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 #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;    /* (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 ((PAGE_SIZE<<MAX_ORDER) / sizeof(struct fib_node *))
 153
 154 static struct fib_node **fz_hash_alloc(int divisor)
 155 {
 156         unsigned long size = divisor * sizeof(struct fib_node *);
 157
 158         if (divisor <= 1024) {
 159                 return kmalloc(size, GFP_KERNEL);
 160         } else {
 161                 return (struct fib_node **)
 162                         __get_free_pages(GFP_KERNEL, get_order(size));
 163         }
 164 }
 165
 166 /* The fib hash lock must be held when this is called. */
 167 static __inline__ void fn_rebuild_zone(struct fn_zone *fz,
 168                                        struct fib_node **old_ht,
 169                                        int old_divisor)
 170 {
 171         int i;
 172         struct fib_node *f, **fp, *next;
 173
 174         for (i=0; i<old_divisor; i++) {
 175                 for (f=old_ht[i]; f; f=next) {
 176                         next = f->fn_next;
 177                         for (fp = fz_chain_p(f->fn_key, fz);
 178                              *fp && fn_key_leq((*fp)->fn_key, f->fn_key);
 179                              fp = &(*fp)->fn_next)
 180                                 /* NONE */;
 181                         f->fn_next = *fp;
 182                         *fp = f;
 183                 }
 184         }
 185 }
 186
 187 static void fz_hash_free(struct fib_node **hash, int divisor)
 188 {
 189         if (divisor <= 1024)
 190                 kfree(hash);
 191         else
 192                 free_pages((unsigned long) hash,
 193                            get_order(divisor * sizeof(struct fib_node *)));
 194 }
 195
 196 static void fn_rehash_zone(struct fn_zone *fz)
 197 {
 198         struct fib_node **ht, **old_ht;
 199         int old_divisor, new_divisor;
 200         u32 new_hashmask;
 201
 202         old_divisor = fz->fz_divisor;
 203
 204         switch (old_divisor) {
 205         case 16:
 206                 new_divisor = 256;
 207                 break;
 208         case 256:
 209                 new_divisor = 1024;
 210                 break;
 211         default:
 212                 if ((old_divisor << 1) > FZ_MAX_DIVISOR) {
 213                         printk(KERN_CRIT "route.c: bad divisor %d!\n", old_divisor);
 214                         return;
 215                 }
 216                 new_divisor = (old_divisor << 1);
 217                 break;
 218         }
 219
 220         new_hashmask = (new_divisor - 1);
 221
 222 #if RT_CACHE_DEBUG >= 2
 223         printk("fn_rehash_zone: hash for zone %d grows from %d\n", fz->fz_order, old_divisor);
 224 #endif
 225
 226         ht = fz_hash_alloc(new_divisor);
 227
 228         if (ht) {
 229                 memset(ht, 0, new_divisor*sizeof(struct fib_node*));
 230
 231                 write_lock_bh(&fib_hash_lock);
 232                 old_ht = fz->fz_hash;
 233                 fz->fz_hash = ht;
 234                 fz->fz_hashmask = new_hashmask;
 235                 fz->fz_divisor = new_divisor;
 236                 fn_rebuild_zone(fz, old_ht, old_divisor);
 237                 write_unlock_bh(&fib_hash_lock);
 238
 239                 fz_hash_free(old_ht, old_divisor);
 240         }
 241 }
 242
 243 static void fn_free_node(struct fib_node * f)
 244 {
 245         fib_release_info(FIB_INFO(f));
 246         kmem_cache_free(fn_hash_kmem, f);
 247 }
 248
 249
 250 static struct fn_zone *
 251 fn_new_zone(struct fn_hash *table, int z)
 252 {
 253         int i;
 254         struct fn_zone *fz = kmalloc(sizeof(struct fn_zone), GFP_KERNEL);
 255         if (!fz)
 256                 return NULL;
 257
 258         memset(fz, 0, sizeof(struct fn_zone));
 259         if (z) {
 260                 fz->fz_divisor = 16;
 261         } else {
 262                 fz->fz_divisor = 1;
 263         }
 264         fz->fz_hashmask = (fz->fz_divisor - 1);
 265         fz->fz_hash = fz_hash_alloc(fz->fz_divisor);
 266         if (!fz->fz_hash) {
 267                 kfree(fz);
 268                 return NULL;
 269         }
 270         memset(fz->fz_hash, 0, fz->fz_divisor*sizeof(struct fib_node*));
 271         fz->fz_order = z;
 272         fz->fz_mask = inet_make_mask(z);
 273
 274         /* Find the first not empty zone with more specific mask */
 275         for (i=z+1; i<=32; i++)
 276                 if (table->fn_zones[i])
 277                         break;
 278         write_lock_bh(&fib_hash_lock);
 279         if (i>32) {
 280                 /* No more specific masks, we are the first. */
 281                 fz->fz_next = table->fn_zone_list;
 282                 table->fn_zone_list = fz;
 283         } else {
 284                 fz->fz_next = table->fn_zones[i]->fz_next;
 285                 table->fn_zones[i]->fz_next = fz;
 286         }
 287         table->fn_zones[z] = fz;
 288         write_unlock_bh(&fib_hash_lock);
 289         return fz;
 290 }
 291
 292 static int
 293 fn_hash_lookup(struct fib_table *tb, const struct rt_key *key, struct fib_result *res)
 294 {
 295         int err;
 296         struct fn_zone *fz;
 297         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 298
 299         read_lock(&fib_hash_lock);
 300         for (fz = t->fn_zone_list; fz; fz = fz->fz_next) {
 301                 struct fib_node *f;
 302                 fn_key_t k = fz_key(key->dst, fz);
 303
 304                 for (f = fz_chain(k, fz); f; f = f->fn_next) {
 305                         if (!fn_key_eq(k, f->fn_key)) {
 306                                 if (fn_key_leq(k, f->fn_key))
 307                                         break;
 308                                 else
 309                                         continue;
 310                         }
 311 #ifdef CONFIG_IP_ROUTE_TOS
 312                         if (f->fn_tos && f->fn_tos != key->tos)
 313                                 continue;
 314 #endif
 315                         f->fn_state |= FN_S_ACCESSED;
 316
 317                         if (f->fn_state&FN_S_ZOMBIE)
 318                                 continue;
 319                         if (f->fn_scope < key->scope)
 320                                 continue;
 321
 322                         err = fib_semantic_match(f->fn_type, FIB_INFO(f), key, res);
 323                         if (err == 0) {
 324                                 res->type = f->fn_type;
 325                                 res->scope = f->fn_scope;
 326                                 res->prefixlen = fz->fz_order;
 327                                 goto out;
 328                         }
 329                         if (err < 0)
 330                                 goto out;
 331                 }
 332         }
 333         err = 1;
 334 out:
 335         read_unlock(&fib_hash_lock);
 336         return err;
 337 }
 338
 339 static int fn_hash_last_dflt=-1;
 340
 341 static int fib_detect_death(struct fib_info *fi, int order,
 342                             struct fib_info **last_resort, int *last_idx)
 343 {
 344         struct neighbour *n;
 345         int state = NUD_NONE;
 346
 347         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 348         if (n) {
 349                 state = n->nud_state;
 350                 neigh_release(n);
 351         }
 352         if (state==NUD_REACHABLE)
 353                 return 0;
 354         if ((state&NUD_VALID) && order != fn_hash_last_dflt)
 355                 return 0;
 356         if ((state&NUD_VALID) ||
 357             (*last_idx<0 && order > fn_hash_last_dflt)) {
 358                 *last_resort = fi;
 359                 *last_idx = order;
 360         }
 361         return 1;
 362 }
 363
 364 static void
 365 fn_hash_select_default(struct fib_table *tb, const struct rt_key *key, struct fib_result *res)
 366 {
 367         int order, last_idx;
 368         struct fib_node *f;
 369         struct fib_info *fi = NULL;
 370         struct fib_info *last_resort;
 371         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 372         struct fn_zone *fz = t->fn_zones[0];
 373
 374         if (fz == NULL)
 375                 return;
 376
 377         last_idx = -1;
 378         last_resort = NULL;
 379         order = -1;
 380
 381         read_lock(&fib_hash_lock);
 382         for (f = fz->fz_hash[0]; f; f = f->fn_next) {
 383                 struct fib_info *next_fi = FIB_INFO(f);
 384
 385                 if ((f->fn_state&FN_S_ZOMBIE) ||
 386                     f->fn_scope != res->scope ||
 387                     f->fn_type != RTN_UNICAST)
 388                         continue;
 389
 390                 if (next_fi->fib_priority > res->fi->fib_priority)
 391                         break;
 392                 if (!next_fi->fib_nh[0].nh_gw || next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
 393                         continue;
 394                 f->fn_state |= FN_S_ACCESSED;
 395
 396                 if (fi == NULL) {
 397                         if (next_fi != res->fi)
 398                                 break;
 399                 } else if (!fib_detect_death(fi, order, &last_resort, &last_idx)) {
 400                         if (res->fi)
 401                                 fib_info_put(res->fi);
 402                         res->fi = fi;
 403                         atomic_inc(&fi->fib_clntref);
 404                         fn_hash_last_dflt = order;
 405                         goto out;
 406                 }
 407                 fi = next_fi;
 408                 order++;
 409         }
 410
 411         if (order<=0 || fi==NULL) {
 412                 fn_hash_last_dflt = -1;
 413                 goto out;
 414         }
 415
 416         if (!fib_detect_death(fi, order, &last_resort, &last_idx)) {
 417                 if (res->fi)
 418                         fib_info_put(res->fi);
 419                 res->fi = fi;
 420                 atomic_inc(&fi->fib_clntref);
 421                 fn_hash_last_dflt = order;
 422                 goto out;
 423         }
 424
 425         if (last_idx >= 0) {
 426                 if (res->fi)
 427                         fib_info_put(res->fi);
 428                 res->fi = last_resort;
 429                 if (last_resort)
 430                         atomic_inc(&last_resort->fib_clntref);
 431         }
 432         fn_hash_last_dflt = last_idx;
 433 out:
 434         read_unlock(&fib_hash_lock);
 435 }
 436
 437 #define FIB_SCAN(f, fp) \
 438 for ( ; ((f) = *(fp)) != NULL; (fp) = &(f)->fn_next)
 439
 440 #define FIB_SCAN_KEY(f, fp, key) \
 441 for ( ; ((f) = *(fp)) != NULL && fn_key_eq((f)->fn_key, (key)); (fp) = &(f)->fn_next)
 442
 443 #ifndef CONFIG_IP_ROUTE_TOS
 444 #define FIB_SCAN_TOS(f, fp, key, tos) FIB_SCAN_KEY(f, fp, key)
 445 #else
 446 #define FIB_SCAN_TOS(f, fp, key, tos) \
 447 for ( ; ((f) = *(fp)) != NULL && fn_key_eq((f)->fn_key, (key)) && \
 448      (f)->fn_tos == (tos) ; (fp) = &(f)->fn_next)
 449 #endif
 450
 451
 452 static void rtmsg_fib(int, struct fib_node*, int, int,
 453                       struct nlmsghdr *n,
 454                       struct netlink_skb_parms *);
 455
 456 static int
 457 fn_hash_insert(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
 458                 struct nlmsghdr *n, struct netlink_skb_parms *req)
 459 {
 460         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 461         struct fib_node *new_f, *f, **fp, **del_fp;
 462         struct fn_zone *fz;
 463         struct fib_info *fi;
 464
 465         int z = r->rtm_dst_len;
 466         int type = r->rtm_type;
 467 #ifdef CONFIG_IP_ROUTE_TOS
 468         u8 tos = r->rtm_tos;
 469 #endif
 470         fn_key_t key;
 471         int err;
 472
 473 FTprint("tb(%d)_insert: %d %08x/%d %d %08x\n", tb->tb_id, r->rtm_type, rta->rta_dst ?
 474 *(u32*)rta->rta_dst : 0, z, rta->rta_oif ? *rta->rta_oif : -1,
 475 rta->rta_prefsrc ? *(u32*)rta->rta_prefsrc : 0);
 476         if (z > 32)
 477                 return -EINVAL;
 478         fz = table->fn_zones[z];
 479         if (!fz && !(fz = fn_new_zone(table, z)))
 480                 return -ENOBUFS;
 481
 482         fz_key_0(key);
 483         if (rta->rta_dst) {
 484                 u32 dst;
 485                 memcpy(&dst, rta->rta_dst, 4);
 486                 if (dst & ~FZ_MASK(fz))
 487                         return -EINVAL;
 488                 key = fz_key(dst, fz);
 489         }
 490
 491         if  ((fi = fib_create_info(r, rta, n, &err)) == NULL)
 492                 return err;
 493
 494         if (fz->fz_nent > (fz->fz_divisor<<1) &&
 495             fz->fz_divisor < FZ_MAX_DIVISOR &&
 496             (z==32 || (1<<z) > fz->fz_divisor))
 497                 fn_rehash_zone(fz);
 498
 499         fp = fz_chain_p(key, fz);
 500
 501
 502         /*
 503          * Scan list to find the first route with the same destination
 504          */
 505         FIB_SCAN(f, fp) {
 506                 if (fn_key_leq(key,f->fn_key))
 507                         break;
 508         }
 509
 510 #ifdef CONFIG_IP_ROUTE_TOS
 511         /*
 512          * Find route with the same destination and tos.
 513          */
 514         FIB_SCAN_KEY(f, fp, key) {
 515                 if (f->fn_tos <= tos)
 516                         break;
 517         }
 518 #endif
 519
 520         del_fp = NULL;
 521
 522         if (f && (f->fn_state&FN_S_ZOMBIE) &&
 523 #ifdef CONFIG_IP_ROUTE_TOS
 524             f->fn_tos == tos &&
 525 #endif
 526             fn_key_eq(f->fn_key, key)) {
 527                 del_fp = fp;
 528                 fp = &f->fn_next;
 529                 f = *fp;
 530                 goto create;
 531         }
 532
 533         FIB_SCAN_TOS(f, fp, key, tos) {
 534                 if (fi->fib_priority <= FIB_INFO(f)->fib_priority)
 535                         break;
 536         }
 537
 538         /* Now f==*fp points to the first node with the same
 539            keys [prefix,tos,priority], if such key already
 540            exists or to the node, before which we will insert new one.
 541          */
 542
 543         if (f &&
 544 #ifdef CONFIG_IP_ROUTE_TOS
 545             f->fn_tos == tos &&
 546 #endif
 547             fn_key_eq(f->fn_key, key) &&
 548             fi->fib_priority == FIB_INFO(f)->fib_priority) {
 549                 struct fib_node **ins_fp;
 550
 551                 err = -EEXIST;
 552                 if (n->nlmsg_flags&NLM_F_EXCL)
 553                         goto out;
 554
 555                 if (n->nlmsg_flags&NLM_F_REPLACE) {
 556                         del_fp = fp;
 557                         fp = &f->fn_next;
 558                         f = *fp;
 559                         goto replace;
 560                 }
 561
 562                 ins_fp = fp;
 563                 err = -EEXIST;
 564
 565                 FIB_SCAN_TOS(f, fp, key, tos) {
 566                         if (fi->fib_priority != FIB_INFO(f)->fib_priority)
 567                                 break;
 568                         if (f->fn_type == type && f->fn_scope == r->rtm_scope
 569                             && FIB_INFO(f) == fi)
 570                                 goto out;
 571                 }
 572
 573                 if (!(n->nlmsg_flags&NLM_F_APPEND)) {
 574                         fp = ins_fp;
 575                         f = *fp;
 576                 }
 577         }
 578
 579 create:
 580         err = -ENOENT;
 581         if (!(n->nlmsg_flags&NLM_F_CREATE))
 582                 goto out;
 583
 584 replace:
 585         err = -ENOBUFS;
 586         new_f = kmem_cache_alloc(fn_hash_kmem, SLAB_KERNEL);
 587         if (new_f == NULL)
 588                 goto out;
 589
 590         memset(new_f, 0, sizeof(struct fib_node));
 591
 592         new_f->fn_key = key;
 593 #ifdef CONFIG_IP_ROUTE_TOS
 594         new_f->fn_tos = tos;
 595 #endif
 596         new_f->fn_type = type;
 597         new_f->fn_scope = r->rtm_scope;
 598         FIB_INFO(new_f) = fi;
 599
 600         /*
 601          * Insert new entry to the list.
 602          */
 603
 604         new_f->fn_next = f;
 605         write_lock_bh(&fib_hash_lock);
 606         *fp = new_f;
 607         write_unlock_bh(&fib_hash_lock);
 608         fz->fz_nent++;
 609
 610         if (del_fp) {
 611                 f = *del_fp;
 612                 /* Unlink replaced node */
 613                 write_lock_bh(&fib_hash_lock);
 614                 *del_fp = f->fn_next;
 615                 write_unlock_bh(&fib_hash_lock);
 616
 617                 if (!(f->fn_state&FN_S_ZOMBIE))
 618                         rtmsg_fib(RTM_DELROUTE, f, z, tb->tb_id, n, req);
 619                 if (f->fn_state&FN_S_ACCESSED)
 620                         rt_cache_flush(-1);
 621                 fn_free_node(f);
 622                 fz->fz_nent--;
 623         } else {
 624                 rt_cache_flush(-1);
 625         }
 626         rtmsg_fib(RTM_NEWROUTE, new_f, z, tb->tb_id, n, req);
 627         return 0;
 628
 629 out:
 630         fib_release_info(fi);
 631         return err;
 632 }
 633
 634
 635 static int
 636 fn_hash_delete(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
 637                 struct nlmsghdr *n, struct netlink_skb_parms *req)
 638 {
 639         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 640         struct fib_node **fp, **del_fp, *f;
 641         int z = r->rtm_dst_len;
 642         struct fn_zone *fz;
 643         fn_key_t key;
 644         int matched;
 645 #ifdef CONFIG_IP_ROUTE_TOS
 646         u8 tos = r->rtm_tos;
 647 #endif
 648
 649 FTprint("tb(%d)_delete: %d %08x/%d %d\n", tb->tb_id, r->rtm_type, rta->rta_dst ?
 650        *(u32*)rta->rta_dst : 0, z, rta->rta_oif ? *rta->rta_oif : -1);
 651         if (z > 32)
 652                 return -EINVAL;
 653         if ((fz  = table->fn_zones[z]) == NULL)
 654                 return -ESRCH;
 655
 656         fz_key_0(key);
 657         if (rta->rta_dst) {
 658                 u32 dst;
 659                 memcpy(&dst, rta->rta_dst, 4);
 660                 if (dst & ~FZ_MASK(fz))
 661                         return -EINVAL;
 662                 key = fz_key(dst, fz);
 663         }
 664
 665         fp = fz_chain_p(key, fz);
 666
 667
 668         FIB_SCAN(f, fp) {
 669                 if (fn_key_eq(f->fn_key, key))
 670                         break;
 671                 if (fn_key_leq(key, f->fn_key)) {
 672                         return -ESRCH;
 673                 }
 674         }
 675 #ifdef CONFIG_IP_ROUTE_TOS
 676         FIB_SCAN_KEY(f, fp, key) {
 677                 if (f->fn_tos == tos)
 678                         break;
 679         }
 680 #endif
 681
 682         matched = 0;
 683         del_fp = NULL;
 684         FIB_SCAN_TOS(f, fp, key, tos) {
 685                 struct fib_info * fi = FIB_INFO(f);
 686
 687                 if (f->fn_state&FN_S_ZOMBIE) {
 688                         return -ESRCH;
 689                 }
 690                 matched++;
 691
 692                 if (del_fp == NULL &&
 693                     (!r->rtm_type || f->fn_type == r->rtm_type) &&
 694                     (r->rtm_scope == RT_SCOPE_NOWHERE || f->fn_scope == r->rtm_scope) &&
 695                     (!r->rtm_protocol || fi->fib_protocol == r->rtm_protocol) &&
 696                     fib_nh_match(r, n, rta, fi) == 0)
 697                         del_fp = fp;
 698         }
 699
 700         if (del_fp) {
 701                 f = *del_fp;
 702                 rtmsg_fib(RTM_DELROUTE, f, z, tb->tb_id, n, req);
 703
 704                 if (matched != 1) {
 705                         write_lock_bh(&fib_hash_lock);
 706                         *del_fp = f->fn_next;
 707                         write_unlock_bh(&fib_hash_lock);
 708
 709                         if (f->fn_state&FN_S_ACCESSED)
 710                                 rt_cache_flush(-1);
 711                         fn_free_node(f);
 712                         fz->fz_nent--;
 713                 } else {
 714                         f->fn_state |= FN_S_ZOMBIE;
 715                         if (f->fn_state&FN_S_ACCESSED) {
 716                                 f->fn_state &= ~FN_S_ACCESSED;
 717                                 rt_cache_flush(-1);
 718                         }
 719                         if (++fib_hash_zombies > 128)
 720                                 fib_flush();
 721                 }
 722
 723                 return 0;
 724         }
 725         return -ESRCH;
 726 }
 727
 728 extern __inline__ int
 729 fn_flush_list(struct fib_node ** fp, int z, struct fn_hash *table)
 730 {
 731         int found = 0;
 732         struct fib_node *f;
 733
 734         while ((f = *fp) != NULL) {
 735                 struct fib_info *fi = FIB_INFO(f);
 736
 737                 if (fi && ((f->fn_state&FN_S_ZOMBIE) || (fi->fib_flags&RTNH_F_DEAD))) {
 738                         write_lock_bh(&fib_hash_lock);
 739                         *fp = f->fn_next;
 740                         write_unlock_bh(&fib_hash_lock);
 741
 742                         fn_free_node(f);
 743                         found++;
 744                         continue;
 745                 }
 746                 fp = &f->fn_next;
 747         }
 748         return found;
 749 }
 750
 751 static int fn_hash_flush(struct fib_table *tb)
 752 {
 753         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 754         struct fn_zone *fz;
 755         int found = 0;
 756
 757         fib_hash_zombies = 0;
 758         for (fz = table->fn_zone_list; fz; fz = fz->fz_next) {
 759                 int i;
 760                 int tmp = 0;
 761                 for (i=fz->fz_divisor-1; i>=0; i--)
 762                         tmp += fn_flush_list(&fz->fz_hash[i], fz->fz_order, table);
 763                 fz->fz_nent -= tmp;
 764                 found += tmp;
 765         }
 766         return found;
 767 }
 768
 769
 770 #ifdef CONFIG_PROC_FS
 771
 772 static int fn_hash_get_info(struct fib_table *tb, char *buffer, int first, int count)
 773 {
 774         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 775         struct fn_zone *fz;
 776         int pos = 0;
 777         int n = 0;
 778
 779         read_lock(&fib_hash_lock);
 780         for (fz=table->fn_zone_list; fz; fz = fz->fz_next) {
 781                 int i;
 782                 struct fib_node *f;
 783                 int maxslot = fz->fz_divisor;
 784                 struct fib_node **fp = fz->fz_hash;
 785
 786                 if (fz->fz_nent == 0)
 787                         continue;
 788
 789                 if (pos + fz->fz_nent <= first) {
 790                         pos += fz->fz_nent;
 791                         continue;
 792                 }
 793
 794                 for (i=0; i < maxslot; i++, fp++) {
 795                         for (f = *fp; f; f = f->fn_next) {
 796                                 if (++pos <= first)
 797                                         continue;
 798                                 fib_node_get_info(f->fn_type,
 799                                                   f->fn_state&FN_S_ZOMBIE,
 800                                                   FIB_INFO(f),
 801                                                   fz_prefix(f->fn_key, fz),
 802                                                   FZ_MASK(fz), buffer);
 803                                 buffer += 128;
 804                                 if (++n >= count)
 805                                         goto out;
 806                         }
 807                 }
 808         }
 809 out:
 810         read_unlock(&fib_hash_lock);
 811         return n;
 812 }
 813 #endif
 814
 815
 816 static __inline__ int
 817 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
 818                      struct fib_table *tb,
 819                      struct fn_zone *fz,
 820                      struct fib_node *f)
 821 {
 822         int i, s_i;
 823
 824         s_i = cb->args[3];
 825         for (i=0; f; i++, f=f->fn_next) {
 826                 if (i < s_i) continue;
 827                 if (f->fn_state&FN_S_ZOMBIE) continue;
 828                 if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid, cb->nlh->nlmsg_seq,
 829                                   RTM_NEWROUTE,
 830                                   tb->tb_id, (f->fn_state&FN_S_ZOMBIE) ? 0 : f->fn_type, f->fn_scope,
 831                                   &f->fn_key, fz->fz_order, f->fn_tos,
 832                                   f->fn_info) < 0) {
 833                         cb->args[3] = i;
 834                         return -1;
 835                 }
 836         }
 837         cb->args[3] = i;
 838         return skb->len;
 839 }
 840
 841 static __inline__ int
 842 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
 843                    struct fib_table *tb,
 844                    struct fn_zone *fz)
 845 {
 846         int h, s_h;
 847
 848         s_h = cb->args[2];
 849         for (h=0; h < fz->fz_divisor; h++) {
 850                 if (h < s_h) continue;
 851                 if (h > s_h)
 852                         memset(&cb->args[3], 0, sizeof(cb->args) - 3*sizeof(cb->args[0]));
 853                 if (fz->fz_hash == NULL || fz->fz_hash[h] == NULL)
 854                         continue;
 855                 if (fn_hash_dump_bucket(skb, cb, tb, fz, fz->fz_hash[h]) < 0) {
 856                         cb->args[2] = h;
 857                         return -1;
 858                 }
 859         }
 860         cb->args[2] = h;
 861         return skb->len;
 862 }
 863
 864 static int fn_hash_dump(struct fib_table *tb, struct sk_buff *skb, struct netlink_callback *cb)
 865 {
 866         int m, s_m;
 867         struct fn_zone *fz;
 868         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 869
 870         s_m = cb->args[1];
 871         read_lock(&fib_hash_lock);
 872         for (fz = table->fn_zone_list, m=0; fz; fz = fz->fz_next, m++) {
 873                 if (m < s_m) continue;
 874                 if (m > s_m)
 875                         memset(&cb->args[2], 0, sizeof(cb->args) - 2*sizeof(cb->args[0]));
 876                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
 877                         cb->args[1] = m;
 878                         read_unlock(&fib_hash_lock);
 879                         return -1;
 880                 }
 881         }
 882         read_unlock(&fib_hash_lock);
 883         cb->args[1] = m;
 884         return skb->len;
 885 }
 886
 887 static void rtmsg_fib(int event, struct fib_node* f, int z, int tb_id,
 888                       struct nlmsghdr *n, struct netlink_skb_parms *req)
 889 {
 890         struct sk_buff *skb;
 891         u32 pid = req ? req->pid : 0;
 892         int size = NLMSG_SPACE(sizeof(struct rtmsg)+256);
 893
 894         skb = alloc_skb(size, GFP_KERNEL);
 895         if (!skb)
 896                 return;
 897
 898         if (fib_dump_info(skb, pid, n->nlmsg_seq, event, tb_id,
 899                           f->fn_type, f->fn_scope, &f->fn_key, z, f->fn_tos,
 900                           FIB_INFO(f)) < 0) {
 901                 kfree_skb(skb);
 902                 return;
 903         }
 904         NETLINK_CB(skb).dst_groups = RTMGRP_IPV4_ROUTE;
 905         if (n->nlmsg_flags&NLM_F_ECHO)
 906                 atomic_inc(&skb->users);
 907         netlink_broadcast(rtnl, skb, pid, RTMGRP_IPV4_ROUTE, GFP_KERNEL);
 908         if (n->nlmsg_flags&NLM_F_ECHO)
 909                 netlink_unicast(rtnl, skb, pid, MSG_DONTWAIT);
 910 }
 911
 912 #ifdef CONFIG_IP_MULTIPLE_TABLES
 913 struct fib_table * fib_hash_init(int id)
 914 #else
 915 struct fib_table * __init fib_hash_init(int id)
 916 #endif
 917 {
 918         struct fib_table *tb;
 919
 920         if (fn_hash_kmem == NULL)
 921                 fn_hash_kmem = kmem_cache_create("ip_fib_hash",
 922                                                  sizeof(struct fib_node),
 923                                                  0, SLAB_HWCACHE_ALIGN,
 924                                                  NULL, NULL);
 925
 926         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash), GFP_KERNEL);
 927         if (tb == NULL)
 928                 return NULL;
 929
 930         tb->tb_id = id;
 931         tb->tb_lookup = fn_hash_lookup;
 932         tb->tb_insert = fn_hash_insert;
 933         tb->tb_delete = fn_hash_delete;
 934         tb->tb_flush = fn_hash_flush;
 935         tb->tb_select_default = fn_hash_select_default;
 936         tb->tb_dump = fn_hash_dump;
 937 #ifdef CONFIG_PROC_FS
 938         tb->tb_get_info = fn_hash_get_info;
 939 #endif
 940         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 941         return tb;
 942 }