Merge git://git.kernel.org/pub/scm/linux/kernel/git/sfrench/cifs-2.6
[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 <linux/slab.h>
17 #include <linux/kmod.h>
18 #include <linux/list.h>
19 #include <linux/spinlock.h>
20 #include <linux/workqueue.h>
21 #include <linux/notifier.h>
22 #include <linux/netdevice.h>
23 #include <linux/netfilter.h>
24 #include <linux/module.h>
25 #include <linux/cache.h>
26 #include <net/xfrm.h>
27 #include <net/ip.h>
28
29 #include "xfrm_hash.h"
30
31 DEFINE_MUTEX(xfrm_cfg_mutex);
32 EXPORT_SYMBOL(xfrm_cfg_mutex);
33
34 static DEFINE_RWLOCK(xfrm_policy_lock);
35
36 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
37 EXPORT_SYMBOL(xfrm_policy_count);
38
39 static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
40 static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
41
42 static kmem_cache_t *xfrm_dst_cache __read_mostly;
43
44 static struct work_struct xfrm_policy_gc_work;
45 static HLIST_HEAD(xfrm_policy_gc_list);
46 static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
47
48 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
49 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
50 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family);
51 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo);
52
53 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
54 {
55         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
56         struct xfrm_type **typemap;
57         int err = 0;
58
59         if (unlikely(afinfo == NULL))
60                 return -EAFNOSUPPORT;
61         typemap = afinfo->type_map;
62
63         if (likely(typemap[type->proto] == NULL))
64                 typemap[type->proto] = type;
65         else
66                 err = -EEXIST;
67         xfrm_policy_unlock_afinfo(afinfo);
68         return err;
69 }
70 EXPORT_SYMBOL(xfrm_register_type);
71
72 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
73 {
74         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
75         struct xfrm_type **typemap;
76         int err = 0;
77
78         if (unlikely(afinfo == NULL))
79                 return -EAFNOSUPPORT;
80         typemap = afinfo->type_map;
81
82         if (unlikely(typemap[type->proto] != type))
83                 err = -ENOENT;
84         else
85                 typemap[type->proto] = NULL;
86         xfrm_policy_unlock_afinfo(afinfo);
87         return err;
88 }
89 EXPORT_SYMBOL(xfrm_unregister_type);
90
91 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
92 {
93         struct xfrm_policy_afinfo *afinfo;
94         struct xfrm_type **typemap;
95         struct xfrm_type *type;
96         int modload_attempted = 0;
97
98 retry:
99         afinfo = xfrm_policy_get_afinfo(family);
100         if (unlikely(afinfo == NULL))
101                 return NULL;
102         typemap = afinfo->type_map;
103
104         type = typemap[proto];
105         if (unlikely(type && !try_module_get(type->owner)))
106                 type = NULL;
107         if (!type && !modload_attempted) {
108                 xfrm_policy_put_afinfo(afinfo);
109                 request_module("xfrm-type-%d-%d",
110                                (int) family, (int) proto);
111                 modload_attempted = 1;
112                 goto retry;
113         }
114
115         xfrm_policy_put_afinfo(afinfo);
116         return type;
117 }
118
119 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl, 
120                     unsigned short family)
121 {
122         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
123         int err = 0;
124
125         if (unlikely(afinfo == NULL))
126                 return -EAFNOSUPPORT;
127
128         if (likely(afinfo->dst_lookup != NULL))
129                 err = afinfo->dst_lookup(dst, fl);
130         else
131                 err = -EINVAL;
132         xfrm_policy_put_afinfo(afinfo);
133         return err;
134 }
135 EXPORT_SYMBOL(xfrm_dst_lookup);
136
137 void xfrm_put_type(struct xfrm_type *type)
138 {
139         module_put(type->owner);
140 }
141
142 int xfrm_register_mode(struct xfrm_mode *mode, int family)
143 {
144         struct xfrm_policy_afinfo *afinfo;
145         struct xfrm_mode **modemap;
146         int err;
147
148         if (unlikely(mode->encap >= XFRM_MODE_MAX))
149                 return -EINVAL;
150
151         afinfo = xfrm_policy_lock_afinfo(family);
152         if (unlikely(afinfo == NULL))
153                 return -EAFNOSUPPORT;
154
155         err = -EEXIST;
156         modemap = afinfo->mode_map;
157         if (likely(modemap[mode->encap] == NULL)) {
158                 modemap[mode->encap] = mode;
159                 err = 0;
160         }
161
162         xfrm_policy_unlock_afinfo(afinfo);
163         return err;
164 }
165 EXPORT_SYMBOL(xfrm_register_mode);
166
167 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
168 {
169         struct xfrm_policy_afinfo *afinfo;
170         struct xfrm_mode **modemap;
171         int err;
172
173         if (unlikely(mode->encap >= XFRM_MODE_MAX))
174                 return -EINVAL;
175
176         afinfo = xfrm_policy_lock_afinfo(family);
177         if (unlikely(afinfo == NULL))
178                 return -EAFNOSUPPORT;
179
180         err = -ENOENT;
181         modemap = afinfo->mode_map;
182         if (likely(modemap[mode->encap] == mode)) {
183                 modemap[mode->encap] = NULL;
184                 err = 0;
185         }
186
187         xfrm_policy_unlock_afinfo(afinfo);
188         return err;
189 }
190 EXPORT_SYMBOL(xfrm_unregister_mode);
191
192 struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
193 {
194         struct xfrm_policy_afinfo *afinfo;
195         struct xfrm_mode *mode;
196         int modload_attempted = 0;
197
198         if (unlikely(encap >= XFRM_MODE_MAX))
199                 return NULL;
200
201 retry:
202         afinfo = xfrm_policy_get_afinfo(family);
203         if (unlikely(afinfo == NULL))
204                 return NULL;
205
206         mode = afinfo->mode_map[encap];
207         if (unlikely(mode && !try_module_get(mode->owner)))
208                 mode = NULL;
209         if (!mode && !modload_attempted) {
210                 xfrm_policy_put_afinfo(afinfo);
211                 request_module("xfrm-mode-%d-%d", family, encap);
212                 modload_attempted = 1;
213                 goto retry;
214         }
215
216         xfrm_policy_put_afinfo(afinfo);
217         return mode;
218 }
219
220 void xfrm_put_mode(struct xfrm_mode *mode)
221 {
222         module_put(mode->owner);
223 }
224
225 static inline unsigned long make_jiffies(long secs)
226 {
227         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
228                 return MAX_SCHEDULE_TIMEOUT-1;
229         else
230                 return secs*HZ;
231 }
232
233 static void xfrm_policy_timer(unsigned long data)
234 {
235         struct xfrm_policy *xp = (struct xfrm_policy*)data;
236         unsigned long now = (unsigned long)xtime.tv_sec;
237         long next = LONG_MAX;
238         int warn = 0;
239         int dir;
240
241         read_lock(&xp->lock);
242
243         if (xp->dead)
244                 goto out;
245
246         dir = xfrm_policy_id2dir(xp->index);
247
248         if (xp->lft.hard_add_expires_seconds) {
249                 long tmo = xp->lft.hard_add_expires_seconds +
250                         xp->curlft.add_time - now;
251                 if (tmo <= 0)
252                         goto expired;
253                 if (tmo < next)
254                         next = tmo;
255         }
256         if (xp->lft.hard_use_expires_seconds) {
257                 long tmo = xp->lft.hard_use_expires_seconds +
258                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
259                 if (tmo <= 0)
260                         goto expired;
261                 if (tmo < next)
262                         next = tmo;
263         }
264         if (xp->lft.soft_add_expires_seconds) {
265                 long tmo = xp->lft.soft_add_expires_seconds +
266                         xp->curlft.add_time - now;
267                 if (tmo <= 0) {
268                         warn = 1;
269                         tmo = XFRM_KM_TIMEOUT;
270                 }
271                 if (tmo < next)
272                         next = tmo;
273         }
274         if (xp->lft.soft_use_expires_seconds) {
275                 long tmo = xp->lft.soft_use_expires_seconds +
276                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
277                 if (tmo <= 0) {
278                         warn = 1;
279                         tmo = XFRM_KM_TIMEOUT;
280                 }
281                 if (tmo < next)
282                         next = tmo;
283         }
284
285         if (warn)
286                 km_policy_expired(xp, dir, 0, 0);
287         if (next != LONG_MAX &&
288             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
289                 xfrm_pol_hold(xp);
290
291 out:
292         read_unlock(&xp->lock);
293         xfrm_pol_put(xp);
294         return;
295
296 expired:
297         read_unlock(&xp->lock);
298         if (!xfrm_policy_delete(xp, dir))
299                 km_policy_expired(xp, dir, 1, 0);
300         xfrm_pol_put(xp);
301 }
302
303
304 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
305  * SPD calls.
306  */
307
308 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
309 {
310         struct xfrm_policy *policy;
311
312         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
313
314         if (policy) {
315                 INIT_HLIST_NODE(&policy->bydst);
316                 INIT_HLIST_NODE(&policy->byidx);
317                 rwlock_init(&policy->lock);
318                 atomic_set(&policy->refcnt, 1);
319                 init_timer(&policy->timer);
320                 policy->timer.data = (unsigned long)policy;
321                 policy->timer.function = xfrm_policy_timer;
322         }
323         return policy;
324 }
325 EXPORT_SYMBOL(xfrm_policy_alloc);
326
327 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
328
329 void __xfrm_policy_destroy(struct xfrm_policy *policy)
330 {
331         BUG_ON(!policy->dead);
332
333         BUG_ON(policy->bundles);
334
335         if (del_timer(&policy->timer))
336                 BUG();
337
338         security_xfrm_policy_free(policy);
339         kfree(policy);
340 }
341 EXPORT_SYMBOL(__xfrm_policy_destroy);
342
343 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
344 {
345         struct dst_entry *dst;
346
347         while ((dst = policy->bundles) != NULL) {
348                 policy->bundles = dst->next;
349                 dst_free(dst);
350         }
351
352         if (del_timer(&policy->timer))
353                 atomic_dec(&policy->refcnt);
354
355         if (atomic_read(&policy->refcnt) > 1)
356                 flow_cache_flush();
357
358         xfrm_pol_put(policy);
359 }
360
361 static void xfrm_policy_gc_task(void *data)
362 {
363         struct xfrm_policy *policy;
364         struct hlist_node *entry, *tmp;
365         struct hlist_head gc_list;
366
367         spin_lock_bh(&xfrm_policy_gc_lock);
368         gc_list.first = xfrm_policy_gc_list.first;
369         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
370         spin_unlock_bh(&xfrm_policy_gc_lock);
371
372         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
373                 xfrm_policy_gc_kill(policy);
374 }
375
376 /* Rule must be locked. Release descentant resources, announce
377  * entry dead. The rule must be unlinked from lists to the moment.
378  */
379
380 static void xfrm_policy_kill(struct xfrm_policy *policy)
381 {
382         int dead;
383
384         write_lock_bh(&policy->lock);
385         dead = policy->dead;
386         policy->dead = 1;
387         write_unlock_bh(&policy->lock);
388
389         if (unlikely(dead)) {
390                 WARN_ON(1);
391                 return;
392         }
393
394         spin_lock(&xfrm_policy_gc_lock);
395         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
396         spin_unlock(&xfrm_policy_gc_lock);
397
398         schedule_work(&xfrm_policy_gc_work);
399 }
400
401 struct xfrm_policy_hash {
402         struct hlist_head       *table;
403         unsigned int            hmask;
404 };
405
406 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
407 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
408 static struct hlist_head *xfrm_policy_byidx __read_mostly;
409 static unsigned int xfrm_idx_hmask __read_mostly;
410 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
411
412 static inline unsigned int idx_hash(u32 index)
413 {
414         return __idx_hash(index, xfrm_idx_hmask);
415 }
416
417 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
418 {
419         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
420         unsigned int hash = __sel_hash(sel, family, hmask);
421
422         return (hash == hmask + 1 ?
423                 &xfrm_policy_inexact[dir] :
424                 xfrm_policy_bydst[dir].table + hash);
425 }
426
427 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
428 {
429         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
430         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
431
432         return xfrm_policy_bydst[dir].table + hash;
433 }
434
435 static void xfrm_dst_hash_transfer(struct hlist_head *list,
436                                    struct hlist_head *ndsttable,
437                                    unsigned int nhashmask)
438 {
439         struct hlist_node *entry, *tmp;
440         struct xfrm_policy *pol;
441
442         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
443                 unsigned int h;
444
445                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
446                                 pol->family, nhashmask);
447                 hlist_add_head(&pol->bydst, ndsttable+h);
448         }
449 }
450
451 static void xfrm_idx_hash_transfer(struct hlist_head *list,
452                                    struct hlist_head *nidxtable,
453                                    unsigned int nhashmask)
454 {
455         struct hlist_node *entry, *tmp;
456         struct xfrm_policy *pol;
457
458         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
459                 unsigned int h;
460
461                 h = __idx_hash(pol->index, nhashmask);
462                 hlist_add_head(&pol->byidx, nidxtable+h);
463         }
464 }
465
466 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
467 {
468         return ((old_hmask + 1) << 1) - 1;
469 }
470
471 static void xfrm_bydst_resize(int dir)
472 {
473         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
474         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
475         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
476         struct hlist_head *odst = xfrm_policy_bydst[dir].table;
477         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
478         int i;
479
480         if (!ndst)
481                 return;
482
483         write_lock_bh(&xfrm_policy_lock);
484
485         for (i = hmask; i >= 0; i--)
486                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
487
488         xfrm_policy_bydst[dir].table = ndst;
489         xfrm_policy_bydst[dir].hmask = nhashmask;
490
491         write_unlock_bh(&xfrm_policy_lock);
492
493         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
494 }
495
496 static void xfrm_byidx_resize(int total)
497 {
498         unsigned int hmask = xfrm_idx_hmask;
499         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
500         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
501         struct hlist_head *oidx = xfrm_policy_byidx;
502         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
503         int i;
504
505         if (!nidx)
506                 return;
507
508         write_lock_bh(&xfrm_policy_lock);
509
510         for (i = hmask; i >= 0; i--)
511                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
512
513         xfrm_policy_byidx = nidx;
514         xfrm_idx_hmask = nhashmask;
515
516         write_unlock_bh(&xfrm_policy_lock);
517
518         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
519 }
520
521 static inline int xfrm_bydst_should_resize(int dir, int *total)
522 {
523         unsigned int cnt = xfrm_policy_count[dir];
524         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
525
526         if (total)
527                 *total += cnt;
528
529         if ((hmask + 1) < xfrm_policy_hashmax &&
530             cnt > hmask)
531                 return 1;
532
533         return 0;
534 }
535
536 static inline int xfrm_byidx_should_resize(int total)
537 {
538         unsigned int hmask = xfrm_idx_hmask;
539
540         if ((hmask + 1) < xfrm_policy_hashmax &&
541             total > hmask)
542                 return 1;
543
544         return 0;
545 }
546
547 static DEFINE_MUTEX(hash_resize_mutex);
548
549 static void xfrm_hash_resize(void *__unused)
550 {
551         int dir, total;
552
553         mutex_lock(&hash_resize_mutex);
554
555         total = 0;
556         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
557                 if (xfrm_bydst_should_resize(dir, &total))
558                         xfrm_bydst_resize(dir);
559         }
560         if (xfrm_byidx_should_resize(total))
561                 xfrm_byidx_resize(total);
562
563         mutex_unlock(&hash_resize_mutex);
564 }
565
566 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize, NULL);
567
568 /* Generate new index... KAME seems to generate them ordered by cost
569  * of an absolute inpredictability of ordering of rules. This will not pass. */
570 static u32 xfrm_gen_index(u8 type, int dir)
571 {
572         static u32 idx_generator;
573
574         for (;;) {
575                 struct hlist_node *entry;
576                 struct hlist_head *list;
577                 struct xfrm_policy *p;
578                 u32 idx;
579                 int found;
580
581                 idx = (idx_generator | dir);
582                 idx_generator += 8;
583                 if (idx == 0)
584                         idx = 8;
585                 list = xfrm_policy_byidx + idx_hash(idx);
586                 found = 0;
587                 hlist_for_each_entry(p, entry, list, byidx) {
588                         if (p->index == idx) {
589                                 found = 1;
590                                 break;
591                         }
592                 }
593                 if (!found)
594                         return idx;
595         }
596 }
597
598 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
599 {
600         u32 *p1 = (u32 *) s1;
601         u32 *p2 = (u32 *) s2;
602         int len = sizeof(struct xfrm_selector) / sizeof(u32);
603         int i;
604
605         for (i = 0; i < len; i++) {
606                 if (p1[i] != p2[i])
607                         return 1;
608         }
609
610         return 0;
611 }
612
613 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
614 {
615         struct xfrm_policy *pol;
616         struct xfrm_policy *delpol;
617         struct hlist_head *chain;
618         struct hlist_node *entry, *newpos, *last;
619         struct dst_entry *gc_list;
620
621         write_lock_bh(&xfrm_policy_lock);
622         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
623         delpol = NULL;
624         newpos = NULL;
625         last = NULL;
626         hlist_for_each_entry(pol, entry, chain, bydst) {
627                 if (!delpol &&
628                     pol->type == policy->type &&
629                     !selector_cmp(&pol->selector, &policy->selector) &&
630                     xfrm_sec_ctx_match(pol->security, policy->security)) {
631                         if (excl) {
632                                 write_unlock_bh(&xfrm_policy_lock);
633                                 return -EEXIST;
634                         }
635                         delpol = pol;
636                         if (policy->priority > pol->priority)
637                                 continue;
638                 } else if (policy->priority >= pol->priority) {
639                         last = &pol->bydst;
640                         continue;
641                 }
642                 if (!newpos)
643                         newpos = &pol->bydst;
644                 if (delpol)
645                         break;
646                 last = &pol->bydst;
647         }
648         if (!newpos)
649                 newpos = last;
650         if (newpos)
651                 hlist_add_after(newpos, &policy->bydst);
652         else
653                 hlist_add_head(&policy->bydst, chain);
654         xfrm_pol_hold(policy);
655         xfrm_policy_count[dir]++;
656         atomic_inc(&flow_cache_genid);
657         if (delpol) {
658                 hlist_del(&delpol->bydst);
659                 hlist_del(&delpol->byidx);
660                 xfrm_policy_count[dir]--;
661         }
662         policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
663         hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
664         policy->curlft.add_time = (unsigned long)xtime.tv_sec;
665         policy->curlft.use_time = 0;
666         if (!mod_timer(&policy->timer, jiffies + HZ))
667                 xfrm_pol_hold(policy);
668         write_unlock_bh(&xfrm_policy_lock);
669
670         if (delpol)
671                 xfrm_policy_kill(delpol);
672         else if (xfrm_bydst_should_resize(dir, NULL))
673                 schedule_work(&xfrm_hash_work);
674
675         read_lock_bh(&xfrm_policy_lock);
676         gc_list = NULL;
677         entry = &policy->bydst;
678         hlist_for_each_entry_continue(policy, entry, bydst) {
679                 struct dst_entry *dst;
680
681                 write_lock(&policy->lock);
682                 dst = policy->bundles;
683                 if (dst) {
684                         struct dst_entry *tail = dst;
685                         while (tail->next)
686                                 tail = tail->next;
687                         tail->next = gc_list;
688                         gc_list = dst;
689
690                         policy->bundles = NULL;
691                 }
692                 write_unlock(&policy->lock);
693         }
694         read_unlock_bh(&xfrm_policy_lock);
695
696         while (gc_list) {
697                 struct dst_entry *dst = gc_list;
698
699                 gc_list = dst->next;
700                 dst_free(dst);
701         }
702
703         return 0;
704 }
705 EXPORT_SYMBOL(xfrm_policy_insert);
706
707 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
708                                           struct xfrm_selector *sel,
709                                           struct xfrm_sec_ctx *ctx, int delete)
710 {
711         struct xfrm_policy *pol, *ret;
712         struct hlist_head *chain;
713         struct hlist_node *entry;
714
715         write_lock_bh(&xfrm_policy_lock);
716         chain = policy_hash_bysel(sel, sel->family, dir);
717         ret = NULL;
718         hlist_for_each_entry(pol, entry, chain, bydst) {
719                 if (pol->type == type &&
720                     !selector_cmp(sel, &pol->selector) &&
721                     xfrm_sec_ctx_match(ctx, pol->security)) {
722                         xfrm_pol_hold(pol);
723                         if (delete) {
724                                 hlist_del(&pol->bydst);
725                                 hlist_del(&pol->byidx);
726                                 xfrm_policy_count[dir]--;
727                         }
728                         ret = pol;
729                         break;
730                 }
731         }
732         write_unlock_bh(&xfrm_policy_lock);
733
734         if (ret && delete) {
735                 atomic_inc(&flow_cache_genid);
736                 xfrm_policy_kill(ret);
737         }
738         return ret;
739 }
740 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
741
742 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete)
743 {
744         struct xfrm_policy *pol, *ret;
745         struct hlist_head *chain;
746         struct hlist_node *entry;
747
748         write_lock_bh(&xfrm_policy_lock);
749         chain = xfrm_policy_byidx + idx_hash(id);
750         ret = NULL;
751         hlist_for_each_entry(pol, entry, chain, byidx) {
752                 if (pol->type == type && pol->index == id) {
753                         xfrm_pol_hold(pol);
754                         if (delete) {
755                                 hlist_del(&pol->bydst);
756                                 hlist_del(&pol->byidx);
757                                 xfrm_policy_count[dir]--;
758                         }
759                         ret = pol;
760                         break;
761                 }
762         }
763         write_unlock_bh(&xfrm_policy_lock);
764
765         if (ret && delete) {
766                 atomic_inc(&flow_cache_genid);
767                 xfrm_policy_kill(ret);
768         }
769         return ret;
770 }
771 EXPORT_SYMBOL(xfrm_policy_byid);
772
773 void xfrm_policy_flush(u8 type)
774 {
775         int dir;
776
777         write_lock_bh(&xfrm_policy_lock);
778         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
779                 struct xfrm_policy *pol;
780                 struct hlist_node *entry;
781                 int i, killed;
782
783                 killed = 0;
784         again1:
785                 hlist_for_each_entry(pol, entry,
786                                      &xfrm_policy_inexact[dir], bydst) {
787                         if (pol->type != type)
788                                 continue;
789                         hlist_del(&pol->bydst);
790                         hlist_del(&pol->byidx);
791                         write_unlock_bh(&xfrm_policy_lock);
792
793                         xfrm_policy_kill(pol);
794                         killed++;
795
796                         write_lock_bh(&xfrm_policy_lock);
797                         goto again1;
798                 }
799
800                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
801         again2:
802                         hlist_for_each_entry(pol, entry,
803                                              xfrm_policy_bydst[dir].table + i,
804                                              bydst) {
805                                 if (pol->type != type)
806                                         continue;
807                                 hlist_del(&pol->bydst);
808                                 hlist_del(&pol->byidx);
809                                 write_unlock_bh(&xfrm_policy_lock);
810
811                                 xfrm_policy_kill(pol);
812                                 killed++;
813
814                                 write_lock_bh(&xfrm_policy_lock);
815                                 goto again2;
816                         }
817                 }
818
819                 xfrm_policy_count[dir] -= killed;
820         }
821         atomic_inc(&flow_cache_genid);
822         write_unlock_bh(&xfrm_policy_lock);
823 }
824 EXPORT_SYMBOL(xfrm_policy_flush);
825
826 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
827                      void *data)
828 {
829         struct xfrm_policy *pol;
830         struct hlist_node *entry;
831         int dir, count, error;
832
833         read_lock_bh(&xfrm_policy_lock);
834         count = 0;
835         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
836                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
837                 int i;
838
839                 hlist_for_each_entry(pol, entry,
840                                      &xfrm_policy_inexact[dir], bydst) {
841                         if (pol->type == type)
842                                 count++;
843                 }
844                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
845                         hlist_for_each_entry(pol, entry, table + i, bydst) {
846                                 if (pol->type == type)
847                                         count++;
848                         }
849                 }
850         }
851
852         if (count == 0) {
853                 error = -ENOENT;
854                 goto out;
855         }
856
857         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
858                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
859                 int i;
860
861                 hlist_for_each_entry(pol, entry,
862                                      &xfrm_policy_inexact[dir], bydst) {
863                         if (pol->type != type)
864                                 continue;
865                         error = func(pol, dir % XFRM_POLICY_MAX, --count, data);
866                         if (error)
867                                 goto out;
868                 }
869                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
870                         hlist_for_each_entry(pol, entry, table + i, bydst) {
871                                 if (pol->type != type)
872                                         continue;
873                                 error = func(pol, dir % XFRM_POLICY_MAX, --count, data);
874                                 if (error)
875                                         goto out;
876                         }
877                 }
878         }
879         error = 0;
880 out:
881         read_unlock_bh(&xfrm_policy_lock);
882         return error;
883 }
884 EXPORT_SYMBOL(xfrm_policy_walk);
885
886 /*
887  * Find policy to apply to this flow.
888  *
889  * Returns 0 if policy found, else an -errno.
890  */
891 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
892                              u8 type, u16 family, int dir)
893 {
894         struct xfrm_selector *sel = &pol->selector;
895         int match, ret = -ESRCH;
896
897         if (pol->family != family ||
898             pol->type != type)
899                 return ret;
900
901         match = xfrm_selector_match(sel, fl, family);
902         if (match)
903                 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
904
905         return ret;
906 }
907
908 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
909                                                      u16 family, u8 dir)
910 {
911         int err;
912         struct xfrm_policy *pol, *ret;
913         xfrm_address_t *daddr, *saddr;
914         struct hlist_node *entry;
915         struct hlist_head *chain;
916         u32 priority = ~0U;
917
918         daddr = xfrm_flowi_daddr(fl, family);
919         saddr = xfrm_flowi_saddr(fl, family);
920         if (unlikely(!daddr || !saddr))
921                 return NULL;
922
923         read_lock_bh(&xfrm_policy_lock);
924         chain = policy_hash_direct(daddr, saddr, family, dir);
925         ret = NULL;
926         hlist_for_each_entry(pol, entry, chain, bydst) {
927                 err = xfrm_policy_match(pol, fl, type, family, dir);
928                 if (err) {
929                         if (err == -ESRCH)
930                                 continue;
931                         else {
932                                 ret = ERR_PTR(err);
933                                 goto fail;
934                         }
935                 } else {
936                         ret = pol;
937                         priority = ret->priority;
938                         break;
939                 }
940         }
941         chain = &xfrm_policy_inexact[dir];
942         hlist_for_each_entry(pol, entry, chain, bydst) {
943                 err = xfrm_policy_match(pol, fl, type, family, dir);
944                 if (err) {
945                         if (err == -ESRCH)
946                                 continue;
947                         else {
948                                 ret = ERR_PTR(err);
949                                 goto fail;
950                         }
951                 } else if (pol->priority < priority) {
952                         ret = pol;
953                         break;
954                 }
955         }
956         if (ret)
957                 xfrm_pol_hold(ret);
958 fail:
959         read_unlock_bh(&xfrm_policy_lock);
960
961         return ret;
962 }
963
964 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
965                                void **objp, atomic_t **obj_refp)
966 {
967         struct xfrm_policy *pol;
968         int err = 0;
969
970 #ifdef CONFIG_XFRM_SUB_POLICY
971         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
972         if (IS_ERR(pol)) {
973                 err = PTR_ERR(pol);
974                 pol = NULL;
975         }
976         if (pol || err)
977                 goto end;
978 #endif
979         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
980         if (IS_ERR(pol)) {
981                 err = PTR_ERR(pol);
982                 pol = NULL;
983         }
984 #ifdef CONFIG_XFRM_SUB_POLICY
985 end:
986 #endif
987         if ((*objp = (void *) pol) != NULL)
988                 *obj_refp = &pol->refcnt;
989         return err;
990 }
991
992 static inline int policy_to_flow_dir(int dir)
993 {
994         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
995             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
996             XFRM_POLICY_FWD == FLOW_DIR_FWD)
997                 return dir;
998         switch (dir) {
999         default:
1000         case XFRM_POLICY_IN:
1001                 return FLOW_DIR_IN;
1002         case XFRM_POLICY_OUT:
1003                 return FLOW_DIR_OUT;
1004         case XFRM_POLICY_FWD:
1005                 return FLOW_DIR_FWD;
1006         };
1007 }
1008
1009 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1010 {
1011         struct xfrm_policy *pol;
1012
1013         read_lock_bh(&xfrm_policy_lock);
1014         if ((pol = sk->sk_policy[dir]) != NULL) {
1015                 int match = xfrm_selector_match(&pol->selector, fl,
1016                                                 sk->sk_family);
1017                 int err = 0;
1018
1019                 if (match) {
1020                         err = security_xfrm_policy_lookup(pol, fl->secid,
1021                                         policy_to_flow_dir(dir));
1022                         if (!err)
1023                                 xfrm_pol_hold(pol);
1024                         else if (err == -ESRCH)
1025                                 pol = NULL;
1026                         else
1027                                 pol = ERR_PTR(err);
1028                 } else
1029                         pol = NULL;
1030         }
1031         read_unlock_bh(&xfrm_policy_lock);
1032         return pol;
1033 }
1034
1035 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1036 {
1037         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1038                                                      pol->family, dir);
1039
1040         hlist_add_head(&pol->bydst, chain);
1041         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1042         xfrm_policy_count[dir]++;
1043         xfrm_pol_hold(pol);
1044
1045         if (xfrm_bydst_should_resize(dir, NULL))
1046                 schedule_work(&xfrm_hash_work);
1047 }
1048
1049 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1050                                                 int dir)
1051 {
1052         if (hlist_unhashed(&pol->bydst))
1053                 return NULL;
1054
1055         hlist_del(&pol->bydst);
1056         hlist_del(&pol->byidx);
1057         xfrm_policy_count[dir]--;
1058
1059         return pol;
1060 }
1061
1062 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1063 {
1064         write_lock_bh(&xfrm_policy_lock);
1065         pol = __xfrm_policy_unlink(pol, dir);
1066         write_unlock_bh(&xfrm_policy_lock);
1067         if (pol) {
1068                 if (dir < XFRM_POLICY_MAX)
1069                         atomic_inc(&flow_cache_genid);
1070                 xfrm_policy_kill(pol);
1071                 return 0;
1072         }
1073         return -ENOENT;
1074 }
1075 EXPORT_SYMBOL(xfrm_policy_delete);
1076
1077 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1078 {
1079         struct xfrm_policy *old_pol;
1080
1081 #ifdef CONFIG_XFRM_SUB_POLICY
1082         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1083                 return -EINVAL;
1084 #endif
1085
1086         write_lock_bh(&xfrm_policy_lock);
1087         old_pol = sk->sk_policy[dir];
1088         sk->sk_policy[dir] = pol;
1089         if (pol) {
1090                 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
1091                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1092                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1093         }
1094         if (old_pol)
1095                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1096         write_unlock_bh(&xfrm_policy_lock);
1097
1098         if (old_pol) {
1099                 xfrm_policy_kill(old_pol);
1100         }
1101         return 0;
1102 }
1103
1104 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1105 {
1106         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1107
1108         if (newp) {
1109                 newp->selector = old->selector;
1110                 if (security_xfrm_policy_clone(old, newp)) {
1111                         kfree(newp);
1112                         return NULL;  /* ENOMEM */
1113                 }
1114                 newp->lft = old->lft;
1115                 newp->curlft = old->curlft;
1116                 newp->action = old->action;
1117                 newp->flags = old->flags;
1118                 newp->xfrm_nr = old->xfrm_nr;
1119                 newp->index = old->index;
1120                 newp->type = old->type;
1121                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1122                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1123                 write_lock_bh(&xfrm_policy_lock);
1124                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1125                 write_unlock_bh(&xfrm_policy_lock);
1126                 xfrm_pol_put(newp);
1127         }
1128         return newp;
1129 }
1130
1131 int __xfrm_sk_clone_policy(struct sock *sk)
1132 {
1133         struct xfrm_policy *p0 = sk->sk_policy[0],
1134                            *p1 = sk->sk_policy[1];
1135
1136         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1137         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1138                 return -ENOMEM;
1139         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1140                 return -ENOMEM;
1141         return 0;
1142 }
1143
1144 static int
1145 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1146                unsigned short family)
1147 {
1148         int err;
1149         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1150
1151         if (unlikely(afinfo == NULL))
1152                 return -EINVAL;
1153         err = afinfo->get_saddr(local, remote);
1154         xfrm_policy_put_afinfo(afinfo);
1155         return err;
1156 }
1157
1158 /* Resolve list of templates for the flow, given policy. */
1159
1160 static int
1161 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1162                       struct xfrm_state **xfrm,
1163                       unsigned short family)
1164 {
1165         int nx;
1166         int i, error;
1167         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1168         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1169         xfrm_address_t tmp;
1170
1171         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1172                 struct xfrm_state *x;
1173                 xfrm_address_t *remote = daddr;
1174                 xfrm_address_t *local  = saddr;
1175                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1176
1177                 if (tmpl->mode == XFRM_MODE_TUNNEL) {
1178                         remote = &tmpl->id.daddr;
1179                         local = &tmpl->saddr;
1180                         if (xfrm_addr_any(local, family)) {
1181                                 error = xfrm_get_saddr(&tmp, remote, family);
1182                                 if (error)
1183                                         goto fail;
1184                                 local = &tmp;
1185                         }
1186                 }
1187
1188                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1189
1190                 if (x && x->km.state == XFRM_STATE_VALID) {
1191                         xfrm[nx++] = x;
1192                         daddr = remote;
1193                         saddr = local;
1194                         continue;
1195                 }
1196                 if (x) {
1197                         error = (x->km.state == XFRM_STATE_ERROR ?
1198                                  -EINVAL : -EAGAIN);
1199                         xfrm_state_put(x);
1200                 }
1201
1202                 if (!tmpl->optional)
1203                         goto fail;
1204         }
1205         return nx;
1206
1207 fail:
1208         for (nx--; nx>=0; nx--)
1209                 xfrm_state_put(xfrm[nx]);
1210         return error;
1211 }
1212
1213 static int
1214 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1215                   struct xfrm_state **xfrm,
1216                   unsigned short family)
1217 {
1218         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1219         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1220         int cnx = 0;
1221         int error;
1222         int ret;
1223         int i;
1224
1225         for (i = 0; i < npols; i++) {
1226                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1227                         error = -ENOBUFS;
1228                         goto fail;
1229                 }
1230
1231                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1232                 if (ret < 0) {
1233                         error = ret;
1234                         goto fail;
1235                 } else
1236                         cnx += ret;
1237         }
1238
1239         /* found states are sorted for outbound processing */
1240         if (npols > 1)
1241                 xfrm_state_sort(xfrm, tpp, cnx, family);
1242
1243         return cnx;
1244
1245  fail:
1246         for (cnx--; cnx>=0; cnx--)
1247                 xfrm_state_put(tpp[cnx]);
1248         return error;
1249
1250 }
1251
1252 /* Check that the bundle accepts the flow and its components are
1253  * still valid.
1254  */
1255
1256 static struct dst_entry *
1257 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1258 {
1259         struct dst_entry *x;
1260         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1261         if (unlikely(afinfo == NULL))
1262                 return ERR_PTR(-EINVAL);
1263         x = afinfo->find_bundle(fl, policy);
1264         xfrm_policy_put_afinfo(afinfo);
1265         return x;
1266 }
1267
1268 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1269  * all the metrics... Shortly, bundle a bundle.
1270  */
1271
1272 static int
1273 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
1274                    struct flowi *fl, struct dst_entry **dst_p,
1275                    unsigned short family)
1276 {
1277         int err;
1278         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1279         if (unlikely(afinfo == NULL))
1280                 return -EINVAL;
1281         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
1282         xfrm_policy_put_afinfo(afinfo);
1283         return err;
1284 }
1285
1286
1287 static int stale_bundle(struct dst_entry *dst);
1288
1289 /* Main function: finds/creates a bundle for given flow.
1290  *
1291  * At the moment we eat a raw IP route. Mostly to speed up lookups
1292  * on interfaces with disabled IPsec.
1293  */
1294 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1295                 struct sock *sk, int flags)
1296 {
1297         struct xfrm_policy *policy;
1298         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1299         int npols;
1300         int pol_dead;
1301         int xfrm_nr;
1302         int pi;
1303         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1304         struct dst_entry *dst, *dst_orig = *dst_p;
1305         int nx = 0;
1306         int err;
1307         u32 genid;
1308         u16 family;
1309         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1310
1311 restart:
1312         genid = atomic_read(&flow_cache_genid);
1313         policy = NULL;
1314         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1315                 pols[pi] = NULL;
1316         npols = 0;
1317         pol_dead = 0;
1318         xfrm_nr = 0;
1319
1320         if (sk && sk->sk_policy[1]) {
1321                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1322                 if (IS_ERR(policy))
1323                         return PTR_ERR(policy);
1324         }
1325
1326         if (!policy) {
1327                 /* To accelerate a bit...  */
1328                 if ((dst_orig->flags & DST_NOXFRM) ||
1329                     !xfrm_policy_count[XFRM_POLICY_OUT])
1330                         return 0;
1331
1332                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1333                                            dir, xfrm_policy_lookup);
1334                 if (IS_ERR(policy))
1335                         return PTR_ERR(policy);
1336         }
1337
1338         if (!policy)
1339                 return 0;
1340
1341         family = dst_orig->ops->family;
1342         policy->curlft.use_time = (unsigned long)xtime.tv_sec;
1343         pols[0] = policy;
1344         npols ++;
1345         xfrm_nr += pols[0]->xfrm_nr;
1346
1347         switch (policy->action) {
1348         case XFRM_POLICY_BLOCK:
1349                 /* Prohibit the flow */
1350                 err = -EPERM;
1351                 goto error;
1352
1353         case XFRM_POLICY_ALLOW:
1354 #ifndef CONFIG_XFRM_SUB_POLICY
1355                 if (policy->xfrm_nr == 0) {
1356                         /* Flow passes not transformed. */
1357                         xfrm_pol_put(policy);
1358                         return 0;
1359                 }
1360 #endif
1361
1362                 /* Try to find matching bundle.
1363                  *
1364                  * LATER: help from flow cache. It is optional, this
1365                  * is required only for output policy.
1366                  */
1367                 dst = xfrm_find_bundle(fl, policy, family);
1368                 if (IS_ERR(dst)) {
1369                         err = PTR_ERR(dst);
1370                         goto error;
1371                 }
1372
1373                 if (dst)
1374                         break;
1375
1376 #ifdef CONFIG_XFRM_SUB_POLICY
1377                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1378                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1379                                                             fl, family,
1380                                                             XFRM_POLICY_OUT);
1381                         if (pols[1]) {
1382                                 if (IS_ERR(pols[1])) {
1383                                         err = PTR_ERR(pols[1]);
1384                                         goto error;
1385                                 }
1386                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1387                                         err = -EPERM;
1388                                         goto error;
1389                                 }
1390                                 npols ++;
1391                                 xfrm_nr += pols[1]->xfrm_nr;
1392                         }
1393                 }
1394
1395                 /*
1396                  * Because neither flowi nor bundle information knows about
1397                  * transformation template size. On more than one policy usage
1398                  * we can realize whether all of them is bypass or not after
1399                  * they are searched. See above not-transformed bypass
1400                  * is surrounded by non-sub policy configuration, too.
1401                  */
1402                 if (xfrm_nr == 0) {
1403                         /* Flow passes not transformed. */
1404                         xfrm_pols_put(pols, npols);
1405                         return 0;
1406                 }
1407
1408 #endif
1409                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1410
1411                 if (unlikely(nx<0)) {
1412                         err = nx;
1413                         if (err == -EAGAIN && flags) {
1414                                 DECLARE_WAITQUEUE(wait, current);
1415
1416                                 add_wait_queue(&km_waitq, &wait);
1417                                 set_current_state(TASK_INTERRUPTIBLE);
1418                                 schedule();
1419                                 set_current_state(TASK_RUNNING);
1420                                 remove_wait_queue(&km_waitq, &wait);
1421
1422                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1423
1424                                 if (nx == -EAGAIN && signal_pending(current)) {
1425                                         err = -ERESTART;
1426                                         goto error;
1427                                 }
1428                                 if (nx == -EAGAIN ||
1429                                     genid != atomic_read(&flow_cache_genid)) {
1430                                         xfrm_pols_put(pols, npols);
1431                                         goto restart;
1432                                 }
1433                                 err = nx;
1434                         }
1435                         if (err < 0)
1436                                 goto error;
1437                 }
1438                 if (nx == 0) {
1439                         /* Flow passes not transformed. */
1440                         xfrm_pols_put(pols, npols);
1441                         return 0;
1442                 }
1443
1444                 dst = dst_orig;
1445                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
1446
1447                 if (unlikely(err)) {
1448                         int i;
1449                         for (i=0; i<nx; i++)
1450                                 xfrm_state_put(xfrm[i]);
1451                         goto error;
1452                 }
1453
1454                 for (pi = 0; pi < npols; pi++) {
1455                         read_lock_bh(&pols[pi]->lock);
1456                         pol_dead |= pols[pi]->dead;
1457                         read_unlock_bh(&pols[pi]->lock);
1458                 }
1459
1460                 write_lock_bh(&policy->lock);
1461                 if (unlikely(pol_dead || stale_bundle(dst))) {
1462                         /* Wow! While we worked on resolving, this
1463                          * policy has gone. Retry. It is not paranoia,
1464                          * we just cannot enlist new bundle to dead object.
1465                          * We can't enlist stable bundles either.
1466                          */
1467                         write_unlock_bh(&policy->lock);
1468                         if (dst)
1469                                 dst_free(dst);
1470
1471                         err = -EHOSTUNREACH;
1472                         goto error;
1473                 }
1474                 dst->next = policy->bundles;
1475                 policy->bundles = dst;
1476                 dst_hold(dst);
1477                 write_unlock_bh(&policy->lock);
1478         }
1479         *dst_p = dst;
1480         dst_release(dst_orig);
1481         xfrm_pols_put(pols, npols);
1482         return 0;
1483
1484 error:
1485         dst_release(dst_orig);
1486         xfrm_pols_put(pols, npols);
1487         *dst_p = NULL;
1488         return err;
1489 }
1490 EXPORT_SYMBOL(xfrm_lookup);
1491
1492 static inline int
1493 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1494 {
1495         struct xfrm_state *x;
1496         int err;
1497
1498         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1499                 return 0;
1500         x = skb->sp->xvec[idx];
1501         if (!x->type->reject)
1502                 return 0;
1503         xfrm_state_hold(x);
1504         err = x->type->reject(x, skb, fl);
1505         xfrm_state_put(x);
1506         return err;
1507 }
1508
1509 /* When skb is transformed back to its "native" form, we have to
1510  * check policy restrictions. At the moment we make this in maximally
1511  * stupid way. Shame on me. :-) Of course, connected sockets must
1512  * have policy cached at them.
1513  */
1514
1515 static inline int
1516 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x, 
1517               unsigned short family)
1518 {
1519         if (xfrm_state_kern(x))
1520                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
1521         return  x->id.proto == tmpl->id.proto &&
1522                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1523                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1524                 x->props.mode == tmpl->mode &&
1525                 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1526                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1527                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1528                   xfrm_state_addr_cmp(tmpl, x, family));
1529 }
1530
1531 /*
1532  * 0 or more than 0 is returned when validation is succeeded (either bypass
1533  * because of optional transport mode, or next index of the mathced secpath
1534  * state with the template.
1535  * -1 is returned when no matching template is found.
1536  * Otherwise "-2 - errored_index" is returned.
1537  */
1538 static inline int
1539 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1540                unsigned short family)
1541 {
1542         int idx = start;
1543
1544         if (tmpl->optional) {
1545                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1546                         return start;
1547         } else
1548                 start = -1;
1549         for (; idx < sp->len; idx++) {
1550                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1551                         return ++idx;
1552                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1553                         if (start == -1)
1554                                 start = -2-idx;
1555                         break;
1556                 }
1557         }
1558         return start;
1559 }
1560
1561 int
1562 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1563 {
1564         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1565         int err;
1566
1567         if (unlikely(afinfo == NULL))
1568                 return -EAFNOSUPPORT;
1569
1570         afinfo->decode_session(skb, fl);
1571         err = security_xfrm_decode_session(skb, &fl->secid);
1572         xfrm_policy_put_afinfo(afinfo);
1573         return err;
1574 }
1575 EXPORT_SYMBOL(xfrm_decode_session);
1576
1577 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1578 {
1579         for (; k < sp->len; k++) {
1580                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1581                         *idxp = k;
1582                         return 1;
1583                 }
1584         }
1585
1586         return 0;
1587 }
1588
1589 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb, 
1590                         unsigned short family)
1591 {
1592         struct xfrm_policy *pol;
1593         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1594         int npols = 0;
1595         int xfrm_nr;
1596         int pi;
1597         struct flowi fl;
1598         u8 fl_dir = policy_to_flow_dir(dir);
1599         int xerr_idx = -1;
1600
1601         if (xfrm_decode_session(skb, &fl, family) < 0)
1602                 return 0;
1603         nf_nat_decode_session(skb, &fl, family);
1604
1605         /* First, check used SA against their selectors. */
1606         if (skb->sp) {
1607                 int i;
1608
1609                 for (i=skb->sp->len-1; i>=0; i--) {
1610                         struct xfrm_state *x = skb->sp->xvec[i];
1611                         if (!xfrm_selector_match(&x->sel, &fl, family))
1612                                 return 0;
1613                 }
1614         }
1615
1616         pol = NULL;
1617         if (sk && sk->sk_policy[dir]) {
1618                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1619                 if (IS_ERR(pol))
1620                         return 0;
1621         }
1622
1623         if (!pol)
1624                 pol = flow_cache_lookup(&fl, family, fl_dir,
1625                                         xfrm_policy_lookup);
1626
1627         if (IS_ERR(pol))
1628                 return 0;
1629
1630         if (!pol) {
1631                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1632                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1633                         return 0;
1634                 }
1635                 return 1;
1636         }
1637
1638         pol->curlft.use_time = (unsigned long)xtime.tv_sec;
1639
1640         pols[0] = pol;
1641         npols ++;
1642 #ifdef CONFIG_XFRM_SUB_POLICY
1643         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1644                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1645                                                     &fl, family,
1646                                                     XFRM_POLICY_IN);
1647                 if (pols[1]) {
1648                         if (IS_ERR(pols[1]))
1649                                 return 0;
1650                         pols[1]->curlft.use_time = (unsigned long)xtime.tv_sec;
1651                         npols ++;
1652                 }
1653         }
1654 #endif
1655
1656         if (pol->action == XFRM_POLICY_ALLOW) {
1657                 struct sec_path *sp;
1658                 static struct sec_path dummy;
1659                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1660                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1661                 struct xfrm_tmpl **tpp = tp;
1662                 int ti = 0;
1663                 int i, k;
1664
1665                 if ((sp = skb->sp) == NULL)
1666                         sp = &dummy;
1667
1668                 for (pi = 0; pi < npols; pi++) {
1669                         if (pols[pi] != pol &&
1670                             pols[pi]->action != XFRM_POLICY_ALLOW)
1671                                 goto reject;
1672                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1673                                 goto reject_error;
1674                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1675                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1676                 }
1677                 xfrm_nr = ti;
1678                 if (npols > 1) {
1679                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1680                         tpp = stp;
1681                 }
1682
1683                 /* For each tunnel xfrm, find the first matching tmpl.
1684                  * For each tmpl before that, find corresponding xfrm.
1685                  * Order is _important_. Later we will implement
1686                  * some barriers, but at the moment barriers
1687                  * are implied between each two transformations.
1688                  */
1689                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1690                         k = xfrm_policy_ok(tpp[i], sp, k, family);
1691                         if (k < 0) {
1692                                 if (k < -1)
1693                                         /* "-2 - errored_index" returned */
1694                                         xerr_idx = -(2+k);
1695                                 goto reject;
1696                         }
1697                 }
1698
1699                 if (secpath_has_nontransport(sp, k, &xerr_idx))
1700                         goto reject;
1701
1702                 xfrm_pols_put(pols, npols);
1703                 return 1;
1704         }
1705
1706 reject:
1707         xfrm_secpath_reject(xerr_idx, skb, &fl);
1708 reject_error:
1709         xfrm_pols_put(pols, npols);
1710         return 0;
1711 }
1712 EXPORT_SYMBOL(__xfrm_policy_check);
1713
1714 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1715 {
1716         struct flowi fl;
1717
1718         if (xfrm_decode_session(skb, &fl, family) < 0)
1719                 return 0;
1720
1721         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1722 }
1723 EXPORT_SYMBOL(__xfrm_route_forward);
1724
1725 /* Optimize later using cookies and generation ids. */
1726
1727 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1728 {
1729         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1730          * to "-1" to force all XFRM destinations to get validated by
1731          * dst_ops->check on every use.  We do this because when a
1732          * normal route referenced by an XFRM dst is obsoleted we do
1733          * not go looking around for all parent referencing XFRM dsts
1734          * so that we can invalidate them.  It is just too much work.
1735          * Instead we make the checks here on every use.  For example:
1736          *
1737          *      XFRM dst A --> IPv4 dst X
1738          *
1739          * X is the "xdst->route" of A (X is also the "dst->path" of A
1740          * in this example).  If X is marked obsolete, "A" will not
1741          * notice.  That's what we are validating here via the
1742          * stale_bundle() check.
1743          *
1744          * When a policy's bundle is pruned, we dst_free() the XFRM
1745          * dst which causes it's ->obsolete field to be set to a
1746          * positive non-zero integer.  If an XFRM dst has been pruned
1747          * like this, we want to force a new route lookup.
1748          */
1749         if (dst->obsolete < 0 && !stale_bundle(dst))
1750                 return dst;
1751
1752         return NULL;
1753 }
1754
1755 static int stale_bundle(struct dst_entry *dst)
1756 {
1757         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1758 }
1759
1760 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1761 {
1762         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1763                 dst->dev = &loopback_dev;
1764                 dev_hold(&loopback_dev);
1765                 dev_put(dev);
1766         }
1767 }
1768 EXPORT_SYMBOL(xfrm_dst_ifdown);
1769
1770 static void xfrm_link_failure(struct sk_buff *skb)
1771 {
1772         /* Impossible. Such dst must be popped before reaches point of failure. */
1773         return;
1774 }
1775
1776 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1777 {
1778         if (dst) {
1779                 if (dst->obsolete) {
1780                         dst_release(dst);
1781                         dst = NULL;
1782                 }
1783         }
1784         return dst;
1785 }
1786
1787 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1788 {
1789         struct dst_entry *dst, **dstp;
1790
1791         write_lock(&pol->lock);
1792         dstp = &pol->bundles;
1793         while ((dst=*dstp) != NULL) {
1794                 if (func(dst)) {
1795                         *dstp = dst->next;
1796                         dst->next = *gc_list_p;
1797                         *gc_list_p = dst;
1798                 } else {
1799                         dstp = &dst->next;
1800                 }
1801         }
1802         write_unlock(&pol->lock);
1803 }
1804
1805 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1806 {
1807         struct dst_entry *gc_list = NULL;
1808         int dir;
1809
1810         read_lock_bh(&xfrm_policy_lock);
1811         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
1812                 struct xfrm_policy *pol;
1813                 struct hlist_node *entry;
1814                 struct hlist_head *table;
1815                 int i;
1816
1817                 hlist_for_each_entry(pol, entry,
1818                                      &xfrm_policy_inexact[dir], bydst)
1819                         prune_one_bundle(pol, func, &gc_list);
1820
1821                 table = xfrm_policy_bydst[dir].table;
1822                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
1823                         hlist_for_each_entry(pol, entry, table + i, bydst)
1824                                 prune_one_bundle(pol, func, &gc_list);
1825                 }
1826         }
1827         read_unlock_bh(&xfrm_policy_lock);
1828
1829         while (gc_list) {
1830                 struct dst_entry *dst = gc_list;
1831                 gc_list = dst->next;
1832                 dst_free(dst);
1833         }
1834 }
1835
1836 static int unused_bundle(struct dst_entry *dst)
1837 {
1838         return !atomic_read(&dst->__refcnt);
1839 }
1840
1841 static void __xfrm_garbage_collect(void)
1842 {
1843         xfrm_prune_bundles(unused_bundle);
1844 }
1845
1846 static int xfrm_flush_bundles(void)
1847 {
1848         xfrm_prune_bundles(stale_bundle);
1849         return 0;
1850 }
1851
1852 void xfrm_init_pmtu(struct dst_entry *dst)
1853 {
1854         do {
1855                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1856                 u32 pmtu, route_mtu_cached;
1857
1858                 pmtu = dst_mtu(dst->child);
1859                 xdst->child_mtu_cached = pmtu;
1860
1861                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1862
1863                 route_mtu_cached = dst_mtu(xdst->route);
1864                 xdst->route_mtu_cached = route_mtu_cached;
1865
1866                 if (pmtu > route_mtu_cached)
1867                         pmtu = route_mtu_cached;
1868
1869                 dst->metrics[RTAX_MTU-1] = pmtu;
1870         } while ((dst = dst->next));
1871 }
1872
1873 EXPORT_SYMBOL(xfrm_init_pmtu);
1874
1875 /* Check that the bundle accepts the flow and its components are
1876  * still valid.
1877  */
1878
1879 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
1880                 struct flowi *fl, int family, int strict)
1881 {
1882         struct dst_entry *dst = &first->u.dst;
1883         struct xfrm_dst *last;
1884         u32 mtu;
1885
1886         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1887             (dst->dev && !netif_running(dst->dev)))
1888                 return 0;
1889
1890         last = NULL;
1891
1892         do {
1893                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1894
1895                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1896                         return 0;
1897                 if (fl && !security_xfrm_flow_state_match(fl, dst->xfrm, pol))
1898                         return 0;
1899                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1900                         return 0;
1901                 if (xdst->genid != dst->xfrm->genid)
1902                         return 0;
1903
1904                 if (strict && fl && dst->xfrm->props.mode != XFRM_MODE_TUNNEL &&
1905                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
1906                         return 0;
1907
1908                 mtu = dst_mtu(dst->child);
1909                 if (xdst->child_mtu_cached != mtu) {
1910                         last = xdst;
1911                         xdst->child_mtu_cached = mtu;
1912                 }
1913
1914                 if (!dst_check(xdst->route, xdst->route_cookie))
1915                         return 0;
1916                 mtu = dst_mtu(xdst->route);
1917                 if (xdst->route_mtu_cached != mtu) {
1918                         last = xdst;
1919                         xdst->route_mtu_cached = mtu;
1920                 }
1921
1922                 dst = dst->child;
1923         } while (dst->xfrm);
1924
1925         if (likely(!last))
1926                 return 1;
1927
1928         mtu = last->child_mtu_cached;
1929         for (;;) {
1930                 dst = &last->u.dst;
1931
1932                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1933                 if (mtu > last->route_mtu_cached)
1934                         mtu = last->route_mtu_cached;
1935                 dst->metrics[RTAX_MTU-1] = mtu;
1936
1937                 if (last == first)
1938                         break;
1939
1940                 last = last->u.next;
1941                 last->child_mtu_cached = mtu;
1942         }
1943
1944         return 1;
1945 }
1946
1947 EXPORT_SYMBOL(xfrm_bundle_ok);
1948
1949 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1950 {
1951         int err = 0;
1952         if (unlikely(afinfo == NULL))
1953                 return -EINVAL;
1954         if (unlikely(afinfo->family >= NPROTO))
1955                 return -EAFNOSUPPORT;
1956         write_lock_bh(&xfrm_policy_afinfo_lock);
1957         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1958                 err = -ENOBUFS;
1959         else {
1960                 struct dst_ops *dst_ops = afinfo->dst_ops;
1961                 if (likely(dst_ops->kmem_cachep == NULL))
1962                         dst_ops->kmem_cachep = xfrm_dst_cache;
1963                 if (likely(dst_ops->check == NULL))
1964                         dst_ops->check = xfrm_dst_check;
1965                 if (likely(dst_ops->negative_advice == NULL))
1966                         dst_ops->negative_advice = xfrm_negative_advice;
1967                 if (likely(dst_ops->link_failure == NULL))
1968                         dst_ops->link_failure = xfrm_link_failure;
1969                 if (likely(afinfo->garbage_collect == NULL))
1970                         afinfo->garbage_collect = __xfrm_garbage_collect;
1971                 xfrm_policy_afinfo[afinfo->family] = afinfo;
1972         }
1973         write_unlock_bh(&xfrm_policy_afinfo_lock);
1974         return err;
1975 }
1976 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
1977
1978 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1979 {
1980         int err = 0;
1981         if (unlikely(afinfo == NULL))
1982                 return -EINVAL;
1983         if (unlikely(afinfo->family >= NPROTO))
1984                 return -EAFNOSUPPORT;
1985         write_lock_bh(&xfrm_policy_afinfo_lock);
1986         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1987                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1988                         err = -EINVAL;
1989                 else {
1990                         struct dst_ops *dst_ops = afinfo->dst_ops;
1991                         xfrm_policy_afinfo[afinfo->family] = NULL;
1992                         dst_ops->kmem_cachep = NULL;
1993                         dst_ops->check = NULL;
1994                         dst_ops->negative_advice = NULL;
1995                         dst_ops->link_failure = NULL;
1996                         afinfo->garbage_collect = NULL;
1997                 }
1998         }
1999         write_unlock_bh(&xfrm_policy_afinfo_lock);
2000         return err;
2001 }
2002 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2003
2004 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2005 {
2006         struct xfrm_policy_afinfo *afinfo;
2007         if (unlikely(family >= NPROTO))
2008                 return NULL;
2009         read_lock(&xfrm_policy_afinfo_lock);
2010         afinfo = xfrm_policy_afinfo[family];
2011         if (unlikely(!afinfo))
2012                 read_unlock(&xfrm_policy_afinfo_lock);
2013         return afinfo;
2014 }
2015
2016 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2017 {
2018         read_unlock(&xfrm_policy_afinfo_lock);
2019 }
2020
2021 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
2022 {
2023         struct xfrm_policy_afinfo *afinfo;
2024         if (unlikely(family >= NPROTO))
2025                 return NULL;
2026         write_lock_bh(&xfrm_policy_afinfo_lock);
2027         afinfo = xfrm_policy_afinfo[family];
2028         if (unlikely(!afinfo))
2029                 write_unlock_bh(&xfrm_policy_afinfo_lock);
2030         return afinfo;
2031 }
2032
2033 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
2034 {
2035         write_unlock_bh(&xfrm_policy_afinfo_lock);
2036 }
2037
2038 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2039 {
2040         switch (event) {
2041         case NETDEV_DOWN:
2042                 xfrm_flush_bundles();
2043         }
2044         return NOTIFY_DONE;
2045 }
2046
2047 static struct notifier_block xfrm_dev_notifier = {
2048         xfrm_dev_event,
2049         NULL,
2050         0
2051 };
2052
2053 static void __init xfrm_policy_init(void)
2054 {
2055         unsigned int hmask, sz;
2056         int dir;
2057
2058         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2059                                            sizeof(struct xfrm_dst),
2060                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2061                                            NULL, NULL);
2062
2063         hmask = 8 - 1;
2064         sz = (hmask+1) * sizeof(struct hlist_head);
2065
2066         xfrm_policy_byidx = xfrm_hash_alloc(sz);
2067         xfrm_idx_hmask = hmask;
2068         if (!xfrm_policy_byidx)
2069                 panic("XFRM: failed to allocate byidx hash\n");
2070
2071         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2072                 struct xfrm_policy_hash *htab;
2073
2074                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2075
2076                 htab = &xfrm_policy_bydst[dir];
2077                 htab->table = xfrm_hash_alloc(sz);
2078                 htab->hmask = hmask;
2079                 if (!htab->table)
2080                         panic("XFRM: failed to allocate bydst hash\n");
2081         }
2082
2083         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
2084         register_netdevice_notifier(&xfrm_dev_notifier);
2085 }
2086
2087 void __init xfrm_init(void)
2088 {
2089         xfrm_state_init();
2090         xfrm_policy_init();
2091         xfrm_input_init();
2092 }
2093