f6c77bd36fddf7f8d597121c156b8d4c0d879c73
[linux-3.10.git] / net / xfrm / xfrm_policy.c
1 /* 
2  * xfrm_policy.c
3  *
4  * Changes:
5  *      Mitsuru KANDA @USAGI
6  *      Kazunori MIYAZAWA @USAGI
7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
8  *              IPv6 support
9  *      Kazunori MIYAZAWA @USAGI
10  *      YOSHIFUJI Hideaki
11  *              Split up af-specific portion
12  *      Derek Atkins <derek@ihtfp.com>          Add the post_input processor
13  *
14  */
15
16 #include <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 static inline int
54 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
55 {
56         return  addr_match(&fl->fl4_dst, &sel->daddr, sel->prefixlen_d) &&
57                 addr_match(&fl->fl4_src, &sel->saddr, sel->prefixlen_s) &&
58                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
59                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
60                 (fl->proto == sel->proto || !sel->proto) &&
61                 (fl->oif == sel->ifindex || !sel->ifindex);
62 }
63
64 static inline int
65 __xfrm6_selector_match(struct xfrm_selector *sel, struct flowi *fl)
66 {
67         return  addr_match(&fl->fl6_dst, &sel->daddr, sel->prefixlen_d) &&
68                 addr_match(&fl->fl6_src, &sel->saddr, sel->prefixlen_s) &&
69                 !((xfrm_flowi_dport(fl) ^ sel->dport) & sel->dport_mask) &&
70                 !((xfrm_flowi_sport(fl) ^ sel->sport) & sel->sport_mask) &&
71                 (fl->proto == sel->proto || !sel->proto) &&
72                 (fl->oif == sel->ifindex || !sel->ifindex);
73 }
74
75 int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
76                     unsigned short family)
77 {
78         switch (family) {
79         case AF_INET:
80                 return __xfrm4_selector_match(sel, fl);
81         case AF_INET6:
82                 return __xfrm6_selector_match(sel, fl);
83         }
84         return 0;
85 }
86
87 int xfrm_register_type(struct xfrm_type *type, unsigned short family)
88 {
89         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
90         struct xfrm_type **typemap;
91         int err = 0;
92
93         if (unlikely(afinfo == NULL))
94                 return -EAFNOSUPPORT;
95         typemap = afinfo->type_map;
96
97         if (likely(typemap[type->proto] == NULL))
98                 typemap[type->proto] = type;
99         else
100                 err = -EEXIST;
101         xfrm_policy_unlock_afinfo(afinfo);
102         return err;
103 }
104 EXPORT_SYMBOL(xfrm_register_type);
105
106 int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
107 {
108         struct xfrm_policy_afinfo *afinfo = xfrm_policy_lock_afinfo(family);
109         struct xfrm_type **typemap;
110         int err = 0;
111
112         if (unlikely(afinfo == NULL))
113                 return -EAFNOSUPPORT;
114         typemap = afinfo->type_map;
115
116         if (unlikely(typemap[type->proto] != type))
117                 err = -ENOENT;
118         else
119                 typemap[type->proto] = NULL;
120         xfrm_policy_unlock_afinfo(afinfo);
121         return err;
122 }
123 EXPORT_SYMBOL(xfrm_unregister_type);
124
125 struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
126 {
127         struct xfrm_policy_afinfo *afinfo;
128         struct xfrm_type **typemap;
129         struct xfrm_type *type;
130         int modload_attempted = 0;
131
132 retry:
133         afinfo = xfrm_policy_get_afinfo(family);
134         if (unlikely(afinfo == NULL))
135                 return NULL;
136         typemap = afinfo->type_map;
137
138         type = typemap[proto];
139         if (unlikely(type && !try_module_get(type->owner)))
140                 type = NULL;
141         if (!type && !modload_attempted) {
142                 xfrm_policy_put_afinfo(afinfo);
143                 request_module("xfrm-type-%d-%d",
144                                (int) family, (int) proto);
145                 modload_attempted = 1;
146                 goto retry;
147         }
148
149         xfrm_policy_put_afinfo(afinfo);
150         return type;
151 }
152
153 int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl, 
154                     unsigned short family)
155 {
156         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
157         int err = 0;
158
159         if (unlikely(afinfo == NULL))
160                 return -EAFNOSUPPORT;
161
162         if (likely(afinfo->dst_lookup != NULL))
163                 err = afinfo->dst_lookup(dst, fl);
164         else
165                 err = -EINVAL;
166         xfrm_policy_put_afinfo(afinfo);
167         return err;
168 }
169 EXPORT_SYMBOL(xfrm_dst_lookup);
170
171 void xfrm_put_type(struct xfrm_type *type)
172 {
173         module_put(type->owner);
174 }
175
176 int xfrm_register_mode(struct xfrm_mode *mode, int family)
177 {
178         struct xfrm_policy_afinfo *afinfo;
179         struct xfrm_mode **modemap;
180         int err;
181
182         if (unlikely(mode->encap >= XFRM_MODE_MAX))
183                 return -EINVAL;
184
185         afinfo = xfrm_policy_lock_afinfo(family);
186         if (unlikely(afinfo == NULL))
187                 return -EAFNOSUPPORT;
188
189         err = -EEXIST;
190         modemap = afinfo->mode_map;
191         if (likely(modemap[mode->encap] == NULL)) {
192                 modemap[mode->encap] = mode;
193                 err = 0;
194         }
195
196         xfrm_policy_unlock_afinfo(afinfo);
197         return err;
198 }
199 EXPORT_SYMBOL(xfrm_register_mode);
200
201 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
202 {
203         struct xfrm_policy_afinfo *afinfo;
204         struct xfrm_mode **modemap;
205         int err;
206
207         if (unlikely(mode->encap >= XFRM_MODE_MAX))
208                 return -EINVAL;
209
210         afinfo = xfrm_policy_lock_afinfo(family);
211         if (unlikely(afinfo == NULL))
212                 return -EAFNOSUPPORT;
213
214         err = -ENOENT;
215         modemap = afinfo->mode_map;
216         if (likely(modemap[mode->encap] == mode)) {
217                 modemap[mode->encap] = NULL;
218                 err = 0;
219         }
220
221         xfrm_policy_unlock_afinfo(afinfo);
222         return err;
223 }
224 EXPORT_SYMBOL(xfrm_unregister_mode);
225
226 struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
227 {
228         struct xfrm_policy_afinfo *afinfo;
229         struct xfrm_mode *mode;
230         int modload_attempted = 0;
231
232         if (unlikely(encap >= XFRM_MODE_MAX))
233                 return NULL;
234
235 retry:
236         afinfo = xfrm_policy_get_afinfo(family);
237         if (unlikely(afinfo == NULL))
238                 return NULL;
239
240         mode = afinfo->mode_map[encap];
241         if (unlikely(mode && !try_module_get(mode->owner)))
242                 mode = NULL;
243         if (!mode && !modload_attempted) {
244                 xfrm_policy_put_afinfo(afinfo);
245                 request_module("xfrm-mode-%d-%d", family, encap);
246                 modload_attempted = 1;
247                 goto retry;
248         }
249
250         xfrm_policy_put_afinfo(afinfo);
251         return mode;
252 }
253
254 void xfrm_put_mode(struct xfrm_mode *mode)
255 {
256         module_put(mode->owner);
257 }
258
259 static inline unsigned long make_jiffies(long secs)
260 {
261         if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
262                 return MAX_SCHEDULE_TIMEOUT-1;
263         else
264                 return secs*HZ;
265 }
266
267 static void xfrm_policy_timer(unsigned long data)
268 {
269         struct xfrm_policy *xp = (struct xfrm_policy*)data;
270         unsigned long now = (unsigned long)xtime.tv_sec;
271         long next = LONG_MAX;
272         int warn = 0;
273         int dir;
274
275         read_lock(&xp->lock);
276
277         if (xp->dead)
278                 goto out;
279
280         dir = xfrm_policy_id2dir(xp->index);
281
282         if (xp->lft.hard_add_expires_seconds) {
283                 long tmo = xp->lft.hard_add_expires_seconds +
284                         xp->curlft.add_time - now;
285                 if (tmo <= 0)
286                         goto expired;
287                 if (tmo < next)
288                         next = tmo;
289         }
290         if (xp->lft.hard_use_expires_seconds) {
291                 long tmo = xp->lft.hard_use_expires_seconds +
292                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
293                 if (tmo <= 0)
294                         goto expired;
295                 if (tmo < next)
296                         next = tmo;
297         }
298         if (xp->lft.soft_add_expires_seconds) {
299                 long tmo = xp->lft.soft_add_expires_seconds +
300                         xp->curlft.add_time - now;
301                 if (tmo <= 0) {
302                         warn = 1;
303                         tmo = XFRM_KM_TIMEOUT;
304                 }
305                 if (tmo < next)
306                         next = tmo;
307         }
308         if (xp->lft.soft_use_expires_seconds) {
309                 long tmo = xp->lft.soft_use_expires_seconds +
310                         (xp->curlft.use_time ? : xp->curlft.add_time) - now;
311                 if (tmo <= 0) {
312                         warn = 1;
313                         tmo = XFRM_KM_TIMEOUT;
314                 }
315                 if (tmo < next)
316                         next = tmo;
317         }
318
319         if (warn)
320                 km_policy_expired(xp, dir, 0, 0);
321         if (next != LONG_MAX &&
322             !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
323                 xfrm_pol_hold(xp);
324
325 out:
326         read_unlock(&xp->lock);
327         xfrm_pol_put(xp);
328         return;
329
330 expired:
331         read_unlock(&xp->lock);
332         if (!xfrm_policy_delete(xp, dir))
333                 km_policy_expired(xp, dir, 1, 0);
334         xfrm_pol_put(xp);
335 }
336
337
338 /* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
339  * SPD calls.
340  */
341
342 struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
343 {
344         struct xfrm_policy *policy;
345
346         policy = kzalloc(sizeof(struct xfrm_policy), gfp);
347
348         if (policy) {
349                 INIT_HLIST_NODE(&policy->bydst);
350                 INIT_HLIST_NODE(&policy->byidx);
351                 rwlock_init(&policy->lock);
352                 atomic_set(&policy->refcnt, 1);
353                 init_timer(&policy->timer);
354                 policy->timer.data = (unsigned long)policy;
355                 policy->timer.function = xfrm_policy_timer;
356         }
357         return policy;
358 }
359 EXPORT_SYMBOL(xfrm_policy_alloc);
360
361 /* Destroy xfrm_policy: descendant resources must be released to this moment. */
362
363 void __xfrm_policy_destroy(struct xfrm_policy *policy)
364 {
365         BUG_ON(!policy->dead);
366
367         BUG_ON(policy->bundles);
368
369         if (del_timer(&policy->timer))
370                 BUG();
371
372         security_xfrm_policy_free(policy);
373         kfree(policy);
374 }
375 EXPORT_SYMBOL(__xfrm_policy_destroy);
376
377 static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
378 {
379         struct dst_entry *dst;
380
381         while ((dst = policy->bundles) != NULL) {
382                 policy->bundles = dst->next;
383                 dst_free(dst);
384         }
385
386         if (del_timer(&policy->timer))
387                 atomic_dec(&policy->refcnt);
388
389         if (atomic_read(&policy->refcnt) > 1)
390                 flow_cache_flush();
391
392         xfrm_pol_put(policy);
393 }
394
395 static void xfrm_policy_gc_task(struct work_struct *work)
396 {
397         struct xfrm_policy *policy;
398         struct hlist_node *entry, *tmp;
399         struct hlist_head gc_list;
400
401         spin_lock_bh(&xfrm_policy_gc_lock);
402         gc_list.first = xfrm_policy_gc_list.first;
403         INIT_HLIST_HEAD(&xfrm_policy_gc_list);
404         spin_unlock_bh(&xfrm_policy_gc_lock);
405
406         hlist_for_each_entry_safe(policy, entry, tmp, &gc_list, bydst)
407                 xfrm_policy_gc_kill(policy);
408 }
409
410 /* Rule must be locked. Release descentant resources, announce
411  * entry dead. The rule must be unlinked from lists to the moment.
412  */
413
414 static void xfrm_policy_kill(struct xfrm_policy *policy)
415 {
416         int dead;
417
418         write_lock_bh(&policy->lock);
419         dead = policy->dead;
420         policy->dead = 1;
421         write_unlock_bh(&policy->lock);
422
423         if (unlikely(dead)) {
424                 WARN_ON(1);
425                 return;
426         }
427
428         spin_lock(&xfrm_policy_gc_lock);
429         hlist_add_head(&policy->bydst, &xfrm_policy_gc_list);
430         spin_unlock(&xfrm_policy_gc_lock);
431
432         schedule_work(&xfrm_policy_gc_work);
433 }
434
435 struct xfrm_policy_hash {
436         struct hlist_head       *table;
437         unsigned int            hmask;
438 };
439
440 static struct hlist_head xfrm_policy_inexact[XFRM_POLICY_MAX*2];
441 static struct xfrm_policy_hash xfrm_policy_bydst[XFRM_POLICY_MAX*2] __read_mostly;
442 static struct hlist_head *xfrm_policy_byidx __read_mostly;
443 static unsigned int xfrm_idx_hmask __read_mostly;
444 static unsigned int xfrm_policy_hashmax __read_mostly = 1 * 1024 * 1024;
445
446 static inline unsigned int idx_hash(u32 index)
447 {
448         return __idx_hash(index, xfrm_idx_hmask);
449 }
450
451 static struct hlist_head *policy_hash_bysel(struct xfrm_selector *sel, unsigned short family, int dir)
452 {
453         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
454         unsigned int hash = __sel_hash(sel, family, hmask);
455
456         return (hash == hmask + 1 ?
457                 &xfrm_policy_inexact[dir] :
458                 xfrm_policy_bydst[dir].table + hash);
459 }
460
461 static struct hlist_head *policy_hash_direct(xfrm_address_t *daddr, xfrm_address_t *saddr, unsigned short family, int dir)
462 {
463         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
464         unsigned int hash = __addr_hash(daddr, saddr, family, hmask);
465
466         return xfrm_policy_bydst[dir].table + hash;
467 }
468
469 static void xfrm_dst_hash_transfer(struct hlist_head *list,
470                                    struct hlist_head *ndsttable,
471                                    unsigned int nhashmask)
472 {
473         struct hlist_node *entry, *tmp;
474         struct xfrm_policy *pol;
475
476         hlist_for_each_entry_safe(pol, entry, tmp, list, bydst) {
477                 unsigned int h;
478
479                 h = __addr_hash(&pol->selector.daddr, &pol->selector.saddr,
480                                 pol->family, nhashmask);
481                 hlist_add_head(&pol->bydst, ndsttable+h);
482         }
483 }
484
485 static void xfrm_idx_hash_transfer(struct hlist_head *list,
486                                    struct hlist_head *nidxtable,
487                                    unsigned int nhashmask)
488 {
489         struct hlist_node *entry, *tmp;
490         struct xfrm_policy *pol;
491
492         hlist_for_each_entry_safe(pol, entry, tmp, list, byidx) {
493                 unsigned int h;
494
495                 h = __idx_hash(pol->index, nhashmask);
496                 hlist_add_head(&pol->byidx, nidxtable+h);
497         }
498 }
499
500 static unsigned long xfrm_new_hash_mask(unsigned int old_hmask)
501 {
502         return ((old_hmask + 1) << 1) - 1;
503 }
504
505 static void xfrm_bydst_resize(int dir)
506 {
507         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
508         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
509         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
510         struct hlist_head *odst = xfrm_policy_bydst[dir].table;
511         struct hlist_head *ndst = xfrm_hash_alloc(nsize);
512         int i;
513
514         if (!ndst)
515                 return;
516
517         write_lock_bh(&xfrm_policy_lock);
518
519         for (i = hmask; i >= 0; i--)
520                 xfrm_dst_hash_transfer(odst + i, ndst, nhashmask);
521
522         xfrm_policy_bydst[dir].table = ndst;
523         xfrm_policy_bydst[dir].hmask = nhashmask;
524
525         write_unlock_bh(&xfrm_policy_lock);
526
527         xfrm_hash_free(odst, (hmask + 1) * sizeof(struct hlist_head));
528 }
529
530 static void xfrm_byidx_resize(int total)
531 {
532         unsigned int hmask = xfrm_idx_hmask;
533         unsigned int nhashmask = xfrm_new_hash_mask(hmask);
534         unsigned int nsize = (nhashmask + 1) * sizeof(struct hlist_head);
535         struct hlist_head *oidx = xfrm_policy_byidx;
536         struct hlist_head *nidx = xfrm_hash_alloc(nsize);
537         int i;
538
539         if (!nidx)
540                 return;
541
542         write_lock_bh(&xfrm_policy_lock);
543
544         for (i = hmask; i >= 0; i--)
545                 xfrm_idx_hash_transfer(oidx + i, nidx, nhashmask);
546
547         xfrm_policy_byidx = nidx;
548         xfrm_idx_hmask = nhashmask;
549
550         write_unlock_bh(&xfrm_policy_lock);
551
552         xfrm_hash_free(oidx, (hmask + 1) * sizeof(struct hlist_head));
553 }
554
555 static inline int xfrm_bydst_should_resize(int dir, int *total)
556 {
557         unsigned int cnt = xfrm_policy_count[dir];
558         unsigned int hmask = xfrm_policy_bydst[dir].hmask;
559
560         if (total)
561                 *total += cnt;
562
563         if ((hmask + 1) < xfrm_policy_hashmax &&
564             cnt > hmask)
565                 return 1;
566
567         return 0;
568 }
569
570 static inline int xfrm_byidx_should_resize(int total)
571 {
572         unsigned int hmask = xfrm_idx_hmask;
573
574         if ((hmask + 1) < xfrm_policy_hashmax &&
575             total > hmask)
576                 return 1;
577
578         return 0;
579 }
580
581 static DEFINE_MUTEX(hash_resize_mutex);
582
583 static void xfrm_hash_resize(struct work_struct *__unused)
584 {
585         int dir, total;
586
587         mutex_lock(&hash_resize_mutex);
588
589         total = 0;
590         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
591                 if (xfrm_bydst_should_resize(dir, &total))
592                         xfrm_bydst_resize(dir);
593         }
594         if (xfrm_byidx_should_resize(total))
595                 xfrm_byidx_resize(total);
596
597         mutex_unlock(&hash_resize_mutex);
598 }
599
600 static DECLARE_WORK(xfrm_hash_work, xfrm_hash_resize);
601
602 /* Generate new index... KAME seems to generate them ordered by cost
603  * of an absolute inpredictability of ordering of rules. This will not pass. */
604 static u32 xfrm_gen_index(u8 type, int dir)
605 {
606         static u32 idx_generator;
607
608         for (;;) {
609                 struct hlist_node *entry;
610                 struct hlist_head *list;
611                 struct xfrm_policy *p;
612                 u32 idx;
613                 int found;
614
615                 idx = (idx_generator | dir);
616                 idx_generator += 8;
617                 if (idx == 0)
618                         idx = 8;
619                 list = xfrm_policy_byidx + idx_hash(idx);
620                 found = 0;
621                 hlist_for_each_entry(p, entry, list, byidx) {
622                         if (p->index == idx) {
623                                 found = 1;
624                                 break;
625                         }
626                 }
627                 if (!found)
628                         return idx;
629         }
630 }
631
632 static inline int selector_cmp(struct xfrm_selector *s1, struct xfrm_selector *s2)
633 {
634         u32 *p1 = (u32 *) s1;
635         u32 *p2 = (u32 *) s2;
636         int len = sizeof(struct xfrm_selector) / sizeof(u32);
637         int i;
638
639         for (i = 0; i < len; i++) {
640                 if (p1[i] != p2[i])
641                         return 1;
642         }
643
644         return 0;
645 }
646
647 int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
648 {
649         struct xfrm_policy *pol;
650         struct xfrm_policy *delpol;
651         struct hlist_head *chain;
652         struct hlist_node *entry, *newpos, *last;
653         struct dst_entry *gc_list;
654
655         write_lock_bh(&xfrm_policy_lock);
656         chain = policy_hash_bysel(&policy->selector, policy->family, dir);
657         delpol = NULL;
658         newpos = NULL;
659         last = NULL;
660         hlist_for_each_entry(pol, entry, chain, bydst) {
661                 if (!delpol &&
662                     pol->type == policy->type &&
663                     !selector_cmp(&pol->selector, &policy->selector) &&
664                     xfrm_sec_ctx_match(pol->security, policy->security)) {
665                         if (excl) {
666                                 write_unlock_bh(&xfrm_policy_lock);
667                                 return -EEXIST;
668                         }
669                         delpol = pol;
670                         if (policy->priority > pol->priority)
671                                 continue;
672                 } else if (policy->priority >= pol->priority) {
673                         last = &pol->bydst;
674                         continue;
675                 }
676                 if (!newpos)
677                         newpos = &pol->bydst;
678                 if (delpol)
679                         break;
680                 last = &pol->bydst;
681         }
682         if (!newpos)
683                 newpos = last;
684         if (newpos)
685                 hlist_add_after(newpos, &policy->bydst);
686         else
687                 hlist_add_head(&policy->bydst, chain);
688         xfrm_pol_hold(policy);
689         xfrm_policy_count[dir]++;
690         atomic_inc(&flow_cache_genid);
691         if (delpol) {
692                 hlist_del(&delpol->bydst);
693                 hlist_del(&delpol->byidx);
694                 xfrm_policy_count[dir]--;
695         }
696         policy->index = delpol ? delpol->index : xfrm_gen_index(policy->type, dir);
697         hlist_add_head(&policy->byidx, xfrm_policy_byidx+idx_hash(policy->index));
698         policy->curlft.add_time = (unsigned long)xtime.tv_sec;
699         policy->curlft.use_time = 0;
700         if (!mod_timer(&policy->timer, jiffies + HZ))
701                 xfrm_pol_hold(policy);
702         write_unlock_bh(&xfrm_policy_lock);
703
704         if (delpol)
705                 xfrm_policy_kill(delpol);
706         else if (xfrm_bydst_should_resize(dir, NULL))
707                 schedule_work(&xfrm_hash_work);
708
709         read_lock_bh(&xfrm_policy_lock);
710         gc_list = NULL;
711         entry = &policy->bydst;
712         hlist_for_each_entry_continue(policy, entry, bydst) {
713                 struct dst_entry *dst;
714
715                 write_lock(&policy->lock);
716                 dst = policy->bundles;
717                 if (dst) {
718                         struct dst_entry *tail = dst;
719                         while (tail->next)
720                                 tail = tail->next;
721                         tail->next = gc_list;
722                         gc_list = dst;
723
724                         policy->bundles = NULL;
725                 }
726                 write_unlock(&policy->lock);
727         }
728         read_unlock_bh(&xfrm_policy_lock);
729
730         while (gc_list) {
731                 struct dst_entry *dst = gc_list;
732
733                 gc_list = dst->next;
734                 dst_free(dst);
735         }
736
737         return 0;
738 }
739 EXPORT_SYMBOL(xfrm_policy_insert);
740
741 struct xfrm_policy *xfrm_policy_bysel_ctx(u8 type, int dir,
742                                           struct xfrm_selector *sel,
743                                           struct xfrm_sec_ctx *ctx, int delete)
744 {
745         struct xfrm_policy *pol, *ret;
746         struct hlist_head *chain;
747         struct hlist_node *entry;
748
749         write_lock_bh(&xfrm_policy_lock);
750         chain = policy_hash_bysel(sel, sel->family, dir);
751         ret = NULL;
752         hlist_for_each_entry(pol, entry, chain, bydst) {
753                 if (pol->type == type &&
754                     !selector_cmp(sel, &pol->selector) &&
755                     xfrm_sec_ctx_match(ctx, pol->security)) {
756                         xfrm_pol_hold(pol);
757                         if (delete) {
758                                 hlist_del(&pol->bydst);
759                                 hlist_del(&pol->byidx);
760                                 xfrm_policy_count[dir]--;
761                         }
762                         ret = pol;
763                         break;
764                 }
765         }
766         write_unlock_bh(&xfrm_policy_lock);
767
768         if (ret && delete) {
769                 atomic_inc(&flow_cache_genid);
770                 xfrm_policy_kill(ret);
771         }
772         return ret;
773 }
774 EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
775
776 struct xfrm_policy *xfrm_policy_byid(u8 type, int dir, u32 id, int delete)
777 {
778         struct xfrm_policy *pol, *ret;
779         struct hlist_head *chain;
780         struct hlist_node *entry;
781
782         write_lock_bh(&xfrm_policy_lock);
783         chain = xfrm_policy_byidx + idx_hash(id);
784         ret = NULL;
785         hlist_for_each_entry(pol, entry, chain, byidx) {
786                 if (pol->type == type && pol->index == id) {
787                         xfrm_pol_hold(pol);
788                         if (delete) {
789                                 hlist_del(&pol->bydst);
790                                 hlist_del(&pol->byidx);
791                                 xfrm_policy_count[dir]--;
792                         }
793                         ret = pol;
794                         break;
795                 }
796         }
797         write_unlock_bh(&xfrm_policy_lock);
798
799         if (ret && delete) {
800                 atomic_inc(&flow_cache_genid);
801                 xfrm_policy_kill(ret);
802         }
803         return ret;
804 }
805 EXPORT_SYMBOL(xfrm_policy_byid);
806
807 void xfrm_policy_flush(u8 type)
808 {
809         int dir;
810
811         write_lock_bh(&xfrm_policy_lock);
812         for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
813                 struct xfrm_policy *pol;
814                 struct hlist_node *entry;
815                 int i, killed;
816
817                 killed = 0;
818         again1:
819                 hlist_for_each_entry(pol, entry,
820                                      &xfrm_policy_inexact[dir], bydst) {
821                         if (pol->type != type)
822                                 continue;
823                         hlist_del(&pol->bydst);
824                         hlist_del(&pol->byidx);
825                         write_unlock_bh(&xfrm_policy_lock);
826
827                         xfrm_policy_kill(pol);
828                         killed++;
829
830                         write_lock_bh(&xfrm_policy_lock);
831                         goto again1;
832                 }
833
834                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
835         again2:
836                         hlist_for_each_entry(pol, entry,
837                                              xfrm_policy_bydst[dir].table + i,
838                                              bydst) {
839                                 if (pol->type != type)
840                                         continue;
841                                 hlist_del(&pol->bydst);
842                                 hlist_del(&pol->byidx);
843                                 write_unlock_bh(&xfrm_policy_lock);
844
845                                 xfrm_policy_kill(pol);
846                                 killed++;
847
848                                 write_lock_bh(&xfrm_policy_lock);
849                                 goto again2;
850                         }
851                 }
852
853                 xfrm_policy_count[dir] -= killed;
854         }
855         atomic_inc(&flow_cache_genid);
856         write_unlock_bh(&xfrm_policy_lock);
857 }
858 EXPORT_SYMBOL(xfrm_policy_flush);
859
860 int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
861                      void *data)
862 {
863         struct xfrm_policy *pol;
864         struct hlist_node *entry;
865         int dir, count, error;
866
867         read_lock_bh(&xfrm_policy_lock);
868         count = 0;
869         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
870                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
871                 int i;
872
873                 hlist_for_each_entry(pol, entry,
874                                      &xfrm_policy_inexact[dir], bydst) {
875                         if (pol->type == type)
876                                 count++;
877                 }
878                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
879                         hlist_for_each_entry(pol, entry, table + i, bydst) {
880                                 if (pol->type == type)
881                                         count++;
882                         }
883                 }
884         }
885
886         if (count == 0) {
887                 error = -ENOENT;
888                 goto out;
889         }
890
891         for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
892                 struct hlist_head *table = xfrm_policy_bydst[dir].table;
893                 int i;
894
895                 hlist_for_each_entry(pol, entry,
896                                      &xfrm_policy_inexact[dir], bydst) {
897                         if (pol->type != type)
898                                 continue;
899                         error = func(pol, dir % XFRM_POLICY_MAX, --count, data);
900                         if (error)
901                                 goto out;
902                 }
903                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
904                         hlist_for_each_entry(pol, entry, table + i, bydst) {
905                                 if (pol->type != type)
906                                         continue;
907                                 error = func(pol, dir % XFRM_POLICY_MAX, --count, data);
908                                 if (error)
909                                         goto out;
910                         }
911                 }
912         }
913         error = 0;
914 out:
915         read_unlock_bh(&xfrm_policy_lock);
916         return error;
917 }
918 EXPORT_SYMBOL(xfrm_policy_walk);
919
920 /*
921  * Find policy to apply to this flow.
922  *
923  * Returns 0 if policy found, else an -errno.
924  */
925 static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
926                              u8 type, u16 family, int dir)
927 {
928         struct xfrm_selector *sel = &pol->selector;
929         int match, ret = -ESRCH;
930
931         if (pol->family != family ||
932             pol->type != type)
933                 return ret;
934
935         match = xfrm_selector_match(sel, fl, family);
936         if (match)
937                 ret = security_xfrm_policy_lookup(pol, fl->secid, dir);
938
939         return ret;
940 }
941
942 static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
943                                                      u16 family, u8 dir)
944 {
945         int err;
946         struct xfrm_policy *pol, *ret;
947         xfrm_address_t *daddr, *saddr;
948         struct hlist_node *entry;
949         struct hlist_head *chain;
950         u32 priority = ~0U;
951
952         daddr = xfrm_flowi_daddr(fl, family);
953         saddr = xfrm_flowi_saddr(fl, family);
954         if (unlikely(!daddr || !saddr))
955                 return NULL;
956
957         read_lock_bh(&xfrm_policy_lock);
958         chain = policy_hash_direct(daddr, saddr, family, dir);
959         ret = NULL;
960         hlist_for_each_entry(pol, entry, chain, bydst) {
961                 err = xfrm_policy_match(pol, fl, type, family, dir);
962                 if (err) {
963                         if (err == -ESRCH)
964                                 continue;
965                         else {
966                                 ret = ERR_PTR(err);
967                                 goto fail;
968                         }
969                 } else {
970                         ret = pol;
971                         priority = ret->priority;
972                         break;
973                 }
974         }
975         chain = &xfrm_policy_inexact[dir];
976         hlist_for_each_entry(pol, entry, chain, bydst) {
977                 err = xfrm_policy_match(pol, fl, type, family, dir);
978                 if (err) {
979                         if (err == -ESRCH)
980                                 continue;
981                         else {
982                                 ret = ERR_PTR(err);
983                                 goto fail;
984                         }
985                 } else if (pol->priority < priority) {
986                         ret = pol;
987                         break;
988                 }
989         }
990         if (ret)
991                 xfrm_pol_hold(ret);
992 fail:
993         read_unlock_bh(&xfrm_policy_lock);
994
995         return ret;
996 }
997
998 static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
999                                void **objp, atomic_t **obj_refp)
1000 {
1001         struct xfrm_policy *pol;
1002         int err = 0;
1003
1004 #ifdef CONFIG_XFRM_SUB_POLICY
1005         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
1006         if (IS_ERR(pol)) {
1007                 err = PTR_ERR(pol);
1008                 pol = NULL;
1009         }
1010         if (pol || err)
1011                 goto end;
1012 #endif
1013         pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
1014         if (IS_ERR(pol)) {
1015                 err = PTR_ERR(pol);
1016                 pol = NULL;
1017         }
1018 #ifdef CONFIG_XFRM_SUB_POLICY
1019 end:
1020 #endif
1021         if ((*objp = (void *) pol) != NULL)
1022                 *obj_refp = &pol->refcnt;
1023         return err;
1024 }
1025
1026 static inline int policy_to_flow_dir(int dir)
1027 {
1028         if (XFRM_POLICY_IN == FLOW_DIR_IN &&
1029             XFRM_POLICY_OUT == FLOW_DIR_OUT &&
1030             XFRM_POLICY_FWD == FLOW_DIR_FWD)
1031                 return dir;
1032         switch (dir) {
1033         default:
1034         case XFRM_POLICY_IN:
1035                 return FLOW_DIR_IN;
1036         case XFRM_POLICY_OUT:
1037                 return FLOW_DIR_OUT;
1038         case XFRM_POLICY_FWD:
1039                 return FLOW_DIR_FWD;
1040         };
1041 }
1042
1043 static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl)
1044 {
1045         struct xfrm_policy *pol;
1046
1047         read_lock_bh(&xfrm_policy_lock);
1048         if ((pol = sk->sk_policy[dir]) != NULL) {
1049                 int match = xfrm_selector_match(&pol->selector, fl,
1050                                                 sk->sk_family);
1051                 int err = 0;
1052
1053                 if (match) {
1054                         err = security_xfrm_policy_lookup(pol, fl->secid,
1055                                         policy_to_flow_dir(dir));
1056                         if (!err)
1057                                 xfrm_pol_hold(pol);
1058                         else if (err == -ESRCH)
1059                                 pol = NULL;
1060                         else
1061                                 pol = ERR_PTR(err);
1062                 } else
1063                         pol = NULL;
1064         }
1065         read_unlock_bh(&xfrm_policy_lock);
1066         return pol;
1067 }
1068
1069 static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
1070 {
1071         struct hlist_head *chain = policy_hash_bysel(&pol->selector,
1072                                                      pol->family, dir);
1073
1074         hlist_add_head(&pol->bydst, chain);
1075         hlist_add_head(&pol->byidx, xfrm_policy_byidx+idx_hash(pol->index));
1076         xfrm_policy_count[dir]++;
1077         xfrm_pol_hold(pol);
1078
1079         if (xfrm_bydst_should_resize(dir, NULL))
1080                 schedule_work(&xfrm_hash_work);
1081 }
1082
1083 static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
1084                                                 int dir)
1085 {
1086         if (hlist_unhashed(&pol->bydst))
1087                 return NULL;
1088
1089         hlist_del(&pol->bydst);
1090         hlist_del(&pol->byidx);
1091         xfrm_policy_count[dir]--;
1092
1093         return pol;
1094 }
1095
1096 int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1097 {
1098         write_lock_bh(&xfrm_policy_lock);
1099         pol = __xfrm_policy_unlink(pol, dir);
1100         write_unlock_bh(&xfrm_policy_lock);
1101         if (pol) {
1102                 if (dir < XFRM_POLICY_MAX)
1103                         atomic_inc(&flow_cache_genid);
1104                 xfrm_policy_kill(pol);
1105                 return 0;
1106         }
1107         return -ENOENT;
1108 }
1109 EXPORT_SYMBOL(xfrm_policy_delete);
1110
1111 int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
1112 {
1113         struct xfrm_policy *old_pol;
1114
1115 #ifdef CONFIG_XFRM_SUB_POLICY
1116         if (pol && pol->type != XFRM_POLICY_TYPE_MAIN)
1117                 return -EINVAL;
1118 #endif
1119
1120         write_lock_bh(&xfrm_policy_lock);
1121         old_pol = sk->sk_policy[dir];
1122         sk->sk_policy[dir] = pol;
1123         if (pol) {
1124                 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
1125                 pol->index = xfrm_gen_index(pol->type, XFRM_POLICY_MAX+dir);
1126                 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
1127         }
1128         if (old_pol)
1129                 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
1130         write_unlock_bh(&xfrm_policy_lock);
1131
1132         if (old_pol) {
1133                 xfrm_policy_kill(old_pol);
1134         }
1135         return 0;
1136 }
1137
1138 static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
1139 {
1140         struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
1141
1142         if (newp) {
1143                 newp->selector = old->selector;
1144                 if (security_xfrm_policy_clone(old, newp)) {
1145                         kfree(newp);
1146                         return NULL;  /* ENOMEM */
1147                 }
1148                 newp->lft = old->lft;
1149                 newp->curlft = old->curlft;
1150                 newp->action = old->action;
1151                 newp->flags = old->flags;
1152                 newp->xfrm_nr = old->xfrm_nr;
1153                 newp->index = old->index;
1154                 newp->type = old->type;
1155                 memcpy(newp->xfrm_vec, old->xfrm_vec,
1156                        newp->xfrm_nr*sizeof(struct xfrm_tmpl));
1157                 write_lock_bh(&xfrm_policy_lock);
1158                 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
1159                 write_unlock_bh(&xfrm_policy_lock);
1160                 xfrm_pol_put(newp);
1161         }
1162         return newp;
1163 }
1164
1165 int __xfrm_sk_clone_policy(struct sock *sk)
1166 {
1167         struct xfrm_policy *p0 = sk->sk_policy[0],
1168                            *p1 = sk->sk_policy[1];
1169
1170         sk->sk_policy[0] = sk->sk_policy[1] = NULL;
1171         if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
1172                 return -ENOMEM;
1173         if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
1174                 return -ENOMEM;
1175         return 0;
1176 }
1177
1178 static int
1179 xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
1180                unsigned short family)
1181 {
1182         int err;
1183         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1184
1185         if (unlikely(afinfo == NULL))
1186                 return -EINVAL;
1187         err = afinfo->get_saddr(local, remote);
1188         xfrm_policy_put_afinfo(afinfo);
1189         return err;
1190 }
1191
1192 /* Resolve list of templates for the flow, given policy. */
1193
1194 static int
1195 xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
1196                       struct xfrm_state **xfrm,
1197                       unsigned short family)
1198 {
1199         int nx;
1200         int i, error;
1201         xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
1202         xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
1203         xfrm_address_t tmp;
1204
1205         for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
1206                 struct xfrm_state *x;
1207                 xfrm_address_t *remote = daddr;
1208                 xfrm_address_t *local  = saddr;
1209                 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
1210
1211                 if (tmpl->mode == XFRM_MODE_TUNNEL) {
1212                         remote = &tmpl->id.daddr;
1213                         local = &tmpl->saddr;
1214                         family = tmpl->encap_family;
1215                         if (xfrm_addr_any(local, family)) {
1216                                 error = xfrm_get_saddr(&tmp, remote, family);
1217                                 if (error)
1218                                         goto fail;
1219                                 local = &tmp;
1220                         }
1221                 }
1222
1223                 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
1224
1225                 if (x && x->km.state == XFRM_STATE_VALID) {
1226                         xfrm[nx++] = x;
1227                         daddr = remote;
1228                         saddr = local;
1229                         continue;
1230                 }
1231                 if (x) {
1232                         error = (x->km.state == XFRM_STATE_ERROR ?
1233                                  -EINVAL : -EAGAIN);
1234                         xfrm_state_put(x);
1235                 }
1236
1237                 if (!tmpl->optional)
1238                         goto fail;
1239         }
1240         return nx;
1241
1242 fail:
1243         for (nx--; nx>=0; nx--)
1244                 xfrm_state_put(xfrm[nx]);
1245         return error;
1246 }
1247
1248 static int
1249 xfrm_tmpl_resolve(struct xfrm_policy **pols, int npols, struct flowi *fl,
1250                   struct xfrm_state **xfrm,
1251                   unsigned short family)
1252 {
1253         struct xfrm_state *tp[XFRM_MAX_DEPTH];
1254         struct xfrm_state **tpp = (npols > 1) ? tp : xfrm;
1255         int cnx = 0;
1256         int error;
1257         int ret;
1258         int i;
1259
1260         for (i = 0; i < npols; i++) {
1261                 if (cnx + pols[i]->xfrm_nr >= XFRM_MAX_DEPTH) {
1262                         error = -ENOBUFS;
1263                         goto fail;
1264                 }
1265
1266                 ret = xfrm_tmpl_resolve_one(pols[i], fl, &tpp[cnx], family);
1267                 if (ret < 0) {
1268                         error = ret;
1269                         goto fail;
1270                 } else
1271                         cnx += ret;
1272         }
1273
1274         /* found states are sorted for outbound processing */
1275         if (npols > 1)
1276                 xfrm_state_sort(xfrm, tpp, cnx, family);
1277
1278         return cnx;
1279
1280  fail:
1281         for (cnx--; cnx>=0; cnx--)
1282                 xfrm_state_put(tpp[cnx]);
1283         return error;
1284
1285 }
1286
1287 /* Check that the bundle accepts the flow and its components are
1288  * still valid.
1289  */
1290
1291 static struct dst_entry *
1292 xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
1293 {
1294         struct dst_entry *x;
1295         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1296         if (unlikely(afinfo == NULL))
1297                 return ERR_PTR(-EINVAL);
1298         x = afinfo->find_bundle(fl, policy);
1299         xfrm_policy_put_afinfo(afinfo);
1300         return x;
1301 }
1302
1303 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
1304  * all the metrics... Shortly, bundle a bundle.
1305  */
1306
1307 static int
1308 xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
1309                    struct flowi *fl, struct dst_entry **dst_p,
1310                    unsigned short family)
1311 {
1312         int err;
1313         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1314         if (unlikely(afinfo == NULL))
1315                 return -EINVAL;
1316         err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
1317         xfrm_policy_put_afinfo(afinfo);
1318         return err;
1319 }
1320
1321
1322 static int stale_bundle(struct dst_entry *dst);
1323
1324 /* Main function: finds/creates a bundle for given flow.
1325  *
1326  * At the moment we eat a raw IP route. Mostly to speed up lookups
1327  * on interfaces with disabled IPsec.
1328  */
1329 int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
1330                 struct sock *sk, int flags)
1331 {
1332         struct xfrm_policy *policy;
1333         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1334         int npols;
1335         int pol_dead;
1336         int xfrm_nr;
1337         int pi;
1338         struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
1339         struct dst_entry *dst, *dst_orig = *dst_p;
1340         int nx = 0;
1341         int err;
1342         u32 genid;
1343         u16 family;
1344         u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
1345
1346 restart:
1347         genid = atomic_read(&flow_cache_genid);
1348         policy = NULL;
1349         for (pi = 0; pi < ARRAY_SIZE(pols); pi++)
1350                 pols[pi] = NULL;
1351         npols = 0;
1352         pol_dead = 0;
1353         xfrm_nr = 0;
1354
1355         if (sk && sk->sk_policy[1]) {
1356                 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
1357                 if (IS_ERR(policy))
1358                         return PTR_ERR(policy);
1359         }
1360
1361         if (!policy) {
1362                 /* To accelerate a bit...  */
1363                 if ((dst_orig->flags & DST_NOXFRM) ||
1364                     !xfrm_policy_count[XFRM_POLICY_OUT])
1365                         return 0;
1366
1367                 policy = flow_cache_lookup(fl, dst_orig->ops->family,
1368                                            dir, xfrm_policy_lookup);
1369                 if (IS_ERR(policy))
1370                         return PTR_ERR(policy);
1371         }
1372
1373         if (!policy)
1374                 return 0;
1375
1376         family = dst_orig->ops->family;
1377         policy->curlft.use_time = (unsigned long)xtime.tv_sec;
1378         pols[0] = policy;
1379         npols ++;
1380         xfrm_nr += pols[0]->xfrm_nr;
1381
1382         switch (policy->action) {
1383         case XFRM_POLICY_BLOCK:
1384                 /* Prohibit the flow */
1385                 err = -EPERM;
1386                 goto error;
1387
1388         case XFRM_POLICY_ALLOW:
1389 #ifndef CONFIG_XFRM_SUB_POLICY
1390                 if (policy->xfrm_nr == 0) {
1391                         /* Flow passes not transformed. */
1392                         xfrm_pol_put(policy);
1393                         return 0;
1394                 }
1395 #endif
1396
1397                 /* Try to find matching bundle.
1398                  *
1399                  * LATER: help from flow cache. It is optional, this
1400                  * is required only for output policy.
1401                  */
1402                 dst = xfrm_find_bundle(fl, policy, family);
1403                 if (IS_ERR(dst)) {
1404                         err = PTR_ERR(dst);
1405                         goto error;
1406                 }
1407
1408                 if (dst)
1409                         break;
1410
1411 #ifdef CONFIG_XFRM_SUB_POLICY
1412                 if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1413                         pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1414                                                             fl, family,
1415                                                             XFRM_POLICY_OUT);
1416                         if (pols[1]) {
1417                                 if (IS_ERR(pols[1])) {
1418                                         err = PTR_ERR(pols[1]);
1419                                         goto error;
1420                                 }
1421                                 if (pols[1]->action == XFRM_POLICY_BLOCK) {
1422                                         err = -EPERM;
1423                                         goto error;
1424                                 }
1425                                 npols ++;
1426                                 xfrm_nr += pols[1]->xfrm_nr;
1427                         }
1428                 }
1429
1430                 /*
1431                  * Because neither flowi nor bundle information knows about
1432                  * transformation template size. On more than one policy usage
1433                  * we can realize whether all of them is bypass or not after
1434                  * they are searched. See above not-transformed bypass
1435                  * is surrounded by non-sub policy configuration, too.
1436                  */
1437                 if (xfrm_nr == 0) {
1438                         /* Flow passes not transformed. */
1439                         xfrm_pols_put(pols, npols);
1440                         return 0;
1441                 }
1442
1443 #endif
1444                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1445
1446                 if (unlikely(nx<0)) {
1447                         err = nx;
1448                         if (err == -EAGAIN && flags) {
1449                                 DECLARE_WAITQUEUE(wait, current);
1450
1451                                 add_wait_queue(&km_waitq, &wait);
1452                                 set_current_state(TASK_INTERRUPTIBLE);
1453                                 schedule();
1454                                 set_current_state(TASK_RUNNING);
1455                                 remove_wait_queue(&km_waitq, &wait);
1456
1457                                 nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
1458
1459                                 if (nx == -EAGAIN && signal_pending(current)) {
1460                                         err = -ERESTART;
1461                                         goto error;
1462                                 }
1463                                 if (nx == -EAGAIN ||
1464                                     genid != atomic_read(&flow_cache_genid)) {
1465                                         xfrm_pols_put(pols, npols);
1466                                         goto restart;
1467                                 }
1468                                 err = nx;
1469                         }
1470                         if (err < 0)
1471                                 goto error;
1472                 }
1473                 if (nx == 0) {
1474                         /* Flow passes not transformed. */
1475                         xfrm_pols_put(pols, npols);
1476                         return 0;
1477                 }
1478
1479                 dst = dst_orig;
1480                 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
1481
1482                 if (unlikely(err)) {
1483                         int i;
1484                         for (i=0; i<nx; i++)
1485                                 xfrm_state_put(xfrm[i]);
1486                         goto error;
1487                 }
1488
1489                 for (pi = 0; pi < npols; pi++) {
1490                         read_lock_bh(&pols[pi]->lock);
1491                         pol_dead |= pols[pi]->dead;
1492                         read_unlock_bh(&pols[pi]->lock);
1493                 }
1494
1495                 write_lock_bh(&policy->lock);
1496                 if (unlikely(pol_dead || stale_bundle(dst))) {
1497                         /* Wow! While we worked on resolving, this
1498                          * policy has gone. Retry. It is not paranoia,
1499                          * we just cannot enlist new bundle to dead object.
1500                          * We can't enlist stable bundles either.
1501                          */
1502                         write_unlock_bh(&policy->lock);
1503                         if (dst)
1504                                 dst_free(dst);
1505
1506                         err = -EHOSTUNREACH;
1507                         goto error;
1508                 }
1509                 dst->next = policy->bundles;
1510                 policy->bundles = dst;
1511                 dst_hold(dst);
1512                 write_unlock_bh(&policy->lock);
1513         }
1514         *dst_p = dst;
1515         dst_release(dst_orig);
1516         xfrm_pols_put(pols, npols);
1517         return 0;
1518
1519 error:
1520         dst_release(dst_orig);
1521         xfrm_pols_put(pols, npols);
1522         *dst_p = NULL;
1523         return err;
1524 }
1525 EXPORT_SYMBOL(xfrm_lookup);
1526
1527 static inline int
1528 xfrm_secpath_reject(int idx, struct sk_buff *skb, struct flowi *fl)
1529 {
1530         struct xfrm_state *x;
1531         int err;
1532
1533         if (!skb->sp || idx < 0 || idx >= skb->sp->len)
1534                 return 0;
1535         x = skb->sp->xvec[idx];
1536         if (!x->type->reject)
1537                 return 0;
1538         xfrm_state_hold(x);
1539         err = x->type->reject(x, skb, fl);
1540         xfrm_state_put(x);
1541         return err;
1542 }
1543
1544 /* When skb is transformed back to its "native" form, we have to
1545  * check policy restrictions. At the moment we make this in maximally
1546  * stupid way. Shame on me. :-) Of course, connected sockets must
1547  * have policy cached at them.
1548  */
1549
1550 static inline int
1551 xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x, 
1552               unsigned short family)
1553 {
1554         if (xfrm_state_kern(x))
1555                 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
1556         return  x->id.proto == tmpl->id.proto &&
1557                 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
1558                 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
1559                 x->props.mode == tmpl->mode &&
1560                 ((tmpl->aalgos & (1<<x->props.aalgo)) ||
1561                  !(xfrm_id_proto_match(tmpl->id.proto, IPSEC_PROTO_ANY))) &&
1562                 !(x->props.mode != XFRM_MODE_TRANSPORT &&
1563                   xfrm_state_addr_cmp(tmpl, x, family));
1564 }
1565
1566 /*
1567  * 0 or more than 0 is returned when validation is succeeded (either bypass
1568  * because of optional transport mode, or next index of the mathced secpath
1569  * state with the template.
1570  * -1 is returned when no matching template is found.
1571  * Otherwise "-2 - errored_index" is returned.
1572  */
1573 static inline int
1574 xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
1575                unsigned short family)
1576 {
1577         int idx = start;
1578
1579         if (tmpl->optional) {
1580                 if (tmpl->mode == XFRM_MODE_TRANSPORT)
1581                         return start;
1582         } else
1583                 start = -1;
1584         for (; idx < sp->len; idx++) {
1585                 if (xfrm_state_ok(tmpl, sp->xvec[idx], family))
1586                         return ++idx;
1587                 if (sp->xvec[idx]->props.mode != XFRM_MODE_TRANSPORT) {
1588                         if (start == -1)
1589                                 start = -2-idx;
1590                         break;
1591                 }
1592         }
1593         return start;
1594 }
1595
1596 int
1597 xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1598 {
1599         struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
1600         int err;
1601
1602         if (unlikely(afinfo == NULL))
1603                 return -EAFNOSUPPORT;
1604
1605         afinfo->decode_session(skb, fl);
1606         err = security_xfrm_decode_session(skb, &fl->secid);
1607         xfrm_policy_put_afinfo(afinfo);
1608         return err;
1609 }
1610 EXPORT_SYMBOL(xfrm_decode_session);
1611
1612 static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp)
1613 {
1614         for (; k < sp->len; k++) {
1615                 if (sp->xvec[k]->props.mode != XFRM_MODE_TRANSPORT) {
1616                         *idxp = k;
1617                         return 1;
1618                 }
1619         }
1620
1621         return 0;
1622 }
1623
1624 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb, 
1625                         unsigned short family)
1626 {
1627         struct xfrm_policy *pol;
1628         struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
1629         int npols = 0;
1630         int xfrm_nr;
1631         int pi;
1632         struct flowi fl;
1633         u8 fl_dir = policy_to_flow_dir(dir);
1634         int xerr_idx = -1;
1635
1636         if (xfrm_decode_session(skb, &fl, family) < 0)
1637                 return 0;
1638         nf_nat_decode_session(skb, &fl, family);
1639
1640         /* First, check used SA against their selectors. */
1641         if (skb->sp) {
1642                 int i;
1643
1644                 for (i=skb->sp->len-1; i>=0; i--) {
1645                         struct xfrm_state *x = skb->sp->xvec[i];
1646                         if (!xfrm_selector_match(&x->sel, &fl, family))
1647                                 return 0;
1648                 }
1649         }
1650
1651         pol = NULL;
1652         if (sk && sk->sk_policy[dir]) {
1653                 pol = xfrm_sk_policy_lookup(sk, dir, &fl);
1654                 if (IS_ERR(pol))
1655                         return 0;
1656         }
1657
1658         if (!pol)
1659                 pol = flow_cache_lookup(&fl, family, fl_dir,
1660                                         xfrm_policy_lookup);
1661
1662         if (IS_ERR(pol))
1663                 return 0;
1664
1665         if (!pol) {
1666                 if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
1667                         xfrm_secpath_reject(xerr_idx, skb, &fl);
1668                         return 0;
1669                 }
1670                 return 1;
1671         }
1672
1673         pol->curlft.use_time = (unsigned long)xtime.tv_sec;
1674
1675         pols[0] = pol;
1676         npols ++;
1677 #ifdef CONFIG_XFRM_SUB_POLICY
1678         if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
1679                 pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
1680                                                     &fl, family,
1681                                                     XFRM_POLICY_IN);
1682                 if (pols[1]) {
1683                         if (IS_ERR(pols[1]))
1684                                 return 0;
1685                         pols[1]->curlft.use_time = (unsigned long)xtime.tv_sec;
1686                         npols ++;
1687                 }
1688         }
1689 #endif
1690
1691         if (pol->action == XFRM_POLICY_ALLOW) {
1692                 struct sec_path *sp;
1693                 static struct sec_path dummy;
1694                 struct xfrm_tmpl *tp[XFRM_MAX_DEPTH];
1695                 struct xfrm_tmpl *stp[XFRM_MAX_DEPTH];
1696                 struct xfrm_tmpl **tpp = tp;
1697                 int ti = 0;
1698                 int i, k;
1699
1700                 if ((sp = skb->sp) == NULL)
1701                         sp = &dummy;
1702
1703                 for (pi = 0; pi < npols; pi++) {
1704                         if (pols[pi] != pol &&
1705                             pols[pi]->action != XFRM_POLICY_ALLOW)
1706                                 goto reject;
1707                         if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH)
1708                                 goto reject_error;
1709                         for (i = 0; i < pols[pi]->xfrm_nr; i++)
1710                                 tpp[ti++] = &pols[pi]->xfrm_vec[i];
1711                 }
1712                 xfrm_nr = ti;
1713                 if (npols > 1) {
1714                         xfrm_tmpl_sort(stp, tpp, xfrm_nr, family);
1715                         tpp = stp;
1716                 }
1717
1718                 /* For each tunnel xfrm, find the first matching tmpl.
1719                  * For each tmpl before that, find corresponding xfrm.
1720                  * Order is _important_. Later we will implement
1721                  * some barriers, but at the moment barriers
1722                  * are implied between each two transformations.
1723                  */
1724                 for (i = xfrm_nr-1, k = 0; i >= 0; i--) {
1725                         k = xfrm_policy_ok(tpp[i], sp, k, family);
1726                         if (k < 0) {
1727                                 if (k < -1)
1728                                         /* "-2 - errored_index" returned */
1729                                         xerr_idx = -(2+k);
1730                                 goto reject;
1731                         }
1732                 }
1733
1734                 if (secpath_has_nontransport(sp, k, &xerr_idx))
1735                         goto reject;
1736
1737                 xfrm_pols_put(pols, npols);
1738                 return 1;
1739         }
1740
1741 reject:
1742         xfrm_secpath_reject(xerr_idx, skb, &fl);
1743 reject_error:
1744         xfrm_pols_put(pols, npols);
1745         return 0;
1746 }
1747 EXPORT_SYMBOL(__xfrm_policy_check);
1748
1749 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1750 {
1751         struct flowi fl;
1752
1753         if (xfrm_decode_session(skb, &fl, family) < 0)
1754                 return 0;
1755
1756         return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1757 }
1758 EXPORT_SYMBOL(__xfrm_route_forward);
1759
1760 /* Optimize later using cookies and generation ids. */
1761
1762 static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1763 {
1764         /* Code (such as __xfrm4_bundle_create()) sets dst->obsolete
1765          * to "-1" to force all XFRM destinations to get validated by
1766          * dst_ops->check on every use.  We do this because when a
1767          * normal route referenced by an XFRM dst is obsoleted we do
1768          * not go looking around for all parent referencing XFRM dsts
1769          * so that we can invalidate them.  It is just too much work.
1770          * Instead we make the checks here on every use.  For example:
1771          *
1772          *      XFRM dst A --> IPv4 dst X
1773          *
1774          * X is the "xdst->route" of A (X is also the "dst->path" of A
1775          * in this example).  If X is marked obsolete, "A" will not
1776          * notice.  That's what we are validating here via the
1777          * stale_bundle() check.
1778          *
1779          * When a policy's bundle is pruned, we dst_free() the XFRM
1780          * dst which causes it's ->obsolete field to be set to a
1781          * positive non-zero integer.  If an XFRM dst has been pruned
1782          * like this, we want to force a new route lookup.
1783          */
1784         if (dst->obsolete < 0 && !stale_bundle(dst))
1785                 return dst;
1786
1787         return NULL;
1788 }
1789
1790 static int stale_bundle(struct dst_entry *dst)
1791 {
1792         return !xfrm_bundle_ok(NULL, (struct xfrm_dst *)dst, NULL, AF_UNSPEC, 0);
1793 }
1794
1795 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1796 {
1797         while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1798                 dst->dev = &loopback_dev;
1799                 dev_hold(&loopback_dev);
1800                 dev_put(dev);
1801         }
1802 }
1803 EXPORT_SYMBOL(xfrm_dst_ifdown);
1804
1805 static void xfrm_link_failure(struct sk_buff *skb)
1806 {
1807         /* Impossible. Such dst must be popped before reaches point of failure. */
1808         return;
1809 }
1810
1811 static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1812 {
1813         if (dst) {
1814                 if (dst->obsolete) {
1815                         dst_release(dst);
1816                         dst = NULL;
1817                 }
1818         }
1819         return dst;
1820 }
1821
1822 static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_entry *), struct dst_entry **gc_list_p)
1823 {
1824         struct dst_entry *dst, **dstp;
1825
1826         write_lock(&pol->lock);
1827         dstp = &pol->bundles;
1828         while ((dst=*dstp) != NULL) {
1829                 if (func(dst)) {
1830                         *dstp = dst->next;
1831                         dst->next = *gc_list_p;
1832                         *gc_list_p = dst;
1833                 } else {
1834                         dstp = &dst->next;
1835                 }
1836         }
1837         write_unlock(&pol->lock);
1838 }
1839
1840 static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1841 {
1842         struct dst_entry *gc_list = NULL;
1843         int dir;
1844
1845         read_lock_bh(&xfrm_policy_lock);
1846         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
1847                 struct xfrm_policy *pol;
1848                 struct hlist_node *entry;
1849                 struct hlist_head *table;
1850                 int i;
1851
1852                 hlist_for_each_entry(pol, entry,
1853                                      &xfrm_policy_inexact[dir], bydst)
1854                         prune_one_bundle(pol, func, &gc_list);
1855
1856                 table = xfrm_policy_bydst[dir].table;
1857                 for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
1858                         hlist_for_each_entry(pol, entry, table + i, bydst)
1859                                 prune_one_bundle(pol, func, &gc_list);
1860                 }
1861         }
1862         read_unlock_bh(&xfrm_policy_lock);
1863
1864         while (gc_list) {
1865                 struct dst_entry *dst = gc_list;
1866                 gc_list = dst->next;
1867                 dst_free(dst);
1868         }
1869 }
1870
1871 static int unused_bundle(struct dst_entry *dst)
1872 {
1873         return !atomic_read(&dst->__refcnt);
1874 }
1875
1876 static void __xfrm_garbage_collect(void)
1877 {
1878         xfrm_prune_bundles(unused_bundle);
1879 }
1880
1881 static int xfrm_flush_bundles(void)
1882 {
1883         xfrm_prune_bundles(stale_bundle);
1884         return 0;
1885 }
1886
1887 void xfrm_init_pmtu(struct dst_entry *dst)
1888 {
1889         do {
1890                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1891                 u32 pmtu, route_mtu_cached;
1892
1893                 pmtu = dst_mtu(dst->child);
1894                 xdst->child_mtu_cached = pmtu;
1895
1896                 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1897
1898                 route_mtu_cached = dst_mtu(xdst->route);
1899                 xdst->route_mtu_cached = route_mtu_cached;
1900
1901                 if (pmtu > route_mtu_cached)
1902                         pmtu = route_mtu_cached;
1903
1904                 dst->metrics[RTAX_MTU-1] = pmtu;
1905         } while ((dst = dst->next));
1906 }
1907
1908 EXPORT_SYMBOL(xfrm_init_pmtu);
1909
1910 /* Check that the bundle accepts the flow and its components are
1911  * still valid.
1912  */
1913
1914 int xfrm_bundle_ok(struct xfrm_policy *pol, struct xfrm_dst *first,
1915                 struct flowi *fl, int family, int strict)
1916 {
1917         struct dst_entry *dst = &first->u.dst;
1918         struct xfrm_dst *last;
1919         u32 mtu;
1920
1921         if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1922             (dst->dev && !netif_running(dst->dev)))
1923                 return 0;
1924
1925         last = NULL;
1926
1927         do {
1928                 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1929
1930                 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1931                         return 0;
1932                 if (fl && pol &&
1933                     !security_xfrm_state_pol_flow_match(dst->xfrm, pol, fl))
1934                         return 0;
1935                 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1936                         return 0;
1937                 if (xdst->genid != dst->xfrm->genid)
1938                         return 0;
1939
1940                 if (strict && fl && dst->xfrm->props.mode != XFRM_MODE_TUNNEL &&
1941                     !xfrm_state_addr_flow_check(dst->xfrm, fl, family))
1942                         return 0;
1943
1944                 mtu = dst_mtu(dst->child);
1945                 if (xdst->child_mtu_cached != mtu) {
1946                         last = xdst;
1947                         xdst->child_mtu_cached = mtu;
1948                 }
1949
1950                 if (!dst_check(xdst->route, xdst->route_cookie))
1951                         return 0;
1952                 mtu = dst_mtu(xdst->route);
1953                 if (xdst->route_mtu_cached != mtu) {
1954                         last = xdst;
1955                         xdst->route_mtu_cached = mtu;
1956                 }
1957
1958                 dst = dst->child;
1959         } while (dst->xfrm);
1960
1961         if (likely(!last))
1962                 return 1;
1963
1964         mtu = last->child_mtu_cached;
1965         for (;;) {
1966                 dst = &last->u.dst;
1967
1968                 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1969                 if (mtu > last->route_mtu_cached)
1970                         mtu = last->route_mtu_cached;
1971                 dst->metrics[RTAX_MTU-1] = mtu;
1972
1973                 if (last == first)
1974                         break;
1975
1976                 last = last->u.next;
1977                 last->child_mtu_cached = mtu;
1978         }
1979
1980         return 1;
1981 }
1982
1983 EXPORT_SYMBOL(xfrm_bundle_ok);
1984
1985 int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1986 {
1987         int err = 0;
1988         if (unlikely(afinfo == NULL))
1989                 return -EINVAL;
1990         if (unlikely(afinfo->family >= NPROTO))
1991                 return -EAFNOSUPPORT;
1992         write_lock_bh(&xfrm_policy_afinfo_lock);
1993         if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1994                 err = -ENOBUFS;
1995         else {
1996                 struct dst_ops *dst_ops = afinfo->dst_ops;
1997                 if (likely(dst_ops->kmem_cachep == NULL))
1998                         dst_ops->kmem_cachep = xfrm_dst_cache;
1999                 if (likely(dst_ops->check == NULL))
2000                         dst_ops->check = xfrm_dst_check;
2001                 if (likely(dst_ops->negative_advice == NULL))
2002                         dst_ops->negative_advice = xfrm_negative_advice;
2003                 if (likely(dst_ops->link_failure == NULL))
2004                         dst_ops->link_failure = xfrm_link_failure;
2005                 if (likely(afinfo->garbage_collect == NULL))
2006                         afinfo->garbage_collect = __xfrm_garbage_collect;
2007                 xfrm_policy_afinfo[afinfo->family] = afinfo;
2008         }
2009         write_unlock_bh(&xfrm_policy_afinfo_lock);
2010         return err;
2011 }
2012 EXPORT_SYMBOL(xfrm_policy_register_afinfo);
2013
2014 int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
2015 {
2016         int err = 0;
2017         if (unlikely(afinfo == NULL))
2018                 return -EINVAL;
2019         if (unlikely(afinfo->family >= NPROTO))
2020                 return -EAFNOSUPPORT;
2021         write_lock_bh(&xfrm_policy_afinfo_lock);
2022         if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
2023                 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
2024                         err = -EINVAL;
2025                 else {
2026                         struct dst_ops *dst_ops = afinfo->dst_ops;
2027                         xfrm_policy_afinfo[afinfo->family] = NULL;
2028                         dst_ops->kmem_cachep = NULL;
2029                         dst_ops->check = NULL;
2030                         dst_ops->negative_advice = NULL;
2031                         dst_ops->link_failure = NULL;
2032                         afinfo->garbage_collect = NULL;
2033                 }
2034         }
2035         write_unlock_bh(&xfrm_policy_afinfo_lock);
2036         return err;
2037 }
2038 EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
2039
2040 static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
2041 {
2042         struct xfrm_policy_afinfo *afinfo;
2043         if (unlikely(family >= NPROTO))
2044                 return NULL;
2045         read_lock(&xfrm_policy_afinfo_lock);
2046         afinfo = xfrm_policy_afinfo[family];
2047         if (unlikely(!afinfo))
2048                 read_unlock(&xfrm_policy_afinfo_lock);
2049         return afinfo;
2050 }
2051
2052 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
2053 {
2054         read_unlock(&xfrm_policy_afinfo_lock);
2055 }
2056
2057 static struct xfrm_policy_afinfo *xfrm_policy_lock_afinfo(unsigned int family)
2058 {
2059         struct xfrm_policy_afinfo *afinfo;
2060         if (unlikely(family >= NPROTO))
2061                 return NULL;
2062         write_lock_bh(&xfrm_policy_afinfo_lock);
2063         afinfo = xfrm_policy_afinfo[family];
2064         if (unlikely(!afinfo))
2065                 write_unlock_bh(&xfrm_policy_afinfo_lock);
2066         return afinfo;
2067 }
2068
2069 static void xfrm_policy_unlock_afinfo(struct xfrm_policy_afinfo *afinfo)
2070 {
2071         write_unlock_bh(&xfrm_policy_afinfo_lock);
2072 }
2073
2074 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
2075 {
2076         switch (event) {
2077         case NETDEV_DOWN:
2078                 xfrm_flush_bundles();
2079         }
2080         return NOTIFY_DONE;
2081 }
2082
2083 static struct notifier_block xfrm_dev_notifier = {
2084         xfrm_dev_event,
2085         NULL,
2086         0
2087 };
2088
2089 static void __init xfrm_policy_init(void)
2090 {
2091         unsigned int hmask, sz;
2092         int dir;
2093
2094         xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
2095                                            sizeof(struct xfrm_dst),
2096                                            0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2097                                            NULL, NULL);
2098
2099         hmask = 8 - 1;
2100         sz = (hmask+1) * sizeof(struct hlist_head);
2101
2102         xfrm_policy_byidx = xfrm_hash_alloc(sz);
2103         xfrm_idx_hmask = hmask;
2104         if (!xfrm_policy_byidx)
2105                 panic("XFRM: failed to allocate byidx hash\n");
2106
2107         for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
2108                 struct xfrm_policy_hash *htab;
2109
2110                 INIT_HLIST_HEAD(&xfrm_policy_inexact[dir]);
2111
2112                 htab = &xfrm_policy_bydst[dir];
2113                 htab->table = xfrm_hash_alloc(sz);
2114                 htab->hmask = hmask;
2115                 if (!htab->table)
2116                         panic("XFRM: failed to allocate bydst hash\n");
2117         }
2118
2119         INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
2120         register_netdevice_notifier(&xfrm_dev_notifier);
2121 }
2122
2123 void __init xfrm_init(void)
2124 {
2125         xfrm_state_init();
2126         xfrm_policy_init();
2127         xfrm_input_init();
2128 }
2129