Merge branch 'upstream' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/libata-dev
[cascardo/linux.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 __read_mostly;
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
122 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl, 
123                     unsigned short family)
124 {
125         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
126         int err = 0;
127
128         if (unlikely(afinfo == NULL))
129                 return -EAFNOSUPPORT;
130
131         if (likely(afinfo->dst_lookup != NULL))
132                 err = afinfo->dst_lookup(dst, fl);
133         else
134                 err = -EINVAL;
135         xfrm_policy_put_afinfo(afinfo);
136         return err;
137 }
138 EXPORT_SYMBOL(xfrm_dst_lookup);
139
140 void xfrm_put_type(struct xfrm_type *type)
141 {
142         module_put(type->owner);
143 }
144
145 static inline unsigned long make_jiffies(long secs)
146 {
147         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
148                 return MAX_SCHEDULE_TIMEOUT-1;
149         else
150                 return secs*HZ;
151 }
152
153 static void xfrm_policy_timer(unsigned long data)
154 {
155         struct xfrm_policy *xp = (struct xfrm_policy*)data;
156         unsigned long now = (unsigned long)xtime.tv_sec;
157         long next = LONG_MAX;
158         int warn = 0;
159         int dir;
160
161         read_lock(&xp->lock);
162
163         if (xp->dead)
164                 goto out;
165
166         dir = xfrm_policy_id2dir(xp->index);
167
168         if (xp->lft.hard_add_expires_seconds) {
169                 long tmo = xp->lft.hard_add_expires_seconds +
170                         xp->curlft.add_time - now;
171                 if (tmo <= 0)
172                         goto expired;
173                 if (tmo < next)
174                         next = tmo;
175         }
176         if (xp->lft.hard_use_expires_seconds) {
177                 long tmo = xp->lft.hard_use_expires_seconds +
178                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
179                 if (tmo <= 0)
180                         goto expired;
181                 if (tmo < next)
182                         next = tmo;
183         }
184         if (xp->lft.soft_add_expires_seconds) {
185                 long tmo = xp->lft.soft_add_expires_seconds +
186                         xp->curlft.add_time - now;
187                 if (tmo <= 0) {
188                         warn = 1;
189                         tmo = XFRM_KM_TIMEOUT;
190                 }
191                 if (tmo < next)
192                         next = tmo;
193         }
194         if (xp->lft.soft_use_expires_seconds) {
195                 long tmo = xp->lft.soft_use_expires_seconds +
196                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
197                 if (tmo <= 0) {
198                         warn = 1;
199                         tmo = XFRM_KM_TIMEOUT;
200                 }
201                 if (tmo < next)
202                         next = tmo;
203         }
204
205         if (warn)
206                 km_policy_expired(xp, dir, 0);
207         if (next != LONG_MAX &&
208             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
209                 xfrm_pol_hold(xp);
210
211 out:
212         read_unlock(&xp->lock);
213         xfrm_pol_put(xp);
214         return;
215
216 expired:
217         read_unlock(&xp->lock);
218         if (!xfrm_policy_delete(xp, dir))
219                 km_policy_expired(xp, dir, 1);
220         xfrm_pol_put(xp);
221 }
222
223
224 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
225  * SPD calls.
226  */
227
228 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
229 {
230         struct xfrm_policy *policy;
231
232         policy = kmalloc(sizeof(struct xfrm_policy), gfp);
233
234         if (policy) {
235                 memset(policy, 0, sizeof(struct xfrm_policy));
236                 atomic_set(&policy->refcnt, 1);
237                 rwlock_init(&policy->lock);
238                 init_timer(&policy->timer);
239                 policy->timer.data = (unsigned long)policy;
240                 policy->timer.function = xfrm_policy_timer;
241         }
242         return policy;
243 }
244 EXPORT_SYMBOL(xfrm_policy_alloc);
245
246 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
247
248 void __xfrm_policy_destroy(struct xfrm_policy *policy)
249 {
250         if (!policy->dead)
251                 BUG();
252
253         if (policy->bundles)
254                 BUG();
255
256         if (del_timer(&policy->timer))
257                 BUG();
258
259         security_xfrm_policy_free(policy);
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         struct dst_entry *gc_list;
351
352         write_lock_bh(&xfrm_policy_lock);
353         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL;) {
354                 if (!delpol && memcmp(&policy->selector, &pol->selector, sizeof(pol->selector)) == 0 &&
355                     xfrm_sec_ctx_match(pol->security, policy->security)) {
356                         if (excl) {
357                                 write_unlock_bh(&xfrm_policy_lock);
358                                 return -EEXIST;
359                         }
360                         *p = pol->next;
361                         delpol = pol;
362                         if (policy->priority > pol->priority)
363                                 continue;
364                 } else if (policy->priority >= pol->priority) {
365                         p = &pol->next;
366                         continue;
367                 }
368                 if (!newpos)
369                         newpos = p;
370                 if (delpol)
371                         break;
372                 p = &pol->next;
373         }
374         if (newpos)
375                 p = newpos;
376         xfrm_pol_hold(policy);
377         policy->next = *p;
378         *p = policy;
379         atomic_inc(&flow_cache_genid);
380         policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
381         policy->curlft.add_time = (unsigned long)xtime.tv_sec;
382         policy->curlft.use_time = 0;
383         if (!mod_timer(&policy->timer, jiffies + HZ))
384                 xfrm_pol_hold(policy);
385         write_unlock_bh(&xfrm_policy_lock);
386
387         if (delpol)
388                 xfrm_policy_kill(delpol);
389
390         read_lock_bh(&xfrm_policy_lock);
391         gc_list = NULL;
392         for (policy = policy->next; policy; policy = policy->next) {
393                 struct dst_entry *dst;
394
395                 write_lock(&policy->lock);
396                 dst = policy->bundles;
397                 if (dst) {
398                         struct dst_entry *tail = dst;
399                         while (tail->next)
400                                 tail = tail->next;
401                         tail->next = gc_list;
402                         gc_list = dst;
403
404                         policy->bundles = NULL;
405                 }
406                 write_unlock(&policy->lock);
407         }
408         read_unlock_bh(&xfrm_policy_lock);
409
410         while (gc_list) {
411                 struct dst_entry *dst = gc_list;
412
413                 gc_list = dst->next;
414                 dst_free(dst);
415         }
416
417         return 0;
418 }
419 EXPORT_SYMBOL(xfrm_policy_insert);
420
421 struct xfrm_policy *xfrm_policy_bysel_ctx(int dir, struct xfrm_selector *sel,
422                                           struct xfrm_sec_ctx *ctx, int delete)
423 {
424         struct xfrm_policy *pol, **p;
425
426         write_lock_bh(&xfrm_policy_lock);
427         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
428                 if ((memcmp(sel, &pol->selector, sizeof(*sel)) == 0) &&
429                     (xfrm_sec_ctx_match(ctx, pol->security))) {
430                         xfrm_pol_hold(pol);
431                         if (delete)
432                                 *p = pol->next;
433                         break;
434                 }
435         }
436         write_unlock_bh(&xfrm_policy_lock);
437
438         if (pol && delete) {
439                 atomic_inc(&flow_cache_genid);
440                 xfrm_policy_kill(pol);
441         }
442         return pol;
443 }
444 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
445
446 struct xfrm_policy *xfrm_policy_byid(int dir, u32 id, int delete)
447 {
448         struct xfrm_policy *pol, **p;
449
450         write_lock_bh(&xfrm_policy_lock);
451         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
452                 if (pol->index == id) {
453                         xfrm_pol_hold(pol);
454                         if (delete)
455                                 *p = pol->next;
456                         break;
457                 }
458         }
459         write_unlock_bh(&xfrm_policy_lock);
460
461         if (pol && delete) {
462                 atomic_inc(&flow_cache_genid);
463                 xfrm_policy_kill(pol);
464         }
465         return pol;
466 }
467 EXPORT_SYMBOL(xfrm_policy_byid);
468
469 void xfrm_policy_flush(void)
470 {
471         struct xfrm_policy *xp;
472         int dir;
473
474         write_lock_bh(&xfrm_policy_lock);
475         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
476                 while ((xp = xfrm_policy_list[dir]) != NULL) {
477                         xfrm_policy_list[dir] = xp->next;
478                         write_unlock_bh(&xfrm_policy_lock);
479
480                         xfrm_policy_kill(xp);
481
482                         write_lock_bh(&xfrm_policy_lock);
483                 }
484         }
485         atomic_inc(&flow_cache_genid);
486         write_unlock_bh(&xfrm_policy_lock);
487 }
488 EXPORT_SYMBOL(xfrm_policy_flush);
489
490 int xfrm_policy_walk(int (*func)(struct xfrm_policy *, int, int, void*),
491                      void *data)
492 {
493         struct xfrm_policy *xp;
494         int dir;
495         int count = 0;
496         int error = 0;
497
498         read_lock_bh(&xfrm_policy_lock);
499         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
500                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next)
501                         count++;
502         }
503
504         if (count == 0) {
505                 error = -ENOENT;
506                 goto out;
507         }
508
509         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
510                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next) {
511                         error = func(xp, dir%XFRM_POLICY_MAX, --count, data);
512                         if (error)
513                                 goto out;
514                 }
515         }
516
517 out:
518         read_unlock_bh(&xfrm_policy_lock);
519         return error;
520 }
521 EXPORT_SYMBOL(xfrm_policy_walk);
522
523 /* Find policy to apply to this flow. */
524
525 static void xfrm_policy_lookup(struct flowi *fl, u32 sk_sid, u16 family, u8 dir,
526                                void **objp, atomic_t **obj_refp)
527 {
528         struct xfrm_policy *pol;
529
530         read_lock_bh(&xfrm_policy_lock);
531         for (pol = xfrm_policy_list[dir]; pol; pol = pol->next) {
532                 struct xfrm_selector *sel = &pol->selector;
533                 int match;
534
535                 if (pol->family != family)
536                         continue;
537
538                 match = xfrm_selector_match(sel, fl, family);
539
540                 if (match) {
541                         if (!security_xfrm_policy_lookup(pol, sk_sid, dir)) {
542                                 xfrm_pol_hold(pol);
543                                 break;
544                         }
545                 }
546         }
547         read_unlock_bh(&xfrm_policy_lock);
548         if ((*objp = (void *) pol) != NULL)
549                 *obj_refp = &pol->refcnt;
550 }
551
552 static inline int policy_to_flow_dir(int dir)
553 {
554         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
555             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
556             XFRM_POLICY_FWD == FLOW_DIR_FWD)
557                 return dir;
558         switch (dir) {
559         default:
560         case XFRM_POLICY_IN:
561                 return FLOW_DIR_IN;
562         case XFRM_POLICY_OUT:
563                 return FLOW_DIR_OUT;
564         case XFRM_POLICY_FWD:
565                 return FLOW_DIR_FWD;
566         };
567 }
568
569 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl, u32 sk_sid)
570 {
571         struct xfrm_policy *pol;
572
573         read_lock_bh(&xfrm_policy_lock);
574         if ((pol = sk->sk_policy[dir]) != NULL) {
575                 int match = xfrm_selector_match(&pol->selector, fl,
576                                                 sk->sk_family);
577                 int err = 0;
578
579                 if (match)
580                   err = security_xfrm_policy_lookup(pol, sk_sid, policy_to_flow_dir(dir));
581
582                 if (match && !err)
583                         xfrm_pol_hold(pol);
584                 else
585                         pol = NULL;
586         }
587         read_unlock_bh(&xfrm_policy_lock);
588         return pol;
589 }
590
591 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
592 {
593         pol->next = xfrm_policy_list[dir];
594         xfrm_policy_list[dir] = pol;
595         xfrm_pol_hold(pol);
596 }
597
598 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
599                                                 int dir)
600 {
601         struct xfrm_policy **polp;
602
603         for (polp = &xfrm_policy_list[dir];
604              *polp != NULL; polp = &(*polp)->next) {
605                 if (*polp == pol) {
606                         *polp = pol->next;
607                         return pol;
608                 }
609         }
610         return NULL;
611 }
612
613 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
614 {
615         write_lock_bh(&xfrm_policy_lock);
616         pol = __xfrm_policy_unlink(pol, dir);
617         write_unlock_bh(&xfrm_policy_lock);
618         if (pol) {
619                 if (dir < XFRM_POLICY_MAX)
620                         atomic_inc(&flow_cache_genid);
621                 xfrm_policy_kill(pol);
622                 return 0;
623         }
624         return -ENOENT;
625 }
626
627 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
628 {
629         struct xfrm_policy *old_pol;
630
631         write_lock_bh(&xfrm_policy_lock);
632         old_pol = sk->sk_policy[dir];
633         sk->sk_policy[dir] = pol;
634         if (pol) {
635                 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
636                 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
637                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
638         }
639         if (old_pol)
640                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
641         write_unlock_bh(&xfrm_policy_lock);
642
643         if (old_pol) {
644                 xfrm_policy_kill(old_pol);
645         }
646         return 0;
647 }
648
649 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
650 {
651         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
652
653         if (newp) {
654                 newp->selector = old->selector;
655                 if (security_xfrm_policy_clone(old, newp)) {
656                         kfree(newp);
657                         return NULL;  /* ENOMEM */
658                 }
659                 newp->lft = old->lft;
660                 newp->curlft = old->curlft;
661                 newp->action = old->action;
662                 newp->flags = old->flags;
663                 newp->xfrm_nr = old->xfrm_nr;
664                 newp->index = old->index;
665                 memcpy(newp->xfrm_vec, old->xfrm_vec,
666                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
667                 write_lock_bh(&xfrm_policy_lock);
668                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
669                 write_unlock_bh(&xfrm_policy_lock);
670                 xfrm_pol_put(newp);
671         }
672         return newp;
673 }
674
675 int __xfrm_sk_clone_policy(struct sock *sk)
676 {
677         struct xfrm_policy *p0 = sk->sk_policy[0],
678                            *p1 = sk->sk_policy[1];
679
680         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
681         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
682                 return -ENOMEM;
683         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
684                 return -ENOMEM;
685         return 0;
686 }
687
688 /* Resolve list of templates for the flow, given policy. */
689
690 static int
691 xfrm_tmpl_resolve(struct xfrm_policy *policy, struct flowi *fl,
692                   struct xfrm_state **xfrm,
693                   unsigned short family)
694 {
695         int nx;
696         int i, error;
697         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
698         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
699
700         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
701                 struct xfrm_state *x;
702                 xfrm_address_t *remote = daddr;
703                 xfrm_address_t *local  = saddr;
704                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
705
706                 if (tmpl->mode) {
707                         remote = &tmpl->id.daddr;
708                         local = &tmpl->saddr;
709                 }
710
711                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
712
713                 if (x && x->km.state == XFRM_STATE_VALID) {
714                         xfrm[nx++] = x;
715                         daddr = remote;
716                         saddr = local;
717                         continue;
718                 }
719                 if (x) {
720                         error = (x->km.state == XFRM_STATE_ERROR ?
721                                  -EINVAL : -EAGAIN);
722                         xfrm_state_put(x);
723                 }
724
725                 if (!tmpl->optional)
726                         goto fail;
727         }
728         return nx;
729
730 fail:
731         for (nx--; nx>=0; nx--)
732                 xfrm_state_put(xfrm[nx]);
733         return error;
734 }
735
736 /* Check that the bundle accepts the flow and its components are
737  * still valid.
738  */
739
740 static struct dst_entry *
741 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
742 {
743         struct dst_entry *x;
744         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
745         if (unlikely(afinfo == NULL))
746                 return ERR_PTR(-EINVAL);
747         x = afinfo->find_bundle(fl, policy);
748         xfrm_policy_put_afinfo(afinfo);
749         return x;
750 }
751
752 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
753  * all the metrics... Shortly, bundle a bundle.
754  */
755
756 static int
757 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
758                    struct flowi *fl, struct dst_entry **dst_p,
759                    unsigned short family)
760 {
761         int err;
762         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
763         if (unlikely(afinfo == NULL))
764                 return -EINVAL;
765         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
766         xfrm_policy_put_afinfo(afinfo);
767         return err;
768 }
769
770
771 static int stale_bundle(struct dst_entry *dst);
772
773 /* Main function: finds/creates a bundle for given flow.
774  *
775  * At the moment we eat a raw IP route. Mostly to speed up lookups
776  * on interfaces with disabled IPsec.
777  */
778 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
779                 struct sock *sk, int flags)
780 {
781         struct xfrm_policy *policy;
782         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
783         struct dst_entry *dst, *dst_orig = *dst_p;
784         int nx = 0;
785         int err;
786         u32 genid;
787         u16 family = dst_orig->ops->family;
788         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
789         u32 sk_sid = security_sk_sid(sk, fl, dir);
790 restart:
791         genid = atomic_read(&flow_cache_genid);
792         policy = NULL;
793         if (sk && sk->sk_policy[1])
794                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, sk_sid);
795
796         if (!policy) {
797                 /* To accelerate a bit...  */
798                 if ((dst_orig->flags & DST_NOXFRM) || !xfrm_policy_list[XFRM_POLICY_OUT])
799                         return 0;
800
801                 policy = flow_cache_lookup(fl, sk_sid, family, dir,
802                                            xfrm_policy_lookup);
803         }
804
805         if (!policy)
806                 return 0;
807
808         policy->curlft.use_time = (unsigned long)xtime.tv_sec;
809
810         switch (policy->action) {
811         case XFRM_POLICY_BLOCK:
812                 /* Prohibit the flow */
813                 err = -EPERM;
814                 goto error;
815
816         case XFRM_POLICY_ALLOW:
817                 if (policy->xfrm_nr == 0) {
818                         /* Flow passes not transformed. */
819                         xfrm_pol_put(policy);
820                         return 0;
821                 }
822
823                 /* Try to find matching bundle.
824                  *
825                  * LATER: help from flow cache. It is optional, this
826                  * is required only for output policy.
827                  */
828                 dst = xfrm_find_bundle(fl, policy, family);
829                 if (IS_ERR(dst)) {
830                         err = PTR_ERR(dst);
831                         goto error;
832                 }
833
834                 if (dst)
835                         break;
836
837                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
838
839                 if (unlikely(nx<0)) {
840                         err = nx;
841                         if (err == -EAGAIN && flags) {
842                                 DECLARE_WAITQUEUE(wait, current);
843
844                                 add_wait_queue(&km_waitq, &wait);
845                                 set_current_state(TASK_INTERRUPTIBLE);
846                                 schedule();
847                                 set_current_state(TASK_RUNNING);
848                                 remove_wait_queue(&km_waitq, &wait);
849
850                                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
851
852                                 if (nx == -EAGAIN && signal_pending(current)) {
853                                         err = -ERESTART;
854                                         goto error;
855                                 }
856                                 if (nx == -EAGAIN ||
857                                     genid != atomic_read(&flow_cache_genid)) {
858                                         xfrm_pol_put(policy);
859                                         goto restart;
860                                 }
861                                 err = nx;
862                         }
863                         if (err < 0)
864                                 goto error;
865                 }
866                 if (nx == 0) {
867                         /* Flow passes not transformed. */
868                         xfrm_pol_put(policy);
869                         return 0;
870                 }
871
872                 dst = dst_orig;
873                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
874
875                 if (unlikely(err)) {
876                         int i;
877                         for (i=0; i<nx; i++)
878                                 xfrm_state_put(xfrm[i]);
879                         goto error;
880                 }
881
882                 write_lock_bh(&policy->lock);
883                 if (unlikely(policy->dead || stale_bundle(dst))) {
884                         /* Wow! While we worked on resolving, this
885                          * policy has gone. Retry. It is not paranoia,
886                          * we just cannot enlist new bundle to dead object.
887                          * We can't enlist stable bundles either.
888                          */
889                         write_unlock_bh(&policy->lock);
890
891                         xfrm_pol_put(policy);
892                         if (dst)
893                                 dst_free(dst);
894                         goto restart;
895                 }
896                 dst->next = policy->bundles;
897                 policy->bundles = dst;
898                 dst_hold(dst);
899                 write_unlock_bh(&policy->lock);
900         }
901         *dst_p = dst;
902         dst_release(dst_orig);
903         xfrm_pol_put(policy);
904         return 0;
905
906 error:
907         dst_release(dst_orig);
908         xfrm_pol_put(policy);
909         *dst_p = NULL;
910         return err;
911 }
912 EXPORT_SYMBOL(xfrm_lookup);
913
914 /* When skb is transformed back to its "native" form, we have to
915  * check policy restrictions. At the moment we make this in maximally
916  * stupid way. Shame on me. :-) Of course, connected sockets must
917  * have policy cached at them.
918  */
919
920 static inline int
921 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x, 
922               unsigned short family)
923 {
924         if (xfrm_state_kern(x))
925                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
926         return  x->id.proto == tmpl->id.proto &&
927                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
928                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
929                 x->props.mode == tmpl->mode &&
930                 (tmpl->aalgos & (1<<x->props.aalgo)) &&
931                 !(x->props.mode && xfrm_state_addr_cmp(tmpl, x, family));
932 }
933
934 static inline int
935 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
936                unsigned short family)
937 {
938         int idx = start;
939
940         if (tmpl->optional) {
941                 if (!tmpl->mode)
942                         return start;
943         } else
944                 start = -1;
945         for (; idx < sp->len; idx++) {
946                 if (xfrm_state_ok(tmpl, sp->x[idx].xvec, family))
947                         return ++idx;
948                 if (sp->x[idx].xvec->props.mode)
949                         break;
950         }
951         return start;
952 }
953
954 static int
955 _decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
956 {
957         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
958
959         if (unlikely(afinfo == NULL))
960                 return -EAFNOSUPPORT;
961
962         afinfo->decode_session(skb, fl);
963         xfrm_policy_put_afinfo(afinfo);
964         return 0;
965 }
966
967 static inline int secpath_has_tunnel(struct sec_path *sp, int k)
968 {
969         for (; k < sp->len; k++) {
970                 if (sp->x[k].xvec->props.mode)
971                         return 1;
972         }
973
974         return 0;
975 }
976
977 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb, 
978                         unsigned short family)
979 {
980         struct xfrm_policy *pol;
981         struct flowi fl;
982         u8 fl_dir = policy_to_flow_dir(dir);
983         u32 sk_sid;
984
985         if (_decode_session(skb, &fl, family) < 0)
986                 return 0;
987
988         sk_sid = security_sk_sid(sk, &fl, fl_dir);
989
990         /* First, check used SA against their selectors. */
991         if (skb->sp) {
992                 int i;
993
994                 for (i=skb->sp->len-1; i>=0; i--) {
995                         struct sec_decap_state *xvec = &(skb->sp->x[i]);
996                         if (!xfrm_selector_match(&xvec->xvec->sel, &fl, family))
997                                 return 0;
998
999                         /* If there is a post_input processor, try running it */
1000                         if (xvec->xvec->type->post_input &&
1001                             (xvec->xvec->type->post_input)(xvec->xvec,
1002                                                            &(xvec->decap),
1003                                                            skb) != 0)
1004                                 return 0;
1005                 }
1006         }
1007
1008         pol = NULL;
1009         if (sk && sk->sk_policy[dir])
1010                 pol = xfrm_sk_policy_lookup(sk, dir, &fl, sk_sid);
1011
1012         if (!pol)
1013                 pol = flow_cache_lookup(&fl, sk_sid, family, fl_dir,
1014                                         xfrm_policy_lookup);
1015
1016         if (!pol)
1017                 return !skb->sp || !secpath_has_tunnel(skb->sp, 0);
1018
1019         pol->curlft.use_time = (unsigned long)xtime.tv_sec;
1020
1021         if (pol->action == XFRM_POLICY_ALLOW) {
1022                 struct sec_path *sp;
1023                 static struct sec_path dummy;
1024                 int i, k;
1025
1026                 if ((sp = skb->sp) == NULL)
1027                         sp = &dummy;
1028
1029                 /* For each tunnel xfrm, find the first matching tmpl.
1030                  * For each tmpl before that, find corresponding xfrm.
1031                  * Order is _important_. Later we will implement
1032                  * some barriers, but at the moment barriers
1033                  * are implied between each two transformations.
1034                  */
1035                 for (i = pol->xfrm_nr-1, k = 0; i >= 0; i--) {
1036                         k = xfrm_policy_ok(pol->xfrm_vec+i, sp, k, family);
1037                         if (k < 0)
1038                                 goto reject;
1039                 }
1040
1041                 if (secpath_has_tunnel(sp, k))
1042                         goto reject;
1043
1044                 xfrm_pol_put(pol);
1045                 return 1;
1046         }
1047
1048 reject:
1049         xfrm_pol_put(pol);
1050         return 0;
1051 }
1052 EXPORT_SYMBOL(__xfrm_policy_check);
1053
1054 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1055 {
1056         struct flowi fl;
1057
1058         if (_decode_session(skb, &fl, family) < 0)
1059                 return 0;
1060
1061         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1062 }
1063 EXPORT_SYMBOL(__xfrm_route_forward);
1064
1065 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1066 {
1067         /* If it is marked obsolete, which is how we even get here,
1068          * then we have purged it from the policy bundle list and we
1069          * did that for a good reason.
1070          */
1071         return NULL;
1072 }
1073
1074 static int stale_bundle(struct dst_entry *dst)
1075 {
1076         return !xfrm_bundle_ok((struct xfrm_dst *)dst, NULL, AF_UNSPEC);
1077 }
1078
1079 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1080 {
1081         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1082                 dst->dev = &loopback_dev;
1083                 dev_hold(&loopback_dev);
1084                 dev_put(dev);
1085         }
1086 }
1087 EXPORT_SYMBOL(xfrm_dst_ifdown);
1088
1089 static void xfrm_link_failure(struct sk_buff *skb)
1090 {
1091         /* Impossible. Such dst must be popped before reaches point of failure. */
1092         return;
1093 }
1094
1095 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1096 {
1097         if (dst) {
1098                 if (dst->obsolete) {
1099                         dst_release(dst);
1100                         dst = NULL;
1101                 }
1102         }
1103         return dst;
1104 }
1105
1106 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1107 {
1108         int i;
1109         struct xfrm_policy *pol;
1110         struct dst_entry *dst, **dstp, *gc_list = NULL;
1111
1112         read_lock_bh(&xfrm_policy_lock);
1113         for (i=0; i<2*XFRM_POLICY_MAX; i++) {
1114                 for (pol = xfrm_policy_list[i]; pol; pol = pol->next) {
1115                         write_lock(&pol->lock);
1116                         dstp = &pol->bundles;
1117                         while ((dst=*dstp) != NULL) {
1118                                 if (func(dst)) {
1119                                         *dstp = dst->next;
1120                                         dst->next = gc_list;
1121                                         gc_list = dst;
1122                                 } else {
1123                                         dstp = &dst->next;
1124                                 }
1125                         }
1126                         write_unlock(&pol->lock);
1127                 }
1128         }
1129         read_unlock_bh(&xfrm_policy_lock);
1130
1131         while (gc_list) {
1132                 dst = gc_list;
1133                 gc_list = dst->next;
1134                 dst_free(dst);
1135         }
1136 }
1137
1138 static int unused_bundle(struct dst_entry *dst)
1139 {
1140         return !atomic_read(&dst->__refcnt);
1141 }
1142
1143 static void __xfrm_garbage_collect(void)
1144 {
1145         xfrm_prune_bundles(unused_bundle);
1146 }
1147
1148 int xfrm_flush_bundles(void)
1149 {
1150         xfrm_prune_bundles(stale_bundle);
1151         return 0;
1152 }
1153
1154 static int always_true(struct dst_entry *dst)
1155 {
1156         return 1;
1157 }
1158
1159 void xfrm_flush_all_bundles(void)
1160 {
1161         xfrm_prune_bundles(always_true);
1162 }
1163
1164 void xfrm_init_pmtu(struct dst_entry *dst)
1165 {
1166         do {
1167                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1168                 u32 pmtu, route_mtu_cached;
1169
1170                 pmtu = dst_mtu(dst->child);
1171                 xdst->child_mtu_cached = pmtu;
1172
1173                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1174
1175                 route_mtu_cached = dst_mtu(xdst->route);
1176                 xdst->route_mtu_cached = route_mtu_cached;
1177
1178                 if (pmtu > route_mtu_cached)
1179                         pmtu = route_mtu_cached;
1180
1181                 dst->metrics[RTAX_MTU-1] = pmtu;
1182         } while ((dst = dst->next));
1183 }
1184
1185 EXPORT_SYMBOL(xfrm_init_pmtu);
1186
1187 /* Check that the bundle accepts the flow and its components are
1188  * still valid.
1189  */
1190
1191 int xfrm_bundle_ok(struct xfrm_dst *first, struct flowi *fl, int family)
1192 {
1193         struct dst_entry *dst = &first->u.dst;
1194         struct xfrm_dst *last;
1195         u32 mtu;
1196
1197         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1198             (dst->dev && !netif_running(dst->dev)))
1199                 return 0;
1200
1201         last = NULL;
1202
1203         do {
1204                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1205
1206                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1207                         return 0;
1208                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1209                         return 0;
1210
1211                 mtu = dst_mtu(dst->child);
1212                 if (xdst->child_mtu_cached != mtu) {
1213                         last = xdst;
1214                         xdst->child_mtu_cached = mtu;
1215                 }
1216
1217                 if (!dst_check(xdst->route, xdst->route_cookie))
1218                         return 0;
1219                 mtu = dst_mtu(xdst->route);
1220                 if (xdst->route_mtu_cached != mtu) {
1221                         last = xdst;
1222                         xdst->route_mtu_cached = mtu;
1223                 }
1224
1225                 dst = dst->child;
1226         } while (dst->xfrm);
1227
1228         if (likely(!last))
1229                 return 1;
1230
1231         mtu = last->child_mtu_cached;
1232         for (;;) {
1233                 dst = &last->u.dst;
1234
1235                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1236                 if (mtu > last->route_mtu_cached)
1237                         mtu = last->route_mtu_cached;
1238                 dst->metrics[RTAX_MTU-1] = mtu;
1239
1240                 if (last == first)
1241                         break;
1242
1243                 last = last->u.next;
1244                 last->child_mtu_cached = mtu;
1245         }
1246
1247         return 1;
1248 }
1249
1250 EXPORT_SYMBOL(xfrm_bundle_ok);
1251
1252 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1253 {
1254         int err = 0;
1255         if (unlikely(afinfo == NULL))
1256                 return -EINVAL;
1257         if (unlikely(afinfo->family >= NPROTO))
1258                 return -EAFNOSUPPORT;
1259         write_lock(&xfrm_policy_afinfo_lock);
1260         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1261                 err = -ENOBUFS;
1262         else {
1263                 struct dst_ops *dst_ops = afinfo->dst_ops;
1264                 if (likely(dst_ops->kmem_cachep == NULL))
1265                         dst_ops->kmem_cachep = xfrm_dst_cache;
1266                 if (likely(dst_ops->check == NULL))
1267                         dst_ops->check = xfrm_dst_check;
1268                 if (likely(dst_ops->negative_advice == NULL))
1269                         dst_ops->negative_advice = xfrm_negative_advice;
1270                 if (likely(dst_ops->link_failure == NULL))
1271                         dst_ops->link_failure = xfrm_link_failure;
1272                 if (likely(afinfo->garbage_collect == NULL))
1273                         afinfo->garbage_collect = __xfrm_garbage_collect;
1274                 xfrm_policy_afinfo[afinfo->family] = afinfo;
1275         }
1276         write_unlock(&xfrm_policy_afinfo_lock);
1277         return err;
1278 }
1279 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
1280
1281 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1282 {
1283         int err = 0;
1284         if (unlikely(afinfo == NULL))
1285                 return -EINVAL;
1286         if (unlikely(afinfo->family >= NPROTO))
1287                 return -EAFNOSUPPORT;
1288         write_lock(&xfrm_policy_afinfo_lock);
1289         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1290                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1291                         err = -EINVAL;
1292                 else {
1293                         struct dst_ops *dst_ops = afinfo->dst_ops;
1294                         xfrm_policy_afinfo[afinfo->family] = NULL;
1295                         dst_ops->kmem_cachep = NULL;
1296                         dst_ops->check = NULL;
1297                         dst_ops->negative_advice = NULL;
1298                         dst_ops->link_failure = NULL;
1299                         afinfo->garbage_collect = NULL;
1300                 }
1301         }
1302         write_unlock(&xfrm_policy_afinfo_lock);
1303         return err;
1304 }
1305 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
1306
1307 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
1308 {
1309         struct xfrm_policy_afinfo *afinfo;
1310         if (unlikely(family >= NPROTO))
1311                 return NULL;
1312         read_lock(&xfrm_policy_afinfo_lock);
1313         afinfo = xfrm_policy_afinfo[family];
1314         if (likely(afinfo != NULL))
1315                 read_lock(&afinfo->lock);
1316         read_unlock(&xfrm_policy_afinfo_lock);
1317         return afinfo;
1318 }
1319
1320 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
1321 {
1322         if (unlikely(afinfo == NULL))
1323                 return;
1324         read_unlock(&afinfo->lock);
1325 }
1326
1327 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
1328 {
1329         switch (event) {
1330         case NETDEV_DOWN:
1331                 xfrm_flush_bundles();
1332         }
1333         return NOTIFY_DONE;
1334 }
1335
1336 static struct notifier_block xfrm_dev_notifier = {
1337         xfrm_dev_event,
1338         NULL,
1339         0
1340 };
1341
1342 static void __init xfrm_policy_init(void)
1343 {
1344         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
1345                                            sizeof(struct xfrm_dst),
1346                                            0, SLAB_HWCACHE_ALIGN,
1347                                            NULL, NULL);
1348         if (!xfrm_dst_cache)
1349                 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1350
1351         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
1352         register_netdevice_notifier(&xfrm_dev_notifier);
1353 }
1354
1355 void __init xfrm_init(void)
1356 {
1357         xfrm_state_init();
1358         xfrm_policy_init();
1359         xfrm_input_init();
1360 }
1361