kernel/linux/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 2005/04/29 01:44:08 echo 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 static __inline__ int fn_key_eq(fn_key_t a, fn_key_t b)
 141 {
 142         return a.datum == b.datum;
 143 }
 144
 145 static __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
 256         if (!fz)
 257                 return NULL;
 258
 259         memset(fz, 0, sizeof(struct fn_zone));
 260         if (z) {
 261                 fz->fz_divisor = 16;
 262         } else {
 263                 fz->fz_divisor = 1;
 264         }
 265         fz->fz_hashmask = (fz->fz_divisor - 1);
 266         fz->fz_hash = fz_hash_alloc(fz->fz_divisor);
 267         if (!fz->fz_hash) {
 268                 kfree(fz);
 269                 return NULL;
 270         }
 271         memset(fz->fz_hash, 0, fz->fz_divisor*sizeof(struct fib_node*));
 272         fz->fz_order = z;
 273         fz->fz_mask = inet_make_mask(z);
 274
 275         /* Find the first not empty zone with more specific mask */
 276         for (i=z+1; i<=32; i++)
 277                 if (table->fn_zones[i])
 278                         break;
 279
 280         write_lock_bh(&fib_hash_lock);
 281         if (i>32) {
 282                 /* No more specific masks, we are the first. */
 283                 fz->fz_next = table->fn_zone_list;
 284                 table->fn_zone_list = fz;
 285         } else {
 286                 fz->fz_next = table->fn_zones[i]->fz_next;
 287                 table->fn_zones[i]->fz_next = fz;
 288         }
 289         table->fn_zones[z] = fz;
 290         write_unlock_bh(&fib_hash_lock);
 291         return fz;
 292 }
 293
 294 static int
 295 fn_hash_lookup(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
 296 {
 297         int err;
 298         struct fn_zone *fz;
 299         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 300
 301         read_lock(&fib_hash_lock);
 302         for (fz = t->fn_zone_list; fz; fz = fz->fz_next) {
 303                 struct fib_node *f;
 304                 fn_key_t k = fz_key(flp->fl4_dst, fz);
 305
 306                 for (f = fz_chain(k, fz); f; f = f->fn_next) {
 307                         if (!fn_key_eq(k, f->fn_key)) {
 308                                 if (fn_key_leq(k, f->fn_key))
 309                                         break;
 310                                 else
 311                                         continue;
 312                         }
 313 #ifdef CONFIG_IP_ROUTE_TOS
 314                         if (f->fn_tos && f->fn_tos != flp->fl4_tos)
 315                                 continue;
 316 #endif
 317                         f->fn_state |= FN_S_ACCESSED;
 318
 319                         if (f->fn_state&FN_S_ZOMBIE)
 320                                 continue;
 321                         if (f->fn_scope < flp->fl4_scope)
 322                                 continue;
 323
 324                         err = fib_semantic_match(f->fn_type, FIB_INFO(f), flp, res);
 325                         if (err == 0) {
 326                                 res->type = f->fn_type;
 327                                 res->scope = f->fn_scope;
 328                                 res->prefixlen = fz->fz_order;
 329                                 goto out;
 330                         }
 331                         if (err < 0)
 332                                 goto out;
 333                 }
 334         }
 335         err = 1;
 336 out:
 337         read_unlock(&fib_hash_lock);
 338         return err;
 339 }
 340
 341 static int fn_hash_last_dflt=-1;
 342
 343 static int fib_detect_death(struct fib_info *fi, int order,
 344                             struct fib_info **last_resort, int *last_idx)
 345 {
 346         struct neighbour *n;
 347         int state = NUD_NONE;
 348
 349         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 350         if (n) {
 351                 state = n->nud_state;
 352                 neigh_release(n);
 353         }
 354         if (state==NUD_REACHABLE)
 355                 return 0;
 356         if ((state&NUD_VALID) && order != fn_hash_last_dflt)
 357                 return 0;
 358         if ((state&NUD_VALID) ||
 359             (*last_idx<0 && order > fn_hash_last_dflt)) {
 360                 *last_resort = fi;
 361                 *last_idx = order;
 362         }
 363         return 1;
 364 }
 365
 366 static void
 367 fn_hash_select_default(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
 368 {
 369         int order, last_idx;
 370         struct fib_node *f;
 371         struct fib_info *fi = NULL;
 372         struct fib_info *last_resort;
 373         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
 374         struct fn_zone *fz = t->fn_zones[0];
 375
 376         if (fz == NULL)
 377                 return;
 378
 379         last_idx = -1;
 380         last_resort = NULL;
 381         order = -1;
 382
 383         read_lock(&fib_hash_lock);
 384         for (f = fz->fz_hash[0]; f; f = f->fn_next) {
 385                 struct fib_info *next_fi = FIB_INFO(f);
 386
 387                 if ((f->fn_state&FN_S_ZOMBIE) ||
 388                     f->fn_scope != res->scope ||
 389                     f->fn_type != RTN_UNICAST)
 390                         continue;
 391
 392                 if (next_fi->fib_priority > res->fi->fib_priority)
 393                         break;
 394                 if (!next_fi->fib_nh[0].nh_gw || next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
 395                         continue;
 396                 f->fn_state |= FN_S_ACCESSED;
 397
 398                 if (fi == NULL) {
 399                         if (next_fi != res->fi)
 400                                 break;
 401                 } else if (!fib_detect_death(fi, order, &last_resort, &last_idx)) {
 402                         if (res->fi)
 403                                 fib_info_put(res->fi);
 404                         res->fi = fi;
 405                         atomic_inc(&fi->fib_clntref);
 406                         fn_hash_last_dflt = order;
 407                         goto out;
 408                 }
 409                 fi = next_fi;
 410                 order++;
 411         }
 412
 413         if (order<=0 || fi==NULL) {
 414                 fn_hash_last_dflt = -1;
 415                 goto out;
 416         }
 417
 418         if (!fib_detect_death(fi, order, &last_resort, &last_idx)) {
 419                 if (res->fi)
 420                         fib_info_put(res->fi);
 421                 res->fi = fi;
 422                 atomic_inc(&fi->fib_clntref);
 423                 fn_hash_last_dflt = order;
 424                 goto out;
 425         }
 426
 427         if (last_idx >= 0) {
 428                 if (res->fi)
 429                         fib_info_put(res->fi);
 430                 res->fi = last_resort;
 431                 if (last_resort)
 432                         atomic_inc(&last_resort->fib_clntref);
 433         }
 434         fn_hash_last_dflt = last_idx;
 435 out:
 436         read_unlock(&fib_hash_lock);
 437 }
 438
 439 #define FIB_SCAN(f, fp) \
 440 for ( ; ((f) = *(fp)) != NULL; (fp) = &(f)->fn_next)
 441
 442 #define FIB_SCAN_KEY(f, fp, key) \
 443 for ( ; ((f) = *(fp)) != NULL && fn_key_eq((f)->fn_key, (key)); (fp) = &(f)->fn_next)
 444
 445 #ifndef CONFIG_IP_ROUTE_TOS
 446 #define FIB_SCAN_TOS(f, fp, key, tos) FIB_SCAN_KEY(f, fp, key)
 447 #else
 448 #define FIB_SCAN_TOS(f, fp, key, tos) \
 449 for ( ; ((f) = *(fp)) != NULL && fn_key_eq((f)->fn_key, (key)) && \
 450      (f)->fn_tos == (tos) ; (fp) = &(f)->fn_next)
 451 #endif
 452
 453
 454 static void rtmsg_fib(int, struct fib_node*, int, int,
 455                       struct nlmsghdr *n,
 456                       struct netlink_skb_parms *);
 457
 458 static int
 459 fn_hash_insert(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
 460                 struct nlmsghdr *n, struct netlink_skb_parms *req)
 461 {
 462         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 463         struct fib_node *new_f, *f, **fp, **del_fp;
 464         struct fn_zone *fz;
 465         struct fib_info *fi;
 466
 467         int z = r->rtm_dst_len;
 468         int type = r->rtm_type;
 469 #ifdef CONFIG_IP_ROUTE_TOS
 470         u8 tos = r->rtm_tos;
 471 #endif
 472         fn_key_t key;
 473         int err;
 474
 475 FTprint("tb(%d)_insert: %d %08x/%d %d %08x\n", tb->tb_id, r->rtm_type, rta->rta_dst ?
 476 *(u32*)rta->rta_dst : 0, z, rta->rta_oif ? *rta->rta_oif : -1,
 477 rta->rta_prefsrc ? *(u32*)rta->rta_prefsrc : 0);
 478
 479         if (z > 32)
 480                 return -EINVAL;
 481         fz = table->fn_zones[z];
 482         if (!fz && !(fz = fn_new_zone(table, z)))
 483                 return -ENOBUFS;
 484
 485         fz_key_0(key);
 486         if (rta->rta_dst) {
 487                 u32 dst;
 488                 memcpy(&dst, rta->rta_dst, 4);
 489                 if (dst & ~FZ_MASK(fz))
 490                         return -EINVAL;
 491                 key = fz_key(dst, fz);
 492         }
 493
 494         if  ((fi = fib_create_info(r, rta, n, &err)) == NULL)
 495                 return err;
 496
 497         if (fz->fz_nent > (fz->fz_divisor<<1) &&
 498             fz->fz_divisor < FZ_MAX_DIVISOR &&
 499             (z==32 || (1<<z) > fz->fz_divisor))
 500                 fn_rehash_zone(fz);
 501
 502         fp = fz_chain_p(key, fz);
 503
 504
 505         /*
 506          * Scan list to find the first route with the same destination
 507          */
 508         FIB_SCAN(f, fp) {
 509                 if (fn_key_leq(key,f->fn_key))
 510                         break;
 511         }
 512
 513 #ifdef CONFIG_IP_ROUTE_TOS
 514         /*
 515          * Find route with the same destination and tos.
 516          */
 517         FIB_SCAN_KEY(f, fp, key) {
 518                 if (f->fn_tos <= tos)
 519                         break;
 520         }
 521 #endif
 522
 523         del_fp = NULL;
 524
 525         if (f && (f->fn_state&FN_S_ZOMBIE) &&
 526 #ifdef CONFIG_IP_ROUTE_TOS
 527             f->fn_tos == tos &&
 528 #endif
 529             fn_key_eq(f->fn_key, key)) {
 530                 del_fp = fp;
 531                 fp = &f->fn_next;
 532                 f = *fp;
 533                 goto create;
 534         }
 535
 536         FIB_SCAN_TOS(f, fp, key, tos) {
 537                 if (fi->fib_priority <= FIB_INFO(f)->fib_priority)
 538                         break;
 539         }
 540
 541         /* Now f==*fp points to the first node with the same
 542            keys [prefix,tos,priority], if such key already
 543            exists or to the node, before which we will insert new one.
 544          */
 545
 546         if (f &&
 547 #ifdef CONFIG_IP_ROUTE_TOS
 548             f->fn_tos == tos &&
 549 #endif
 550             fn_key_eq(f->fn_key, key) &&
 551             fi->fib_priority == FIB_INFO(f)->fib_priority) {
 552                 struct fib_node **ins_fp;
 553
 554                 err = -EEXIST;
 555                 if (n->nlmsg_flags&NLM_F_EXCL)
 556                         goto out;
 557
 558                 if (n->nlmsg_flags&NLM_F_REPLACE) {
 559                         del_fp = fp;
 560                         fp = &f->fn_next;
 561                         f = *fp;
 562                         goto replace;
 563                 }
 564
 565                 ins_fp = fp;
 566                 err = -EEXIST;
 567
 568                 FIB_SCAN_TOS(f, fp, key, tos) {
 569                         if (fi->fib_priority != FIB_INFO(f)->fib_priority)
 570                                 break;
 571                         if (f->fn_type == type && f->fn_scope == r->rtm_scope
 572                             && FIB_INFO(f) == fi)
 573                                 goto out;
 574                 }
 575
 576                 if (!(n->nlmsg_flags&NLM_F_APPEND)) {
 577                         fp = ins_fp;
 578                         f = *fp;
 579                 }
 580         }
 581
 582 create:
 583         err = -ENOENT;
 584         if (!(n->nlmsg_flags&NLM_F_CREATE))
 585                 goto out;
 586
 587 replace:
 588         err = -ENOBUFS;
 589         new_f = kmem_cache_alloc(fn_hash_kmem, SLAB_KERNEL);
 590         if (new_f == NULL)
 591                 goto out;
 592
 593         memset(new_f, 0, sizeof(struct fib_node));
 594
 595         new_f->fn_key = key;
 596 #ifdef CONFIG_IP_ROUTE_TOS
 597         new_f->fn_tos = tos;
 598 #endif
 599         new_f->fn_type = type;
 600         new_f->fn_scope = r->rtm_scope;
 601         FIB_INFO(new_f) = fi;
 602
 603         /*
 604          * Insert new entry to the list.
 605          */
 606
 607         new_f->fn_next = f;
 608         write_lock_bh(&fib_hash_lock);
 609         *fp = new_f;
 610         write_unlock_bh(&fib_hash_lock);
 611         fz->fz_nent++;
 612
 613         if (del_fp) {
 614                 f = *del_fp;
 615                 /* Unlink replaced node */
 616                 write_lock_bh(&fib_hash_lock);
 617                 *del_fp = f->fn_next;
 618                 write_unlock_bh(&fib_hash_lock);
 619
 620                 if (!(f->fn_state&FN_S_ZOMBIE))
 621                         rtmsg_fib(RTM_DELROUTE, f, z, tb->tb_id, n, req);
 622                 if (f->fn_state&FN_S_ACCESSED)
 623                         rt_cache_flush(-1);
 624                 fn_free_node(f);
 625                 fz->fz_nent--;
 626         } else {
 627                 rt_cache_flush(-1);
 628         }
 629         rtmsg_fib(RTM_NEWROUTE, new_f, z, tb->tb_id, n, req);
 630         return 0;
 631
 632 out:
 633         fib_release_info(fi);
 634         return err;
 635 }
 636
 637
 638 static int
 639 fn_hash_delete(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
 640                 struct nlmsghdr *n, struct netlink_skb_parms *req)
 641 {
 642         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 643         struct fib_node **fp, **del_fp, *f;
 644         int z = r->rtm_dst_len;
 645         struct fn_zone *fz;
 646         fn_key_t key;
 647         int matched;
 648 #ifdef CONFIG_IP_ROUTE_TOS
 649         u8 tos = r->rtm_tos;
 650 #endif
 651
 652 FTprint("tb(%d)_delete: %d %08x/%d %d\n", tb->tb_id, r->rtm_type, rta->rta_dst ?
 653        *(u32*)rta->rta_dst : 0, z, rta->rta_oif ? *rta->rta_oif : -1);
 654
 655         if (z > 32)
 656                 return -EINVAL;
 657         if ((fz  = table->fn_zones[z]) == NULL)
 658                 return -ESRCH;
 659
 660         fz_key_0(key);
 661         if (rta->rta_dst) {
 662                 u32 dst;
 663                 memcpy(&dst, rta->rta_dst, 4);
 664                 if (dst & ~FZ_MASK(fz))
 665                         return -EINVAL;
 666                 key = fz_key(dst, fz);
 667         }
 668
 669         fp = fz_chain_p(key, fz);
 670
 671
 672         FIB_SCAN(f, fp) {
 673                 if (fn_key_eq(f->fn_key, key))
 674                         break;
 675                 if (fn_key_leq(key, f->fn_key)) {
 676                         return -ESRCH;
 677                 }
 678         }
 679 #ifdef CONFIG_IP_ROUTE_TOS
 680         FIB_SCAN_KEY(f, fp, key) {
 681                 if (f->fn_tos == tos)
 682                         break;
 683         }
 684 #endif
 685
 686         matched = 0;
 687         del_fp = NULL;
 688         FIB_SCAN_TOS(f, fp, key, tos) {
 689                 struct fib_info * fi = FIB_INFO(f);
 690
 691                 if (f->fn_state&FN_S_ZOMBIE) {
 692                         return -ESRCH;
 693                 }
 694                 matched++;
 695
 696                 if (del_fp == NULL &&
 697                     (!r->rtm_type || f->fn_type == r->rtm_type) &&
 698                     (r->rtm_scope == RT_SCOPE_NOWHERE || f->fn_scope == r->rtm_scope) &&
 699                     (!r->rtm_protocol || fi->fib_protocol == r->rtm_protocol) &&
 700                     fib_nh_match(r, n, rta, fi) == 0)
 701                         del_fp = fp;
 702         }
 703
 704         if (del_fp) {
 705                 f = *del_fp;
 706                 rtmsg_fib(RTM_DELROUTE, f, z, tb->tb_id, n, req);
 707
 708                 if (matched != 1) {
 709                         write_lock_bh(&fib_hash_lock);
 710                         *del_fp = f->fn_next;
 711                         write_unlock_bh(&fib_hash_lock);
 712
 713                         if (f->fn_state&FN_S_ACCESSED)
 714                                 rt_cache_flush(-1);
 715                         fn_free_node(f);
 716                         fz->fz_nent--;
 717                 } else {
 718                         f->fn_state |= FN_S_ZOMBIE;
 719                         if (f->fn_state&FN_S_ACCESSED) {
 720                                 f->fn_state &= ~FN_S_ACCESSED;
 721                                 rt_cache_flush(-1);
 722                         }
 723                         if (++fib_hash_zombies > 128)
 724                                 fib_flush();
 725                 }
 726
 727                 return 0;
 728         }
 729         return -ESRCH;
 730 }
 731
 732 static __inline__ int
 733 fn_flush_list(struct fib_node ** fp, int z, struct fn_hash *table)
 734 {
 735         int found = 0;
 736         struct fib_node *f;
 737
 738         while ((f = *fp) != NULL) {
 739                 struct fib_info *fi = FIB_INFO(f);
 740
 741                 if (fi && ((f->fn_state&FN_S_ZOMBIE) || (fi->fib_flags&RTNH_F_DEAD))) {
 742                         write_lock_bh(&fib_hash_lock);
 743                         *fp = f->fn_next;
 744                         write_unlock_bh(&fib_hash_lock);
 745
 746                         fn_free_node(f);
 747                         found++;
 748                         continue;
 749                 }
 750                 fp = &f->fn_next;
 751         }
 752         return found;
 753 }
 754
 755 static int fn_hash_flush(struct fib_table *tb)
 756 {
 757         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 758         struct fn_zone *fz;
 759         int found = 0;
 760
 761         fib_hash_zombies = 0;
 762         for (fz = table->fn_zone_list; fz; fz = fz->fz_next) {
 763                 int i;
 764                 int tmp = 0;
 765                 for (i=fz->fz_divisor-1; i>=0; i--)
 766                         tmp += fn_flush_list(&fz->fz_hash[i], fz->fz_order, table);
 767                 fz->fz_nent -= tmp;
 768                 found += tmp;
 769         }
 770         return found;
 771 }
 772
 773
 774 static __inline__ int
 775 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
 776                      struct fib_table *tb,
 777                      struct fn_zone *fz,
 778                      struct fib_node *f)
 779 {
 780         int i, s_i;
 781
 782         s_i = cb->args[3];
 783         for (i=0; f; i++, f=f->fn_next) {
 784                 if (i < s_i) continue;
 785                 if (f->fn_state&FN_S_ZOMBIE) continue;
 786                 if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid, cb->nlh->nlmsg_seq,
 787                                   RTM_NEWROUTE,
 788                                   tb->tb_id, (f->fn_state&FN_S_ZOMBIE) ? 0 : f->fn_type, f->fn_scope,
 789                                   &f->fn_key, fz->fz_order, f->fn_tos,
 790                                   f->fn_info) < 0) {
 791                         cb->args[3] = i;
 792                         return -1;
 793                 }
 794         }
 795         cb->args[3] = i;
 796         return skb->len;
 797 }
 798
 799 static __inline__ int
 800 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
 801                    struct fib_table *tb,
 802                    struct fn_zone *fz)
 803 {
 804         int h, s_h;
 805
 806         s_h = cb->args[2];
 807         for (h=0; h < fz->fz_divisor; h++) {
 808                 if (h < s_h) continue;
 809                 if (h > s_h)
 810                         memset(&cb->args[3], 0, sizeof(cb->args) - 3*sizeof(cb->args[0]));
 811                 if (fz->fz_hash == NULL || fz->fz_hash[h] == NULL)
 812                         continue;
 813                 if (fn_hash_dump_bucket(skb, cb, tb, fz, fz->fz_hash[h]) < 0) {
 814                         cb->args[2] = h;
 815                         return -1;
 816                 }
 817         }
 818         cb->args[2] = h;
 819         return skb->len;
 820 }
 821
 822 static int fn_hash_dump(struct fib_table *tb, struct sk_buff *skb, struct netlink_callback *cb)
 823 {
 824         int m, s_m;
 825         struct fn_zone *fz;
 826         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
 827
 828         s_m = cb->args[1];
 829         read_lock(&fib_hash_lock);
 830         for (fz = table->fn_zone_list, m=0; fz; fz = fz->fz_next, m++) {
 831                 if (m < s_m) continue;
 832                 if (m > s_m)
 833                         memset(&cb->args[2], 0, sizeof(cb->args) - 2*sizeof(cb->args[0]));
 834                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
 835                         cb->args[1] = m;
 836                         read_unlock(&fib_hash_lock);
 837                         return -1;
 838                 }
 839         }
 840         read_unlock(&fib_hash_lock);
 841         cb->args[1] = m;
 842         return skb->len;
 843 }
 844
 845 static void rtmsg_fib(int event, struct fib_node* f, int z, int tb_id,
 846                       struct nlmsghdr *n, struct netlink_skb_parms *req)
 847 {
 848         struct sk_buff *skb;
 849         u32 pid = req ? req->pid : 0;
 850         int size = NLMSG_SPACE(sizeof(struct rtmsg)+256);
 851
 852         skb = alloc_skb(size, GFP_KERNEL);
 853         if (!skb)
 854                 return;
 855
 856         if (fib_dump_info(skb, pid, n->nlmsg_seq, event, tb_id,
 857                           f->fn_type, f->fn_scope, &f->fn_key, z, f->fn_tos,
 858                           FIB_INFO(f)) < 0) {
 859                 kfree_skb(skb);
 860                 return;
 861         }
 862         NETLINK_CB(skb).dst_groups = RTMGRP_IPV4_ROUTE;
 863         if (n->nlmsg_flags&NLM_F_ECHO)
 864                 atomic_inc(&skb->users);
 865         netlink_broadcast(rtnl, skb, pid, RTMGRP_IPV4_ROUTE, GFP_KERNEL);
 866         if (n->nlmsg_flags&NLM_F_ECHO)
 867                 netlink_unicast(rtnl, skb, pid, MSG_DONTWAIT);
 868 }
 869
 870 #ifdef CONFIG_IP_MULTIPLE_TABLES
 871 struct fib_table * fib_hash_init(int id)
 872 #else
 873 struct fib_table * __init fib_hash_init(int id)
 874 #endif
 875 {
 876         struct fib_table *tb;
 877
 878         if (fn_hash_kmem == NULL)
 879                 fn_hash_kmem = kmem_cache_create("ip_fib_hash",
 880                                                  sizeof(struct fib_node),
 881                                                  0, SLAB_HWCACHE_ALIGN,
 882                                                  NULL, NULL);
 883
 884         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash), GFP_KERNEL);
 885         if (tb == NULL)
 886                 return NULL;
 887
 888         tb->tb_id = id;
 889         tb->tb_lookup = fn_hash_lookup;
 890         tb->tb_insert = fn_hash_insert;
 891         tb->tb_delete = fn_hash_delete;
 892         tb->tb_flush = fn_hash_flush;
 893         tb->tb_select_default = fn_hash_select_default;
 894         tb->tb_dump = fn_hash_dump;
 895         memset(tb->tb_data, 0, sizeof(struct fn_hash));
 896         return tb;
 897 }
 898
 899 /* ------------------------------------------------------------------------ */
 900 #ifdef CONFIG_PROC_FS
 901
 902 struct fib_iter_state {
 903         struct fn_zone  *zone;
 904         int             bucket;
 905         struct fib_node **hash;
 906         struct fib_node *node;
 907 };
 908
 909 static __inline__ struct fib_node *fib_get_first(struct seq_file *seq)
 910 {
 911         struct fib_iter_state* iter = seq->private;
 912         struct fn_hash *table = (struct fn_hash *)ip_fib_main_table->tb_data;
 913
 914         iter->bucket = 0;
 915         iter->hash   = NULL;
 916         iter->node   = NULL;
 917
 918         for (iter->zone = table->fn_zone_list; iter->zone;
 919              iter->zone = iter->zone->fz_next) {
 920                 int maxslot;
 921 #if 0 //BRCM_BEGIN
 922                 if (!iter->zone->fz_next)
 923                         continue;
 924 #endif //BRCM_END
 925
 926                 iter->hash = iter->zone->fz_hash;
 927                 maxslot = iter->zone->fz_divisor;
 928
 929                 for (iter->bucket = 0; iter->bucket < maxslot;
 930                      ++iter->bucket, ++iter->hash) {
 931                         iter->node = *iter->hash;
 932                         if (iter->node)
 933                                 goto out;
 934                 }
 935         }
 936 out:
 937         return iter->node;
 938 }
 939
 940 static __inline__ struct fib_node *fib_get_next(struct seq_file *seq)
 941 {
 942         struct fib_iter_state* iter = seq->private;
 943
 944         if (iter->node)
 945                 iter->node = iter->node->fn_next;
 946
 947         if (iter->node)
 948                 goto out;
 949
 950         if (!iter->zone)
 951                 goto out;
 952
 953         for (;;) {
 954                 int maxslot;
 955
 956                 maxslot = iter->zone->fz_divisor;
 957
 958                 while (++iter->bucket < maxslot) {
 959                         iter->node = *++iter->hash;
 960
 961                         if (iter->node)
 962                                 goto out;
 963                 }
 964
 965                 iter->zone = iter->zone->fz_next;
 966
 967                 if (!iter->zone)
 968                         goto out;
 969
 970                 iter->hash = iter->zone->fz_hash;
 971                 iter->bucket = 0;
 972                 iter->node = *iter->hash;
 973                 if (iter->node)
 974                         break;
 975         }
 976 out:
 977         return iter->node;
 978 }
 979
 980 static void *fib_seq_start(struct seq_file *seq, loff_t *pos)
 981 {
 982         void *v = NULL;
 983
 984         read_lock(&fib_hash_lock);
 985         if (ip_fib_main_table)
 986                 v = *pos ? fib_get_next(seq) : SEQ_START_TOKEN;
 987         return v;
 988 }
 989
 990 static void *fib_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 991 {
 992         ++*pos;
 993         return v == SEQ_START_TOKEN ? fib_get_first(seq) : fib_get_next(seq);
 994 }
 995
 996 static void fib_seq_stop(struct seq_file *seq, void *v)
 997 {
 998         read_unlock(&fib_hash_lock);
 999 }
1000
1001 static unsigned fib_flag_trans(int type, int dead, u32 mask, struct fib_info *fi)
1002 {
1003         static unsigned type2flags[RTN_MAX + 1] = {
1004                 [7] = RTF_REJECT, [8] = RTF_REJECT,
1005         };
1006         unsigned flags = type2flags[type];
1007
1008         if (fi && fi->fib_nh->nh_gw)
1009                 flags |= RTF_GATEWAY;
1010         if (mask == 0xFFFFFFFF)
1011                 flags |= RTF_HOST;
1012         if (!dead)
1013                 flags |= RTF_UP;
1014         return flags;
1015 }
1016
1017 /*
1018  *      This outputs /proc/net/route.
1019  *
1020  *      It always works in backward compatibility mode.
1021  *      The format of the file is not supposed to be changed.
1022  */
1023 static int fib_seq_show(struct seq_file *seq, void *v)
1024 {
1025         struct fib_iter_state* iter;
1026         char bf[128];
1027         u32 prefix, mask;
1028         unsigned flags;
1029         struct fib_node *f;
1030         struct fib_info *fi;
1031
1032         if (v == SEQ_START_TOKEN) {
1033                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
1034                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
1035                            "\tWindow\tIRTT");
1036                 goto out;
1037         }
1038
1039         f       = v;
1040         fi      = FIB_INFO(f);
1041         iter    = seq->private;
1042         prefix  = fz_prefix(f->fn_key, iter->zone);
1043         mask    = FZ_MASK(iter->zone);
1044         flags   = fib_flag_trans(f->fn_type, f->fn_state & FN_S_ZOMBIE,
1045                                  mask, fi);
1046         if (fi)
1047                 snprintf(bf, sizeof(bf),
1048                          "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1049                          fi->fib_dev ? fi->fib_dev->name : "*", prefix,
1050                          fi->fib_nh->nh_gw, flags, 0, 0, fi->fib_priority,
1051                          mask, (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
1052                          fi->fib_window,
1053                          fi->fib_rtt >> 3);
1054         else
1055                 snprintf(bf, sizeof(bf),
1056                          "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1057                          prefix, 0, flags, 0, 0, 0, mask, 0, 0, 0);
1058         seq_printf(seq, "%-127s\n", bf);
1059 out:
1060         return 0;
1061 }
1062
1063 static struct seq_operations fib_seq_ops = {
1064         .start  = fib_seq_start,
1065         .next   = fib_seq_next,
1066         .stop   = fib_seq_stop,
1067         .show   = fib_seq_show,
1068 };
1069
1070 static int fib_seq_open(struct inode *inode, struct file *file)
1071 {
1072         struct seq_file *seq;
1073         int rc = -ENOMEM;
1074         struct fib_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
1075
1076         if (!s)
1077                 goto out;
1078
1079         rc = seq_open(file, &fib_seq_ops);
1080         if (rc)
1081                 goto out_kfree;
1082
1083         seq          = file->private_data;
1084         seq->private = s;
1085         memset(s, 0, sizeof(*s));
1086 out:
1087         return rc;
1088 out_kfree:
1089         kfree(s);
1090         goto out;
1091 }
1092
1093 static struct file_operations fib_seq_fops = {
1094         .owner          = THIS_MODULE,
1095         .open           = fib_seq_open,
1096         .read           = seq_read,
1097         .llseek         = seq_lseek,
1098         .release        = seq_release_private,
1099 };
1100
1101 int __init fib_proc_init(void)
1102 {
1103         if (!proc_net_fops_create("route", S_IRUGO, &fib_seq_fops))
1104                 return -ENOMEM;
1105         return 0;
1106 }
1107
1108 void __init fib_proc_exit(void)
1109 {
1110         proc_net_remove("route");
1111 }
1112 #endif /* CONFIG_PROC_FS */