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 <linux/config.h>
  19 #include <asm/uaccess.h>
  20 #include <asm/system.h>
  21 #include <asm/bitops.h>
  22 #include <linux/types.h>
  23 #include <linux/kernel.h>
  24 #include <linux/sched.h>
  25 #include <linux/mm.h>
  26 #include <linux/string.h>
  27 #include <linux/socket.h>
  28 #include <linux/sockios.h>
  29 #include <linux/errno.h>
  30 #include <linux/in.h>
  31 #include <linux/inet.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;
  49 static kmem_cache_t *fn_alias_kmem;
  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 rwlock_t fib_hash_lock = RW_LOCK_UNLOCKED;
  96
  97 #define FZ_MAX_DIVISOR ((PAGE_SIZE<<MAX_ORDER) / sizeof(struct hlist_head))
  98
  99 static struct hlist_head *fz_hash_alloc(int divisor)
 100 {
 101         unsigned long size = divisor * sizeof(struct hlist_head);
 102
 103         if (size <= PAGE_SIZE) {
 104                 return kmalloc(size, GFP_KERNEL);
 105         } else {
 106                 return (struct hlist_head *)
 107                         __get_free_pages(GFP_KERNEL, get_order(size));
 108         }
 109 }
 110
 111 /* The fib hash lock must be held when this is called. */
 112 static inline void fn_rebuild_zone(struct fn_zone *fz,
 113                                    struct hlist_head *old_ht,
 114                                    int old_divisor)
 115 {
 116         int i;
 117
 118         for (i = 0; i < old_divisor; i++) {
 119                 struct hlist_node *node, *n;
 120                 struct fib_node *f;
 121
 122                 hlist_for_each_entry_safe(f, node, n, &old_ht[i], fn_hash) {
 123                         struct hlist_head *new_head;
 124
 125                         hlist_del(&f->fn_hash);
 126
 127                         new_head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 128                         hlist_add_head(&f->fn_hash, new_head);
 129                 }
 130         }
 131 }
 132
 133 static void fz_hash_free(struct hlist_head *hash, int divisor)
 134 {
 135         unsigned long size = divisor * sizeof(struct hlist_head);
 136
 137         if (size <= PAGE_SIZE)
 138                 kfree(hash);
 139         else
 140                 free_pages((unsigned long)hash, get_order(size));
 141 }
 142
 143 static void fn_rehash_zone(struct fn_zone *fz)
 144 {
 145         struct hlist_head *ht, *old_ht;
 146         int old_divisor, new_divisor;
 147         u32 new_hashmask;
 148
 149         old_divisor = fz->fz_divisor;
 150
 151         switch (old_divisor) {
 152         case 16:
 153                 new_divisor = 256;
 154                 break;
 155         case 256:
 156                 new_divisor = 1024;
 157                 break;
 158         default:
 159                 if ((old_divisor << 1) > FZ_MAX_DIVISOR) {
 160                         printk(KERN_CRIT "route.c: bad divisor %d!\n", old_divisor);
 161                         return;
 162                 }
 163                 new_divisor = (old_divisor << 1);
 164                 break;
 165         }
 166
 167         new_hashmask = (new_divisor - 1);
 168
 169 #if RT_CACHE_DEBUG >= 2
 170         printk("fn_rehash_zone: hash for zone %d grows from %d\n", fz->fz_order, old_divisor);
 171 #endif
 172
 173         ht = fz_hash_alloc(new_divisor);
 174
 175         if (ht) {
 176                 memset(ht, 0, new_divisor * sizeof(struct hlist_head));
 177
 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                 write_unlock_bh(&fib_hash_lock);
 185
 186                 fz_hash_free(old_ht, old_divisor);
 187         }
 188 }
 189
 190 static inline void fn_free_node(struct fib_node * f)
 191 {
 192         kmem_cache_free(fn_hash_kmem, f);
 193 }
 194
 195 static inline void fn_free_alias(struct fib_alias *fa)
 196 {
 197         fib_release_info(fa->fa_info);
 198         kmem_cache_free(fn_alias_kmem, fa);
 199 }
 200
 201 static struct fn_zone *
 202 fn_new_zone(struct fn_hash *table, int z)
 203 {
 204         int i;
 205         struct fn_zone *fz = kmalloc(sizeof(struct fn_zone), GFP_KERNEL);
 206         if (!fz)
 207                 return NULL;
 208
 209         memset(fz, 0, sizeof(struct fn_zone));
 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         memset(fz->fz_hash, 0, fz->fz_divisor * sizeof(struct hlist_head *));
 222         fz->fz_order = z;
 223         fz->fz_mask = inet_make_mask(z);
 224
 225         /* Find the first not empty zone with more specific mask */
 226         for (i=z+1; i<=32; i++)
 227                 if (table->fn_zones[i])
 228                         break;
 229         write_lock_bh(&fib_hash_lock);
 230         if (i>32) {
 231                 /* No more specific masks, we are the first. */
 232                 fz->fz_next = table->fn_zone_list;
 233                 table->fn_zone_list = fz;
 234         } else {
 235                 fz->fz_next = table->fn_zones[i]->fz_next;
 236                 table->fn_zones[i]->fz_next = fz;
 237         }
 238         table->fn_zones[z] = fz;
 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                 u32 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                                                  fz->fz_order);
 265                         if (err <= 0)
 266                                 goto out;
 267                 }
 268         }
 269         err = 1;
 270 out:
 271         read_unlock(&fib_hash_lock);
 272         return err;
 273 }
 274
 275 static int fn_hash_last_dflt=-1;
 276
 277 static int fib_detect_death(struct fib_info *fi, int order,
 278                             struct fib_info **last_resort, int *last_idx)
 279 {
 280         struct neighbour *n;
 281         int state = NUD_NONE;
 282
 283         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 284         if (n) {
 285                 state = n->nud_state;
 286                 neigh_release(n);
 287         }
 288         if (state==NUD_REACHABLE)
 289                 return 0;
 290         if ((state&NUD_VALID) && order != fn_hash_last_dflt)
 291                 return 0;
 292         if ((state&NUD_VALID) ||
 293             (*last_idx<0 && order > fn_hash_last_dflt)) {
 294                 *last_resort = fi;
 295                 *last_idx = order;
 296         }
 297         return 1;
 298 }
 299
 300 static void
 301 fn_hash_select_default(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
 302 {
 303         int order, last_idx;
 304         struct hlist_node *node;
 305         struct fib_node *f;
 306         struct fib_info *fi = NULL;
 307         struct fib_info *last_resort;
 308         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 309         struct fn_zone *fz = t->fn_zones[0];
 310
 311         if (fz == NULL)
 312                 return;
 313
 314         last_idx = -1;
 315         last_resort = NULL;
 316         order = -1;
 317
 318         read_lock(&fib_hash_lock);
 319         hlist_for_each_entry(f, node, &fz->fz_hash[0], fn_hash) {
 320                 struct fib_alias *fa;
 321
 322                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 323                         struct fib_info *next_fi = fa->fa_info;
 324
 325                         if (fa->fa_scope != res->scope ||
 326                             fa->fa_type != RTN_UNICAST)
 327                                 continue;
 328
 329                         if (next_fi->fib_priority > res->fi->fib_priority)
 330                                 break;
 331                         if (!next_fi->fib_nh[0].nh_gw ||
 332                             next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
 333                                 continue;
 334                         fa->fa_state |= FA_S_ACCESSED;
 335
 336                         if (fi == NULL) {
 337                                 if (next_fi != res->fi)
 338                                         break;
 339                         } else if (!fib_detect_death(fi, order, &last_resort,
 340                                                      &last_idx)) {
 341                                 if (res->fi)
 342                                         fib_info_put(res->fi);
 343                                 res->fi = fi;
 344                                 atomic_inc(&fi->fib_clntref);
 345                                 fn_hash_last_dflt = order;
 346                                 goto out;
 347                         }
 348                         fi = next_fi;
 349                         order++;
 350                 }
 351         }
 352
 353         if (order <= 0 || fi == NULL) {
 354                 fn_hash_last_dflt = -1;
 355                 goto out;
 356         }
 357
 358         if (!fib_detect_death(fi, order, &last_resort, &last_idx)) {
 359                 if (res->fi)
 360                         fib_info_put(res->fi);
 361                 res->fi = fi;
 362                 atomic_inc(&fi->fib_clntref);
 363                 fn_hash_last_dflt = order;
 364                 goto out;
 365         }
 366
 367         if (last_idx >= 0) {
 368                 if (res->fi)
 369                         fib_info_put(res->fi);
 370                 res->fi = last_resort;
 371                 if (last_resort)
 372                         atomic_inc(&last_resort->fib_clntref);
 373         }
 374         fn_hash_last_dflt = last_idx;
 375 out:
 376         read_unlock(&fib_hash_lock);
 377 }
 378
 379 static void rtmsg_fib(int, struct fib_node *, struct fib_alias *,
 380                       int, int,
 381                       struct nlmsghdr *n,
 382                       struct netlink_skb_parms *);
 383
 384 /* Insert node F to FZ. */
 385 static inline void fib_insert_node(struct fn_zone *fz, struct fib_node *f)
 386 {
 387         struct hlist_head *head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
 388
 389         hlist_add_head(&f->fn_hash, head);
 390 }
 391
 392 /* Return the node in FZ matching KEY. */
 393 static struct fib_node *fib_find_node(struct fn_zone *fz, u32 key)
 394 {
 395         struct hlist_head *head = &fz->fz_hash[fn_hash(key, fz)];
 396         struct hlist_node *node;
 397         struct fib_node *f;
 398
 399         hlist_for_each_entry(f, node, head, fn_hash) {
 400                 if (f->fn_key == key)
 401                         return f;
 402         }
 403
 404         return NULL;
 405 }
 406
 407 /* Return the first fib alias matching TOS with
 408  * priority less than or equal to PRIO.
 409  */
 410 static struct fib_alias *fib_find_alias(struct fib_node *fn, u8 tos, u32 prio)
 411 {
 412         if (fn) {
 413                 struct list_head *head = &fn->fn_alias;
 414                 struct fib_alias *fa;
 415
 416                 list_for_each_entry(fa, head, fa_list) {
 417                         if (fa->fa_tos > tos)
 418                                 continue;
 419                         if (fa->fa_info->fib_priority >= prio ||
 420                             fa->fa_tos < tos)
 421                                 return fa;
 422                 }
 423         }
 424         return NULL;
 425 }
 426
 427 static int
 428 fn_hash_insert(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
 429                struct nlmsghdr *n, struct netlink_skb_parms *req)
 430 {
 431         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 432         struct fib_node *new_f, *f;
 433         struct fib_alias *fa, *new_fa;
 434         struct fn_zone *fz;
 435         struct fib_info *fi;
 436         int z = r->rtm_dst_len;
 437         int type = r->rtm_type;
 438         u8 tos = r->rtm_tos;
 439         u32 key;
 440         int err;
 441
 442         if (z > 32)
 443                 return -EINVAL;
 444         fz = table->fn_zones[z];
 445         if (!fz && !(fz = fn_new_zone(table, z)))
 446                 return -ENOBUFS;
 447
 448         key = 0;
 449         if (rta->rta_dst) {
 450                 u32 dst;
 451                 memcpy(&dst, rta->rta_dst, 4);
 452                 if (dst & ~FZ_MASK(fz))
 453                         return -EINVAL;
 454                 key = fz_key(dst, fz);
 455         }
 456
 457         if  ((fi = fib_create_info(r, rta, n, &err)) == NULL)
 458                 return err;
 459
 460         if (fz->fz_nent > (fz->fz_divisor<<1) &&
 461             fz->fz_divisor < FZ_MAX_DIVISOR &&
 462             (z==32 || (1<<z) > fz->fz_divisor))
 463                 fn_rehash_zone(fz);
 464
 465         f = fib_find_node(fz, key);
 466         fa = fib_find_alias(f, tos, fi->fib_priority);
 467
 468         /* Now fa, if non-NULL, points to the first fib alias
 469          * with the same keys [prefix,tos,priority], if such key already
 470          * exists or to the node before which we will insert new one.
 471          *
 472          * If fa is NULL, we will need to allocate a new one and
 473          * insert to the head of f.
 474          *
 475          * If f is NULL, no fib node matched the destination key
 476          * and we need to allocate a new one of those as well.
 477          */
 478
 479         if (fa && fa->fa_tos == tos &&
 480             fa->fa_info->fib_priority == fi->fib_priority) {
 481                 struct fib_alias *fa_orig;
 482
 483                 err = -EEXIST;
 484                 if (n->nlmsg_flags & NLM_F_EXCL)
 485                         goto out;
 486
 487                 if (n->nlmsg_flags & NLM_F_REPLACE) {
 488                         struct fib_info *fi_drop;
 489                         u8 state;
 490
 491                         write_lock_bh(&fib_hash_lock);
 492                         fi_drop = fa->fa_info;
 493                         fa->fa_info = fi;
 494                         fa->fa_type = type;
 495                         fa->fa_scope = r->rtm_scope;
 496                         state = fa->fa_state;
 497                         fa->fa_state &= ~FA_S_ACCESSED;
 498                         write_unlock_bh(&fib_hash_lock);
 499
 500                         fib_release_info(fi_drop);
 501                         if (state & FA_S_ACCESSED)
 502                                 rt_cache_flush(-1);
 503                         return 0;
 504                 }
 505
 506                 /* Error if we find a perfect match which
 507                  * uses the same scope, type, and nexthop
 508                  * information.
 509                  */
 510                 fa_orig = fa;
 511                 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 512                 list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 513                         if (fa->fa_tos != tos)
 514                                 break;
 515                         if (fa->fa_info->fib_priority != fi->fib_priority)
 516                                 break;
 517                         if (fa->fa_type == type &&
 518                             fa->fa_scope == r->rtm_scope &&
 519                             fa->fa_info == fi)
 520                                 goto out;
 521                 }
 522                 if (!(n->nlmsg_flags & NLM_F_APPEND))
 523                         fa = fa_orig;
 524         }
 525
 526         err = -ENOENT;
 527         if (!(n->nlmsg_flags&NLM_F_CREATE))
 528                 goto out;
 529
 530         err = -ENOBUFS;
 531         new_fa = kmem_cache_alloc(fn_alias_kmem, SLAB_KERNEL);
 532         if (new_fa == NULL)
 533                 goto out;
 534
 535         new_f = NULL;
 536         if (!f) {
 537                 new_f = kmem_cache_alloc(fn_hash_kmem, SLAB_KERNEL);
 538                 if (new_f == NULL)
 539                         goto out_free_new_fa;
 540
 541                 INIT_HLIST_NODE(&new_f->fn_hash);
 542                 INIT_LIST_HEAD(&new_f->fn_alias);
 543                 new_f->fn_key = key;
 544                 f = new_f;
 545         }
 546
 547         new_fa->fa_info = fi;
 548         new_fa->fa_tos = tos;
 549         new_fa->fa_type = type;
 550         new_fa->fa_scope = r->rtm_scope;
 551         new_fa->fa_state = 0;
 552
 553         /*
 554          * Insert new entry to the list.
 555          */
 556
 557         write_lock_bh(&fib_hash_lock);
 558         if (new_f)
 559                 fib_insert_node(fz, new_f);
 560         list_add_tail(&new_fa->fa_list,
 561                  (fa ? &fa->fa_list : &f->fn_alias));
 562         write_unlock_bh(&fib_hash_lock);
 563
 564         if (new_f)
 565                 fz->fz_nent++;
 566         rt_cache_flush(-1);
 567
 568         rtmsg_fib(RTM_NEWROUTE, f, new_fa, z, tb->tb_id, n, req);
 569         return 0;
 570
 571 out_free_new_fa:
 572         kmem_cache_free(fn_alias_kmem, new_fa);
 573 out:
 574         fib_release_info(fi);
 575         return err;
 576 }
 577
 578
 579 static int
 580 fn_hash_delete(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
 581                struct nlmsghdr *n, struct netlink_skb_parms *req)
 582 {
 583         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 584         struct fib_node *f;
 585         struct fib_alias *fa, *fa_to_delete;
 586         int z = r->rtm_dst_len;
 587         struct fn_zone *fz;
 588         u32 key;
 589         u8 tos = r->rtm_tos;
 590
 591         if (z > 32)
 592                 return -EINVAL;
 593         if ((fz  = table->fn_zones[z]) == NULL)
 594                 return -ESRCH;
 595
 596         key = 0;
 597         if (rta->rta_dst) {
 598                 u32 dst;
 599                 memcpy(&dst, rta->rta_dst, 4);
 600                 if (dst & ~FZ_MASK(fz))
 601                         return -EINVAL;
 602                 key = fz_key(dst, fz);
 603         }
 604
 605         f = fib_find_node(fz, key);
 606         fa = fib_find_alias(f, tos, 0);
 607         if (!fa)
 608                 return -ESRCH;
 609
 610         fa_to_delete = NULL;
 611         fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
 612         list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
 613                 struct fib_info *fi = fa->fa_info;
 614
 615                 if (fa->fa_tos != tos)
 616                         break;
 617
 618                 if ((!r->rtm_type ||
 619                      fa->fa_type == r->rtm_type) &&
 620                     (r->rtm_scope == RT_SCOPE_NOWHERE ||
 621                      fa->fa_scope == r->rtm_scope) &&
 622                     (!r->rtm_protocol ||
 623                      fi->fib_protocol == r->rtm_protocol) &&
 624                     fib_nh_match(r, n, rta, fi) == 0) {
 625                         fa_to_delete = fa;
 626                         break;
 627                 }
 628         }
 629
 630         if (fa_to_delete) {
 631                 int kill_fn;
 632
 633                 fa = fa_to_delete;
 634                 rtmsg_fib(RTM_DELROUTE, f, fa, z, tb->tb_id, n, req);
 635
 636                 kill_fn = 0;
 637                 write_lock_bh(&fib_hash_lock);
 638                 list_del(&fa->fa_list);
 639                 if (list_empty(&f->fn_alias)) {
 640                         hlist_del(&f->fn_hash);
 641                         kill_fn = 1;
 642                 }
 643                 write_unlock_bh(&fib_hash_lock);
 644
 645                 if (fa->fa_state & FA_S_ACCESSED)
 646                         rt_cache_flush(-1);
 647                 fn_free_alias(fa);
 648                 if (kill_fn) {
 649                         fn_free_node(f);
 650                         fz->fz_nent--;
 651                 }
 652
 653                 return 0;
 654         }
 655         return -ESRCH;
 656 }
 657
 658 static int fn_flush_list(struct fn_zone *fz, int idx)
 659 {
 660         struct hlist_head *head = &fz->fz_hash[idx];
 661         struct hlist_node *node, *n;
 662         struct fib_node *f;
 663         int found = 0;
 664
 665         hlist_for_each_entry_safe(f, node, n, head, fn_hash) {
 666                 struct fib_alias *fa, *fa_node;
 667                 int kill_f;
 668
 669                 kill_f = 0;
 670                 list_for_each_entry_safe(fa, fa_node, &f->fn_alias, fa_list) {
 671                         struct fib_info *fi = fa->fa_info;
 672
 673                         if (fi && (fi->fib_flags&RTNH_F_DEAD)) {
 674                                 write_lock_bh(&fib_hash_lock);
 675                                 list_del(&fa->fa_list);
 676                                 if (list_empty(&f->fn_alias)) {
 677                                         hlist_del(&f->fn_hash);
 678                                         kill_f = 1;
 679                                 }
 680                                 write_unlock_bh(&fib_hash_lock);
 681
 682                                 fn_free_alias(fa);
 683                                 found++;
 684                         }
 685                 }
 686                 if (kill_f) {
 687                         fn_free_node(f);
 688                         fz->fz_nent--;
 689                 }
 690         }
 691         return found;
 692 }
 693
 694 static int fn_hash_flush(struct fib_table *tb)
 695 {
 696         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
 697         struct fn_zone *fz;
 698         int found = 0;
 699
 700         for (fz = table->fn_zone_list; fz; fz = fz->fz_next) {
 701                 int i;
 702
 703                 for (i = fz->fz_divisor - 1; i >= 0; i--)
 704                         found += fn_flush_list(fz, i);
 705         }
 706         return found;
 707 }
 708
 709
 710 static inline int
 711 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
 712                      struct fib_table *tb,
 713                      struct fn_zone *fz,
 714                      struct hlist_head *head)
 715 {
 716         struct hlist_node *node;
 717         struct fib_node *f;
 718         int i, s_i;
 719
 720         s_i = cb->args[3];
 721         i = 0;
 722         hlist_for_each_entry(f, node, head, fn_hash) {
 723                 struct fib_alias *fa;
 724
 725                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
 726                         if (i < s_i)
 727                                 continue;
 728
 729                         if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
 730                                           cb->nlh->nlmsg_seq,
 731                                           RTM_NEWROUTE,
 732                                           tb->tb_id,
 733                                           fa->fa_type,
 734                                           fa->fa_scope,
 735                                           &f->fn_key,
 736                                           fz->fz_order,
 737                                           fa->fa_tos,
 738                                           fa->fa_info) < 0) {
 739                                 cb->args[3] = i;
 740                                 return -1;
 741                         }
 742
 743                         i++;
 744                 }
 745         }
 746         cb->args[3] = i;
 747         return skb->len;
 748 }
 749
 750 static inline int
 751 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
 752                    struct fib_table *tb,
 753                    struct fn_zone *fz)
 754 {
 755         int h, s_h;
 756
 757         s_h = cb->args[2];
 758         for (h=0; h < fz->fz_divisor; h++) {
 759                 if (h < s_h) continue;
 760                 if (h > s_h)
 761                         memset(&cb->args[3], 0,
 762                                sizeof(cb->args) - 3*sizeof(cb->args[0]));
 763                 if (fz->fz_hash == NULL ||
 764                     hlist_empty(&fz->fz_hash[h]))
 765                         continue;
 766                 if (fn_hash_dump_bucket(skb, cb, tb, fz, &fz->fz_hash[h])<0) {
 767                         cb->args[2] = h;
 768                         return -1;
 769                 }
 770         }
 771         cb->args[2] = h;
 772         return skb->len;
 773 }
 774
 775 static int fn_hash_dump(struct fib_table *tb, struct sk_buff *skb, struct netlink_callback *cb)
 776 {
 777         int m, s_m;
 778         struct fn_zone *fz;
 779         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 780
 781         s_m = cb->args[1];
 782         read_lock(&fib_hash_lock);
 783         for (fz = table->fn_zone_list, m=0; fz; fz = fz->fz_next, m++) {
 784                 if (m < s_m) continue;
 785                 if (m > s_m)
 786                         memset(&cb->args[2], 0,
 787                                sizeof(cb->args) - 2*sizeof(cb->args[0]));
 788                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
 789                         cb->args[1] = m;
 790                         read_unlock(&fib_hash_lock);
 791                         return -1;
 792                 }
 793         }
 794         read_unlock(&fib_hash_lock);
 795         cb->args[1] = m;
 796         return skb->len;
 797 }
 798
 799 static void rtmsg_fib(int event, struct fib_node *f, struct fib_alias *fa,
 800                       int z, int tb_id,
 801                       struct nlmsghdr *n, struct netlink_skb_parms *req)
 802 {
 803         struct sk_buff *skb;
 804         u32 pid = req ? req->pid : 0;
 805         int size = NLMSG_SPACE(sizeof(struct rtmsg)+256);
 806
 807         skb = alloc_skb(size, GFP_KERNEL);
 808         if (!skb)
 809                 return;
 810
 811         if (fib_dump_info(skb, pid, n->nlmsg_seq, event, tb_id,
 812                           fa->fa_type, fa->fa_scope, &f->fn_key, z,
 813                           fa->fa_tos,
 814                           fa->fa_info) < 0) {
 815                 kfree_skb(skb);
 816                 return;
 817         }
 818         NETLINK_CB(skb).dst_groups = RTMGRP_IPV4_ROUTE;
 819         if (n->nlmsg_flags&NLM_F_ECHO)
 820                 atomic_inc(&skb->users);
 821         netlink_broadcast(rtnl, skb, pid, RTMGRP_IPV4_ROUTE, GFP_KERNEL);
 822         if (n->nlmsg_flags&NLM_F_ECHO)
 823                 netlink_unicast(rtnl, skb, pid, MSG_DONTWAIT);
 824 }
 825
 826 #ifdef CONFIG_IP_MULTIPLE_TABLES
 827 struct fib_table * fib_hash_init(int id)
 828 #else
 829 struct fib_table * __init fib_hash_init(int id)
 830 #endif
 831 {
 832         struct fib_table *tb;
 833
 834         if (fn_hash_kmem == NULL)
 835                 fn_hash_kmem = kmem_cache_create("ip_fib_hash",
 836                                                  sizeof(struct fib_node),
 837                                                  0, SLAB_HWCACHE_ALIGN,
 838                                                  NULL, NULL);
 839
 840         if (fn_alias_kmem == NULL)
 841                 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
 842                                                   sizeof(struct fib_alias),
 843                                                   0, SLAB_HWCACHE_ALIGN,
 844                                                   NULL, NULL);
 845
 846         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash),
 847                      GFP_KERNEL);
 848         if (tb == NULL)
 849                 return NULL;
 850
 851         tb->tb_id = id;
 852         tb->tb_lookup = fn_hash_lookup;
 853         tb->tb_insert = fn_hash_insert;
 854         tb->tb_delete = fn_hash_delete;
 855         tb->tb_flush = fn_hash_flush;
 856         tb->tb_select_default = fn_hash_select_default;
 857         tb->tb_dump = fn_hash_dump;
 858         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 859         return tb;
 860 }
 861
 862 /* ------------------------------------------------------------------------ */
 863 #ifdef CONFIG_PROC_FS
 864
 865 struct fib_iter_state {
 866         struct fn_zone  *zone;
 867         int             bucket;
 868         struct hlist_head *hash_head;
 869         struct fib_node *fn;
 870         struct fib_alias *fa;
 871 };
 872
 873 static struct fib_alias *fib_get_first(struct seq_file *seq)
 874 {
 875         struct fib_iter_state *iter = seq->private;
 876         struct fn_hash *table = (struct fn_hash *) ip_fib_main_table->tb_data;
 877
 878         iter->bucket    = 0;
 879         iter->hash_head = NULL;
 880         iter->fn        = NULL;
 881         iter->fa        = NULL;
 882
 883         for (iter->zone = table->fn_zone_list; iter->zone;
 884              iter->zone = iter->zone->fz_next) {
 885                 int maxslot;
 886
 887                 if (!iter->zone->fz_nent)
 888                         continue;
 889
 890                 iter->hash_head = iter->zone->fz_hash;
 891                 maxslot = iter->zone->fz_divisor;
 892
 893                 for (iter->bucket = 0; iter->bucket < maxslot;
 894                      ++iter->bucket, ++iter->hash_head) {
 895                         struct hlist_node *node;
 896                         struct fib_node *fn;
 897
 898                         hlist_for_each_entry(fn,node,iter->hash_head,fn_hash) {
 899                                 struct fib_alias *fa;
 900
 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 out:
 910         return iter->fa;
 911 }
 912
 913 static struct fib_alias *fib_get_next(struct seq_file *seq)
 914 {
 915         struct fib_iter_state *iter = seq->private;
 916         struct fib_node *fn;
 917         struct fib_alias *fa;
 918
 919         /* Advance FA, if any. */
 920         fn = iter->fn;
 921         fa = iter->fa;
 922         if (fa) {
 923                 BUG_ON(!fn);
 924                 list_for_each_entry_continue(fa, &fn->fn_alias, fa_list) {
 925                         iter->fa = fa;
 926                         goto out;
 927                 }
 928         }
 929
 930         fa = iter->fa = NULL;
 931
 932         /* Advance FN. */
 933         if (fn) {
 934                 struct hlist_node *node = &fn->fn_hash;
 935                 hlist_for_each_entry_continue(fn, node, fn_hash) {
 936                         iter->fn = fn;
 937
 938                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 939                                 iter->fa = fa;
 940                                 goto out;
 941                         }
 942                 }
 943         }
 944
 945         fn = iter->fn = NULL;
 946
 947         /* Advance hash chain. */
 948         if (!iter->zone)
 949                 goto out;
 950
 951         for (;;) {
 952                 struct hlist_node *node;
 953                 int maxslot;
 954
 955                 maxslot = iter->zone->fz_divisor;
 956
 957                 while (++iter->bucket < maxslot) {
 958                         iter->hash_head++;
 959
 960                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 961                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 962                                         iter->fn = fn;
 963                                         iter->fa = fa;
 964                                         goto out;
 965                                 }
 966                         }
 967                 }
 968
 969                 iter->zone = iter->zone->fz_next;
 970
 971                 if (!iter->zone)
 972                         goto out;
 973
 974                 iter->bucket = 0;
 975                 iter->hash_head = iter->zone->fz_hash;
 976
 977                 hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
 978                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
 979                                 iter->fn = fn;
 980                                 iter->fa = fa;
 981                                 goto out;
 982                         }
 983                 }
 984         }
 985 out:
 986         return fa;
 987 }
 988
 989 static void *fib_seq_start(struct seq_file *seq, loff_t *pos)
 990 {
 991         void *v = NULL;
 992
 993         read_lock(&fib_hash_lock);
 994         if (ip_fib_main_table)
 995                 v = *pos ? fib_get_next(seq) : SEQ_START_TOKEN;
 996         return v;
 997 }
 998
 999 static void *fib_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1000 {
1001         ++*pos;
1002         return v == SEQ_START_TOKEN ? fib_get_first(seq) : fib_get_next(seq);
1003 }
1004
1005 static void fib_seq_stop(struct seq_file *seq, void *v)
1006 {
1007         read_unlock(&fib_hash_lock);
1008 }
1009
1010 static unsigned fib_flag_trans(int type, u32 mask, struct fib_info *fi)
1011 {
1012         static unsigned type2flags[RTN_MAX + 1] = {
1013                 [7] = RTF_REJECT, [8] = RTF_REJECT,
1014         };
1015         unsigned flags = type2flags[type];
1016
1017         if (fi && fi->fib_nh->nh_gw)
1018                 flags |= RTF_GATEWAY;
1019         if (mask == 0xFFFFFFFF)
1020                 flags |= RTF_HOST;
1021         flags |= RTF_UP;
1022         return flags;
1023 }
1024
1025 extern int dev_in_nx_info(struct net_device *, struct nx_info *);
1026
1027 /*
1028  *      This outputs /proc/net/route.
1029  *
1030  *      It always works in backward compatibility mode.
1031  *      The format of the file is not supposed to be changed.
1032  */
1033 static int fib_seq_show(struct seq_file *seq, void *v)
1034 {
1035         struct fib_iter_state *iter;
1036         char bf[128];
1037         u32 prefix, mask;
1038         unsigned flags;
1039         struct fib_node *f;
1040         struct fib_alias *fa;
1041         struct fib_info *fi;
1042
1043         if (v == SEQ_START_TOKEN) {
1044                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
1045                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
1046                            "\tWindow\tIRTT");
1047                 goto out;
1048         }
1049
1050         iter    = seq->private;
1051         f       = iter->fn;
1052         fa      = iter->fa;
1053         fi      = fa->fa_info;
1054         prefix  = f->fn_key;
1055         mask    = FZ_MASK(iter->zone);
1056         flags   = fib_flag_trans(fa->fa_type, mask, fi);
1057         if (fi && (!vx_flags(VXF_HIDE_NETIF, 0) ||
1058                 dev_in_nx_info(fi->fib_dev, current->nx_info)))
1059                 snprintf(bf, sizeof(bf),
1060                          "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1061                          fi->fib_dev ? fi->fib_dev->name : "*", prefix,
1062                          fi->fib_nh->nh_gw, flags, 0, 0, fi->fib_priority,
1063                          mask, (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
1064                          fi->fib_window,
1065                          fi->fib_rtt >> 3);
1066         else
1067                 snprintf(bf, sizeof(bf),
1068                          "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1069                          prefix, 0, flags, 0, 0, 0, mask, 0, 0, 0);
1070         seq_printf(seq, "%-127s\n", bf);
1071 out:
1072         return 0;
1073 }
1074
1075 static struct seq_operations fib_seq_ops = {
1076         .start  = fib_seq_start,
1077         .next   = fib_seq_next,
1078         .stop   = fib_seq_stop,
1079         .show   = fib_seq_show,
1080 };
1081
1082 static int fib_seq_open(struct inode *inode, struct file *file)
1083 {
1084         struct seq_file *seq;
1085         int rc = -ENOMEM;
1086         struct fib_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
1087
1088         if (!s)
1089                 goto out;
1090
1091         rc = seq_open(file, &fib_seq_ops);
1092         if (rc)
1093                 goto out_kfree;
1094
1095         seq          = file->private_data;
1096         seq->private = s;
1097         memset(s, 0, sizeof(*s));
1098 out:
1099         return rc;
1100 out_kfree:
1101         kfree(s);
1102         goto out;
1103 }
1104
1105 static struct file_operations fib_seq_fops = {
1106         .owner          = THIS_MODULE,
1107         .open           = fib_seq_open,
1108         .read           = seq_read,
1109         .llseek         = seq_lseek,
1110         .release        = seq_release_private,
1111 };
1112
1113 int __init fib_proc_init(void)
1114 {
1115         if (!proc_net_fops_create("route", S_IRUGO, &fib_seq_fops))
1116                 return -ENOMEM;
1117         return 0;
1118 }
1119
1120 void __init fib_proc_exit(void)
1121 {
1122         proc_net_remove("route");
1123 }
1124 #endif /* CONFIG_PROC_FS */