update from upstream
[linux-2.6] / 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 = xp->index & 7;
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(int 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         kfree(policy);
260 }
261 EXPORT_SYMBOL(__xfrm_policy_destroy);
262
263 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
264 {
265         struct dst_entry *dst;
266
267         while ((dst = policy->bundles) != NULL) {
268                 policy->bundles = dst->next;
269                 dst_free(dst);
270         }
271
272         if (del_timer(&policy->timer))
273                 atomic_dec(&policy->refcnt);
274
275         if (atomic_read(&policy->refcnt) > 1)
276                 flow_cache_flush();
277
278         xfrm_pol_put(policy);
279 }
280
281 static void xfrm_policy_gc_task(void *data)
282 {
283         struct xfrm_policy *policy;
284         struct list_head *entry, *tmp;
285         struct list_head gc_list = LIST_HEAD_INIT(gc_list);
286
287         spin_lock_bh(&xfrm_policy_gc_lock);
288         list_splice_init(&xfrm_policy_gc_list, &gc_list);
289         spin_unlock_bh(&xfrm_policy_gc_lock);
290
291         list_for_each_safe(entry, tmp, &gc_list) {
292                 policy = list_entry(entry, struct xfrm_policy, list);
293                 xfrm_policy_gc_kill(policy);
294         }
295 }
296
297 /* Rule must be locked. Release descentant resources, announce
298  * entry dead. The rule must be unlinked from lists to the moment.
299  */
300
301 static void xfrm_policy_kill(struct xfrm_policy *policy)
302 {
303         int dead;
304
305         write_lock_bh(&policy->lock);
306         dead = policy->dead;
307         policy->dead = 1;
308         write_unlock_bh(&policy->lock);
309
310         if (unlikely(dead)) {
311                 WARN_ON(1);
312                 return;
313         }
314
315         spin_lock(&xfrm_policy_gc_lock);
316         list_add(&policy->list, &xfrm_policy_gc_list);
317         spin_unlock(&xfrm_policy_gc_lock);
318
319         schedule_work(&xfrm_policy_gc_work);
320 }
321
322 /* Generate new index... KAME seems to generate them ordered by cost
323  * of an absolute inpredictability of ordering of rules. This will not pass. */
324 static u32 xfrm_gen_index(int dir)
325 {
326         u32 idx;
327         struct xfrm_policy *p;
328         static u32 idx_generator;
329
330         for (;;) {
331                 idx = (idx_generator | dir);
332                 idx_generator += 8;
333                 if (idx == 0)
334                         idx = 8;
335                 for (p = xfrm_policy_list[dir]; p; p = p->next) {
336                         if (p->index == idx)
337                                 break;
338                 }
339                 if (!p)
340                         return idx;
341         }
342 }
343
344 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
345 {
346         struct xfrm_policy *pol, **p;
347         struct xfrm_policy *delpol = NULL;
348         struct xfrm_policy **newpos = NULL;
349
350         write_lock_bh(&xfrm_policy_lock);
351         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL;) {
352                 if (!delpol && memcmp(&policy->selector, &pol->selector, sizeof(pol->selector)) == 0) {
353                         if (excl) {
354                                 write_unlock_bh(&xfrm_policy_lock);
355                                 return -EEXIST;
356                         }
357                         *p = pol->next;
358                         delpol = pol;
359                         if (policy->priority > pol->priority)
360                                 continue;
361                 } else if (policy->priority >= pol->priority) {
362                         p = &pol->next;
363                         continue;
364                 }
365                 if (!newpos)
366                         newpos = p;
367                 if (delpol)
368                         break;
369                 p = &pol->next;
370         }
371         if (newpos)
372                 p = newpos;
373         xfrm_pol_hold(policy);
374         policy->next = *p;
375         *p = policy;
376         atomic_inc(&flow_cache_genid);
377         policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
378         policy->curlft.add_time = (unsigned long)xtime.tv_sec;
379         policy->curlft.use_time = 0;
380         if (!mod_timer(&policy->timer, jiffies + HZ))
381                 xfrm_pol_hold(policy);
382         write_unlock_bh(&xfrm_policy_lock);
383
384         if (delpol) {
385                 xfrm_policy_kill(delpol);
386         }
387         return 0;
388 }
389 EXPORT_SYMBOL(xfrm_policy_insert);
390
391 struct xfrm_policy *xfrm_policy_bysel(int dir, struct xfrm_selector *sel,
392                                       int delete)
393 {
394         struct xfrm_policy *pol, **p;
395
396         write_lock_bh(&xfrm_policy_lock);
397         for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
398                 if (memcmp(sel, &pol->selector, sizeof(*sel)) == 0) {
399                         xfrm_pol_hold(pol);
400                         if (delete)
401                                 *p = pol->next;
402                         break;
403                 }
404         }
405         write_unlock_bh(&xfrm_policy_lock);
406
407         if (pol && delete) {
408                 atomic_inc(&flow_cache_genid);
409                 xfrm_policy_kill(pol);
410         }
411         return pol;
412 }
413 EXPORT_SYMBOL(xfrm_policy_bysel);
414
415 struct xfrm_policy *xfrm_policy_byid(int dir, u32 id, int delete)
416 {
417         struct xfrm_policy *pol, **p;
418
419         write_lock_bh(&xfrm_policy_lock);
420         for (p = &xfrm_policy_list[id & 7]; (pol=*p)!=NULL; p = &pol->next) {
421                 if (pol->index == id) {
422                         xfrm_pol_hold(pol);
423                         if (delete)
424                                 *p = pol->next;
425                         break;
426                 }
427         }
428         write_unlock_bh(&xfrm_policy_lock);
429
430         if (pol && delete) {
431                 atomic_inc(&flow_cache_genid);
432                 xfrm_policy_kill(pol);
433         }
434         return pol;
435 }
436 EXPORT_SYMBOL(xfrm_policy_byid);
437
438 void xfrm_policy_flush(void)
439 {
440         struct xfrm_policy *xp;
441         int dir;
442
443         write_lock_bh(&xfrm_policy_lock);
444         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
445                 while ((xp = xfrm_policy_list[dir]) != NULL) {
446                         xfrm_policy_list[dir] = xp->next;
447                         write_unlock_bh(&xfrm_policy_lock);
448
449                         xfrm_policy_kill(xp);
450
451                         write_lock_bh(&xfrm_policy_lock);
452                 }
453         }
454         atomic_inc(&flow_cache_genid);
455         write_unlock_bh(&xfrm_policy_lock);
456 }
457 EXPORT_SYMBOL(xfrm_policy_flush);
458
459 int xfrm_policy_walk(int (*func)(struct xfrm_policy *, int, int, void*),
460                      void *data)
461 {
462         struct xfrm_policy *xp;
463         int dir;
464         int count = 0;
465         int error = 0;
466
467         read_lock_bh(&xfrm_policy_lock);
468         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
469                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next)
470                         count++;
471         }
472
473         if (count == 0) {
474                 error = -ENOENT;
475                 goto out;
476         }
477
478         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
479                 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next) {
480                         error = func(xp, dir%XFRM_POLICY_MAX, --count, data);
481                         if (error)
482                                 goto out;
483                 }
484         }
485
486 out:
487         read_unlock_bh(&xfrm_policy_lock);
488         return error;
489 }
490 EXPORT_SYMBOL(xfrm_policy_walk);
491
492 /* Find policy to apply to this flow. */
493
494 static void xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
495                                void **objp, atomic_t **obj_refp)
496 {
497         struct xfrm_policy *pol;
498
499         read_lock_bh(&xfrm_policy_lock);
500         for (pol = xfrm_policy_list[dir]; pol; pol = pol->next) {
501                 struct xfrm_selector *sel = &pol->selector;
502                 int match;
503
504                 if (pol->family != family)
505                         continue;
506
507                 match = xfrm_selector_match(sel, fl, family);
508                 if (match) {
509                         xfrm_pol_hold(pol);
510                         break;
511                 }
512         }
513         read_unlock_bh(&xfrm_policy_lock);
514         if ((*objp = (void *) pol) != NULL)
515                 *obj_refp = &pol->refcnt;
516 }
517
518 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
519 {
520         struct xfrm_policy *pol;
521
522         read_lock_bh(&xfrm_policy_lock);
523         if ((pol = sk->sk_policy[dir]) != NULL) {
524                 int match = xfrm_selector_match(&pol->selector, fl,
525                                                 sk->sk_family);
526                 if (match)
527                         xfrm_pol_hold(pol);
528                 else
529                         pol = NULL;
530         }
531         read_unlock_bh(&xfrm_policy_lock);
532         return pol;
533 }
534
535 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
536 {
537         pol->next = xfrm_policy_list[dir];
538         xfrm_policy_list[dir] = pol;
539         xfrm_pol_hold(pol);
540 }
541
542 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
543                                                 int dir)
544 {
545         struct xfrm_policy **polp;
546
547         for (polp = &xfrm_policy_list[dir];
548              *polp != NULL; polp = &(*polp)->next) {
549                 if (*polp == pol) {
550                         *polp = pol->next;
551                         return pol;
552                 }
553         }
554         return NULL;
555 }
556
557 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
558 {
559         write_lock_bh(&xfrm_policy_lock);
560         pol = __xfrm_policy_unlink(pol, dir);
561         write_unlock_bh(&xfrm_policy_lock);
562         if (pol) {
563                 if (dir < XFRM_POLICY_MAX)
564                         atomic_inc(&flow_cache_genid);
565                 xfrm_policy_kill(pol);
566                 return 0;
567         }
568         return -ENOENT;
569 }
570
571 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
572 {
573         struct xfrm_policy *old_pol;
574
575         write_lock_bh(&xfrm_policy_lock);
576         old_pol = sk->sk_policy[dir];
577         sk->sk_policy[dir] = pol;
578         if (pol) {
579                 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
580                 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
581                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
582         }
583         if (old_pol)
584                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
585         write_unlock_bh(&xfrm_policy_lock);
586
587         if (old_pol) {
588                 xfrm_policy_kill(old_pol);
589         }
590         return 0;
591 }
592
593 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
594 {
595         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
596
597         if (newp) {
598                 newp->selector = old->selector;
599                 newp->lft = old->lft;
600                 newp->curlft = old->curlft;
601                 newp->action = old->action;
602                 newp->flags = old->flags;
603                 newp->xfrm_nr = old->xfrm_nr;
604                 newp->index = old->index;
605                 memcpy(newp->xfrm_vec, old->xfrm_vec,
606                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
607                 write_lock_bh(&xfrm_policy_lock);
608                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
609                 write_unlock_bh(&xfrm_policy_lock);
610                 xfrm_pol_put(newp);
611         }
612         return newp;
613 }
614
615 int __xfrm_sk_clone_policy(struct sock *sk)
616 {
617         struct xfrm_policy *p0 = sk->sk_policy[0],
618                            *p1 = sk->sk_policy[1];
619
620         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
621         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
622                 return -ENOMEM;
623         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
624                 return -ENOMEM;
625         return 0;
626 }
627
628 /* Resolve list of templates for the flow, given policy. */
629
630 static int
631 xfrm_tmpl_resolve(struct xfrm_policy *policy, struct flowi *fl,
632                   struct xfrm_state **xfrm,
633                   unsigned short family)
634 {
635         int nx;
636         int i, error;
637         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
638         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
639
640         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
641                 struct xfrm_state *x;
642                 xfrm_address_t *remote = daddr;
643                 xfrm_address_t *local  = saddr;
644                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
645
646                 if (tmpl->mode) {
647                         remote = &tmpl->id.daddr;
648                         local = &tmpl->saddr;
649                 }
650
651                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
652
653                 if (x && x->km.state == XFRM_STATE_VALID) {
654                         xfrm[nx++] = x;
655                         daddr = remote;
656                         saddr = local;
657                         continue;
658                 }
659                 if (x) {
660                         error = (x->km.state == XFRM_STATE_ERROR ?
661                                  -EINVAL : -EAGAIN);
662                         xfrm_state_put(x);
663                 }
664
665                 if (!tmpl->optional)
666                         goto fail;
667         }
668         return nx;
669
670 fail:
671         for (nx--; nx>=0; nx--)
672                 xfrm_state_put(xfrm[nx]);
673         return error;
674 }
675
676 /* Check that the bundle accepts the flow and its components are
677  * still valid.
678  */
679
680 static struct dst_entry *
681 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
682 {
683         struct dst_entry *x;
684         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
685         if (unlikely(afinfo == NULL))
686                 return ERR_PTR(-EINVAL);
687         x = afinfo->find_bundle(fl, policy);
688         xfrm_policy_put_afinfo(afinfo);
689         return x;
690 }
691
692 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
693  * all the metrics... Shortly, bundle a bundle.
694  */
695
696 static int
697 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
698                    struct flowi *fl, struct dst_entry **dst_p,
699                    unsigned short family)
700 {
701         int err;
702         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
703         if (unlikely(afinfo == NULL))
704                 return -EINVAL;
705         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
706         xfrm_policy_put_afinfo(afinfo);
707         return err;
708 }
709
710 static inline int policy_to_flow_dir(int dir)
711 {
712         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
713             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
714             XFRM_POLICY_FWD == FLOW_DIR_FWD)
715                 return dir;
716         switch (dir) {
717         default:
718         case XFRM_POLICY_IN:
719                 return FLOW_DIR_IN;
720         case XFRM_POLICY_OUT:
721                 return FLOW_DIR_OUT;
722         case XFRM_POLICY_FWD:
723                 return FLOW_DIR_FWD;
724         };
725 }
726
727 static int stale_bundle(struct dst_entry *dst);
728
729 /* Main function: finds/creates a bundle for given flow.
730  *
731  * At the moment we eat a raw IP route. Mostly to speed up lookups
732  * on interfaces with disabled IPsec.
733  */
734 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
735                 struct sock *sk, int flags)
736 {
737         struct xfrm_policy *policy;
738         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
739         struct dst_entry *dst, *dst_orig = *dst_p;
740         int nx = 0;
741         int err;
742         u32 genid;
743         u16 family = dst_orig->ops->family;
744 restart:
745         genid = atomic_read(&flow_cache_genid);
746         policy = NULL;
747         if (sk && sk->sk_policy[1])
748                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
749
750         if (!policy) {
751                 /* To accelerate a bit...  */
752                 if ((dst_orig->flags & DST_NOXFRM) || !xfrm_policy_list[XFRM_POLICY_OUT])
753                         return 0;
754
755                 policy = flow_cache_lookup(fl, family,
756                                            policy_to_flow_dir(XFRM_POLICY_OUT),
757                                            xfrm_policy_lookup);
758         }
759
760         if (!policy)
761                 return 0;
762
763         policy->curlft.use_time = (unsigned long)xtime.tv_sec;
764
765         switch (policy->action) {
766         case XFRM_POLICY_BLOCK:
767                 /* Prohibit the flow */
768                 err = -EPERM;
769                 goto error;
770
771         case XFRM_POLICY_ALLOW:
772                 if (policy->xfrm_nr == 0) {
773                         /* Flow passes not transformed. */
774                         xfrm_pol_put(policy);
775                         return 0;
776                 }
777
778                 /* Try to find matching bundle.
779                  *
780                  * LATER: help from flow cache. It is optional, this
781                  * is required only for output policy.
782                  */
783                 dst = xfrm_find_bundle(fl, policy, family);
784                 if (IS_ERR(dst)) {
785                         err = PTR_ERR(dst);
786                         goto error;
787                 }
788
789                 if (dst)
790                         break;
791
792                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
793
794                 if (unlikely(nx<0)) {
795                         err = nx;
796                         if (err == -EAGAIN && flags) {
797                                 DECLARE_WAITQUEUE(wait, current);
798
799                                 add_wait_queue(&km_waitq, &wait);
800                                 set_current_state(TASK_INTERRUPTIBLE);
801                                 schedule();
802                                 set_current_state(TASK_RUNNING);
803                                 remove_wait_queue(&km_waitq, &wait);
804
805                                 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
806
807                                 if (nx == -EAGAIN && signal_pending(current)) {
808                                         err = -ERESTART;
809                                         goto error;
810                                 }
811                                 if (nx == -EAGAIN ||
812                                     genid != atomic_read(&flow_cache_genid)) {
813                                         xfrm_pol_put(policy);
814                                         goto restart;
815                                 }
816                                 err = nx;
817                         }
818                         if (err < 0)
819                                 goto error;
820                 }
821                 if (nx == 0) {
822                         /* Flow passes not transformed. */
823                         xfrm_pol_put(policy);
824                         return 0;
825                 }
826
827                 dst = dst_orig;
828                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
829
830                 if (unlikely(err)) {
831                         int i;
832                         for (i=0; i<nx; i++)
833                                 xfrm_state_put(xfrm[i]);
834                         goto error;
835                 }
836
837                 write_lock_bh(&policy->lock);
838                 if (unlikely(policy->dead || stale_bundle(dst))) {
839                         /* Wow! While we worked on resolving, this
840                          * policy has gone. Retry. It is not paranoia,
841                          * we just cannot enlist new bundle to dead object.
842                          * We can't enlist stable bundles either.
843                          */
844                         write_unlock_bh(&policy->lock);
845
846                         xfrm_pol_put(policy);
847                         if (dst)
848                                 dst_free(dst);
849                         goto restart;
850                 }
851                 dst->next = policy->bundles;
852                 policy->bundles = dst;
853                 dst_hold(dst);
854                 write_unlock_bh(&policy->lock);
855         }
856         *dst_p = dst;
857         dst_release(dst_orig);
858         xfrm_pol_put(policy);
859         return 0;
860
861 error:
862         dst_release(dst_orig);
863         xfrm_pol_put(policy);
864         *dst_p = NULL;
865         return err;
866 }
867 EXPORT_SYMBOL(xfrm_lookup);
868
869 /* When skb is transformed back to its "native" form, we have to
870  * check policy restrictions. At the moment we make this in maximally
871  * stupid way. Shame on me. :-) Of course, connected sockets must
872  * have policy cached at them.
873  */
874
875 static inline int
876 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x, 
877               unsigned short family)
878 {
879         if (xfrm_state_kern(x))
880                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
881         return  x->id.proto == tmpl->id.proto &&
882                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
883                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
884                 x->props.mode == tmpl->mode &&
885                 (tmpl->aalgos & (1<<x->props.aalgo)) &&
886                 !(x->props.mode && xfrm_state_addr_cmp(tmpl, x, family));
887 }
888
889 static inline int
890 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
891                unsigned short family)
892 {
893         int idx = start;
894
895         if (tmpl->optional) {
896                 if (!tmpl->mode)
897                         return start;
898         } else
899                 start = -1;
900         for (; idx < sp->len; idx++) {
901                 if (xfrm_state_ok(tmpl, sp->x[idx].xvec, family))
902                         return ++idx;
903                 if (sp->x[idx].xvec->props.mode)
904                         break;
905         }
906         return start;
907 }
908
909 static int
910 _decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
911 {
912         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
913
914         if (unlikely(afinfo == NULL))
915                 return -EAFNOSUPPORT;
916
917         afinfo->decode_session(skb, fl);
918         xfrm_policy_put_afinfo(afinfo);
919         return 0;
920 }
921
922 static inline int secpath_has_tunnel(struct sec_path *sp, int k)
923 {
924         for (; k < sp->len; k++) {
925                 if (sp->x[k].xvec->props.mode)
926                         return 1;
927         }
928
929         return 0;
930 }
931
932 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb, 
933                         unsigned short family)
934 {
935         struct xfrm_policy *pol;
936         struct flowi fl;
937
938         if (_decode_session(skb, &fl, family) < 0)
939                 return 0;
940
941         /* First, check used SA against their selectors. */
942         if (skb->sp) {
943                 int i;
944
945                 for (i=skb->sp->len-1; i>=0; i--) {
946                   struct sec_decap_state *xvec = &(skb->sp->x[i]);
947                         if (!xfrm_selector_match(&xvec->xvec->sel, &fl, family))
948                                 return 0;
949
950                         /* If there is a post_input processor, try running it */
951                         if (xvec->xvec->type->post_input &&
952                             (xvec->xvec->type->post_input)(xvec->xvec,
953                                                            &(xvec->decap),
954                                                            skb) != 0)
955                                 return 0;
956                 }
957         }
958
959         pol = NULL;
960         if (sk && sk->sk_policy[dir])
961                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
962
963         if (!pol)
964                 pol = flow_cache_lookup(&fl, family,
965                                         policy_to_flow_dir(dir),
966                                         xfrm_policy_lookup);
967
968         if (!pol)
969                 return !skb->sp || !secpath_has_tunnel(skb->sp, 0);
970
971         pol->curlft.use_time = (unsigned long)xtime.tv_sec;
972
973         if (pol->action == XFRM_POLICY_ALLOW) {
974                 struct sec_path *sp;
975                 static struct sec_path dummy;
976                 int i, k;
977
978                 if ((sp = skb->sp) == NULL)
979                         sp = &dummy;
980
981                 /* For each tunnel xfrm, find the first matching tmpl.
982                  * For each tmpl before that, find corresponding xfrm.
983                  * Order is _important_. Later we will implement
984                  * some barriers, but at the moment barriers
985                  * are implied between each two transformations.
986                  */
987                 for (i = pol->xfrm_nr-1, k = 0; i >= 0; i--) {
988                         k = xfrm_policy_ok(pol->xfrm_vec+i, sp, k, family);
989                         if (k < 0)
990                                 goto reject;
991                 }
992
993                 if (secpath_has_tunnel(sp, k))
994                         goto reject;
995
996                 xfrm_pol_put(pol);
997                 return 1;
998         }
999
1000 reject:
1001         xfrm_pol_put(pol);
1002         return 0;
1003 }
1004 EXPORT_SYMBOL(__xfrm_policy_check);
1005
1006 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1007 {
1008         struct flowi fl;
1009
1010         if (_decode_session(skb, &fl, family) < 0)
1011                 return 0;
1012
1013         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1014 }
1015 EXPORT_SYMBOL(__xfrm_route_forward);
1016
1017 /* Optimize later using cookies and generation ids. */
1018
1019 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1020 {
1021         if (!stale_bundle(dst))
1022                 return dst;
1023
1024         return NULL;
1025 }
1026
1027 static int stale_bundle(struct dst_entry *dst)
1028 {
1029         return !xfrm_bundle_ok((struct xfrm_dst *)dst, NULL, AF_UNSPEC);
1030 }
1031
1032 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1033 {
1034         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1035                 dst->dev = &loopback_dev;
1036                 dev_hold(&loopback_dev);
1037                 dev_put(dev);
1038         }
1039 }
1040 EXPORT_SYMBOL(xfrm_dst_ifdown);
1041
1042 static void xfrm_link_failure(struct sk_buff *skb)
1043 {
1044         /* Impossible. Such dst must be popped before reaches point of failure. */
1045         return;
1046 }
1047
1048 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1049 {
1050         if (dst) {
1051                 if (dst->obsolete) {
1052                         dst_release(dst);
1053                         dst = NULL;
1054                 }
1055         }
1056         return dst;
1057 }
1058
1059 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1060 {
1061         int i;
1062         struct xfrm_policy *pol;
1063         struct dst_entry *dst, **dstp, *gc_list = NULL;
1064
1065         read_lock_bh(&xfrm_policy_lock);
1066         for (i=0; i<2*XFRM_POLICY_MAX; i++) {
1067                 for (pol = xfrm_policy_list[i]; pol; pol = pol->next) {
1068                         write_lock(&pol->lock);
1069                         dstp = &pol->bundles;
1070                         while ((dst=*dstp) != NULL) {
1071                                 if (func(dst)) {
1072                                         *dstp = dst->next;
1073                                         dst->next = gc_list;
1074                                         gc_list = dst;
1075                                 } else {
1076                                         dstp = &dst->next;
1077                                 }
1078                         }
1079                         write_unlock(&pol->lock);
1080                 }
1081         }
1082         read_unlock_bh(&xfrm_policy_lock);
1083
1084         while (gc_list) {
1085                 dst = gc_list;
1086                 gc_list = dst->next;
1087                 dst_free(dst);
1088         }
1089 }
1090
1091 static int unused_bundle(struct dst_entry *dst)
1092 {
1093         return !atomic_read(&dst->__refcnt);
1094 }
1095
1096 static void __xfrm_garbage_collect(void)
1097 {
1098         xfrm_prune_bundles(unused_bundle);
1099 }
1100
1101 int xfrm_flush_bundles(void)
1102 {
1103         xfrm_prune_bundles(stale_bundle);
1104         return 0;
1105 }
1106
1107 void xfrm_init_pmtu(struct dst_entry *dst)
1108 {
1109         do {
1110                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1111                 u32 pmtu, route_mtu_cached;
1112
1113                 pmtu = dst_mtu(dst->child);
1114                 xdst->child_mtu_cached = pmtu;
1115
1116                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1117
1118                 route_mtu_cached = dst_mtu(xdst->route);
1119                 xdst->route_mtu_cached = route_mtu_cached;
1120
1121                 if (pmtu > route_mtu_cached)
1122                         pmtu = route_mtu_cached;
1123
1124                 dst->metrics[RTAX_MTU-1] = pmtu;
1125         } while ((dst = dst->next));
1126 }
1127
1128 EXPORT_SYMBOL(xfrm_init_pmtu);
1129
1130 /* Check that the bundle accepts the flow and its components are
1131  * still valid.
1132  */
1133
1134 int xfrm_bundle_ok(struct xfrm_dst *first, struct flowi *fl, int family)
1135 {
1136         struct dst_entry *dst = &first->u.dst;
1137         struct xfrm_dst *last;
1138         u32 mtu;
1139
1140         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1141             (dst->dev && !netif_running(dst->dev)))
1142                 return 0;
1143
1144         last = NULL;
1145
1146         do {
1147                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1148
1149                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1150                         return 0;
1151                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1152                         return 0;
1153
1154                 mtu = dst_mtu(dst->child);
1155                 if (xdst->child_mtu_cached != mtu) {
1156                         last = xdst;
1157                         xdst->child_mtu_cached = mtu;
1158                 }
1159
1160                 if (!dst_check(xdst->route, xdst->route_cookie))
1161                         return 0;
1162                 mtu = dst_mtu(xdst->route);
1163                 if (xdst->route_mtu_cached != mtu) {
1164                         last = xdst;
1165                         xdst->route_mtu_cached = mtu;
1166                 }
1167
1168                 dst = dst->child;
1169         } while (dst->xfrm);
1170
1171         if (likely(!last))
1172                 return 1;
1173
1174         mtu = last->child_mtu_cached;
1175         for (;;) {
1176                 dst = &last->u.dst;
1177
1178                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1179                 if (mtu > last->route_mtu_cached)
1180                         mtu = last->route_mtu_cached;
1181                 dst->metrics[RTAX_MTU-1] = mtu;
1182
1183                 if (last == first)
1184                         break;
1185
1186                 last = last->u.next;
1187                 last->child_mtu_cached = mtu;
1188         }
1189
1190         return 1;
1191 }
1192
1193 EXPORT_SYMBOL(xfrm_bundle_ok);
1194
1195 /* Well... that's _TASK_. We need to scan through transformation
1196  * list and figure out what mss tcp should generate in order to
1197  * final datagram fit to mtu. Mama mia... :-)
1198  *
1199  * Apparently, some easy way exists, but we used to choose the most
1200  * bizarre ones. :-) So, raising Kalashnikov... tra-ta-ta.
1201  *
1202  * Consider this function as something like dark humour. :-)
1203  */
1204 static int xfrm_get_mss(struct dst_entry *dst, u32 mtu)
1205 {
1206         int res = mtu - dst->header_len;
1207
1208         for (;;) {
1209                 struct dst_entry *d = dst;
1210                 int m = res;
1211
1212                 do {
1213                         struct xfrm_state *x = d->xfrm;
1214                         if (x) {
1215                                 spin_lock_bh(&x->lock);
1216                                 if (x->km.state == XFRM_STATE_VALID &&
1217                                     x->type && x->type->get_max_size)
1218                                         m = x->type->get_max_size(d->xfrm, m);
1219                                 else
1220                                         m += x->props.header_len;
1221                                 spin_unlock_bh(&x->lock);
1222                         }
1223                 } while ((d = d->child) != NULL);
1224
1225                 if (m <= mtu)
1226                         break;
1227                 res -= (m - mtu);
1228                 if (res < 88)
1229                         return mtu;
1230         }
1231
1232         return res + dst->header_len;
1233 }
1234
1235 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1236 {
1237         int err = 0;
1238         if (unlikely(afinfo == NULL))
1239                 return -EINVAL;
1240         if (unlikely(afinfo->family >= NPROTO))
1241                 return -EAFNOSUPPORT;
1242         write_lock(&xfrm_policy_afinfo_lock);
1243         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1244                 err = -ENOBUFS;
1245         else {
1246                 struct dst_ops *dst_ops = afinfo->dst_ops;
1247                 if (likely(dst_ops->kmem_cachep == NULL))
1248                         dst_ops->kmem_cachep = xfrm_dst_cache;
1249                 if (likely(dst_ops->check == NULL))
1250                         dst_ops->check = xfrm_dst_check;
1251                 if (likely(dst_ops->negative_advice == NULL))
1252                         dst_ops->negative_advice = xfrm_negative_advice;
1253                 if (likely(dst_ops->link_failure == NULL))
1254                         dst_ops->link_failure = xfrm_link_failure;
1255                 if (likely(dst_ops->get_mss == NULL))
1256                         dst_ops->get_mss = xfrm_get_mss;
1257                 if (likely(afinfo->garbage_collect == NULL))
1258                         afinfo->garbage_collect = __xfrm_garbage_collect;
1259                 xfrm_policy_afinfo[afinfo->family] = afinfo;
1260         }
1261         write_unlock(&xfrm_policy_afinfo_lock);
1262         return err;
1263 }
1264 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
1265
1266 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1267 {
1268         int err = 0;
1269         if (unlikely(afinfo == NULL))
1270                 return -EINVAL;
1271         if (unlikely(afinfo->family >= NPROTO))
1272                 return -EAFNOSUPPORT;
1273         write_lock(&xfrm_policy_afinfo_lock);
1274         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1275                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1276                         err = -EINVAL;
1277                 else {
1278                         struct dst_ops *dst_ops = afinfo->dst_ops;
1279                         xfrm_policy_afinfo[afinfo->family] = NULL;
1280                         dst_ops->kmem_cachep = NULL;
1281                         dst_ops->check = NULL;
1282                         dst_ops->negative_advice = NULL;
1283                         dst_ops->link_failure = NULL;
1284                         dst_ops->get_mss = NULL;
1285                         afinfo->garbage_collect = NULL;
1286                 }
1287         }
1288         write_unlock(&xfrm_policy_afinfo_lock);
1289         return err;
1290 }
1291 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
1292
1293 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
1294 {
1295         struct xfrm_policy_afinfo *afinfo;
1296         if (unlikely(family >= NPROTO))
1297                 return NULL;
1298         read_lock(&xfrm_policy_afinfo_lock);
1299         afinfo = xfrm_policy_afinfo[family];
1300         if (likely(afinfo != NULL))
1301                 read_lock(&afinfo->lock);
1302         read_unlock(&xfrm_policy_afinfo_lock);
1303         return afinfo;
1304 }
1305
1306 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
1307 {
1308         if (unlikely(afinfo == NULL))
1309                 return;
1310         read_unlock(&afinfo->lock);
1311 }
1312
1313 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
1314 {
1315         switch (event) {
1316         case NETDEV_DOWN:
1317                 xfrm_flush_bundles();
1318         }
1319         return NOTIFY_DONE;
1320 }
1321
1322 static struct notifier_block xfrm_dev_notifier = {
1323         xfrm_dev_event,
1324         NULL,
1325         0
1326 };
1327
1328 static void __init xfrm_policy_init(void)
1329 {
1330         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
1331                                            sizeof(struct xfrm_dst),
1332                                            0, SLAB_HWCACHE_ALIGN,
1333                                            NULL, NULL);
1334         if (!xfrm_dst_cache)
1335                 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1336
1337         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
1338         register_netdevice_notifier(&xfrm_dev_notifier);
1339 }
1340
1341 void __init xfrm_init(void)
1342 {
1343         xfrm_state_init();
1344         xfrm_policy_init();
1345         xfrm_input_init();
1346 }
1347