Pull fix-cpu-possible-map into release branch
[linux-2.6] / 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 <linux/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/inetdevice.h>
33 #include <linux/netdevice.h>
34 #include <linux/if_arp.h>
35 #include <linux/proc_fs.h>
36 #include <linux/skbuff.h>
37 #include <linux/netlink.h>
38 #include <linux/init.h>
39
40 #include <net/ip.h>
41 #include <net/protocol.h>
42 #include <net/route.h>
43 #include <net/tcp.h>
44 #include <net/sock.h>
45 #include <net/ip_fib.h>
46
47 #include "fib_lookup.h"
48
49 static kmem_cache_t *fn_hash_kmem __read_mostly;
50 static kmem_cache_t *fn_alias_kmem __read_mostly;
51
52 struct fib_node {
53         struct hlist_node       fn_hash;
54         struct list_head        fn_alias;
55         u32                     fn_key;
56 };
57
58 struct fn_zone {
59         struct fn_zone          *fz_next;       /* Next not empty zone  */
60         struct hlist_head       *fz_hash;       /* Hash table pointer   */
61         int                     fz_nent;        /* Number of entries    */
62
63         int                     fz_divisor;     /* Hash divisor         */
64         u32                     fz_hashmask;    /* (fz_divisor - 1)     */
65 #define FZ_HASHMASK(fz)         ((fz)->fz_hashmask)
66
67         int                     fz_order;       /* Zone order           */
68         u32                     fz_mask;
69 #define FZ_MASK(fz)             ((fz)->fz_mask)
70 };
71
72 /* NOTE. On fast computers evaluation of fz_hashmask and fz_mask
73  * can be cheaper than memory lookup, so that FZ_* macros are used.
74  */
75
76 struct fn_hash {
77         struct fn_zone  *fn_zones[33];
78         struct fn_zone  *fn_zone_list;
79 };
80
81 static inline u32 fn_hash(u32 key, struct fn_zone *fz)
82 {
83         u32 h = ntohl(key)>>(32 - fz->fz_order);
84         h ^= (h>>20);
85         h ^= (h>>10);
86         h ^= (h>>5);
87         h &= FZ_HASHMASK(fz);
88         return h;
89 }
90
91 static inline u32 fz_key(u32 dst, struct fn_zone *fz)
92 {
93         return dst & FZ_MASK(fz);
94 }
95
96 static DEFINE_RWLOCK(fib_hash_lock);
97 static unsigned int fib_hash_genid;
98
99 #define FZ_MAX_DIVISOR ((PAGE_SIZE<<MAX_ORDER) / sizeof(struct hlist_head))
100
101 static struct hlist_head *fz_hash_alloc(int divisor)
102 {
103         unsigned long size = divisor * sizeof(struct hlist_head);
104
105         if (size <= PAGE_SIZE) {
106                 return kmalloc(size, GFP_KERNEL);
107         } else {
108                 return (struct hlist_head *)
109                         __get_free_pages(GFP_KERNEL, get_order(size));
110         }
111 }
112
113 /* The fib hash lock must be held when this is called. */
114 static inline void fn_rebuild_zone(struct fn_zone *fz,
115                                    struct hlist_head *old_ht,
116                                    int old_divisor)
117 {
118         int i;
119
120         for (i = 0; i < old_divisor; i++) {
121                 struct hlist_node *node, *n;
122                 struct fib_node *f;
123
124                 hlist_for_each_entry_safe(f, node, n, &old_ht[i], fn_hash) {
125                         struct hlist_head *new_head;
126
127                         hlist_del(&f->fn_hash);
128
129                         new_head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
130                         hlist_add_head(&f->fn_hash, new_head);
131                 }
132         }
133 }
134
135 static void fz_hash_free(struct hlist_head *hash, int divisor)
136 {
137         unsigned long size = divisor * sizeof(struct hlist_head);
138
139         if (size <= PAGE_SIZE)
140                 kfree(hash);
141         else
142                 free_pages((unsigned long)hash, get_order(size));
143 }
144
145 static void fn_rehash_zone(struct fn_zone *fz)
146 {
147         struct hlist_head *ht, *old_ht;
148         int old_divisor, new_divisor;
149         u32 new_hashmask;
150                 
151         old_divisor = fz->fz_divisor;
152
153         switch (old_divisor) {
154         case 16:
155                 new_divisor = 256;
156                 break;
157         case 256:
158                 new_divisor = 1024;
159                 break;
160         default:
161                 if ((old_divisor << 1) > FZ_MAX_DIVISOR) {
162                         printk(KERN_CRIT "route.c: bad divisor %d!\n", old_divisor);
163                         return;
164                 }
165                 new_divisor = (old_divisor << 1);
166                 break;
167         }
168
169         new_hashmask = (new_divisor - 1);
170
171 #if RT_CACHE_DEBUG >= 2
172         printk("fn_rehash_zone: hash for zone %d grows from %d\n", fz->fz_order, old_divisor);
173 #endif
174
175         ht = fz_hash_alloc(new_divisor);
176
177         if (ht) {
178                 memset(ht, 0, new_divisor * sizeof(struct hlist_head));
179
180                 write_lock_bh(&fib_hash_lock);
181                 old_ht = fz->fz_hash;
182                 fz->fz_hash = ht;
183                 fz->fz_hashmask = new_hashmask;
184                 fz->fz_divisor = new_divisor;
185                 fn_rebuild_zone(fz, old_ht, old_divisor);
186                 fib_hash_genid++;
187                 write_unlock_bh(&fib_hash_lock);
188
189                 fz_hash_free(old_ht, old_divisor);
190         }
191 }
192
193 static inline void fn_free_node(struct fib_node * f)
194 {
195         kmem_cache_free(fn_hash_kmem, f);
196 }
197
198 static inline void fn_free_alias(struct fib_alias *fa)
199 {
200         fib_release_info(fa->fa_info);
201         kmem_cache_free(fn_alias_kmem, fa);
202 }
203
204 static struct fn_zone *
205 fn_new_zone(struct fn_hash *table, int z)
206 {
207         int i;
208         struct fn_zone *fz = kmalloc(sizeof(struct fn_zone), GFP_KERNEL);
209         if (!fz)
210                 return NULL;
211
212         memset(fz, 0, sizeof(struct fn_zone));
213         if (z) {
214                 fz->fz_divisor = 16;
215         } else {
216                 fz->fz_divisor = 1;
217         }
218         fz->fz_hashmask = (fz->fz_divisor - 1);
219         fz->fz_hash = fz_hash_alloc(fz->fz_divisor);
220         if (!fz->fz_hash) {
221                 kfree(fz);
222                 return NULL;
223         }
224         memset(fz->fz_hash, 0, fz->fz_divisor * sizeof(struct hlist_head *));
225         fz->fz_order = z;
226         fz->fz_mask = inet_make_mask(z);
227
228         /* Find the first not empty zone with more specific mask */
229         for (i=z+1; i<=32; i++)
230                 if (table->fn_zones[i])
231                         break;
232         write_lock_bh(&fib_hash_lock);
233         if (i>32) {
234                 /* No more specific masks, we are the first. */
235                 fz->fz_next = table->fn_zone_list;
236                 table->fn_zone_list = fz;
237         } else {
238                 fz->fz_next = table->fn_zones[i]->fz_next;
239                 table->fn_zones[i]->fz_next = fz;
240         }
241         table->fn_zones[z] = fz;
242         fib_hash_genid++;
243         write_unlock_bh(&fib_hash_lock);
244         return fz;
245 }
246
247 static int
248 fn_hash_lookup(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
249 {
250         int err;
251         struct fn_zone *fz;
252         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
253
254         read_lock(&fib_hash_lock);
255         for (fz = t->fn_zone_list; fz; fz = fz->fz_next) {
256                 struct hlist_head *head;
257                 struct hlist_node *node;
258                 struct fib_node *f;
259                 u32 k = fz_key(flp->fl4_dst, fz);
260
261                 head = &fz->fz_hash[fn_hash(k, fz)];
262                 hlist_for_each_entry(f, node, head, fn_hash) {
263                         if (f->fn_key != k)
264                                 continue;
265
266                         err = fib_semantic_match(&f->fn_alias,
267                                                  flp, res,
268                                                  f->fn_key, fz->fz_mask,
269                                                  fz->fz_order);
270                         if (err <= 0)
271                                 goto out;
272                 }
273         }
274         err = 1;
275 out:
276         read_unlock(&fib_hash_lock);
277         return err;
278 }
279
280 static int fn_hash_last_dflt=-1;
281
282 static void
283 fn_hash_select_default(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
284 {
285         int order, last_idx;
286         struct hlist_node *node;
287         struct fib_node *f;
288         struct fib_info *fi = NULL;
289         struct fib_info *last_resort;
290         struct fn_hash *t = (struct fn_hash*)tb->tb_data;
291         struct fn_zone *fz = t->fn_zones[0];
292
293         if (fz == NULL)
294                 return;
295
296         last_idx = -1;
297         last_resort = NULL;
298         order = -1;
299
300         read_lock(&fib_hash_lock);
301         hlist_for_each_entry(f, node, &fz->fz_hash[0], fn_hash) {
302                 struct fib_alias *fa;
303
304                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
305                         struct fib_info *next_fi = fa->fa_info;
306
307                         if (fa->fa_scope != res->scope ||
308                             fa->fa_type != RTN_UNICAST)
309                                 continue;
310
311                         if (next_fi->fib_priority > res->fi->fib_priority)
312                                 break;
313                         if (!next_fi->fib_nh[0].nh_gw ||
314                             next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
315                                 continue;
316                         fa->fa_state |= FA_S_ACCESSED;
317
318                         if (fi == NULL) {
319                                 if (next_fi != res->fi)
320                                         break;
321                         } else if (!fib_detect_death(fi, order, &last_resort,
322                                                      &last_idx, &fn_hash_last_dflt)) {
323                                 if (res->fi)
324                                         fib_info_put(res->fi);
325                                 res->fi = fi;
326                                 atomic_inc(&fi->fib_clntref);
327                                 fn_hash_last_dflt = order;
328                                 goto out;
329                         }
330                         fi = next_fi;
331                         order++;
332                 }
333         }
334
335         if (order <= 0 || fi == NULL) {
336                 fn_hash_last_dflt = -1;
337                 goto out;
338         }
339
340         if (!fib_detect_death(fi, order, &last_resort, &last_idx, &fn_hash_last_dflt)) {
341                 if (res->fi)
342                         fib_info_put(res->fi);
343                 res->fi = fi;
344                 atomic_inc(&fi->fib_clntref);
345                 fn_hash_last_dflt = order;
346                 goto out;
347         }
348
349         if (last_idx >= 0) {
350                 if (res->fi)
351                         fib_info_put(res->fi);
352                 res->fi = last_resort;
353                 if (last_resort)
354                         atomic_inc(&last_resort->fib_clntref);
355         }
356         fn_hash_last_dflt = last_idx;
357 out:
358         read_unlock(&fib_hash_lock);
359 }
360
361 /* Insert node F to FZ. */
362 static inline void fib_insert_node(struct fn_zone *fz, struct fib_node *f)
363 {
364         struct hlist_head *head = &fz->fz_hash[fn_hash(f->fn_key, fz)];
365
366         hlist_add_head(&f->fn_hash, head);
367 }
368
369 /* Return the node in FZ matching KEY. */
370 static struct fib_node *fib_find_node(struct fn_zone *fz, u32 key)
371 {
372         struct hlist_head *head = &fz->fz_hash[fn_hash(key, fz)];
373         struct hlist_node *node;
374         struct fib_node *f;
375
376         hlist_for_each_entry(f, node, head, fn_hash) {
377                 if (f->fn_key == key)
378                         return f;
379         }
380
381         return NULL;
382 }
383
384 static int
385 fn_hash_insert(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
386                struct nlmsghdr *n, struct netlink_skb_parms *req)
387 {
388         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
389         struct fib_node *new_f, *f;
390         struct fib_alias *fa, *new_fa;
391         struct fn_zone *fz;
392         struct fib_info *fi;
393         int z = r->rtm_dst_len;
394         int type = r->rtm_type;
395         u8 tos = r->rtm_tos;
396         u32 key;
397         int err;
398
399         if (z > 32)
400                 return -EINVAL;
401         fz = table->fn_zones[z];
402         if (!fz && !(fz = fn_new_zone(table, z)))
403                 return -ENOBUFS;
404
405         key = 0;
406         if (rta->rta_dst) {
407                 u32 dst;
408                 memcpy(&dst, rta->rta_dst, 4);
409                 if (dst & ~FZ_MASK(fz))
410                         return -EINVAL;
411                 key = fz_key(dst, fz);
412         }
413
414         if  ((fi = fib_create_info(r, rta, n, &err)) == NULL)
415                 return err;
416
417         if (fz->fz_nent > (fz->fz_divisor<<1) &&
418             fz->fz_divisor < FZ_MAX_DIVISOR &&
419             (z==32 || (1<<z) > fz->fz_divisor))
420                 fn_rehash_zone(fz);
421
422         f = fib_find_node(fz, key);
423
424         if (!f)
425                 fa = NULL;
426         else
427                 fa = fib_find_alias(&f->fn_alias, tos, fi->fib_priority);
428
429         /* Now fa, if non-NULL, points to the first fib alias
430          * with the same keys [prefix,tos,priority], if such key already
431          * exists or to the node before which we will insert new one.
432          *
433          * If fa is NULL, we will need to allocate a new one and
434          * insert to the head of f.
435          *
436          * If f is NULL, no fib node matched the destination key
437          * and we need to allocate a new one of those as well.
438          */
439
440         if (fa && fa->fa_tos == tos &&
441             fa->fa_info->fib_priority == fi->fib_priority) {
442                 struct fib_alias *fa_orig;
443
444                 err = -EEXIST;
445                 if (n->nlmsg_flags & NLM_F_EXCL)
446                         goto out;
447
448                 if (n->nlmsg_flags & NLM_F_REPLACE) {
449                         struct fib_info *fi_drop;
450                         u8 state;
451
452                         write_lock_bh(&fib_hash_lock);
453                         fi_drop = fa->fa_info;
454                         fa->fa_info = fi;
455                         fa->fa_type = type;
456                         fa->fa_scope = r->rtm_scope;
457                         state = fa->fa_state;
458                         fa->fa_state &= ~FA_S_ACCESSED;
459                         fib_hash_genid++;
460                         write_unlock_bh(&fib_hash_lock);
461
462                         fib_release_info(fi_drop);
463                         if (state & FA_S_ACCESSED)
464                                 rt_cache_flush(-1);
465                         return 0;
466                 }
467
468                 /* Error if we find a perfect match which
469                  * uses the same scope, type, and nexthop
470                  * information.
471                  */
472                 fa_orig = fa;
473                 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
474                 list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
475                         if (fa->fa_tos != tos)
476                                 break;
477                         if (fa->fa_info->fib_priority != fi->fib_priority)
478                                 break;
479                         if (fa->fa_type == type &&
480                             fa->fa_scope == r->rtm_scope &&
481                             fa->fa_info == fi)
482                                 goto out;
483                 }
484                 if (!(n->nlmsg_flags & NLM_F_APPEND))
485                         fa = fa_orig;
486         }
487
488         err = -ENOENT;
489         if (!(n->nlmsg_flags&NLM_F_CREATE))
490                 goto out;
491
492         err = -ENOBUFS;
493         new_fa = kmem_cache_alloc(fn_alias_kmem, SLAB_KERNEL);
494         if (new_fa == NULL)
495                 goto out;
496
497         new_f = NULL;
498         if (!f) {
499                 new_f = kmem_cache_alloc(fn_hash_kmem, SLAB_KERNEL);
500                 if (new_f == NULL)
501                         goto out_free_new_fa;
502
503                 INIT_HLIST_NODE(&new_f->fn_hash);
504                 INIT_LIST_HEAD(&new_f->fn_alias);
505                 new_f->fn_key = key;
506                 f = new_f;
507         }
508
509         new_fa->fa_info = fi;
510         new_fa->fa_tos = tos;
511         new_fa->fa_type = type;
512         new_fa->fa_scope = r->rtm_scope;
513         new_fa->fa_state = 0;
514
515         /*
516          * Insert new entry to the list.
517          */
518
519         write_lock_bh(&fib_hash_lock);
520         if (new_f)
521                 fib_insert_node(fz, new_f);
522         list_add_tail(&new_fa->fa_list,
523                  (fa ? &fa->fa_list : &f->fn_alias));
524         fib_hash_genid++;
525         write_unlock_bh(&fib_hash_lock);
526
527         if (new_f)
528                 fz->fz_nent++;
529         rt_cache_flush(-1);
530
531         rtmsg_fib(RTM_NEWROUTE, key, new_fa, z, tb->tb_id, n, req);
532         return 0;
533
534 out_free_new_fa:
535         kmem_cache_free(fn_alias_kmem, new_fa);
536 out:
537         fib_release_info(fi);
538         return err;
539 }
540
541
542 static int
543 fn_hash_delete(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
544                struct nlmsghdr *n, struct netlink_skb_parms *req)
545 {
546         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
547         struct fib_node *f;
548         struct fib_alias *fa, *fa_to_delete;
549         int z = r->rtm_dst_len;
550         struct fn_zone *fz;
551         u32 key;
552         u8 tos = r->rtm_tos;
553
554         if (z > 32)
555                 return -EINVAL;
556         if ((fz  = table->fn_zones[z]) == NULL)
557                 return -ESRCH;
558
559         key = 0;
560         if (rta->rta_dst) {
561                 u32 dst;
562                 memcpy(&dst, rta->rta_dst, 4);
563                 if (dst & ~FZ_MASK(fz))
564                         return -EINVAL;
565                 key = fz_key(dst, fz);
566         }
567
568         f = fib_find_node(fz, key);
569
570         if (!f)
571                 fa = NULL;
572         else
573                 fa = fib_find_alias(&f->fn_alias, tos, 0);
574         if (!fa)
575                 return -ESRCH;
576
577         fa_to_delete = NULL;
578         fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
579         list_for_each_entry_continue(fa, &f->fn_alias, fa_list) {
580                 struct fib_info *fi = fa->fa_info;
581
582                 if (fa->fa_tos != tos)
583                         break;
584
585                 if ((!r->rtm_type ||
586                      fa->fa_type == r->rtm_type) &&
587                     (r->rtm_scope == RT_SCOPE_NOWHERE ||
588                      fa->fa_scope == r->rtm_scope) &&
589                     (!r->rtm_protocol ||
590                      fi->fib_protocol == r->rtm_protocol) &&
591                     fib_nh_match(r, n, rta, fi) == 0) {
592                         fa_to_delete = fa;
593                         break;
594                 }
595         }
596
597         if (fa_to_delete) {
598                 int kill_fn;
599
600                 fa = fa_to_delete;
601                 rtmsg_fib(RTM_DELROUTE, key, fa, z, tb->tb_id, n, req);
602
603                 kill_fn = 0;
604                 write_lock_bh(&fib_hash_lock);
605                 list_del(&fa->fa_list);
606                 if (list_empty(&f->fn_alias)) {
607                         hlist_del(&f->fn_hash);
608                         kill_fn = 1;
609                 }
610                 fib_hash_genid++;
611                 write_unlock_bh(&fib_hash_lock);
612
613                 if (fa->fa_state & FA_S_ACCESSED)
614                         rt_cache_flush(-1);
615                 fn_free_alias(fa);
616                 if (kill_fn) {
617                         fn_free_node(f);
618                         fz->fz_nent--;
619                 }
620
621                 return 0;
622         }
623         return -ESRCH;
624 }
625
626 static int fn_flush_list(struct fn_zone *fz, int idx)
627 {
628         struct hlist_head *head = &fz->fz_hash[idx];
629         struct hlist_node *node, *n;
630         struct fib_node *f;
631         int found = 0;
632
633         hlist_for_each_entry_safe(f, node, n, head, fn_hash) {
634                 struct fib_alias *fa, *fa_node;
635                 int kill_f;
636
637                 kill_f = 0;
638                 list_for_each_entry_safe(fa, fa_node, &f->fn_alias, fa_list) {
639                         struct fib_info *fi = fa->fa_info;
640
641                         if (fi && (fi->fib_flags&RTNH_F_DEAD)) {
642                                 write_lock_bh(&fib_hash_lock);
643                                 list_del(&fa->fa_list);
644                                 if (list_empty(&f->fn_alias)) {
645                                         hlist_del(&f->fn_hash);
646                                         kill_f = 1;
647                                 }
648                                 fib_hash_genid++;
649                                 write_unlock_bh(&fib_hash_lock);
650
651                                 fn_free_alias(fa);
652                                 found++;
653                         }
654                 }
655                 if (kill_f) {
656                         fn_free_node(f);
657                         fz->fz_nent--;
658                 }
659         }
660         return found;
661 }
662
663 static int fn_hash_flush(struct fib_table *tb)
664 {
665         struct fn_hash *table = (struct fn_hash *) tb->tb_data;
666         struct fn_zone *fz;
667         int found = 0;
668
669         for (fz = table->fn_zone_list; fz; fz = fz->fz_next) {
670                 int i;
671
672                 for (i = fz->fz_divisor - 1; i >= 0; i--)
673                         found += fn_flush_list(fz, i);
674         }
675         return found;
676 }
677
678
679 static inline int
680 fn_hash_dump_bucket(struct sk_buff *skb, struct netlink_callback *cb,
681                      struct fib_table *tb,
682                      struct fn_zone *fz,
683                      struct hlist_head *head)
684 {
685         struct hlist_node *node;
686         struct fib_node *f;
687         int i, s_i;
688
689         s_i = cb->args[3];
690         i = 0;
691         hlist_for_each_entry(f, node, head, fn_hash) {
692                 struct fib_alias *fa;
693
694                 list_for_each_entry(fa, &f->fn_alias, fa_list) {
695                         if (i < s_i)
696                                 goto next;
697
698                         if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
699                                           cb->nlh->nlmsg_seq,
700                                           RTM_NEWROUTE,
701                                           tb->tb_id,
702                                           fa->fa_type,
703                                           fa->fa_scope,
704                                           &f->fn_key,
705                                           fz->fz_order,
706                                           fa->fa_tos,
707                                           fa->fa_info,
708                                           NLM_F_MULTI) < 0) {
709                                 cb->args[3] = i;
710                                 return -1;
711                         }
712                 next:
713                         i++;
714                 }
715         }
716         cb->args[3] = i;
717         return skb->len;
718 }
719
720 static inline int
721 fn_hash_dump_zone(struct sk_buff *skb, struct netlink_callback *cb,
722                    struct fib_table *tb,
723                    struct fn_zone *fz)
724 {
725         int h, s_h;
726
727         s_h = cb->args[2];
728         for (h=0; h < fz->fz_divisor; h++) {
729                 if (h < s_h) continue;
730                 if (h > s_h)
731                         memset(&cb->args[3], 0,
732                                sizeof(cb->args) - 3*sizeof(cb->args[0]));
733                 if (fz->fz_hash == NULL ||
734                     hlist_empty(&fz->fz_hash[h]))
735                         continue;
736                 if (fn_hash_dump_bucket(skb, cb, tb, fz, &fz->fz_hash[h])<0) {
737                         cb->args[2] = h;
738                         return -1;
739                 }
740         }
741         cb->args[2] = h;
742         return skb->len;
743 }
744
745 static int fn_hash_dump(struct fib_table *tb, struct sk_buff *skb, struct netlink_callback *cb)
746 {
747         int m, s_m;
748         struct fn_zone *fz;
749         struct fn_hash *table = (struct fn_hash*)tb->tb_data;
750
751         s_m = cb->args[1];
752         read_lock(&fib_hash_lock);
753         for (fz = table->fn_zone_list, m=0; fz; fz = fz->fz_next, m++) {
754                 if (m < s_m) continue;
755                 if (m > s_m)
756                         memset(&cb->args[2], 0,
757                                sizeof(cb->args) - 2*sizeof(cb->args[0]));
758                 if (fn_hash_dump_zone(skb, cb, tb, fz) < 0) {
759                         cb->args[1] = m;
760                         read_unlock(&fib_hash_lock);
761                         return -1;
762                 }
763         }
764         read_unlock(&fib_hash_lock);
765         cb->args[1] = m;
766         return skb->len;
767 }
768
769 #ifdef CONFIG_IP_MULTIPLE_TABLES
770 struct fib_table * fib_hash_init(int id)
771 #else
772 struct fib_table * __init fib_hash_init(int id)
773 #endif
774 {
775         struct fib_table *tb;
776
777         if (fn_hash_kmem == NULL)
778                 fn_hash_kmem = kmem_cache_create("ip_fib_hash",
779                                                  sizeof(struct fib_node),
780                                                  0, SLAB_HWCACHE_ALIGN,
781                                                  NULL, NULL);
782
783         if (fn_alias_kmem == NULL)
784                 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
785                                                   sizeof(struct fib_alias),
786                                                   0, SLAB_HWCACHE_ALIGN,
787                                                   NULL, NULL);
788
789         tb = kmalloc(sizeof(struct fib_table) + sizeof(struct fn_hash),
790                      GFP_KERNEL);
791         if (tb == NULL)
792                 return NULL;
793
794         tb->tb_id = id;
795         tb->tb_lookup = fn_hash_lookup;
796         tb->tb_insert = fn_hash_insert;
797         tb->tb_delete = fn_hash_delete;
798         tb->tb_flush = fn_hash_flush;
799         tb->tb_select_default = fn_hash_select_default;
800         tb->tb_dump = fn_hash_dump;
801         memset(tb->tb_data, 0, sizeof(struct fn_hash));
802         return tb;
803 }
804
805 /* ------------------------------------------------------------------------ */
806 #ifdef CONFIG_PROC_FS
807
808 struct fib_iter_state {
809         struct fn_zone  *zone;
810         int             bucket;
811         struct hlist_head *hash_head;
812         struct fib_node *fn;
813         struct fib_alias *fa;
814         loff_t pos;
815         unsigned int genid;
816         int valid;
817 };
818
819 static struct fib_alias *fib_get_first(struct seq_file *seq)
820 {
821         struct fib_iter_state *iter = seq->private;
822         struct fn_hash *table = (struct fn_hash *) ip_fib_main_table->tb_data;
823
824         iter->bucket    = 0;
825         iter->hash_head = NULL;
826         iter->fn        = NULL;
827         iter->fa        = NULL;
828         iter->pos       = 0;
829         iter->genid     = fib_hash_genid;
830         iter->valid     = 1;
831
832         for (iter->zone = table->fn_zone_list; iter->zone;
833              iter->zone = iter->zone->fz_next) {
834                 int maxslot;
835
836                 if (!iter->zone->fz_nent)
837                         continue;
838
839                 iter->hash_head = iter->zone->fz_hash;
840                 maxslot = iter->zone->fz_divisor;
841
842                 for (iter->bucket = 0; iter->bucket < maxslot;
843                      ++iter->bucket, ++iter->hash_head) {
844                         struct hlist_node *node;
845                         struct fib_node *fn;
846
847                         hlist_for_each_entry(fn,node,iter->hash_head,fn_hash) {
848                                 struct fib_alias *fa;
849
850                                 list_for_each_entry(fa,&fn->fn_alias,fa_list) {
851                                         iter->fn = fn;
852                                         iter->fa = fa;
853                                         goto out;
854                                 }
855                         }
856                 }
857         }
858 out:
859         return iter->fa;
860 }
861
862 static struct fib_alias *fib_get_next(struct seq_file *seq)
863 {
864         struct fib_iter_state *iter = seq->private;
865         struct fib_node *fn;
866         struct fib_alias *fa;
867
868         /* Advance FA, if any. */
869         fn = iter->fn;
870         fa = iter->fa;
871         if (fa) {
872                 BUG_ON(!fn);
873                 list_for_each_entry_continue(fa, &fn->fn_alias, fa_list) {
874                         iter->fa = fa;
875                         goto out;
876                 }
877         }
878
879         fa = iter->fa = NULL;
880
881         /* Advance FN. */
882         if (fn) {
883                 struct hlist_node *node = &fn->fn_hash;
884                 hlist_for_each_entry_continue(fn, node, fn_hash) {
885                         iter->fn = fn;
886
887                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
888                                 iter->fa = fa;
889                                 goto out;
890                         }
891                 }
892         }
893
894         fn = iter->fn = NULL;
895
896         /* Advance hash chain. */
897         if (!iter->zone)
898                 goto out;
899
900         for (;;) {
901                 struct hlist_node *node;
902                 int maxslot;
903
904                 maxslot = iter->zone->fz_divisor;
905
906                 while (++iter->bucket < maxslot) {
907                         iter->hash_head++;
908
909                         hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
910                                 list_for_each_entry(fa, &fn->fn_alias, fa_list) {
911                                         iter->fn = fn;
912                                         iter->fa = fa;
913                                         goto out;
914                                 }
915                         }
916                 }
917
918                 iter->zone = iter->zone->fz_next;
919
920                 if (!iter->zone)
921                         goto out;
922                 
923                 iter->bucket = 0;
924                 iter->hash_head = iter->zone->fz_hash;
925
926                 hlist_for_each_entry(fn, node, iter->hash_head, fn_hash) {
927                         list_for_each_entry(fa, &fn->fn_alias, fa_list) {
928                                 iter->fn = fn;
929                                 iter->fa = fa;
930                                 goto out;
931                         }
932                 }
933         }
934 out:
935         iter->pos++;
936         return fa;
937 }
938
939 static struct fib_alias *fib_get_idx(struct seq_file *seq, loff_t pos)
940 {
941         struct fib_iter_state *iter = seq->private;
942         struct fib_alias *fa;
943         
944         if (iter->valid && pos >= iter->pos && iter->genid == fib_hash_genid) {
945                 fa   = iter->fa;
946                 pos -= iter->pos;
947         } else
948                 fa = fib_get_first(seq);
949
950         if (fa)
951                 while (pos && (fa = fib_get_next(seq)))
952                         --pos;
953         return pos ? NULL : fa;
954 }
955
956 static void *fib_seq_start(struct seq_file *seq, loff_t *pos)
957 {
958         void *v = NULL;
959
960         read_lock(&fib_hash_lock);
961         if (ip_fib_main_table)
962                 v = *pos ? fib_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
963         return v;
964 }
965
966 static void *fib_seq_next(struct seq_file *seq, void *v, loff_t *pos)
967 {
968         ++*pos;
969         return v == SEQ_START_TOKEN ? fib_get_first(seq) : fib_get_next(seq);
970 }
971
972 static void fib_seq_stop(struct seq_file *seq, void *v)
973 {
974         read_unlock(&fib_hash_lock);
975 }
976
977 static unsigned fib_flag_trans(int type, u32 mask, struct fib_info *fi)
978 {
979         static const unsigned type2flags[RTN_MAX + 1] = {
980                 [7] = RTF_REJECT, [8] = RTF_REJECT,
981         };
982         unsigned flags = type2flags[type];
983
984         if (fi && fi->fib_nh->nh_gw)
985                 flags |= RTF_GATEWAY;
986         if (mask == 0xFFFFFFFF)
987                 flags |= RTF_HOST;
988         flags |= RTF_UP;
989         return flags;
990 }
991
992 /* 
993  *      This outputs /proc/net/route.
994  *
995  *      It always works in backward compatibility mode.
996  *      The format of the file is not supposed to be changed.
997  */
998 static int fib_seq_show(struct seq_file *seq, void *v)
999 {
1000         struct fib_iter_state *iter;
1001         char bf[128];
1002         u32 prefix, mask;
1003         unsigned flags;
1004         struct fib_node *f;
1005         struct fib_alias *fa;
1006         struct fib_info *fi;
1007
1008         if (v == SEQ_START_TOKEN) {
1009                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
1010                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
1011                            "\tWindow\tIRTT");
1012                 goto out;
1013         }
1014
1015         iter    = seq->private;
1016         f       = iter->fn;
1017         fa      = iter->fa;
1018         fi      = fa->fa_info;
1019         prefix  = f->fn_key;
1020         mask    = FZ_MASK(iter->zone);
1021         flags   = fib_flag_trans(fa->fa_type, mask, fi);
1022         if (fi)
1023                 snprintf(bf, sizeof(bf),
1024                          "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1025                          fi->fib_dev ? fi->fib_dev->name : "*", prefix,
1026                          fi->fib_nh->nh_gw, flags, 0, 0, fi->fib_priority,
1027                          mask, (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
1028                          fi->fib_window,
1029                          fi->fib_rtt >> 3);
1030         else
1031                 snprintf(bf, sizeof(bf),
1032                          "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
1033                          prefix, 0, flags, 0, 0, 0, mask, 0, 0, 0);
1034         seq_printf(seq, "%-127s\n", bf);
1035 out:
1036         return 0;
1037 }
1038
1039 static struct seq_operations fib_seq_ops = {
1040         .start  = fib_seq_start,
1041         .next   = fib_seq_next,
1042         .stop   = fib_seq_stop,
1043         .show   = fib_seq_show,
1044 };
1045
1046 static int fib_seq_open(struct inode *inode, struct file *file)
1047 {
1048         struct seq_file *seq;
1049         int rc = -ENOMEM;
1050         struct fib_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
1051        
1052         if (!s)
1053                 goto out;
1054
1055         rc = seq_open(file, &fib_seq_ops);
1056         if (rc)
1057                 goto out_kfree;
1058
1059         seq          = file->private_data;
1060         seq->private = s;
1061         memset(s, 0, sizeof(*s));
1062 out:
1063         return rc;
1064 out_kfree:
1065         kfree(s);
1066         goto out;
1067 }
1068
1069 static struct file_operations fib_seq_fops = {
1070         .owner          = THIS_MODULE,
1071         .open           = fib_seq_open,
1072         .read           = seq_read,
1073         .llseek         = seq_lseek,
1074         .release        = seq_release_private,
1075 };
1076
1077 int __init fib_proc_init(void)
1078 {
1079         if (!proc_net_fops_create("route", S_IRUGO, &fib_seq_fops))
1080                 return -ENOMEM;
1081         return 0;
1082 }
1083
1084 void __init fib_proc_exit(void)
1085 {
1086         proc_net_remove("route");
1087 }
1088 #endif /* CONFIG_PROC_FS */