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