 639                         if (fi && (fi->fib_flags&RTNH_F_DEAD)) {
 640                                 write_lock_bh(&fib_hash_lock);
 641                                 list_del(&fa->fa_list);
 642                                 if (list_empty(&f->fn_alias)) {
 643                                         hlist_del(&f->fn_hash);
 644                                         kill_f = 1;
 645                                 }
 646                                 fib_hash_genid++;
 647                                 write_unlock_bh(&fib_hash_lock);
 648
 649                                 fn_free_alias(fa);
 650                                 found++;
 651                         }
 652                 }
 653                 if (kill_f) {
 654                         fn_free_node(f);
 655                         fz->fz_nent--;
 656                 }
 657         }
 658         return found;
 659 }
 660
 661 static int fn_hash_flush(struct fib_table *tb)
 662 {
 663         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 664         struct fn_zone *fz;
 665         int found = 0;
 666
 667         for (fz = table->fn_zone_list; fz; fz = fz->fz_next) {
 668                 int i;
 669
 670                 for (i = fz->fz_divisor - 1; i >= 0; i--)
 671                         found += fn_flush_list(fz, i);
 672         }
 673         return found;
 674 }
 675
 676
 677 static inline int
 678 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
 679                      struct fib_table *tb,
 680                      struct fn_zone *fz,
 681                      struct hlist_head *head)
 682 {
 683         struct hlist_node *node;
 684         struct fib_node *f;
 685         int i, s_i;
 686
 687         s_i = cb->args[3];
 688         i = 0;
 689         hlist_for_each_entry(f, node, head, fn_hash) {
 690                 struct fib_alias *fa;
 691
 692                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 693                         if (i < s_i)
 694                                 goto next;
 695
 696                         if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
 697                                           cb->nlh->nlmsg_seq,
 698                                           RTM_NEWROUTE,
 699                                           tb->tb_id,
 700                                           fa->fa_type,
 701                                           fa->fa_scope,
 702                                           &f->fn_key,
 703                                           fz->fz_order,
 704                                           fa->fa_tos,
 705                                           fa->fa_info,
 706                                           NLM_F_MULTI) < 0) {
 707                                 cb->args[3] = i;
 708                                 return -1;
 709                         }
 710                 next:
 711                         i++;
 712                 }
 713         }
 714         cb->args[3] = i;
 715         return skb->len;
 716 }
 717
 718 static inline int
 719 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
 720                    struct fib_table *tb,
 721                    struct fn_zone *fz)
 722 {
 723         int h, s_h;
 724
 725         s_h = cb->args[2];
 726         for (h=0; h < fz->fz_divisor; h++) {
 727                 if (h < s_h) continue;
 728                 if (h > s_h)
 729                         memset(&cb->args[3], 0,
 730                                sizeof(cb->args) - 3*sizeof(cb->args[0]));
 731                 if (fz->fz_hash == NULL ||
 732                     hlist_empty(&fz->fz_hash[h]))
 733                         continue;
 734                 if (fn_hash_dump_bucket(skb, cb, tb, fz, &fz->fz_hash[h])<0) {
 735                         cb->args[2] = h;
 736                         return -1;
 737                 }
 738         }
 739         cb->args[2] = h;
 740         return skb->len;
 741 }
 742
 743 static int fn_hash_dump(struct fib_table *tb, struct sk_buff *skb, struct netlink_callback *cb)
 744 {
 745         int m, s_m;
 746         struct fn_zone *fz;
 747         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 748
 749         s_m = cb->args[1];
 750         read_lock(&fib_hash_lock);
 751         for (fz = table->fn_zone_list, m=0; fz; fz = fz->fz_next, m++) {
 752                 if (m < s_m) continue;
 753                 if (m > s_m)
 754                         memset(&cb->args[2], 0,
 755                                sizeof(cb->args) - 2*sizeof(cb->args[0]));
 756                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
 757                         cb->args[1] = m;
 758                         read_unlock(&fib_hash_lock);
 759                         return -1;
 760                 }
 761         }
 762         read_unlock(&fib_hash_lock);
 763         cb->args[1] = m;
 764         return skb->len;
 765 }
 766
 767 #ifdef CONFIG_IP_MULTIPLE_TABLES
 768 struct fib_table * fib_hash_init(int id)
 769 #else
 770 struct fib_table * __init fib_hash_init(int id)
 771 #endif
 772 {
 773         struct fib_table *tb;
 774
 775         if (fn_hash_kmem == NULL)
 776                 fn_hash_kmem = kmem_cache_create("ip_fib_hash",
 777                                                  sizeof(struct fib_node),
 778                                                  0, SLAB_HWCACHE_ALIGN,
 779                                                  NULL, NULL);
 780
 781         if (fn_alias_kmem == NULL)
 782                 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
 783                                                   sizeof(struct fib_alias),
 784                                                   0, SLAB_HWCACHE_ALIGN,
 785                                                   NULL, NULL);
 786
 787         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash),
 788                      GFP_KERNEL);
 789         if (tb == NULL)
 790                 return NULL;
 791
 792         tb->tb_id = id;
 793         tb->tb_lookup = fn_hash_lookup;
 794         tb->tb_insert = fn_hash_insert;
 795         tb->tb_delete = fn_hash_delete;
 796         tb->tb_flush = fn_hash_flush;
 797         tb->tb_select_default = fn_hash_select_default;
 798         tb->tb_dump = fn_hash_dump;
 799         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 800         return tb;
 801 }
 802
 803 /* ------------------------------------------------------------------------ */
 804 #ifdef CONFIG_PROC_FS
 805
 806 struct fib_iter_state {
 807         struct fn_zone  *zone;
 808         int             bucket;
 809         struct hlist_head *hash_head;
 810         struct fib_node *fn;
 811         struct fib_alias *fa;
 812         loff_t pos;
 813         unsigned int genid;
 814         int valid;
 815 };
 816
 817 static struct fib_alias *fib_get_first(struct seq_file *seq)
 818 {
 819         struct fib_iter_state *iter = seq->private;
 820         struct fn_hash *table = (struct fn_hash *) ip_fib_main_table->tb_data;
 821
 822         iter->bucket    = 0;
 823         iter->hash_head = NULL;
 824         iter->fn        = NULL;
 825         iter->fa        = NULL;
 826         iter->pos       = 0;
 827         iter->genid     = fib_hash_genid;
 828         iter->valid     = 1;
 829
 830         for (iter->zone = table->fn_zone_list; iter->zone;
 831              iter->zone = iter->zone->fz_next) {
 832                 int maxslot;
 833
 834                 if (!iter->zone->fz_nent)
 835                         continue;
 836
 837                 iter->hash_head = iter->zone->fz_hash;
 838                 maxslot = iter->zone->fz_divisor;
 839
 840                 for (iter->bucket = 0; iter->bucket < maxslot;
 841                      ++iter->bucket, ++iter->hash_head) {
 842                         struct hlist_node *node;
 843                         struct fib_node *fn;
 844
 845                         hlist_for_each_entry(fn,node,iter->hash_head,fn_hash) {
 846                                 struct fib_alias *fa;
 847
 848                                 list_for_each_entry(fa,&fn->fn_alias,fa_list) {
 849                                         iter->fn = fn;
 850                                         iter->fa = fa;
 851                                         goto out;
 852                                 }
 853                         }
 854                 }
 855         }
 856 out:
 857         return iter->fa;
 858 }
 859
 860 static struct fib_alias *fib_get_next(struct seq_file *seq)
 861 {
 862         struct fib_iter_state *iter = seq->private;
 863         struct fib_node *fn;
 864         struct fib_alias *fa;
 865
 866         /* Advance FA, if any. */
 867         fn = iter->fn;
 868         fa = iter->fa;
 869         if (fa) {
 870                 BUG_ON(!fn);
 871                 list_for_each_entry_continue(fa, &fn->fn_alias, fa_list) {
 872                         iter->fa = fa;
 873                         goto out;
 874                 }
 875         }
 876
 877         fa = iter->fa = NULL;
 878
 879         /* Advance FN. */
 880         if (fn) {
 881                 struct hlist_node *node = &fn->fn_hash;
 882                 hlist_for_each_entry_continue(fn, node, fn_hash) {
 883                         iter->fn = fn;
 884
 885                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 886                                 iter->fa = fa;
 887                                 goto out;
 888                         }
 889                 }
 890         }
 891
 892         fn = iter->fn = NULL;
 893
 894         /* Advance hash chain. */
 895         if (!iter->zone)
 896                 goto out;
 897
 898         for (;;) {
 899                 struct hlist_node *node;
 900                 int maxslot;
 901
 902                 maxslot = iter->zone->fz_divisor;
 903
 904                 while (++iter->bucket < maxslot) {
 905                         iter->hash_head++;
 906
 907                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 908                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 909                                         iter->fn = fn;
 910                                         iter->fa = fa;
 911                                         goto out;
 912                                 }
 913                         }
 914                 }
 915
 916                 iter->zone = iter->zone->fz_next;
 917
 918                 if (!iter->zone)
 919                         goto out;
 920
 921                 iter->bucket = 0;
 922                 iter->hash_head = iter->zone->fz_hash;
 923
 924                 hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 925                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 926                                 iter->fn = fn;
 927                                 iter->fa = fa;
 928                                 goto out;
 929                         }
 930                 }
 931         }
 932 out:
 933         iter->pos++;
 934         return fa;
 935 }
 936
 937 static struct fib_alias *fib_get_idx(struct seq_file *seq, loff_t pos)
 938 {
 939         struct fib_iter_state *iter = seq->private;
 940         struct fib_alias *fa;
 941
 942         if (iter->valid && pos >= iter->pos && iter->genid == fib_hash_genid) {
 943                 fa   = iter->fa;
 944                 pos -= iter->pos;
 945         } else
 946                 fa = fib_get_first(seq);
 947
 948         if (fa)
 949                 while (pos && (fa = fib_get_next(seq)))
 950                         --pos;
 951         return pos ? NULL : fa;
 952 }
 953
 954 static void *fib_seq_start(struct seq_file *seq, loff_t *pos)
 955 {
 956         void *v = NULL;
 957
 958         read_lock(&fib_hash_lock);
 959         if (ip_fib_main_table)
 960                 v = *pos ? fib_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 961         return v;
 962 }
 963
 964 static void *fib_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 965 {
 966         ++*pos;
 967         return v == SEQ_START_TOKEN ? fib_get_first(seq) : fib_get_next(seq);
 968 }
 969
 970 static void fib_seq_stop(struct seq_file *seq, void *v)
 971 {
 972         read_unlock(&fib_hash_lock);
 973 }
 974
 975 static unsigned fib_flag_trans(int type, u32 mask, struct fib_info *fi)
 976 {
 977         static const unsigned type2flags[RTN_MAX + 1] = {
 978                 [7] = RTF_REJECT, [8] = RTF_REJECT,
 979         };
 980         unsigned flags = type2flags[type];
 981
 982         if (fi && fi->fib_nh->nh_gw)
 983                 flags |= RTF_GATEWAY;
 984         if (mask == 0xFFFFFFFF)
 985                 flags |= RTF_HOST;
 986         flags |= RTF_UP;
 987         return flags;
 988 }
 989
 990 extern int dev_in_nx_info(struct net_device *, struct nx_info *);
 991
 992 /*
 993  *      This outputs /proc/net/route.
 994  *
 995  *      It always works in backward compatibility mode.
 996  *      The format of the file is not supposed to be changed.
 997  */
 998 static int fib_seq_show(struct seq_file *seq, void *v)
 999 {
1000         struct fib_iter_state *iter;
1001         char bf[128];
1002         u32 prefix, mask;
1003         unsigned flags;
1004         struct fib_node *f;
1005         struct fib_alias *fa;
1006         struct fib_info *fi;
1007
1008         if (v == SEQ_START_TOKEN) {
1009                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
1010                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
1011                            "\tWindow\tIRTT");
1012                 goto out;
1013         }
1014
1015         iter    = seq->private;
1016         f       = iter->fn;
1017         fa      = iter->fa;
1018         fi      = fa->fa_info;
1019         prefix  = f->fn_key;
1020         mask    = FZ_MASK(iter->zone);
1021         flags   = fib_flag_trans(fa->fa_type, mask, fi);
1022         if (fi && (!vx_flags(VXF_HIDE_NETIF, 0) ||
1023                 dev_in_nx_info(fi->fib_dev, current->nx_info)))
1024                 snprintf(bf, sizeof(bf),
1025                          "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1026                          fi->fib_dev ? fi->fib_dev->name : "*", prefix,
1027                          fi->fib_nh->nh_gw, flags, 0, 0, fi->fib_priority,
1028                          mask, (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
1029                          fi->fib_window,
1030                          fi->fib_rtt >> 3);
1031         else
1032                 snprintf(bf, sizeof(bf),
1033                          "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1034                          prefix, 0, flags, 0, 0, 0, mask, 0, 0, 0);
1035         seq_printf(seq, "%-127s\n", bf);
1036 out:
1037         return 0;
1038 }
1039
1040 static struct seq_operations fib_seq_ops = {
1041         .start  = fib_seq_start,
1042         .next   = fib_seq_next,
1043         .stop   = fib_seq_stop,
1044         .show   = fib_seq_show,
1045 };
1046
1047 static int fib_seq_open(struct inode *inode, struct file *file)
1048 {
1049         struct seq_file *seq;
1050         int rc = -ENOMEM;
1051         struct fib_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1052
1053         if (!s)
1054                 goto out;
1055
1056         rc = seq_open(file, &fib_seq_ops);
1057         if (rc)
1058                 goto out_kfree;
1059
1060         seq          = file->private_data;
1061         seq->private = s;
1062 out:
1063         return rc;
1064 out_kfree:
1065         kfree(s);
1066         goto out;
1067 }
1068
1069 static struct file_operations fib_seq_fops = {
1070         .owner          = THIS_MODULE,
1071         .open           = fib_seq_open,
1072         .read           = seq_read,
1073         .llseek         = seq_lseek,
1074         .release        = seq_release_private,
1075 };
1076
1077 int __init fib_proc_init(void)
1078 {
1079         if (!proc_net_fops_create("route", S_IRUGO, &fib_seq_fops))
1080                 return -ENOMEM;
1081         return 0;
1082 }
1083
1084 void __init fib_proc_exit(void)
1085 {
1086         proc_net_remove("route");
1087 }
1088 #endif /* CONFIG_PROC_FS */