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 1.1.1.1 2005/04/11 02:51:13 jack 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 #define FTprint(a...)
  47 /*
  48    printk(KERN_DEBUG a)
  49  */
  50
  51 static kmem_cache_t * fn_hash_kmem;
  52
  53 /*
  54    These bizarre types are just to force strict type checking.
  55    When I reversed order of bytes and changed to natural mask lengths,
  56    I forgot to make fixes in several places. Now I am lazy to return
  57    it back.
  58  */
  59
  60 typedef struct {
  61         u32     datum;
  62 } fn_key_t;
  63
  64 typedef struct {
  65         u32     datum;
  66 } fn_hash_idx_t;
  67
  68 struct fib_node
  69 {
  70         struct fib_node         *fn_next;
  71         struct fib_info         *fn_info;
  72 #define FIB_INFO(f)     ((f)->fn_info)
  73         fn_key_t                fn_key;
  74         u8                      fn_tos;
  75         u8                      fn_type;
  76         u8                      fn_scope;
  77         u8                      fn_state;
  78 };
  79
  80 #define FN_S_ZOMBIE     1
  81 #define FN_S_ACCESSED   2
  82
  83 static int fib_hash_zombies;
  84
  85 struct fn_zone
  86 {
  87         struct fn_zone  *fz_next;       /* Next not empty zone  */
  88         struct fib_node **fz_hash;      /* Hash table pointer   */
  89         int             fz_nent;        /* Number of entries    */
  90
  91         int             fz_divisor;     /* Hash divisor         */
  92         u32             fz_hashmask;    /* (1<<fz_divisor) - 1  */
  93 #define FZ_HASHMASK(fz) ((fz)->fz_hashmask)
  94
  95         int             fz_order;       /* Zone order           */
  96         u32             fz_mask;
  97 #define FZ_MASK(fz)     ((fz)->fz_mask)
  98 };
  99
 100 /* NOTE. On fast computers evaluation of fz_hashmask and fz_mask
 101    can be cheaper than memory lookup, so that FZ_* macros are used.
 102  */
 103
 104 struct fn_hash
 105 {
 106         struct fn_zone  *fn_zones[33];
 107         struct fn_zone  *fn_zone_list;
 108 };
 109
 110 static __inline__ fn_hash_idx_t fn_hash(fn_key_t key, struct fn_zone *fz)
 111 {
 112         u32 h = ntohl(key.datum)>>(32 - fz->fz_order);
 113         h ^= (h>>20);
 114         h ^= (h>>10);
 115         h ^= (h>>5);
 116         h &= FZ_HASHMASK(fz);
 117         return *(fn_hash_idx_t*)&h;
 118 }
 119
 120 #define fz_key_0(key)           ((key).datum = 0)
 121 #define fz_prefix(key,fz)       ((key).datum)
 122
 123 static __inline__ fn_key_t fz_key(u32 dst, struct fn_zone *fz)
 124 {
 125         fn_key_t k;
 126         k.datum = dst & FZ_MASK(fz);
 127         return k;
 128 }
 129
 130 static __inline__ struct fib_node ** fz_chain_p(fn_key_t key, struct fn_zone *fz)
 131 {
 132         return &fz->fz_hash[fn_hash(key, fz).datum];
 133 }
 134
 135 static __inline__ struct fib_node * fz_chain(fn_key_t key, struct fn_zone *fz)
 136 {
 137         return fz->fz_hash[fn_hash(key, fz).datum];
 138 }
 139
 140 extern __inline__ int fn_key_eq(fn_key_t a, fn_key_t b)
 141 {
 142         return a.datum == b.datum;
 143 }
 144
 145 extern __inline__ int fn_key_leq(fn_key_t a, fn_key_t b)
 146 {
 147         return a.datum <= b.datum;
 148 }
 149
 150 static rwlock_t fib_hash_lock = RW_LOCK_UNLOCKED;
 151
 152 #define FZ_MAX_DIVISOR 1024
 153
 154 #ifdef CONFIG_IP_ROUTE_LARGE_TABLES
 155
 156 /* The fib hash lock must be held when this is called. */
 157 static __inline__ void fn_rebuild_zone(struct fn_zone *fz,
 158                                        struct fib_node **old_ht,
 159                                        int old_divisor)
 160 {
 161         int i;
 162         struct fib_node *f, **fp, *next;
 163
 164         for (i=0; i<old_divisor; i++) {
 165                 for (f=old_ht[i]; f; f=next) {
 166                         next = f->fn_next;
 167                         for (fp = fz_chain_p(f->fn_key, fz);
 168                              *fp && fn_key_leq((*fp)->fn_key, f->fn_key);
 169                              fp = &(*fp)->fn_next)
 170                                 /* NONE */;
 171                         f->fn_next = *fp;
 172                         *fp = f;
 173                 }
 174         }
 175 }
 176
 177 static void fn_rehash_zone(struct fn_zone *fz)
 178 {
 179         struct fib_node **ht, **old_ht;
 180         int old_divisor, new_divisor;
 181         u32 new_hashmask;
 182
 183         old_divisor = fz->fz_divisor;
 184
 185         switch (old_divisor) {
 186         case 16:
 187                 new_divisor = 256;
 188                 new_hashmask = 0xFF;
 189                 break;
 190         case 256:
 191                 new_divisor = 1024;
 192                 new_hashmask = 0x3FF;
 193                 break;
 194         default:
 195                 printk(KERN_CRIT "route.c: bad divisor %d!\n", old_divisor);
 196                 return;
 197         }
 198 #if RT_CACHE_DEBUG >= 2
 199         printk("fn_rehash_zone: hash for zone %d grows from %d\n", fz->fz_order, old_divisor);
 200 #endif
 201
 202         ht = kmalloc(new_divisor*sizeof(struct fib_node*), GFP_KERNEL);
 203
 204         if (ht) {
 205                 memset(ht, 0, new_divisor*sizeof(struct fib_node*));
 206                 write_lock_bh(&fib_hash_lock);
 207                 old_ht = fz->fz_hash;
 208                 fz->fz_hash = ht;
 209                 fz->fz_hashmask = new_hashmask;
 210                 fz->fz_divisor = new_divisor;
 211                 fn_rebuild_zone(fz, old_ht, old_divisor);
 212                 write_unlock_bh(&fib_hash_lock);
 213                 kfree(old_ht);
 214         }
 215 }
 216 #endif /* CONFIG_IP_ROUTE_LARGE_TABLES */
 217
 218 static void fn_free_node(struct fib_node * f)
 219 {
 220         fib_release_info(FIB_INFO(f));
 221         kmem_cache_free(fn_hash_kmem, f);
 222 }
 223
 224
 225 static struct fn_zone *
 226 fn_new_zone(struct fn_hash *table, int z)
 227 {
 228         int i;
 229         struct fn_zone *fz = kmalloc(sizeof(struct fn_zone), GFP_KERNEL);
 230         if (!fz)
 231                 return NULL;
 232
 233         memset(fz, 0, sizeof(struct fn_zone));
 234         if (z) {
 235                 fz->fz_divisor = 16;
 236                 fz->fz_hashmask = 0xF;
 237         } else {
 238                 fz->fz_divisor = 1;
 239                 fz->fz_hashmask = 0;
 240         }
 241         fz->fz_hash = kmalloc(fz->fz_divisor*sizeof(struct fib_node*), GFP_KERNEL);
 242         if (!fz->fz_hash) {
 243                 kfree(fz);
 244                 return NULL;
 245         }
 246         memset(fz->fz_hash, 0, fz->fz_divisor*sizeof(struct fib_node*));
 247         fz->fz_order = z;
 248         fz->fz_mask = inet_make_mask(z);
 249
 250         /* Find the first not empty zone with more specific mask */
 251         for (i=z+1; i<=32; i++)
 252                 if (table->fn_zones[i])
 253                         break;
 254         write_lock_bh(&fib_hash_lock);
 255         if (i>32) {
 256                 /* No more specific masks, we are the first. */
 257                 fz->fz_next = table->fn_zone_list;
 258                 table->fn_zone_list = fz;
 259         } else {
 260                 fz->fz_next = table->fn_zones[i]->fz_next;
 261                 table->fn_zones[i]->fz_next = fz;
 262         }
 263         table->fn_zones[z] = fz;
 264         write_unlock_bh(&fib_hash_lock);
 265         return fz;
 266 }
 267
 268 static int
 269 fn_hash_lookup(struct fib_table *tb, const struct rt_key *key, struct fib_result *res)
 270 {
 271         int err;
 272         struct fn_zone *fz;
 273         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 274
 275         read_lock(&fib_hash_lock);
 276         for (fz = t->fn_zone_list; fz; fz = fz->fz_next) {
 277                 struct fib_node *f;
 278                 fn_key_t k = fz_key(key->dst, fz);
 279
 280                 for (f = fz_chain(k, fz); f; f = f->fn_next) {
 281                         if (!fn_key_eq(k, f->fn_key)) {
 282                                 if (fn_key_leq(k, f->fn_key))
 283                                         break;
 284                                 else
 285                                         continue;
 286                         }
 287 #ifdef CONFIG_IP_ROUTE_TOS
 288                         if (f->fn_tos && f->fn_tos != key->tos)
 289                                 continue;
 290 #endif
 291                         f->fn_state |= FN_S_ACCESSED;
 292
 293                         if (f->fn_state&FN_S_ZOMBIE)
 294                                 continue;
 295                         if (f->fn_scope < key->scope)
 296                                 continue;
 297
 298                         err = fib_semantic_match(f->fn_type, FIB_INFO(f), key, res);
 299                         if (err == 0) {
 300                                 res->type = f->fn_type;
 301                                 res->scope = f->fn_scope;
 302                                 res->prefixlen = fz->fz_order;
 303                                 goto out;
 304                         }
 305                         if (err < 0)
 306                                 goto out;
 307                 }
 308         }
 309         err = 1;
 310 out:
 311         read_unlock(&fib_hash_lock);
 312         return err;
 313 }
 314
 315 static int fn_hash_last_dflt=-1;
 316
 317 static int fib_detect_death(struct fib_info *fi, int order,
 318                             struct fib_info **last_resort, int *last_idx)
 319 {
 320         struct neighbour *n;
 321         int state = NUD_NONE;
 322
 323         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 324         if (n) {
 325                 state = n->nud_state;
 326                 neigh_release(n);
 327         }
 328         if (state==NUD_REACHABLE)
 329                 return 0;
 330         if ((state&NUD_VALID) && order != fn_hash_last_dflt)
 331                 return 0;
 332         if ((state&NUD_VALID) ||
 333             (*last_idx<0 && order > fn_hash_last_dflt)) {
 334                 *last_resort = fi;
 335                 *last_idx = order;
 336         }
 337         return 1;
 338 }
 339
 340 static void
 341 fn_hash_select_default(struct fib_table *tb, const struct rt_key *key, struct fib_result *res)
 342 {
 343         int order, last_idx;
 344         struct fib_node *f;
 345         struct fib_info *fi = NULL;
 346         struct fib_info *last_resort;
 347         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 348         struct fn_zone *fz = t->fn_zones[0];
 349
 350         if (fz == NULL)
 351                 return;
 352
 353         last_idx = -1;
 354         last_resort = NULL;
 355         order = -1;
 356
 357         read_lock(&fib_hash_lock);
 358         for (f = fz->fz_hash[0]; f; f = f->fn_next) {
 359                 struct fib_info *next_fi = FIB_INFO(f);
 360
 361                 if ((f->fn_state&FN_S_ZOMBIE) ||
 362                     f->fn_scope != res->scope ||
 363                     f->fn_type != RTN_UNICAST)
 364                         continue;
 365
 366                 if (next_fi->fib_priority > res->fi->fib_priority)
 367                         break;
 368                 if (!next_fi->fib_nh[0].nh_gw || next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
 369                         continue;
 370                 f->fn_state |= FN_S_ACCESSED;
 371
 372                 if (fi == NULL) {
 373                         if (next_fi != res->fi)
 374                                 break;
 375                 } else if (!fib_detect_death(fi, order, &last_resort, &last_idx)) {
 376                         if (res->fi)
 377                                 fib_info_put(res->fi);
 378                         res->fi = fi;
 379                         atomic_inc(&fi->fib_clntref);
 380                         fn_hash_last_dflt = order;
 381                         goto out;
 382                 }
 383                 fi = next_fi;
 384                 order++;
 385         }
 386
 387         if (order<=0 || fi==NULL) {
 388                 fn_hash_last_dflt = -1;
 389                 goto out;
 390         }
 391
 392         if (!fib_detect_death(fi, order, &last_resort, &last_idx)) {
 393                 if (res->fi)
 394                         fib_info_put(res->fi);
 395                 res->fi = fi;
 396                 atomic_inc(&fi->fib_clntref);
 397                 fn_hash_last_dflt = order;
 398                 goto out;
 399         }
 400
 401         if (last_idx >= 0) {
 402                 if (res->fi)
 403                         fib_info_put(res->fi);
 404                 res->fi = last_resort;
 405                 if (last_resort)
 406                         atomic_inc(&last_resort->fib_clntref);
 407         }
 408         fn_hash_last_dflt = last_idx;
 409 out:
 410         read_unlock(&fib_hash_lock);
 411 }
 412
 413 #define FIB_SCAN(f, fp) \
 414 for ( ; ((f) = *(fp)) != NULL; (fp) = &(f)->fn_next)
 415
 416 #define FIB_SCAN_KEY(f, fp, key) \
 417 for ( ; ((f) = *(fp)) != NULL && fn_key_eq((f)->fn_key, (key)); (fp) = &(f)->fn_next)
 418
 419 #ifndef CONFIG_IP_ROUTE_TOS
 420 #define FIB_SCAN_TOS(f, fp, key, tos) FIB_SCAN_KEY(f, fp, key)
 421 #else
 422 #define FIB_SCAN_TOS(f, fp, key, tos) \
 423 for ( ; ((f) = *(fp)) != NULL && fn_key_eq((f)->fn_key, (key)) && \
 424      (f)->fn_tos == (tos) ; (fp) = &(f)->fn_next)
 425 #endif
 426
 427
 428 static void rtmsg_fib(int, struct fib_node*, int, int,
 429                       struct nlmsghdr *n,
 430                       struct netlink_skb_parms *);
 431
 432 static int
 433 fn_hash_insert(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
 434                 struct nlmsghdr *n, struct netlink_skb_parms *req)
 435 {
 436         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 437         struct fib_node *new_f, *f, **fp, **del_fp;
 438         struct fn_zone *fz;
 439         struct fib_info *fi;
 440
 441         int z = r->rtm_dst_len;
 442         int type = r->rtm_type;
 443 #ifdef CONFIG_IP_ROUTE_TOS
 444         u8 tos = r->rtm_tos;
 445 #endif
 446         fn_key_t key;
 447         int err;
 448
 449 FTprint("tb(%d)_insert: %d %08x/%d %d %08x\n", tb->tb_id, r->rtm_type, rta->rta_dst ?
 450 *(u32*)rta->rta_dst : 0, z, rta->rta_oif ? *rta->rta_oif : -1,
 451 rta->rta_prefsrc ? *(u32*)rta->rta_prefsrc : 0);
 452         if (z > 32)
 453                 return -EINVAL;
 454         fz = table->fn_zones[z];
 455         if (!fz && !(fz = fn_new_zone(table, z)))
 456                 return -ENOBUFS;
 457
 458         fz_key_0(key);
 459         if (rta->rta_dst) {
 460                 u32 dst;
 461                 memcpy(&dst, rta->rta_dst, 4);
 462                 if (dst & ~FZ_MASK(fz))
 463                         return -EINVAL;
 464                 key = fz_key(dst, fz);
 465         }
 466
 467         if  ((fi = fib_create_info(r, rta, n, &err)) == NULL)
 468                 return err;
 469
 470 #ifdef CONFIG_IP_ROUTE_LARGE_TABLES
 471         if (fz->fz_nent > (fz->fz_divisor<<2) &&
 472             fz->fz_divisor < FZ_MAX_DIVISOR &&
 473             (z==32 || (1<<z) > fz->fz_divisor))
 474                 fn_rehash_zone(fz);
 475 #endif
 476
 477         fp = fz_chain_p(key, fz);
 478
 479
 480         /*
 481          * Scan list to find the first route with the same destination
 482          */
 483         FIB_SCAN(f, fp) {
 484                 if (fn_key_leq(key,f->fn_key))
 485                         break;
 486         }
 487
 488 #ifdef CONFIG_IP_ROUTE_TOS
 489         /*
 490          * Find route with the same destination and tos.
 491          */
 492         FIB_SCAN_KEY(f, fp, key) {
 493                 if (f->fn_tos <= tos)
 494                         break;
 495         }
 496 #endif
 497
 498         del_fp = NULL;
 499
 500         if (f && (f->fn_state&FN_S_ZOMBIE) &&
 501 #ifdef CONFIG_IP_ROUTE_TOS
 502             f->fn_tos == tos &&
 503 #endif
 504             fn_key_eq(f->fn_key, key)) {
 505                 del_fp = fp;
 506                 fp = &f->fn_next;
 507                 f = *fp;
 508                 goto create;
 509         }
 510
 511         FIB_SCAN_TOS(f, fp, key, tos) {
 512                 if (fi->fib_priority <= FIB_INFO(f)->fib_priority)
 513                         break;
 514         }
 515
 516         /* Now f==*fp points to the first node with the same
 517            keys [prefix,tos,priority], if such key already
 518            exists or to the node, before which we will insert new one.
 519          */
 520
 521         if (f &&
 522 #ifdef CONFIG_IP_ROUTE_TOS
 523             f->fn_tos == tos &&
 524 #endif
 525             fn_key_eq(f->fn_key, key) &&
 526             fi->fib_priority == FIB_INFO(f)->fib_priority) {
 527                 struct fib_node **ins_fp;
 528
 529                 err = -EEXIST;
 530                 if (n->nlmsg_flags&NLM_F_EXCL)
 531                         goto out;
 532
 533                 if (n->nlmsg_flags&NLM_F_REPLACE) {
 534                         del_fp = fp;
 535                         fp = &f->fn_next;
 536                         f = *fp;
 537                         goto replace;
 538                 }
 539
 540                 ins_fp = fp;
 541                 err = -EEXIST;
 542
 543                 FIB_SCAN_TOS(f, fp, key, tos) {
 544                         if (fi->fib_priority != FIB_INFO(f)->fib_priority)
 545                                 break;
 546                         if (f->fn_type == type && f->fn_scope == r->rtm_scope
 547                             && FIB_INFO(f) == fi)
 548                                 goto out;
 549                 }
 550
 551                 if (!(n->nlmsg_flags&NLM_F_APPEND)) {
 552                         fp = ins_fp;
 553                         f = *fp;
 554                 }
 555         }
 556
 557 create:
 558         err = -ENOENT;
 559         if (!(n->nlmsg_flags&NLM_F_CREATE))
 560                 goto out;
 561
 562 replace:
 563         err = -ENOBUFS;
 564         new_f = kmem_cache_alloc(fn_hash_kmem, SLAB_KERNEL);
 565         if (new_f == NULL)
 566                 goto out;
 567
 568         memset(new_f, 0, sizeof(struct fib_node));
 569
 570         new_f->fn_key = key;
 571 #ifdef CONFIG_IP_ROUTE_TOS
 572         new_f->fn_tos = tos;
 573 #endif
 574         new_f->fn_type = type;
 575         new_f->fn_scope = r->rtm_scope;
 576         FIB_INFO(new_f) = fi;
 577
 578         /*
 579          * Insert new entry to the list.
 580          */
 581
 582         new_f->fn_next = f;
 583         write_lock_bh(&fib_hash_lock);
 584         *fp = new_f;
 585         write_unlock_bh(&fib_hash_lock);
 586         fz->fz_nent++;
 587
 588         if (del_fp) {
 589                 f = *del_fp;
 590                 /* Unlink replaced node */
 591                 write_lock_bh(&fib_hash_lock);
 592                 *del_fp = f->fn_next;
 593                 write_unlock_bh(&fib_hash_lock);
 594
 595                 if (!(f->fn_state&FN_S_ZOMBIE))
 596                         rtmsg_fib(RTM_DELROUTE, f, z, tb->tb_id, n, req);
 597                 if (f->fn_state&FN_S_ACCESSED)
 598                         rt_cache_flush(-1);
 599                 fn_free_node(f);
 600                 fz->fz_nent--;
 601         } else {
 602                 rt_cache_flush(-1);
 603         }
 604         rtmsg_fib(RTM_NEWROUTE, new_f, z, tb->tb_id, n, req);
 605         return 0;
 606
 607 out:
 608         fib_release_info(fi);
 609         return err;
 610 }
 611
 612
 613 static int
 614 fn_hash_delete(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
 615                 struct nlmsghdr *n, struct netlink_skb_parms *req)
 616 {
 617         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 618         struct fib_node **fp, **del_fp, *f;
 619         int z = r->rtm_dst_len;
 620         struct fn_zone *fz;
 621         fn_key_t key;
 622         int matched;
 623 #ifdef CONFIG_IP_ROUTE_TOS
 624         u8 tos = r->rtm_tos;
 625 #endif
 626
 627 FTprint("tb(%d)_delete: %d %08x/%d %d\n", tb->tb_id, r->rtm_type, rta->rta_dst ?
 628        *(u32*)rta->rta_dst : 0, z, rta->rta_oif ? *rta->rta_oif : -1);
 629         if (z > 32)
 630                 return -EINVAL;
 631         if ((fz  = table->fn_zones[z]) == NULL)
 632                 return -ESRCH;
 633
 634         fz_key_0(key);
 635         if (rta->rta_dst) {
 636                 u32 dst;
 637                 memcpy(&dst, rta->rta_dst, 4);
 638                 if (dst & ~FZ_MASK(fz))
 639                         return -EINVAL;
 640                 key = fz_key(dst, fz);
 641         }
 642
 643         fp = fz_chain_p(key, fz);
 644
 645
 646         FIB_SCAN(f, fp) {
 647                 if (fn_key_eq(f->fn_key, key))
 648                         break;
 649                 if (fn_key_leq(key, f->fn_key)) {
 650                         return -ESRCH;
 651                 }
 652         }
 653 #ifdef CONFIG_IP_ROUTE_TOS
 654         FIB_SCAN_KEY(f, fp, key) {
 655                 if (f->fn_tos == tos)
 656                         break;
 657         }
 658 #endif
 659
 660         matched = 0;
 661         del_fp = NULL;
 662         FIB_SCAN_TOS(f, fp, key, tos) {
 663                 struct fib_info * fi = FIB_INFO(f);
 664
 665                 if (f->fn_state&FN_S_ZOMBIE) {
 666                         return -ESRCH;
 667                 }
 668                 matched++;
 669
 670                 if (del_fp == NULL &&
 671                     (!r->rtm_type || f->fn_type == r->rtm_type) &&
 672                     (r->rtm_scope == RT_SCOPE_NOWHERE || f->fn_scope == r->rtm_scope) &&
 673                     (!r->rtm_protocol || fi->fib_protocol == r->rtm_protocol) &&
 674                     fib_nh_match(r, n, rta, fi) == 0)
 675                         del_fp = fp;
 676         }
 677
 678         if (del_fp) {
 679                 f = *del_fp;
 680                 rtmsg_fib(RTM_DELROUTE, f, z, tb->tb_id, n, req);
 681
 682                 if (matched != 1) {
 683                         write_lock_bh(&fib_hash_lock);
 684                         *del_fp = f->fn_next;
 685                         write_unlock_bh(&fib_hash_lock);
 686
 687                         if (f->fn_state&FN_S_ACCESSED)
 688                                 rt_cache_flush(-1);
 689                         fn_free_node(f);
 690                         fz->fz_nent--;
 691                 } else {
 692                         f->fn_state |= FN_S_ZOMBIE;
 693                         if (f->fn_state&FN_S_ACCESSED) {
 694                                 f->fn_state &= ~FN_S_ACCESSED;
 695                                 rt_cache_flush(-1);
 696                         }
 697                         if (++fib_hash_zombies > 128)
 698                                 fib_flush();
 699                 }
 700
 701                 return 0;
 702         }
 703         return -ESRCH;
 704 }
 705
 706 extern __inline__ int
 707 fn_flush_list(struct fib_node ** fp, int z, struct fn_hash *table)
 708 {
 709         int found = 0;
 710         struct fib_node *f;
 711
 712         while ((f = *fp) != NULL) {
 713                 struct fib_info *fi = FIB_INFO(f);
 714
 715                 if (fi && ((f->fn_state&FN_S_ZOMBIE) || (fi->fib_flags&RTNH_F_DEAD))) {
 716                         write_lock_bh(&fib_hash_lock);
 717                         *fp = f->fn_next;
 718                         write_unlock_bh(&fib_hash_lock);
 719
 720                         fn_free_node(f);
 721                         found++;
 722                         continue;
 723                 }
 724                 fp = &f->fn_next;
 725         }
 726         return found;
 727 }
 728
 729 static int fn_hash_flush(struct fib_table *tb)
 730 {
 731         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 732         struct fn_zone *fz;
 733         int found = 0;
 734
 735         fib_hash_zombies = 0;
 736         for (fz = table->fn_zone_list; fz; fz = fz->fz_next) {
 737                 int i;
 738                 int tmp = 0;
 739                 for (i=fz->fz_divisor-1; i>=0; i--)
 740                         tmp += fn_flush_list(&fz->fz_hash[i], fz->fz_order, table);
 741                 fz->fz_nent -= tmp;
 742                 found += tmp;
 743         }
 744         return found;
 745 }
 746
 747
 748 #ifdef CONFIG_PROC_FS
 749
 750 static int fn_hash_get_info(struct fib_table *tb, char *buffer, int first, int count)
 751 {
 752         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 753         struct fn_zone *fz;
 754         int pos = 0;
 755         int n = 0;
 756
 757         read_lock(&fib_hash_lock);
 758         for (fz=table->fn_zone_list; fz; fz = fz->fz_next) {
 759                 int i;
 760                 struct fib_node *f;
 761                 int maxslot = fz->fz_divisor;
 762                 struct fib_node **fp = fz->fz_hash;
 763
 764                 if (fz->fz_nent == 0)
 765                         continue;
 766
 767                 if (pos + fz->fz_nent <= first) {
 768                         pos += fz->fz_nent;
 769                         continue;
 770                 }
 771
 772                 for (i=0; i < maxslot; i++, fp++) {
 773                         for (f = *fp; f; f = f->fn_next) {
 774                                 if (++pos <= first)
 775                                         continue;
 776                                 fib_node_get_info(f->fn_type,
 777                                                   f->fn_state&FN_S_ZOMBIE,
 778                                                   FIB_INFO(f),
 779                                                   fz_prefix(f->fn_key, fz),
 780                                                   FZ_MASK(fz), buffer);
 781                                 buffer += 128;
 782                                 if (++n >= count)
 783                                         goto out;
 784                         }
 785                 }
 786         }
 787 out:
 788         read_unlock(&fib_hash_lock);
 789         return n;
 790 }
 791 #endif
 792
 793
 794 static __inline__ int
 795 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
 796                      struct fib_table *tb,
 797                      struct fn_zone *fz,
 798                      struct fib_node *f)
 799 {
 800         int i, s_i;
 801
 802         s_i = cb->args[3];
 803         for (i=0; f; i++, f=f->fn_next) {
 804                 if (i < s_i) continue;
 805                 if (f->fn_state&FN_S_ZOMBIE) continue;
 806                 if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid, cb->nlh->nlmsg_seq,
 807                                   RTM_NEWROUTE,
 808                                   tb->tb_id, (f->fn_state&FN_S_ZOMBIE) ? 0 : f->fn_type, f->fn_scope,
 809                                   &f->fn_key, fz->fz_order, f->fn_tos,
 810                                   f->fn_info) < 0) {
 811                         cb->args[3] = i;
 812                         return -1;
 813                 }
 814         }
 815         cb->args[3] = i;
 816         return skb->len;
 817 }
 818
 819 static __inline__ int
 820 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
 821                    struct fib_table *tb,
 822                    struct fn_zone *fz)
 823 {
 824         int h, s_h;
 825
 826         s_h = cb->args[2];
 827         for (h=0; h < fz->fz_divisor; h++) {
 828                 if (h < s_h) continue;
 829                 if (h > s_h)
 830                         memset(&cb->args[3], 0, sizeof(cb->args) - 3*sizeof(cb->args[0]));
 831                 if (fz->fz_hash == NULL || fz->fz_hash[h] == NULL)
 832                         continue;
 833                 if (fn_hash_dump_bucket(skb, cb, tb, fz, fz->fz_hash[h]) < 0) {
 834                         cb->args[2] = h;
 835                         return -1;
 836                 }
 837         }
 838         cb->args[2] = h;
 839         return skb->len;
 840 }
 841
 842 static int fn_hash_dump(struct fib_table *tb, struct sk_buff *skb, struct netlink_callback *cb)
 843 {
 844         int m, s_m;
 845         struct fn_zone *fz;
 846         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 847
 848         s_m = cb->args[1];
 849         read_lock(&fib_hash_lock);
 850         for (fz = table->fn_zone_list, m=0; fz; fz = fz->fz_next, m++) {
 851                 if (m < s_m) continue;
 852                 if (m > s_m)
 853                         memset(&cb->args[2], 0, sizeof(cb->args) - 2*sizeof(cb->args[0]));
 854                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
 855                         cb->args[1] = m;
 856                         read_unlock(&fib_hash_lock);
 857                         return -1;
 858                 }
 859         }
 860         read_unlock(&fib_hash_lock);
 861         cb->args[1] = m;
 862         return skb->len;
 863 }
 864
 865 static void rtmsg_fib(int event, struct fib_node* f, int z, int tb_id,
 866                       struct nlmsghdr *n, struct netlink_skb_parms *req)
 867 {
 868         struct sk_buff *skb;
 869         u32 pid = req ? req->pid : 0;
 870         int size = NLMSG_SPACE(sizeof(struct rtmsg)+256);
 871
 872         skb = alloc_skb(size, GFP_KERNEL);
 873         if (!skb)
 874                 return;
 875
 876         if (fib_dump_info(skb, pid, n->nlmsg_seq, event, tb_id,
 877                           f->fn_type, f->fn_scope, &f->fn_key, z, f->fn_tos,
 878                           FIB_INFO(f)) < 0) {
 879                 kfree_skb(skb);
 880                 return;
 881         }
 882         NETLINK_CB(skb).dst_groups = RTMGRP_IPV4_ROUTE;
 883         if (n->nlmsg_flags&NLM_F_ECHO)
 884                 atomic_inc(&skb->users);
 885         netlink_broadcast(rtnl, skb, pid, RTMGRP_IPV4_ROUTE, GFP_KERNEL);
 886         if (n->nlmsg_flags&NLM_F_ECHO)
 887                 netlink_unicast(rtnl, skb, pid, MSG_DONTWAIT);
 888 }
 889
 890 #ifdef CONFIG_IP_MULTIPLE_TABLES
 891 struct fib_table * fib_hash_init(int id)
 892 #else
 893 struct fib_table * __init fib_hash_init(int id)
 894 #endif
 895 {
 896         struct fib_table *tb;
 897
 898         if (fn_hash_kmem == NULL)
 899                 fn_hash_kmem = kmem_cache_create("ip_fib_hash",
 900                                                  sizeof(struct fib_node),
 901                                                  0, SLAB_HWCACHE_ALIGN,
 902                                                  NULL, NULL);
 903
 904         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash), GFP_KERNEL);
 905         if (tb == NULL)
 906                 return NULL;
 907
 908         tb->tb_id = id;
 909         tb->tb_lookup = fn_hash_lookup;
 910         tb->tb_insert = fn_hash_insert;
 911         tb->tb_delete = fn_hash_delete;
 912         tb->tb_flush = fn_hash_flush;
 913         tb->tb_select_default = fn_hash_select_default;
 914         tb->tb_dump = fn_hash_dump;
 915 #ifdef CONFIG_PROC_FS
 916         tb->tb_get_info = fn_hash_get_info;
 917 #endif
 918         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 919         return tb;
 920 }