Merge with /home/shaggy/git/linus-clean/
[pandora-kernel.git] / net / xfrm / xfrm_policy.c
1 /* 
2  * xfrm_policy.c
3  *
4  * Changes:
5  *      Mitsuru KANDA @USAGI
6  *      Kazunori MIYAZAWA @USAGI
7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
8  *              IPv6 support
9  *      Kazunori MIYAZAWA @USAGI
10  *      YOSHIFUJI Hideaki
11  *              Split up af-specific portion
12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
13  *      
14  */
15
16 #include <asm/bug.h>
17 #include <linux/config.h>
18 #include <linux/slab.h>
19 #include <linux/kmod.h>
20 #include <linux/list.h>
21 #include <linux/spinlock.h>
22 #include <linux/workqueue.h>
23 #include <linux/notifier.h>
24 #include <linux/netdevice.h>
25 #include <linux/module.h>
26 #include <net/xfrm.h>
27 #include <net/ip.h>
28
29 DECLARE_MUTEX(xfrm_cfg_sem);
30 EXPORT_SYMBOL(xfrm_cfg_sem);
31
32 static DEFINE_RWLOCK(xfrm_policy_lock);
33
34 struct xfrm_policy *xfrm_policy_list[XFRM_POLICY_MAX*2];
35 EXPORT_SYMBOL(xfrm_policy_list);
36
37 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
38 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
39
40 static kmem_cache_t *xfrm_dst_cache;
41
42 static struct work_struct xfrm_policy_gc_work;
43 static struct list_head xfrm_policy_gc_list =
44         LIST_HEAD_INIT(xfrm_policy_gc_list);
45 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
46
47 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
48 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
49
50 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
51 {
52         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
53         struct xfrm_type_map *typemap;
54         int err = 0;
55
56         if (unlikely(afinfo == NULL))
57                 return -EAFNOSUPPORT;
58         typemap = afinfo->type_map;
59
60         write_lock(&typemap->lock);
61         if (likely(typemap->map[type->proto] == NULL))
62                 typemap->map[type->proto] = type;
63         else
64                 err = -EEXIST;
65         write_unlock(&typemap->lock);
66         xfrm_policy_put_afinfo(afinfo);
67         return err;
68 }
69 EXPORT_SYMBOL(xfrm_register_type);
70
71 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
72 {
73         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
74         struct xfrm_type_map *typemap;
75         int err = 0;
76
77         if (unlikely(afinfo == NULL))
78                 return -EAFNOSUPPORT;
79         typemap = afinfo->type_map;
80
81         write_lock(&typemap->lock);
82         if (unlikely(typemap->map[type->proto] != type))
83                 err = -ENOENT;
84         else
85                 typemap->map[type->proto] = NULL;
86         write_unlock(&typemap->lock);
87         xfrm_policy_put_afinfo(afinfo);
88         return err;
89 }
90 EXPORT_SYMBOL(xfrm_unregister_type);
91
92 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
93 {
94         struct xfrm_policy_afinfo *afinfo;
95         struct xfrm_type_map *typemap;
96         struct xfrm_type *type;
97         int modload_attempted = 0;
98
99 retry:
100         afinfo = xfrm_policy_get_afinfo(family);
101         if (unlikely(afinfo == NULL))
102                 return NULL;
103         typemap = afinfo->type_map;
104
105         read_lock(&typemap->lock);
106         type = typemap->map[proto];
107         if (unlikely(type && !try_module_get(type->owner)))
108                 type = NULL;
109         read_unlock(&typemap->lock);
110         if (!type && !modload_attempted) {
111                 xfrm_policy_put_afinfo(afinfo);
112                 request_module("xfrm-type-%d-%d",
113                                (int) family, (int) proto);
114                 modload_attempted = 1;
115                 goto retry;
116         }
117
118         xfrm_policy_put_afinfo(afinfo);
119         return type;
120 }
121 EXPORT_SYMBOL(xfrm_get_type);
122
123 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl, 
124                     unsigned short family)
125 {
126         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
127         int err = 0;
128
129         if (unlikely(afinfo == NULL))
130                 return -EAFNOSUPPORT;
131
132         if (likely(afinfo->dst_lookup != NULL))
133                 err = afinfo->dst_lookup(dst, fl);
134         else
135                 err = -EINVAL;
136         xfrm_policy_put_afinfo(afinfo);
137         return err;
138 }
139 EXPORT_SYMBOL(xfrm_dst_lookup);
140
141 void xfrm_put_type(struct xfrm_type *type)
142 {
143         module_put(type->owner);
144 }
145
146 static inline unsigned long make_jiffies(long secs)
147 {
148         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
149                 return MAX_SCHEDULE_TIMEOUT-1;
150         else
151                 return secs*HZ;
152 }
153
154 static void xfrm_policy_timer(unsigned long data)
155 {
156         struct xfrm_policy *xp = (struct xfrm_policy*)data;
157         unsigned long now = (unsigned long)xtime.tv_sec;
158         long next = LONG_MAX;
159         int warn = 0;
160         int dir;
161
162         read_lock(&xp->lock);
163
164         if (xp->dead)
165                 goto out;
166
167         dir = xp->index & 7;
168
169         if (xp->lft.hard_add_expires_seconds) {
170                 long tmo = xp->lft.hard_add_expires_seconds +
171                         xp->curlft.add_time - now;
172                 if (tmo <= 0)
173                         goto expired;
174                 if (tmo < next)
175                         next = tmo;
176         }
177         if (xp->lft.hard_use_expires_seconds) {
178                 long tmo = xp->lft.hard_use_expires_seconds +
179                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
180                 if (tmo <= 0)
181                         goto expired;
182                 if (tmo < next)
183                         next = tmo;
184         }
185         if (xp->lft.soft_add_expires_seconds) {
186                 long tmo = xp->lft.soft_add_expires_seconds +
187                         xp->curlft.add_time - now;
188                 if (tmo <= 0) {
189                         warn = 1;
190                         tmo = XFRM_KM_TIMEOUT;
191                 }
192                 if (tmo < next)
193                         next = tmo;
194         }
195         if (xp->lft.soft_use_expires_seconds) {
196                 long tmo = xp->lft.soft_use_expires_seconds +
197                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
198                 if (tmo <= 0) {
199                         warn = 1;
200                         tmo = XFRM_KM_TIMEOUT;
201                 }
202                 if (tmo < next)
203                         next = tmo;
204         }
205
206         if (warn)
207                 km_policy_expired(xp, dir, 0);
208         if (next != LONG_MAX &&
209             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
210                 xfrm_pol_hold(xp);
211
212 out:
213         read_unlock(&xp->lock);
214         xfrm_pol_put(xp);
215         return;
216
217 expired:
218         read_unlock(&xp->lock);
219         if (!xfrm_policy_delete(xp, dir))
220                 km_policy_expired(xp, dir, 1);
221         xfrm_pol_put(xp);
222 }
223
224
225 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
226  * SPD calls.
227  */
228
229 struct xfrm_policy *xfrm_policy_alloc(int gfp)
230 {
231         struct xfrm_policy *policy;
232
233         policy = kmalloc(sizeof(struct xfrm_policy), gfp);
234
235         if (policy) {
236                 memset(policy, 0, sizeof(struct xfrm_policy));
237                 atomic_set(&policy->refcnt, 1);
238                 rwlock_init(&policy->lock);
239                 init_timer(&policy->timer);
240                 policy->timer.data = (unsigned long)policy;
241                 policy->timer.function = xfrm_policy_timer;
242         }
243         return policy;
244 }
245 EXPORT_SYMBOL(xfrm_policy_alloc);
246
247 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
248
249 void __xfrm_policy_destroy(struct xfrm_policy *policy)
250 {
251         if (!policy->dead)
252                 BUG();
253
254         if (policy->bundles)
255                 BUG();
256
257         if (del_timer(&policy->timer))
258                 BUG();
259
260         kfree(policy);
261 }
262 EXPORT_SYMBOL(__xfrm_policy_destroy);
263
264 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
265 {
266         struct dst_entry *dst;
267
268         while ((dst = policy->bundles) != NULL) {
269                 policy->bundles = dst->next;
270                 dst_free(dst);
271         }
272
273         if (del_timer(&policy->timer))
274                 atomic_dec(&policy->refcnt);
275
276         if (atomic_read(&policy->refcnt) > 1)
277                 flow_cache_flush();
278
279         xfrm_pol_put(policy);
280 }
281
282 static void xfrm_policy_gc_task(void *data)
283 {
284         struct xfrm_policy *policy;
285         struct list_head *entry, *tmp;
286         struct list_head gc_list = LIST_HEAD_INIT(gc_list);
287
288         spin_lock_bh(&xfrm_policy_gc_lock);
289         list_splice_init(&xfrm_policy_gc_list, &gc_list);
290         spin_unlock_bh(&xfrm_policy_gc_lock);
291
292         list_for_each_safe(entry, tmp, &gc_list) {
293                 policy = list_entry(entry, struct xfrm_policy, list);
294                 xfrm_policy_gc_kill(policy);
295         }
296 }
297
298 /* Rule must be locked. Release descentant resources, announce
299  * entry dead. The rule must be unlinked from lists to the moment.
300  */
301
302 static void xfrm_policy_kill(struct xfrm_policy *policy)
303 {
304         int dead;
305
306         write_lock_bh(&policy->lock);
307         dead = policy->dead;
308         policy->dead = 1;
309         write_unlock_bh(&policy->lock);
310
311         if (unlikely(dead)) {
312                 WARN_ON(1);
313                 return;
314         }
315
316         spin_lock(&xfrm_policy_gc_lock);
317         list_add(&policy->list, &xfrm_policy_gc_list);
318         spin_unlock(&xfrm_policy_gc_lock);
319
320         schedule_work(&xfrm_policy_gc_work);
321 }
322
323 /* Generate new index... KAME seems to generate them ordered by cost
324  * of an absolute inpredictability of ordering of rules. This will not pass. */
325 static u32 xfrm_gen_index(int dir)
326 {
327         u32 idx;
328         struct xfrm_policy *p;
329         static u32 idx_generator;
330
331         for (;;) {
332                 idx = (idx_generator | dir);
333                 idx_generator += 8;
334                 if (idx == 0)
335                         idx = 8;
336                 for (p = xfrm_policy_list[dir]; p; p = p->next) {
337                         if (p->index == idx)
338                                 break;
339                 }
340                 if (!p)
341                         return idx;
342         }
343 }
344
345 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
346 {
347         struct xfrm_policy *pol, **p;
348         struct xfrm_policy *delpol = NULL;
349         struct xfrm_policy **newpos = NULL;
350
351         write_lock_bh(&xfrm_policy_lock);
352         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL;) {
353                 if (!delpol && memcmp(&policy->selector, &pol->selector, sizeof(pol->selector)) == 0) {
354                         if (excl) {
355                                 write_unlock_bh(&xfrm_policy_lock);
356                                 return -EEXIST;
357                         }
358                         *p = pol->next;
359                         delpol = pol;
360                         if (policy->priority > pol->priority)
361                                 continue;
362                 } else if (policy->priority >= pol->priority) {
363                         p = &pol->next;
364                         continue;
365                 }
366                 if (!newpos)
367                         newpos = p;
368                 if (delpol)
369                         break;
370                 p = &pol->next;
371         }
372         if (newpos)
373                 p = newpos;
374         xfrm_pol_hold(policy);
375         policy->next = *p;
376         *p = policy;
377         atomic_inc(&flow_cache_genid);
378         policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
379         policy->curlft.add_time = (unsigned long)xtime.tv_sec;
380         policy->curlft.use_time = 0;
381         if (!mod_timer(&policy->timer, jiffies + HZ))
382                 xfrm_pol_hold(policy);
383         write_unlock_bh(&xfrm_policy_lock);
384
385         if (delpol) {
386                 xfrm_policy_kill(delpol);
387         }
388         return 0;
389 }
390 EXPORT_SYMBOL(xfrm_policy_insert);
391
392 struct xfrm_policy *xfrm_policy_bysel(int dir, struct xfrm_selector *sel,
393                                       int delete)
394 {
395         struct xfrm_policy *pol, **p;
396
397         write_lock_bh(&xfrm_policy_lock);
398         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
399                 if (memcmp(sel, &pol->selector, sizeof(*sel)) == 0) {
400                         xfrm_pol_hold(pol);
401                         if (delete)
402                                 *p = pol->next;
403                         break;
404                 }
405         }
406         write_unlock_bh(&xfrm_policy_lock);
407
408         if (pol && delete) {
409                 atomic_inc(&flow_cache_genid);
410                 xfrm_policy_kill(pol);
411         }
412         return pol;
413 }
414 EXPORT_SYMBOL(xfrm_policy_bysel);
415
416 struct xfrm_policy *xfrm_policy_byid(int dir, u32 id, int delete)
417 {
418         struct xfrm_policy *pol, **p;
419
420         write_lock_bh(&xfrm_policy_lock);
421         for (p = &xfrm_policy_list[id & 7]; (pol=*p)!=NULL; p = &pol->next) {
422                 if (pol->index == id) {
423                         xfrm_pol_hold(pol);
424                         if (delete)
425                                 *p = pol->next;
426                         break;
427                 }
428         }
429         write_unlock_bh(&xfrm_policy_lock);
430
431         if (pol && delete) {
432                 atomic_inc(&flow_cache_genid);
433                 xfrm_policy_kill(pol);
434         }
435         return pol;
436 }
437 EXPORT_SYMBOL(xfrm_policy_byid);
438
439 void xfrm_policy_flush(void)
440 {
441         struct xfrm_policy *xp;
442         int dir;
443
444         write_lock_bh(&xfrm_policy_lock);
445         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
446                 while ((xp = xfrm_policy_list[dir]) != NULL) {
447                         xfrm_policy_list[dir] = xp->next;
448                         write_unlock_bh(&xfrm_policy_lock);
449
450                         xfrm_policy_kill(xp);
451
452                         write_lock_bh(&xfrm_policy_lock);
453                 }
454         }
455         atomic_inc(&flow_cache_genid);
456         write_unlock_bh(&xfrm_policy_lock);
457 }
458 EXPORT_SYMBOL(xfrm_policy_flush);
459
460 int xfrm_policy_walk(int (*func)(struct xfrm_policy *, int, int, void*),
461                      void *data)
462 {
463         struct xfrm_policy *xp;
464         int dir;
465         int count = 0;
466         int error = 0;
467
468         read_lock_bh(&xfrm_policy_lock);
469         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
470                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next)
471                         count++;
472         }
473
474         if (count == 0) {
475                 error = -ENOENT;
476                 goto out;
477         }
478
479         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
480                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next) {
481                         error = func(xp, dir%XFRM_POLICY_MAX, --count, data);
482                         if (error)
483                                 goto out;
484                 }
485         }
486
487 out:
488         read_unlock_bh(&xfrm_policy_lock);
489         return error;
490 }
491 EXPORT_SYMBOL(xfrm_policy_walk);
492
493 /* Find policy to apply to this flow. */
494
495 static void xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
496                                void **objp, atomic_t **obj_refp)
497 {
498         struct xfrm_policy *pol;
499
500         read_lock_bh(&xfrm_policy_lock);
501         for (pol = xfrm_policy_list[dir]; pol; pol = pol->next) {
502                 struct xfrm_selector *sel = &pol->selector;
503                 int match;
504
505                 if (pol->family != family)
506                         continue;
507
508                 match = xfrm_selector_match(sel, fl, family);
509                 if (match) {
510                         xfrm_pol_hold(pol);
511                         break;
512                 }
513         }
514         read_unlock_bh(&xfrm_policy_lock);
515         if ((*objp = (void *) pol) != NULL)
516                 *obj_refp = &pol->refcnt;
517 }
518
519 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
520 {
521         struct xfrm_policy *pol;
522
523         read_lock_bh(&xfrm_policy_lock);
524         if ((pol = sk->sk_policy[dir]) != NULL) {
525                 int match = xfrm_selector_match(&pol->selector, fl,
526                                                 sk->sk_family);
527                 if (match)
528                         xfrm_pol_hold(pol);
529                 else
530                         pol = NULL;
531         }
532         read_unlock_bh(&xfrm_policy_lock);
533         return pol;
534 }
535
536 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
537 {
538         pol->next = xfrm_policy_list[dir];
539         xfrm_policy_list[dir] = pol;
540         xfrm_pol_hold(pol);
541 }
542
543 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
544                                                 int dir)
545 {
546         struct xfrm_policy **polp;
547
548         for (polp = &xfrm_policy_list[dir];
549              *polp != NULL; polp = &(*polp)->next) {
550                 if (*polp == pol) {
551                         *polp = pol->next;
552                         return pol;
553                 }
554         }
555         return NULL;
556 }
557
558 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
559 {
560         write_lock_bh(&xfrm_policy_lock);
561         pol = __xfrm_policy_unlink(pol, dir);
562         write_unlock_bh(&xfrm_policy_lock);
563         if (pol) {
564                 if (dir < XFRM_POLICY_MAX)
565                         atomic_inc(&flow_cache_genid);
566                 xfrm_policy_kill(pol);
567                 return 0;
568         }
569         return -ENOENT;
570 }
571
572 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
573 {
574         struct xfrm_policy *old_pol;
575
576         write_lock_bh(&xfrm_policy_lock);
577         old_pol = sk->sk_policy[dir];
578         sk->sk_policy[dir] = pol;
579         if (pol) {
580                 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
581                 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
582                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
583         }
584         if (old_pol)
585                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
586         write_unlock_bh(&xfrm_policy_lock);
587
588         if (old_pol) {
589                 xfrm_policy_kill(old_pol);
590         }
591         return 0;
592 }
593
594 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
595 {
596         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
597
598         if (newp) {
599                 newp->selector = old->selector;
600                 newp->lft = old->lft;
601                 newp->curlft = old->curlft;
602                 newp->action = old->action;
603                 newp->flags = old->flags;
604                 newp->xfrm_nr = old->xfrm_nr;
605                 newp->index = old->index;
606                 memcpy(newp->xfrm_vec, old->xfrm_vec,
607                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
608                 write_lock_bh(&xfrm_policy_lock);
609                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
610                 write_unlock_bh(&xfrm_policy_lock);
611                 xfrm_pol_put(newp);
612         }
613         return newp;
614 }
615
616 int __xfrm_sk_clone_policy(struct sock *sk)
617 {
618         struct xfrm_policy *p0 = sk->sk_policy[0],
619                            *p1 = sk->sk_policy[1];
620
621         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
622         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
623                 return -ENOMEM;
624         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
625                 return -ENOMEM;
626         return 0;
627 }
628
629 /* Resolve list of templates for the flow, given policy. */
630
631 static int
632 xfrm_tmpl_resolve(struct xfrm_policy *policy, struct flowi *fl,
633                   struct xfrm_state **xfrm,
634                   unsigned short family)
635 {
636         int nx;
637         int i, error;
638         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
639         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
640
641         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
642                 struct xfrm_state *x;
643                 xfrm_address_t *remote = daddr;
644                 xfrm_address_t *local  = saddr;
645                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
646
647                 if (tmpl->mode) {
648                         remote = &tmpl->id.daddr;
649                         local = &tmpl->saddr;
650                 }
651
652                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
653
654                 if (x && x->km.state == XFRM_STATE_VALID) {
655                         xfrm[nx++] = x;
656                         daddr = remote;
657                         saddr = local;
658                         continue;
659                 }
660                 if (x) {
661                         error = (x->km.state == XFRM_STATE_ERROR ?
662                                  -EINVAL : -EAGAIN);
663                         xfrm_state_put(x);
664                 }
665
666                 if (!tmpl->optional)
667                         goto fail;
668         }
669         return nx;
670
671 fail:
672         for (nx--; nx>=0; nx--)
673                 xfrm_state_put(xfrm[nx]);
674         return error;
675 }
676
677 /* Check that the bundle accepts the flow and its components are
678  * still valid.
679  */
680
681 static struct dst_entry *
682 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
683 {
684         struct dst_entry *x;
685         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
686         if (unlikely(afinfo == NULL))
687                 return ERR_PTR(-EINVAL);
688         x = afinfo->find_bundle(fl, policy);
689         xfrm_policy_put_afinfo(afinfo);
690         return x;
691 }
692
693 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
694  * all the metrics... Shortly, bundle a bundle.
695  */
696
697 static int
698 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
699                    struct flowi *fl, struct dst_entry **dst_p,
700                    unsigned short family)
701 {
702         int err;
703         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
704         if (unlikely(afinfo == NULL))
705                 return -EINVAL;
706         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
707         xfrm_policy_put_afinfo(afinfo);
708         return err;
709 }
710
711 static inline int policy_to_flow_dir(int dir)
712 {
713         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
714             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
715             XFRM_POLICY_FWD == FLOW_DIR_FWD)
716                 return dir;
717         switch (dir) {
718         default:
719         case XFRM_POLICY_IN:
720                 return FLOW_DIR_IN;
721         case XFRM_POLICY_OUT:
722                 return FLOW_DIR_OUT;
723         case XFRM_POLICY_FWD:
724                 return FLOW_DIR_FWD;
725         };
726 }
727
728 static int stale_bundle(struct dst_entry *dst);
729
730 /* Main function: finds/creates a bundle for given flow.
731  *
732  * At the moment we eat a raw IP route. Mostly to speed up lookups
733  * on interfaces with disabled IPsec.
734  */
735 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
736                 struct sock *sk, int flags)
737 {
738         struct xfrm_policy *policy;
739         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
740         struct dst_entry *dst, *dst_orig = *dst_p;
741         int nx = 0;
742         int err;
743         u32 genid;
744         u16 family = dst_orig->ops->family;
745 restart:
746         genid = atomic_read(&flow_cache_genid);
747         policy = NULL;
748         if (sk && sk->sk_policy[1])
749                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
750
751         if (!policy) {
752                 /* To accelerate a bit...  */
753                 if ((dst_orig->flags & DST_NOXFRM) || !xfrm_policy_list[XFRM_POLICY_OUT])
754                         return 0;
755
756                 policy = flow_cache_lookup(fl, family,
757                                            policy_to_flow_dir(XFRM_POLICY_OUT),
758                                            xfrm_policy_lookup);
759         }
760
761         if (!policy)
762                 return 0;
763
764         policy->curlft.use_time = (unsigned long)xtime.tv_sec;
765
766         switch (policy->action) {
767         case XFRM_POLICY_BLOCK:
768                 /* Prohibit the flow */
769                 xfrm_pol_put(policy);
770                 return -EPERM;
771
772         case XFRM_POLICY_ALLOW:
773                 if (policy->xfrm_nr == 0) {
774                         /* Flow passes not transformed. */
775                         xfrm_pol_put(policy);
776                         return 0;
777                 }
778
779                 /* Try to find matching bundle.
780                  *
781                  * LATER: help from flow cache. It is optional, this
782                  * is required only for output policy.
783                  */
784                 dst = xfrm_find_bundle(fl, policy, family);
785                 if (IS_ERR(dst)) {
786                         xfrm_pol_put(policy);
787                         return PTR_ERR(dst);
788                 }
789
790                 if (dst)
791                         break;
792
793                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
794
795                 if (unlikely(nx<0)) {
796                         err = nx;
797                         if (err == -EAGAIN && flags) {
798                                 DECLARE_WAITQUEUE(wait, current);
799
800                                 add_wait_queue(&km_waitq, &wait);
801                                 set_current_state(TASK_INTERRUPTIBLE);
802                                 schedule();
803                                 set_current_state(TASK_RUNNING);
804                                 remove_wait_queue(&km_waitq, &wait);
805
806                                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
807
808                                 if (nx == -EAGAIN && signal_pending(current)) {
809                                         err = -ERESTART;
810                                         goto error;
811                                 }
812                                 if (nx == -EAGAIN ||
813                                     genid != atomic_read(&flow_cache_genid)) {
814                                         xfrm_pol_put(policy);
815                                         goto restart;
816                                 }
817                                 err = nx;
818                         }
819                         if (err < 0)
820                                 goto error;
821                 }
822                 if (nx == 0) {
823                         /* Flow passes not transformed. */
824                         xfrm_pol_put(policy);
825                         return 0;
826                 }
827
828                 dst = dst_orig;
829                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
830
831                 if (unlikely(err)) {
832                         int i;
833                         for (i=0; i<nx; i++)
834                                 xfrm_state_put(xfrm[i]);
835                         goto error;
836                 }
837
838                 write_lock_bh(&policy->lock);
839                 if (unlikely(policy->dead || stale_bundle(dst))) {
840                         /* Wow! While we worked on resolving, this
841                          * policy has gone. Retry. It is not paranoia,
842                          * we just cannot enlist new bundle to dead object.
843                          * We can't enlist stable bundles either.
844                          */
845                         write_unlock_bh(&policy->lock);
846
847                         xfrm_pol_put(policy);
848                         if (dst)
849                                 dst_free(dst);
850                         goto restart;
851                 }
852                 dst->next = policy->bundles;
853                 policy->bundles = dst;
854                 dst_hold(dst);
855                 write_unlock_bh(&policy->lock);
856         }
857         *dst_p = dst;
858         dst_release(dst_orig);
859         xfrm_pol_put(policy);
860         return 0;
861
862 error:
863         dst_release(dst_orig);
864         xfrm_pol_put(policy);
865         *dst_p = NULL;
866         return err;
867 }
868 EXPORT_SYMBOL(xfrm_lookup);
869
870 /* When skb is transformed back to its "native" form, we have to
871  * check policy restrictions. At the moment we make this in maximally
872  * stupid way. Shame on me. :-) Of course, connected sockets must
873  * have policy cached at them.
874  */
875
876 static inline int
877 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x, 
878               unsigned short family)
879 {
880         if (xfrm_state_kern(x))
881                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
882         return  x->id.proto == tmpl->id.proto &&
883                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
884                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
885                 x->props.mode == tmpl->mode &&
886                 (tmpl->aalgos & (1<<x->props.aalgo)) &&
887                 !(x->props.mode && xfrm_state_addr_cmp(tmpl, x, family));
888 }
889
890 static inline int
891 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
892                unsigned short family)
893 {
894         int idx = start;
895
896         if (tmpl->optional) {
897                 if (!tmpl->mode)
898                         return start;
899         } else
900                 start = -1;
901         for (; idx < sp->len; idx++) {
902                 if (xfrm_state_ok(tmpl, sp->x[idx].xvec, family))
903                         return ++idx;
904                 if (sp->x[idx].xvec->props.mode)
905                         break;
906         }
907         return start;
908 }
909
910 static int
911 _decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
912 {
913         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
914
915         if (unlikely(afinfo == NULL))
916                 return -EAFNOSUPPORT;
917
918         afinfo->decode_session(skb, fl);
919         xfrm_policy_put_afinfo(afinfo);
920         return 0;
921 }
922
923 static inline int secpath_has_tunnel(struct sec_path *sp, int k)
924 {
925         for (; k < sp->len; k++) {
926                 if (sp->x[k].xvec->props.mode)
927                         return 1;
928         }
929
930         return 0;
931 }
932
933 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb, 
934                         unsigned short family)
935 {
936         struct xfrm_policy *pol;
937         struct flowi fl;
938
939         if (_decode_session(skb, &fl, family) < 0)
940                 return 0;
941
942         /* First, check used SA against their selectors. */
943         if (skb->sp) {
944                 int i;
945
946                 for (i=skb->sp->len-1; i>=0; i--) {
947                   struct sec_decap_state *xvec = &(skb->sp->x[i]);
948                         if (!xfrm_selector_match(&xvec->xvec->sel, &fl, family))
949                                 return 0;
950
951                         /* If there is a post_input processor, try running it */
952                         if (xvec->xvec->type->post_input &&
953                             (xvec->xvec->type->post_input)(xvec->xvec,
954                                                            &(xvec->decap),
955                                                            skb) != 0)
956                                 return 0;
957                 }
958         }
959
960         pol = NULL;
961         if (sk && sk->sk_policy[dir])
962                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
963
964         if (!pol)
965                 pol = flow_cache_lookup(&fl, family,
966                                         policy_to_flow_dir(dir),
967                                         xfrm_policy_lookup);
968
969         if (!pol)
970                 return !skb->sp || !secpath_has_tunnel(skb->sp, 0);
971
972         pol->curlft.use_time = (unsigned long)xtime.tv_sec;
973
974         if (pol->action == XFRM_POLICY_ALLOW) {
975                 struct sec_path *sp;
976                 static struct sec_path dummy;
977                 int i, k;
978
979                 if ((sp = skb->sp) == NULL)
980                         sp = &dummy;
981
982                 /* For each tunnel xfrm, find the first matching tmpl.
983                  * For each tmpl before that, find corresponding xfrm.
984                  * Order is _important_. Later we will implement
985                  * some barriers, but at the moment barriers
986                  * are implied between each two transformations.
987                  */
988                 for (i = pol->xfrm_nr-1, k = 0; i >= 0; i--) {
989                         k = xfrm_policy_ok(pol->xfrm_vec+i, sp, k, family);
990                         if (k < 0)
991                                 goto reject;
992                 }
993
994                 if (secpath_has_tunnel(sp, k))
995                         goto reject;
996
997                 xfrm_pol_put(pol);
998                 return 1;
999         }
1000
1001 reject:
1002         xfrm_pol_put(pol);
1003         return 0;
1004 }
1005 EXPORT_SYMBOL(__xfrm_policy_check);
1006
1007 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1008 {
1009         struct flowi fl;
1010
1011         if (_decode_session(skb, &fl, family) < 0)
1012                 return 0;
1013
1014         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1015 }
1016 EXPORT_SYMBOL(__xfrm_route_forward);
1017
1018 /* Optimize later using cookies and generation ids. */
1019
1020 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1021 {
1022         if (!stale_bundle(dst))
1023                 return dst;
1024
1025         return NULL;
1026 }
1027
1028 static int stale_bundle(struct dst_entry *dst)
1029 {
1030         return !xfrm_bundle_ok((struct xfrm_dst *)dst, NULL, AF_UNSPEC);
1031 }
1032
1033 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1034 {
1035         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1036                 dst->dev = &loopback_dev;
1037                 dev_hold(&loopback_dev);
1038                 dev_put(dev);
1039         }
1040 }
1041 EXPORT_SYMBOL(xfrm_dst_ifdown);
1042
1043 static void xfrm_link_failure(struct sk_buff *skb)
1044 {
1045         /* Impossible. Such dst must be popped before reaches point of failure. */
1046         return;
1047 }
1048
1049 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1050 {
1051         if (dst) {
1052                 if (dst->obsolete) {
1053                         dst_release(dst);
1054                         dst = NULL;
1055                 }
1056         }
1057         return dst;
1058 }
1059
1060 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1061 {
1062         int i;
1063         struct xfrm_policy *pol;
1064         struct dst_entry *dst, **dstp, *gc_list = NULL;
1065
1066         read_lock_bh(&xfrm_policy_lock);
1067         for (i=0; i<2*XFRM_POLICY_MAX; i++) {
1068                 for (pol = xfrm_policy_list[i]; pol; pol = pol->next) {
1069                         write_lock(&pol->lock);
1070                         dstp = &pol->bundles;
1071                         while ((dst=*dstp) != NULL) {
1072                                 if (func(dst)) {
1073                                         *dstp = dst->next;
1074                                         dst->next = gc_list;
1075                                         gc_list = dst;
1076                                 } else {
1077                                         dstp = &dst->next;
1078                                 }
1079                         }
1080                         write_unlock(&pol->lock);
1081                 }
1082         }
1083         read_unlock_bh(&xfrm_policy_lock);
1084
1085         while (gc_list) {
1086                 dst = gc_list;
1087                 gc_list = dst->next;
1088                 dst_free(dst);
1089         }
1090 }
1091
1092 static int unused_bundle(struct dst_entry *dst)
1093 {
1094         return !atomic_read(&dst->__refcnt);
1095 }
1096
1097 static void __xfrm_garbage_collect(void)
1098 {
1099         xfrm_prune_bundles(unused_bundle);
1100 }
1101
1102 int xfrm_flush_bundles(void)
1103 {
1104         xfrm_prune_bundles(stale_bundle);
1105         return 0;
1106 }
1107
1108 void xfrm_init_pmtu(struct dst_entry *dst)
1109 {
1110         do {
1111                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1112                 u32 pmtu, route_mtu_cached;
1113
1114                 pmtu = dst_mtu(dst->child);
1115                 xdst->child_mtu_cached = pmtu;
1116
1117                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1118
1119                 route_mtu_cached = dst_mtu(xdst->route);
1120                 xdst->route_mtu_cached = route_mtu_cached;
1121
1122                 if (pmtu > route_mtu_cached)
1123                         pmtu = route_mtu_cached;
1124
1125                 dst->metrics[RTAX_MTU-1] = pmtu;
1126         } while ((dst = dst->next));
1127 }
1128
1129 EXPORT_SYMBOL(xfrm_init_pmtu);
1130
1131 /* Check that the bundle accepts the flow and its components are
1132  * still valid.
1133  */
1134
1135 int xfrm_bundle_ok(struct xfrm_dst *first, struct flowi *fl, int family)
1136 {
1137         struct dst_entry *dst = &first->u.dst;
1138         struct xfrm_dst *last;
1139         u32 mtu;
1140
1141         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1142             (dst->dev && !netif_running(dst->dev)))
1143                 return 0;
1144
1145         last = NULL;
1146
1147         do {
1148                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1149
1150                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1151                         return 0;
1152                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1153                         return 0;
1154
1155                 mtu = dst_mtu(dst->child);
1156                 if (xdst->child_mtu_cached != mtu) {
1157                         last = xdst;
1158                         xdst->child_mtu_cached = mtu;
1159                 }
1160
1161                 if (!dst_check(xdst->route, xdst->route_cookie))
1162                         return 0;
1163                 mtu = dst_mtu(xdst->route);
1164                 if (xdst->route_mtu_cached != mtu) {
1165                         last = xdst;
1166                         xdst->route_mtu_cached = mtu;
1167                 }
1168
1169                 dst = dst->child;
1170         } while (dst->xfrm);
1171
1172         if (likely(!last))
1173                 return 1;
1174
1175         mtu = last->child_mtu_cached;
1176         for (;;) {
1177                 dst = &last->u.dst;
1178
1179                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1180                 if (mtu > last->route_mtu_cached)
1181                         mtu = last->route_mtu_cached;
1182                 dst->metrics[RTAX_MTU-1] = mtu;
1183
1184                 if (last == first)
1185                         break;
1186
1187                 last = last->u.next;
1188                 last->child_mtu_cached = mtu;
1189         }
1190
1191         return 1;
1192 }
1193
1194 EXPORT_SYMBOL(xfrm_bundle_ok);
1195
1196 /* Well... that's _TASK_. We need to scan through transformation
1197  * list and figure out what mss tcp should generate in order to
1198  * final datagram fit to mtu. Mama mia... :-)
1199  *
1200  * Apparently, some easy way exists, but we used to choose the most
1201  * bizarre ones. :-) So, raising Kalashnikov... tra-ta-ta.
1202  *
1203  * Consider this function as something like dark humour. :-)
1204  */
1205 static int xfrm_get_mss(struct dst_entry *dst, u32 mtu)
1206 {
1207         int res = mtu - dst->header_len;
1208
1209         for (;;) {
1210                 struct dst_entry *d = dst;
1211                 int m = res;
1212
1213                 do {
1214                         struct xfrm_state *x = d->xfrm;
1215                         if (x) {
1216                                 spin_lock_bh(&x->lock);
1217                                 if (x->km.state == XFRM_STATE_VALID &&
1218                                     x->type && x->type->get_max_size)
1219                                         m = x->type->get_max_size(d->xfrm, m);
1220                                 else
1221                                         m += x->props.header_len;
1222                                 spin_unlock_bh(&x->lock);
1223                         }
1224                 } while ((d = d->child) != NULL);
1225
1226                 if (m <= mtu)
1227                         break;
1228                 res -= (m - mtu);
1229                 if (res < 88)
1230                         return mtu;
1231         }
1232
1233         return res + dst->header_len;
1234 }
1235
1236 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1237 {
1238         int err = 0;
1239         if (unlikely(afinfo == NULL))
1240                 return -EINVAL;
1241         if (unlikely(afinfo->family >= NPROTO))
1242                 return -EAFNOSUPPORT;
1243         write_lock(&xfrm_policy_afinfo_lock);
1244         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1245                 err = -ENOBUFS;
1246         else {
1247                 struct dst_ops *dst_ops = afinfo->dst_ops;
1248                 if (likely(dst_ops->kmem_cachep == NULL))
1249                         dst_ops->kmem_cachep = xfrm_dst_cache;
1250                 if (likely(dst_ops->check == NULL))
1251                         dst_ops->check = xfrm_dst_check;
1252                 if (likely(dst_ops->negative_advice == NULL))
1253                         dst_ops->negative_advice = xfrm_negative_advice;
1254                 if (likely(dst_ops->link_failure == NULL))
1255                         dst_ops->link_failure = xfrm_link_failure;
1256                 if (likely(dst_ops->get_mss == NULL))
1257                         dst_ops->get_mss = xfrm_get_mss;
1258                 if (likely(afinfo->garbage_collect == NULL))
1259                         afinfo->garbage_collect = __xfrm_garbage_collect;
1260                 xfrm_policy_afinfo[afinfo->family] = afinfo;
1261         }
1262         write_unlock(&xfrm_policy_afinfo_lock);
1263         return err;
1264 }
1265 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
1266
1267 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1268 {
1269         int err = 0;
1270         if (unlikely(afinfo == NULL))
1271                 return -EINVAL;
1272         if (unlikely(afinfo->family >= NPROTO))
1273                 return -EAFNOSUPPORT;
1274         write_lock(&xfrm_policy_afinfo_lock);
1275         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1276                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1277                         err = -EINVAL;
1278                 else {
1279                         struct dst_ops *dst_ops = afinfo->dst_ops;
1280                         xfrm_policy_afinfo[afinfo->family] = NULL;
1281                         dst_ops->kmem_cachep = NULL;
1282                         dst_ops->check = NULL;
1283                         dst_ops->negative_advice = NULL;
1284                         dst_ops->link_failure = NULL;
1285                         dst_ops->get_mss = NULL;
1286                         afinfo->garbage_collect = NULL;
1287                 }
1288         }
1289         write_unlock(&xfrm_policy_afinfo_lock);
1290         return err;
1291 }
1292 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
1293
1294 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
1295 {
1296         struct xfrm_policy_afinfo *afinfo;
1297         if (unlikely(family >= NPROTO))
1298                 return NULL;
1299         read_lock(&xfrm_policy_afinfo_lock);
1300         afinfo = xfrm_policy_afinfo[family];
1301         if (likely(afinfo != NULL))
1302                 read_lock(&afinfo->lock);
1303         read_unlock(&xfrm_policy_afinfo_lock);
1304         return afinfo;
1305 }
1306
1307 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
1308 {
1309         if (unlikely(afinfo == NULL))
1310                 return;
1311         read_unlock(&afinfo->lock);
1312 }
1313
1314 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
1315 {
1316         switch (event) {
1317         case NETDEV_DOWN:
1318                 xfrm_flush_bundles();
1319         }
1320         return NOTIFY_DONE;
1321 }
1322
1323 static struct notifier_block xfrm_dev_notifier = {
1324         xfrm_dev_event,
1325         NULL,
1326         0
1327 };
1328
1329 static void __init xfrm_policy_init(void)
1330 {
1331         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
1332                                            sizeof(struct xfrm_dst),
1333                                            0, SLAB_HWCACHE_ALIGN,
1334                                            NULL, NULL);
1335         if (!xfrm_dst_cache)
1336                 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1337
1338         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
1339         register_netdevice_notifier(&xfrm_dev_notifier);
1340 }
1341
1342 void __init xfrm_init(void)
1343 {
1344         xfrm_state_init();
1345         xfrm_policy_init();
1346         xfrm_input_init();
1347 }
1348